Agricultural

Economics
and Policy
World Scientific Lecture Notes in Economics and Policy

ISSN: 2630-4872

Series Editors: Felix Munoz-Garcia (Washington State University, USA)

Ariel Dinar (University of California, Riverside, USA)
Dirk Bergemann (Yale University, USA)
George Mailath (University of Pennsylvania, USA)
Devashish Mitra (Syracuse University, USA)
Kar-yiu Wong (University of Washington, USA)
Richard Carpiano (University of California, Riverside, USA)
Chetan Dave (University of Alberta, Canada)
Malik Shukayev (University of Alberta, Canada)
George C Davis (Virginia Tech University, USA)
Marco M Sorge (University of Salerno, Italy)
Alessia Paccagnini (University College Dublin, Ireland)
Luca Lambertini (Bologna University, Italy)
Konstantinos Georgalos (Lancaster University, UK)
Aart de Zeeuw (Tilburg University, The Netherlands)

The World Scientific Lecture Notes in Economics and Policy series is aimed to
produce lecture note texts for a wide range of economics disciplines, both theoretical
and applied at the undergraduate and graduate levels. Contributors to the series are
highly ranked and experienced professors of economics who see in publication of
their lectures a mission to disseminate the teaching of economics in an affordable
manner to students and other readers interested in enriching their knowledge of
economic topics. The series was formerly titled World Scientific Lecture Notes in
Economics.

Published:

Vol. 26: Agricultural Economics and Policy

by David Zilberman, Ruiqing Miao and Jian Rong

Vol. 25: Handbook of Economy-wide Microeconomics:

Theory, Applications, Tests and Extensions
(In 4 Volumes)
Volume 1: Existence and Uniqueness of Equilibrium
Volume 2: Stability and Optimality of Equilibrium
Volume 3: Comparative Statics and Applications
Volume 4: Tests and Extensions
by W. D. A. Bryant

For the complete list of volumes in this series, please visit

www.worldscientific.com/series/wslnep
 World Scientific Lecture Notes in Economics and Policy – Vol. 26

Agricultural
Economics
and Policy

David Zilberman
University of California, Berkeley, USA

Ruiqing Miao
Auburn University, USA

Jian Rong
Auburn University, USA

World Scientific
NEW JERSEY • LONDON • SINGAPORE • BEIJING • SHANGHAI • TAIPEI • CHENNAI
Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

Library of Congress Cataloging-in-Publication Data

Names: Zilberman, David, 1947– author. | Miao, Ruiqing, author. |
Rong, Jian (Economist), author.
Title: Agricultural economics and policy / David Zilberman, University of California, Berkeley, USA,
Ruiqing Miao, Auburn University, USA, Jian Rong, Auburn University, USA.
Description: New Jersey : World Scientific, [2025] | Series: World Scientific lecture notes in
economics and policy, 2630-4872 ; vol. 26 | Includes bibliographical references and index.
Identifiers: LCCN 2024046677 | ISBN 9789811299162 (hardcover) |
ISBN 9789819801121 (paperback) | ISBN 9789811299179 (ebook) |
ISBN 9789811299186 (ebook other)
Subjects: LCSH: Agriculture--Economic aspects--United States.
Classification: LCC HD1761 .Z55 2025 | DDC 338.10973--dc23/eng/20241115
LC record available at https://lccn.loc.gov/2024046677

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

Copyright © 2025 by World Scientific Publishing Co. Pte. Ltd.

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance
Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy
is not required from the publisher.

For any available supplementary material, please visit

https://www.worldscientific.com/worldscibooks/10.1142/14016#t=suppl

Desk Editors: Nambirajan Karuppiah/Geysilla Jean/Analyn Alcala

Typeset by Stallion Press

Email: [email protected]

Printed in Singapore
I dedicate this book to my teachers, and especially my students.
I learned more from them than they learned from me.
—David Zilberman
We dedicate this book to our parents.
—Ruiqing Miao and Jian Rong
This page intentionally left blank
 About the Authors

David Zilberman holds the Robinson Chair in the Agricultural

and Resource Economics Department at the University of California
at Berkeley. He is the recipient of the 2019 Wolf Prize in Agriculture
and was elected a member of the U.S. National Academy of Science
in 2019. David served as the 2018–2019 President of the Agricultural
& Applied Economics Association (AAEA). He’s a fellow of multiple
professional associations and has published in both professional and
popular outlets. He has over 400 refereed articles in journals ranging
from Science to ARE Update and has edited 25 books. In addition,
he has served as a consultant to the U.S. Environmental Protection
Agency, the World Bank, and the FAO. David’s BA is from Tel Aviv
University and his Ph.D. is from Berkeley. David is the co-founder
of the Beahrs Environmental Leadership Program and the academic
director of the Berkeley Master of Development Practice program.
David’s research analyzes water, innovation, supply chains, and the
interactions between agriculture, energy, and the environment. He
has researched the economics and political economy of agricultural
biotechnology and the potential of the bioeconomy. In addition, he
has been working on water policy programs and the economic impacts
of the COVID-19 pandemic.

vii
viii Agricultural Economics and Policy

Ruiqing Miao is an associate professor in agricultural economics

in the Department of Agricultural Economics and Rural Sociology
at Auburn University. His research focuses on agricultural sustain-
ability, innovation, and decision-making under risk and uncertainty.
He studies the interaction between agricultural production and its
environment, aiming to understand and quantify agriculture’s impact
on land use, water use, water quality, and biodiversity, as well as
how agricultural production is affected by farmers’ behaviors, public
policies, agricultural innovation, technology adoption, and climate
change. Ruiqing has served as a co-editor for Choices Magazine,
and as an associate editor for American Journal of Agricultural
Economics, Agricultural Economics, and Natural Resource Modeling.
Ruiqing received his PhD in economics from Iowa State University
in 2012.

Jian Rong is a senior lecturer in the Department of Agricultural

Economics and Rural Sociology at Auburn University. In 2022, Dr.
Rong was a visiting scholar in the Department of Agricultural and
Resource Economics at the University of California, Berkeley, and
a visiting lecturer in the Department of Agricultural and Resource
Economics at the University of California, Davis. Dr. Rong has rich
experience in teaching agricultural policy, agribusiness management
and agribusiness marketing. She holds a PhD in Economics from
Fudan University and an MBA from Iowa State University.
 Contents

About the Authors vii

Prologue: The Uniqueness of Agricultural Economics xi

Chapter 1 History of U.S. Agriculture: A Lesson in

Development 1

Chapter 2 Review of Production Economics 17

Chapter 3 Consumer Demand 55

Chapter 4 Economic Analysis of Behavior under Risk

and Uncertainty 81

Chapter 5 Welfare Analysis of Agricultural Policies 149

Chapter 6 The Political Economy of Agricultural Policies 171

Chapter 7 Agricultural Innovation 197

Chapter 8 Technology Adoption 227

Chapter 9 Agricultural Supply Chain 259

Chapter 10 Health Risks, Food Security, and Food Safety 307

ix
x Agricultural Economics and Policy

Chapter 11 Economics of Pesticides 329

Chapter 12 Economics of Space: Interaction of Urban

and Rural Sectors 347

Chapter 13 Climate Change and Agriculture 363

Epilogue: Sustainable Development 383

Appendix A. Elements of Optimization Theory 399

Appendix B. Problem Sets 421

Appendix C. Solutions to Selected Problem Sets 433

Index 455
 Prologue: The Uniqueness of
Agricultural Economics

Agricultural economics has evolved as a distinguished discipline of

economics. Economics, like any other discipline, has a core and
many subdisciplines that tend to emphasize specific issues and
develop their own agenda and methods to address specific challenges
of the subdiscipline. Agricultural economics is among the largest
economic subdisciplines. Agricultural economics has evolved over
time as analytical tools and economic and political realities change.
It applies, as well as develops, economic methodologies, and it
incorporates notions and theories from other disciplines (e.g., life
science, biophysical science, and psychology) to address the problems
of agriculture and natural resources (Zilberman, 2019).
This book provides a review and some original results on various
issues of agricultural economics and policy. The book contains lecture
notes taught by David Zilberman in his PhD-level agricultural
economics and policy course sequence at the University of Cali-
fornia, Berkeley, augmented by lectures taught by Ruiqing Miao
and Jian Rong at Auburn University. The book provides students
and researchers in agricultural economics and related fields with a
U.S. agriculture background, economic theories, applied models, and
analysis of major agricultural policies in the United States and some
other countries. The content of this book is broad, covering topics of
agricultural history, production, demand, risk, technical innovation,

xi
xii Agricultural Economics and Policy

adoption, welfare analysis, payment for ecosystem service, agri-

cultural supply chain, the political economy of agricultural poli-
cies, and sustainable development in agriculture. Before we dive
into these contents, let us first appreciate the uniqueness of agricul-
tural economics. In this prologue, we will address the link between
agricultural economics and economics, introduce some of the major
fields of agricultural economics, commemorate the range of scholars
who laid the foundation for advanced agricultural economics, and
provide a perspective on its present and future.

The Diverse Field of Agricultural Economics

Let us start with the beginning of

the economic profession. Economics
was spun off philosophy, and Adam
Smith1 was a professor of moral philos-
ophy. Mathematics was introduced to
explain the economy’s behavior since
“the laws of Nature are written in
the language of mathematics” (Smith,
1776). Adam Smith introduced the
notion of the invisible hand, show-
ing that under perfect competition,
markets will result in socially desir-
able outcomes. Later, economists used
Adam Smith
mathematical tools to refine the con-
ditions under which Smith’s results worked, and to understand the
policy implications. The mathematical sophistication of economics
was improved when von Neumann2 and Morgenstern3 introduced
game theory, and Kuznets4 and Frisch5 introduced econometrics.

1
https://www.econlib.org/library/Enc/bios/Smith.html.
2
https://mathshistory.st-andrews.ac.uk/Biographies/Von Neumann/.
3
https://www.informs.org/Explore/History-of-O.R.-Excellence/Biographical-
Profiles/Morgenstern-Oskar.
4
https://www.nobelprize.org/prizes/economic-sciences/1971/kuznets/facts/.
5
https://www.nobelprize.org/prizes/economic-sciences/1969/frisch/facts/.
 Prologue: The Uniqueness of Agricultural Economics xiii

John Maynard Keynes6

was an economist and a
trader. He developed an
applied theory based on
his experience in the real
world. Based on observa-
tions, he developed new
economic concepts, includ-
ing risk premiums, non-
equilibrating markets, and
John Maynard Keynes
the role of government pol-
icy in stabilizing the econ-
omy. Keynes’s macroeconomics became a major part of economic
thought and policy analysis.

Agricultural economics was estab-

lished in the early 20th century and
was the result of a merger of the
disciplines of farm economics and
farm management. Farm manage-
ment was more practical, empha-
sizing understanding farm behavior
and helping farmers solve practical
problems. Farm economics aimed
to understand agricultural markets,
prices, policy, and the “macro” of
agriculture. The Keynesian think-
ing that contributed to govern-
John Kenneth Galbraith ment intervention during the Great
Depression also led to agricultural
policies introduced and shaped by farm economists. The new field
of agricultural economics has combined theory and empirical work,
and evolved with events and technology. One of the early leading
agricultural economists was John Kenneth Galbraith.7 He obtained

6
https://www.econlib.org/library/Enc/bios/Keynes.html.
7
https://www.econlib.org/library/Enc/bios/Galbraith.html.
xiv Agricultural Economics and Policy

his agricultural economics degree from the University of California,

Berkeley, worked at the University of California, Davis on agri-
cultural research issues, then moved to Harvard, played a major
role in the government during the Second World War, and made
major contributions to mainstream economics. He became a public
intellectual, and his 1958 book, The Affluent Society, was influential.
His career exemplified the major edge of agricultural economics,
which is an applied discipline that works with farmers, businesses,
and researchers, learns about the major problems on the ground,
and develops ideas and techniques to modify them. In the process,
agricultural economists have made major advancements to expand
the tools and methodologies of economics.

Agricultural economics has multi-

ple fields of research, and agricul-
tural policy is a major field that
has evolved over the years. One
of the field’s giants was Willard
Cochrane,8 who argued that U.S.
agricultural policy evolved from
land policies to research policies
to income support and environ-
mental policies. He introduced the
technological treadmill that has
been used throughout economics,
which emphasizes that most of the
adopters of new technology may not
gain much because of price effects,
Bruce Gardner and most of the rent from new
technologies goes to early adopters
(Cochrane, 1979). Another leading
scholar of agricultural economics

8
https://www.aaea.org/trust/appreciation-clubs/willard-w-cochrane.
 Prologue: The Uniqueness of Agricultural Economics xv

was Frederick Waugh,9 a prolific microeconomist who developed a

methodological approach to assess the economics of variability and
unstable prices and evaluate the impact of stabilization policies.
He identified conditions wherein certain groups may benefit from
price instability (Waugh, 1944). Earl Heady10 pioneered the use of
mathematical programming for planning government policy, espe-
cially in agriculture, and was among the first to harness the power of
computers to assess how alternative policies would affect distribution
throughout the economy. Many have built on his approach. Bruce
Gardner11 analyzed the evolution of U.S. agricultural policy in
the 20th century, recognizing that during this period, because of
migration and technological change, farmers have ceased to be
relatively disadvantaged groups and the support for policies reflects
their political power. He suggested policy reform that would be as
efficient as possible given the policy constraint (Gardner, 2009).
Agricultural economists also had a major impact on the eco-
nomics of labor and community. Nobel laureate Theodore Schultz12
introduced the notions of human capital and the ability to deal with
disequilibrium, and his work revolutionized labor economics. Nobel
laureate Eleanor Ostrom13 introduced new insight to understand
the important role of institutions and political arrangements in
allocating resources, especially those shared by the community.
Modern institutional economic research follows the path blazed by
Ostrom. Agricultural economic research has addressed major labor
issues that become prominent over time, including the economics
of migration (Taylor, 1999) and the implication of technological
innovation on workers (Schmitz and Seckler, 1970).

9
https://www.aaea.org/trust/appreciation-clubs/frederick-v-waugh.
10
https://www.aaea.org/trust/appreciation-clubs/earl-o-heady.
11
https://prabook.com/web/bruce l.gardner/1393823.
12
https://www.nobelprize.org/prizes/economic-sciences/1979/schultz/facts/.
13
https://en.wikipedia.org/wiki/Elinor Ostrom.
xvi Agricultural Economics and Policy

Theodore Schultz Eleanor Ostrom

Consumer economics has always been a major area of emphasis

by agricultural economists. It evolved from a simple estimation of
food demand to analyzing the economics of nutrition and product
quality in agriculture, which led to the establishment of food stamps
and other nutritional programs, where agricultural economists such
as Sylvia Lane14 played an important role. She was a scholar who
contributed to the development and implementation of food and
nutrition policies in the U.S. Over the years, research on consumer
economics was expanded to understand the economics of local and
organic food, the economics of food labeling, and the linkages
between food nutrition and health. There is a large body of research
on the global economics of food, addressing issues of global food
security (Tweeten, 1999) and issues of food aid (Barrett, 2002).
Agricultural economists, such as World Food Prize winner Per
Pinstrup-Andersen,15 developed a systematic way to assess the global
food situation (Pinstrup-Andersen et al., 1999), others developed
programs to assess the impact of climate change on food security
(Wheeler and von Braun, 2013).

14
https://senate.universityofcalifornia.edu/ files/inmemoriam/html/SylviaLane.
html.
15
https://www.worldfoodprize.org/en/laureates/20002009 laureates/2001
pinstrupandersen/.
 Prologue: The Uniqueness of Agricultural Economics xvii

Sylvia Lane Per Pinstrup-Andersen

Much of the emphasis in agricultural

economics is on understanding agri-
cultural production, technology, and
innovation, where the data availabil-
ity and demand from clientele groups
afforded agricultural economists a
head start compared to other disci-
plines. Agricultural economists were
among the first to estimate the agri-
cultural production function and use
programming methods to improve
production practices (Heady et al.,
1960). Several scholars have studied
Zvi Grilliches the evolution of agricultural systems
from traditional to modern (Pingali,
1997; Antle, 1999; Chavas, 2008). With the advance of big data,
improved monitoring and precision, agricultural economists have
studied how information technology and this new capability will
affect agriculture and food markets (Miao and Khanna, 2020; Finger
et al., 2019; Weersink et al., 2018). Probably one of the most
important contributions in this area to economics is the study of
xviii Agricultural Economics and Policy

technology adoption. Grilliches’s16 (1957) seminal paper introduced a

quantitative approach to estimating the factors that affect adoption,
becoming a baseline for future papers, such as Feder et al. (1985).

Yujiro Hayami Vernon Ruttan

Much of the research on the economics of innovation and the

link between research, innovation, and technology has been done by
agricultural economists (Evenson et al., 1975). The notion of induced
innovation, where differences in conditions and location result in
differences in innovation, was articulated by agricultural economists
Hayami17 and Ruttan18 (1971), and agricultural economists have
developed measurements of research productivity and documented
the high return rate and underinvestment in agricultural research
(Huffman and Evenson, 2008; Alston and Pardey, 2021).
International trade is a key element of the agricultural economy,
and agricultural trade issues were behind the development of com-
parative advantage by David Ricardo19 (Ricardo, 1821). Agricultural

16
https://nap.nationalacademies.org/read/10269/chapter/6#82.
17
https://fukuoka-prize.org/en/laureates/detail/e28dfce3-1d2f-4b61-b739-
666efa631e57.
18
http://www.nasonline.org/member-directory/deceased-members/50953.html.
19
https://www.econlib.org/library/Enc/bios/Ricardo.html.
 Prologue: The Uniqueness of Agricultural Economics xix

economists documented how trade barriers and protectionism in

developed countries harmed developing countries and advocated
for freer trade that would assume affordable and healthy diets.
Agricultural economists like D. Gale Johnson20 adjusted and applied
the ideas of Adam Smith and David Ricardo to a modern world and
provided an empirical understanding of trade relationships needed for
improved policy (Johnson, 1997). Study by Anderson et al. (2013)
has found that, over time, agricultural markets have become more
competitive, and many trade-distorted policies have been reduced.
Agriculture plays a much bigger role in
developing economies. There is, there-
fore, an influential synchronization between
agricultural and development economics.
Irma Adleman21 was a pioneer in develop-
ing computable general equilibrium mod-
els that assess the impact of proposed
policies on the economy, as well as
the efficiency and distributional alloca-
tion (Adelman et al., 2014). Agricultural
economists have used various quantitative
methods to understand farmers’ behavior
Irma Adelman and address poverty and malnourishment,
enabling better decision-making for eco-
nomic agents and promoting policy change (de Janvry and Sadoulet,
2022). Research in agricultural development also provides new
insights into value chains and the mechanisms that lead to structural
changes within economies and the emergence of modern value chains
(Barrett et al., 2022).

20
https://nap.nationalacademies.org/read/11429/chapter/13.
21
https://senate.universityofcalifornia.edu/in-memoriam/files/irma-adelman.
html.
xx Agricultural Economics and Policy

Agricultural economics gave rise to

environmental and resource economics.
Agriculture must deal with challenges
of drought, dust bowls, and floods.
Agricultural economists, such as Oscar
Burt22 developed dynamic models to
address natural resource management
over time (Burt, 1964). Similarly,
there is a large body of literature
on the economics of soil manage-
ment (Stevens, 2018) and pest control
(Sexton et al., 2007). Siegfried von
Oscar R. Burt Ciriacy-Wantrup23 was a pioneer of
environmental economics and many of his concepts were modeled
and estimated by economists over the years. Tony Fisher,24 working
with Kenneth Arrow,25 developed a method to address concerns for
irreversible damage in assessing new projects (Arrow and Fisher,
1974). Agricultural economists introduced multiple methods to assess
the non-market benefits of environmental amenities and incorpo-
rate them into social decision-making (Hanemann et al., 1991;
Freeman III et al., 2014; Randall et al., 1974).
The development of agriculture depends on the availability of
finance, and many financial institutions, including futures markets,
originated in agriculture. Therefore, agricultural economists have
made a great contribution to finance. Working (1961) was a major
contributor to the economics of futures markets, and agricultural
economists significantly contributed to this field (Carter, 1999).
Agricultural economists were among the first to recognize that credit
markets tend to be imperfect, and there is a role for different types of
institutions to provide credit to agriculture and agribusiness. Barry
and Robison (2001) provided an early survey of the literature on

22
https://www.tributearchive.com/obituaries/664377/Oscar-Burt.
23
https://en.wikipedia.org/wiki/Siegfried von Ciriacy-Wantrup.
24
https://are.berkeley.edu/users/anthony-fisher.
25
https://www.nobelprize.org/prizes/economic-sciences/1972/arrow/facts/.
 Prologue: The Uniqueness of Agricultural Economics xxi

agricultural finance mechanisms, policies, and institutions and their

implications.
Much of the produc-
tion of agriculture is
done outside farms,
in processing, whole-
saling, and retailing.
The high rate of inno-
vation in agriculture
leads to new prod-
ucts and organizations
that are implemented
Ray Goldberg through modern sup-
ply chains (Zilberman
et al., 2022). Ray Goldberg26 (1968) developed agribusiness research
agenda that relies on case studies incorporating microeconomic con-
cepts with economic reality. Michael Boehlje27 (1999) documented
and provided a framework to analyze the transition from traditional
agriculture to modern agribusiness. Richard Sexton (2013) combined
modern industrial economic thinking with a keen eye for economic
reality to develop new approaches for analyses of agribusiness. Jeff
Perloff and his collaborators wrote the classic text on industrial
organization that serves the economic perspective (Carlton and
Perloff, 2005).
Quantification has been a crucial element in agricultural eco-
nomics, and agricultural economists developed and expanded new
estimation and prediction methods. For instance, they developed
advanced methods to assess the dynamics of supply and demand.
Marc Nerlove28 introduced the notion of adaptive expectation that
allows for the assessment and prediction of supply response param-
eters over time (Nerlove, 1956). Shankar Subramanian collaborated

26
https://www.hbs.edu/faculty/Pages/profile.aspx?facId=12285.
27
https://agribusiness.purdue.edu/people/michael-boehlje/.
28
https://www.aeaweb.org/about-aea/honors-awards/distinguished-fellows/marc-
nerlove.
xxii Agricultural Economics and Policy

with Sir Agnus Deaton29 to develop an

applied system of equations for demand
for food and calories (Subramanian and
Deaton, 1996). Aigner et al. (1977) pio-
neered estimating stochastic frontiers of
production functions. Mundlak (1978)
developed creative methods to analyze
pooled cross-section and time series
data. George Judge’s30 research and
writing integrated economic methods to
make them approachable and coherent
and introduced new methods (Judge,
Marc Nerlove 1982; Judge and Mittelhammer, 2011).
The emphasis on applied policymaking
in agriculture and the environment led
agricultural economists to contribute
to the development of an applied wel-
fare analysis framework (Just et al.,
2005). The importance of political con-
sideration led to the emergence of
important contributions in political
economy, which enables the assessment
of complex issues; for example, atti-
tudes toward biotechnology (Swinnen,
2010). Agricultural economists have
George G. Judge developed many numerical systems over
the years (e.g., McCarl and Spreen,
1997). Richard Howitt (1995) developed new, creative methods using
constrained optimization for quantitative policy design. Agricultural
economists have been crucial in developing an international trade
model. Irma Adleman was crucial in developing the computer general
equilibrium model, and Thomas Hertel31 introduced the Global

29
https://scholar.princeton.edu/deaton/home.
30
https://are.berkeley.edu/users/george-g-judge.
31
https://web.ics.purdue.edu/∼hertel/.
 Prologue: The Uniqueness of Agricultural Economics xxiii

Trade Analysis Project (GTAP), which includes both software and

a database that allows the assessment of the economic impact of
multiple policies (Hertel, 1997).
Because of the constant contact of agricultural economists with
practitioners in the field, they recognized some of the limitations
of neoclassical economics. They were pioneers in the development
of models of “behavioral economics.” Agricultural economics were
among the first to use experiments to assess risk preferences,
estimate imitation models of technology adoption, develop models of
adaptive expectation, and introduce different ad hoc rules to resource
allocation and risk management (Wuepper et al., 2023). Consumer
food choices have become an area with multiple applications of
behavioral economics (Lusk and McClusky, 2018; Caputo and Just,
2022). Agricultural economists were among the first to apply the
notion of social capital to economic choices (Schmid and Robison,
1995). Recent studies assess the impact of culture, beliefs, and self-
efficacy on farmers’ behavior and use methods from other social
sciences to understand economic behavior.

The Present and the Future of Agricultural

Economics
Agricultural economics is part of, but simultaneously distinct from,
economics. It is inspired by reality, evolving over time; it is multi-
disciplinary in the sense that it presents results of cross-pollination
of economics and other disciplines, and therefore has its own applied
theory and unique tools that have influenced economics and other
disciplines; and it guides agricultural and environmental policies.
Any discipline needs sources of financial support, and in the case
of agricultural economics, it is the U.S. Department of Agricul-
ture and other federal and state agencies that provide resources
for research. Other international organizations, such as the Food
and Agriculture Organization (FAO) and the World Bank, and
other non-governmental organizations and private sector companies,
have invested in agricultural research that they are interested in
implementing. Agricultural economists have several associations,
xxiv Agricultural Economics and Policy

including the International Agricultural Economics Association, the

Agricultural and Applied Economics Association, the European
Agricultural Economics Association, and many national and regional
associations. As the share of agriculture in the GDP and employment
in agriculture declined, agricultural economists expanded their reach
to address issues of natural resources, food, energy, and development.
Today, the title “Agricultural and Applied Economics” is much more
descriptive of the research activities by organizations that started as
a narrowly defined agricultural economics unit.
We expect that these dynamics will continue, and the agricultural
and applied economic discipline will expand to cover multiple areas,
adapting to reality. There will be a larger emphasis beyond the
farm gate, with more emphasis on agribusiness. With the new tools
of biotechnology and the challenges of climate change, we expect
that agriculture and natural resources will become a major part
of a modern bioeconomy that applies tools of biotechnology and
information technology with land and other natural resources to
produce food, fuel, energy, and biochemicals. It is crucial to transition
from a non-renewable to a renewable economy (Zilberman et al.,
2013). Agricultural and Applied Economics will continue to be an
integrating discipline between economics and the natural sciences
and will contribute to the emergence of more integrated social science
tools. The capacity and approach of the discipline will be affected by
the development of technology and computational capabilities, with
enhanced utilization of expansive databases, modern analytical tools,
and improved monitoring and interaction.
We hope that this book will provide a good introduction
to some major issues and methods in agricultural and applied
economics, and that it will be useful for graduate students and
researchers who are studying, researching, or teaching agricultural
economics and policy. Chapter 1 offers a brief history of U.S.
agriculture and agricultural policy. Following it, we then present
three chapters (i.e., Chapters 2–4) that summarize some core models
related to production, demand, as well as decision-making under
risk and uncertainty. Chapters 5 and 6 discuss welfare analysis
and political economy of agricultural policies. Chapters 7 and 8
 Prologue: The Uniqueness of Agricultural Economics xxv

focus on technical innovation and adoption, respectively, on which

Chapter 9 (Agricultural Supply Chain) is built. Chapter 10 examines
food security and safety, and Chapter 11 studies the economics of
pesticides. Chapter 12 investigates some major conservation policies
and payments for ecosystem services in agriculture, followed by
Chapter 13 in which the relationship between agriculture and climate
change is discussed. Finally, the epilogue discusses the economics of
sustainable development in agriculture.
The book has several advanced chapters that require some back-
ground in calculus, but we believe that much of the book is accessible
to individuals with some quantitative background. The notes in this
book do not cover all the fields of agricultural economics equally; it is
based on classes that we actually taught and thus, for example, has a
minimal discussion of agricultural trade, finance, labor, or nutrition.
The text has evolved over time and is based on classes and knowledge
that we obtained from our teachers (in particular, in the case of
David Zilberman: Richard Just, Gordon Rausser, Andrew Schmitz,
and Alain de Janvry) and many generations of students. We hope
you will enjoy the book.

Acknowledgments
The book would not be possible without the help and support
of many people. We thank Carlo Cafiero, who contributed to the
early draft of the class note. We thank the late David Buschena
for useful advice and comments. Comments and suggestions from
Madhu Khanna, Jeff LeFrance, and from students in the course
sequences at both University of California, Berkeley and Auburn
University are deeply appreciated. We are particularly thankful for
Benjamin Shapiro’s thorough editing of the book. We thank Eugene
Adjei, Nabin Bhandari, Brian Cornish, Bijesh Mishra, Pritam Mitra,
Sadie Shoemaker, and Pathmanathan Sivashankar for reading some
of the chapters in this book. Research assistance offered by Ngbede
Musa, Sadie Shoemaker, and Azaz Zaman is deeply appreciated.
The authors also thank Giannini Foundation at the University of
California, Berkeley, Alabama Agricultural Experiment Station, and
xxvi Agricultural Economics and Policy

the Hatch program of the National Institute of Food and Agriculture

within the USDA for financial support.

References
Adelman, I., D. Zilberman, and E. Kim. 2014. A Conversation with Irma
Adelman. Annual Review of Resource Economics 6(1): 1–16.
Aigner, D., CA K. Lovell, and P. Schmidt. 1977. Formulation and Estimation
of Stochastic Frontier Production Function Models. Journal of Economet-
rics 6(1): 21–37.
Alston, J.M. and P.G. Pardey. 2021. The Economics of Agricultural Innovation.
Handbook of Agricultural Economics 5: 3895–3980.
Anderson, K., G. Rausser, and J. Swinnen. 2013. Political Economy of Public
Policies: Insights from Distortions to Agricultural and Food Markets. Journal
of Economic Literature 51(2): 423–477.
Antle, J.M. 1999. The New Economics of Agriculture. American Journal of
Agricultural Economics 81(5): 993–1010.
Arrow, K.J. and A.C. Fisher. 1974. Environmental Preservation, Uncertainty, of
Irreversibility. The Quarterly Journal of Economics 88(2): 312–319.
Barrett, C.B. 2002. Food Security and Food Assistance Programs. Handbook of
Agricultural Economics 2: 2103–2190.
Barrett, C.B., T. Reardon, J. Swinnen, and D. Zilberman. 2022. Agri-Food
Value Chain Revolutions in Low- and Middle-Income Countries. Journal of
Economic Literature 60(4): 1316–1377.
Barry, P.J. and L.J. Robison. 2001. Agricultural Finance: Credit, Credit Con-
straints, and Consequences. Handbook of Agricultural Economics 1: 513–571.
Boehlje, M. 1999. Structural Changes in the Agricultural Industries: How Do We
Measure, Analyze and Understand Them? American Journal of Agricultural
Economics 81(5): 1028–1041.
Burt, O. 1964. The Economics of Conjunctive Use of Ground and Surface Water.
Hilgardia 36(2): 31–111.
Caputo, V. and D. Just. 2022. The Economics of Food Related Policies: Consid-
ering Public Health and Malnutrition. Handbook of Agricultural Economics
6: 5117.
Carlton, D.W. and J.M. Perloff. 2005. Modern Industrial Organization, 4th Ed.
Pearson Addison-Wesley, Boston.
Carter, C.A. 1999. Commodity Futures Markets: A Survey. Australian Journal of
Agricultural and Resource Economics 43(2): 209–247.
Chavas, J.-P. 2008. On the Economics of Agricultural Production. Australian
Journal of Agricultural and Resource Economics 52(4): 365–380.
Cochrane, W.W. 1979. The Development of American Agriculture: A Historical
Analysis. University of Minnesota Press.
de Janvry, A. and E. Sadoulet. 2022. Agriculture for Development: Analytics and
Action. Annual Review of Resource Economics 14: 1–16.
 Prologue: The Uniqueness of Agricultural Economics xxvii

Evenson, R.E. and Y. Kislev. 1975. Agricultural Research and Productivity. Yale
University Press.
Feder, G., R.E. Just, and D. Zilberman. 1985. Adoption of Agricultural Innova-
tions in Developing Countries: A Survey. Economic Development and Cultural
Change 33(2): 255–298.
Finger, R., S.M. Swinton, N. El Benni, and A. Walter. 2019. Precision Farming at
the Nexus of Agricultural Production and the Environment. Annual Review
of Resource Economics 11: 313–335.
Freeman III, A.M, J.A. Herriges, and C.L. Kling. 2014. The Measurement of
Environmental and Resource Values: Theory and Methods. Routledge.
Galbraith, J.K. 1998. The Affluent Society. Houghton Mifflin Harcourt.
Gardner, B.L. 2009. American Agriculture in the Twentieth Century: How It
Flourished and What It Cost. Harvard University Press.
Goldberg, R.A. 1968. Agribusiness Coordination: A Systems Approach to the
Wheat, Soybean, and Florida Orange Economies. Agribusiness Coordination:
A Systems Approach to the Wheat, Soybean, and Florida Orange Economies
50(3): 782–783.
Griliches, Z. 1957. Hybrid Corn: An Exploration in the Economics of Techno-
logical Change. Econometrica, Journal of the Econometric Society (1957):
501–522.
Hanemann, M., J. Loomis, and B. Kanninen. 1991. Statistical Efficiency of
Double-Bounded Dichotomous Choice Contingent Valuation. American Jour-
nal of Agricultural Economics 73(4): 1255–1263.
Hayami, Y. and V.W. Ruttan. 1971. Agricultural Development: An International
Perspective. Baltimore, Md/London: The Johns Hopkins Press.
Heady, E.O., C.F. Curtiss, and L.D. John. 1960. Agricultural Production Func-
tions. Kalyani Publishers, Chennai.
Hertel, T.W. 1997. Global Trade Analysis: Modeling and Applications. Cambridge
University Press.
Howitt, R.E. 1995. Positive Mathematical Programming. American Journal of
Agricultural Economics 77(2): 329–342.
Huffman, W.E. and R.E. Evenson. 2008. Science for Agriculture: A Long-Term
Perspective. John Wiley & Sons, Hoboken.
Johnson, D.G. 1997. Agriculture and the Wealth of Nations. The American
Economic Review 87(2): 1–12.
Judge, G.G. 1982. Introduction to the Theory and Practice of Econometrics.
Wiley, Hoboken.
Judge, G.G. and R.C. Mittelhammer. 2011. An Information Theoretic Approach
to Econometrics. Cambridge University Press.
Just, R.E., D.L. Hueth, and A. Schmitz. 2005. The Welfare Economics of Public
Policy: A Practical Approach to Project and Policy Evaluation. Edward Elgar
Publishing.
Lusk, J.L. and J. McCluskey. 2018. Understanding the Impacts of Food Consumer
Choice and Food Policy Outcomes. Applied Economic Perspectives and Policy
40(1): 5–21.
McCarl, B.A. and T.H. Spreen. 1997. Applied Mathematical Programming Using
Algebraic Systems. Cambridge, MA.
xxviii Agricultural Economics and Policy

Miao, R. and M. Khanna. 2020. Harnessing Advances in Agricultural Technologies

to Optimize Resource Utilization in the Food-Energy-Water Nexus. Annual
Review of Resource Economics 12: 65–85.
Mundlak, Y. 1978. On the Pooling of Time Series and Cross Section Data.
Econometrica: Journal of the Econometric Society 46: 69–85.
Nerlove, M. 1956. Estimates of the Elasticities of Supply of Selected Agricultural
Commodities. American Journal of Agricultural Economics 38(2): 496–509.
Pingali, P.L. 1997. From Subsistence to Commercial Production Systems: the
Transformation of Asian Agriculture. American Journal of Agricultural
Economics 79(2): 628–634.
Pinstrup-Andersen, P., R. Pandya-Lorch, and M.W. Rosegrant. 1999. World Food
Prospects: Critical Issues For the Early Twenty-First Century. Food Policy
Report.
Randall, A., B. Ives, and C. Eastman. 1974. Bidding Games for Valuation of Aes-
thetic Environmental Improvements. Journal of Environmental Economics
and Management 1(2): 132–149.
Ricardo, D. 1821. On the Principles of Political Economy. J. Murray, London.
Schmid, A.A. and L.J. Robison. 1995. Applications of Social Capital Theory.
Journal of Agricultural and Applied Economics 27(1): 59–66.
Schmitz, A. and D. Seckler. 1970. Mechanized Agriculture and Social Welfare:
The Case of the Tomato Harvester. American Journal of Agricultural
Economics 52(4): 569–577.
Schultz, T.W. 1961. Investment in Human Capital. The American Economic
Review 51(1): 1–17.
Sexton, S.E., Z. Lei, and D. Zilberman. 2007. The Economics of Pesticides
and Pest Control. International Review of Environmental and Resource
Economics 1(3): 271–326.
Sexton, R.J. 2013. Market Power, Misconceptions, and Modern Agricultural
Markets. American Journal of Agricultural Economics 95(2): 209–219.
Smith, A. 1776. An Inquiry into the Nature and Causes of the Wealth of Nations.
W. Strahan and T. Cadell, London.
Stevens, A.W. 2018. The Economics of Soil Health. Food Policy 80: 1–9.
Subramanian, S. and A. Deaton. 1996. The Demand for Food and Calories.
Journal of Political Economy 104(1): 133–162.
Swinnen, J.F.M. 2010. The Political Economy of Agricultural of Food Policies:
Recent Contributions, New Insights, and Areas for Further Research. Applied
Economic Perspectives and Policy 32(1): 33–58.
Taylor, E.J. 1999. The New Economics of Labour Migration and the Role of
Remittances in the Migration Process. International Migration 37(1): 63–88.
Tweeten, L. 1999. The Economics of Global Food Security. Applied Economic
Perspectives and Policy 21(2): 473–488.
Waugh, F.V. 1944. Does the Consumer Benefit from Price Instability? The
Quarterly Journal of Economics 58(4): 602–614.
Weersink, A., E. Fraser, D. Pannell, E. Duncan, and S. Rotz. 2018. Opportunities
and Challenges for Big Data in Agricultural and Environmental Analysis.
Annual Review of Resource Economics 10: 19–37.
 Prologue: The Uniqueness of Agricultural Economics xxix

Wheeler, T. and J. von Braun. 2013. Climate Change Impacts on Global Food
Security. Science 341(6145): 508–513.
Working, H. 1961. New Concepts Concerning Futures Markets and Prices. The
American Economic Review 51(2): 160–163.
Wuepper, D., S. Bukchin-Peles, D. Just, and D. Zilberman. 2023. Behavioral
Agricultural Economics. Applied Economic Perspectives and Policy 45:
2094–2105.
Zilberman, D. 2019. Agricultural Economics as a Poster Child of Applied
Economics: Big Data & Big Issues. American Journal of Agricultural
Economics 101(2): 353–364.
Zilberman, D., E. Kim, S. Kirschner, S. Kaplan, and J. Reeves. 2013. Technology
and the Future Bioeconomy. Agricultural Economics 44(s1): 95–102.
Zilberman, D., T. Reardon, J. Silver, L. Lu, and A. Heiman. 2022. From the
Laboratory to the Consumer: Innovation, Supply Chain, and Adoption with
Applications to Natural Resources. Proceedings of the National Academy of
Sciences 119(23): e2115880119.
This page intentionally left blank
 Chapter 1

History of U.S. Agriculture:

A Lesson in Development

The settlement of the United States and the evolution of U.S.

agriculture offer unique case studies in economic development. In
this chapter, we will discuss agricultural policies in the United
States beginning around 1620 when settlers from Europe established
new agricultural policies and systems. The discussion draws much
from Cochrane (1979). Currently, when we speak about agricultural
policies, we mostly speak about price supports, subsidies, as well
as environmental regulations and payments, but agricultural policies
evolved from land and settlement policies to farm support policies
to landscape and environmental quality policies. We can distinguish
between four stages in the U.S. agricultural development process:
1620–1740 (search for appropriate farming practices), 1740–1810
(establishment of viable production modes), 1810–1900 (the west-
ward movement), and 1900–present (the intensification of farming
practices). We now briefly describe some key evolvement in U.S.
agriculture during these four stages.

1. Search for Appropriate Farming

Practices — 1620–1740
In the early 1600s, groups of aspiring settlers migrated from Europe
to find their luck in the new world. They brought farm animals, seeds,

1
2 Agricultural Economics and Policy

equipment, and some know-how, and started settling. But not many
made it; the weather and soil conditions were not accommodating,
and therefore many settlers perished. We mainly read about the
success stories, though. To survive, the settlers became involved in
fishing, hunting, and gathering, and their food production activities
resembled those in the hunting-and-gathering stage of human devel-
opment. Slowly, the early settlers learned how to farm in the New
World. They adapted to the local conditions, borrowed practices
and crops from Native Americans, and established some modes of
operation that became successful.
The early settlers learned to plant quite a few crops from the
Indigenous, including corn (or maize), pumpkins, squash, beans,
and tobacco. With the exception of corn, crops during this period
were specialized by geographic area and stage of settlement. In the
early 1700s, New England had become the center of a sea-based
fishing and whaling industry. In the middle colonies (New Jersey and
Pennsylvania), settlers produced wheat, corn, rye, oats, and barley. In
Maryland, Virginia, and North Carolina, the preeminent commercial
crop was tobacco. Production of tobacco developed rapidly since the
early settlement and provided an important cash crop for export
(Cochrane, 1979). Rice was introduced in the late 1600s and became
an established crop around Charleston, South Carolina by the early
1700s. In the 1600s, animal husbandry virtually did not exist. The
livestock industry began to take shape in the 1700s due to the
increased demand for higher-quality livestock.
Colonial agriculture was a hand-labor agriculture, and early
settlers used little or no animal power. The basic tools were hoes,
axes, and scythes. There was a continuing scarcity of human labor in
the colonies. Farmers in New England depended almost entirely on
family labor and sought to increase their labor supply by raising large
families. Large farms imported indentured servants from Europe.
Large tobacco or rice plantations in Maryland, Virginia, and South
Carolina turned increasingly to the importation of slaves. The system
of slavery was firmly institutionalized by the 1700s.
 History of U.S. Agriculture: A Lesson in Development 3

2. Establishment of Viable Production

Modes — 1740–1810
By 1740, the older, settled areas became highly specialized in
the production of agricultural commodities and food. In the
New England states, farmers produced corn, wheat, hay, apples,
and livestock. In Pennsylvania and New Jersey, farmers produced
a surplus of wheat and potatoes, and developed a small livestock-
producing and fattening industry. In the Southern states, tobacco,
rice, cotton, and sugar cane were the major crops. The surplus of
agricultural products was available for sale to the growing towns and
export to Europe.
The modes of operation differed in the North and South of
the New World. While the North attracted many migrants from
Europe, who established relatively small farms and villages, the
South established many large plantations that relied on slave labor.
We must recall that the popular mode of settlement in South
America, which attracted a relatively small number of European
settlers, was also the plantation model. Brazil was a major importer
of slaves in the production of sugar cane. The earlier slaves in the
New World arrived in South Carolina for the production of sugar
cane, but once the demand for cotton took off, the South transitioned
to it. The division between the North and the South resulted in a
continuous conflict that led to the U.S. Civil War.
During the American Revolution period, a freehold land tenure
system was established, and the United States acquired the national
public domain. The federal lands were sold in public auctions
to private persons in cash initially, and then in both cash and
credits through a series of Land Acts in 1785, 1796, 1800, and
1804. Purchases of the public domain were typically made by
speculators who hoped to resell that land to settlers at higher prices.
Pioneers who could not afford to buy land simply “squatted” on
vacant public lands, which was a popular means of acquiring a
farm during this period. In addition, large amounts of land were
4 Agricultural Economics and Policy

distributed as grants or bounties to people who had served in the

Continental Army.
From 1775 to 1810, some pioneers started moving across the
Appalachians. Hunters and woodsmen showed the way, followed
by some small pioneer farmers who had a patch of corn and
some livestock. Life for the pioneer farmers and their families was
both hard and cruel as they traversed the Appalachian Mountains
(Cochrane, 1979).

3. The Westward Movement — 1810–1900

While the early settlements were established on the East Coast,
during the 1800s, the government promoted populating further west.
After the Louisiana Purchase in 1803, the U.S. government controlled
the majority of what we recognize as the continental U.S. today.
While some people moved west and occupied land with no official
rights (i.e., “squatters”), to attract people to move en masse, the
government established a system of homesteading whereby a family
was granted the rights to a parcel of land in a designated area of
the west. The basic idea was that first arrivals ensured rights, but
these rights were only held up when families occupied the land. Later
this system was applied to water rights, where it is called the prior
appropriation system based on the principle of “first in time, first
in right,” as well as “use it or lose it.” These systems of land and
resource allocation proved effective early on, but later became sources
of inefficiency.
The government played a crucial role in the settlement processes
because it owned much of the new land, allowing it to direct the
processes. A series of land policies during this period promoted the
westward movement. Under the provisions of the Preemption Act
of 1841, a farmer could settle legally in the public domain before
he purchased his land. Further, he could buy 160 acres of the land
on which he had settled for $1.25 per acre once the land in that
area was opened for sale. This was close to a policy of free land
disposal. The government also granted land rights to farmers through
homesteading. The Homestead Act was enacted on May 20, 1862,
to distribute lands in the public domain to settlers free of charge
 History of U.S. Agriculture: A Lesson in Development 5

under a few conditions. More and more people migrated west of the
Mississippi River into the Great West. The final settlement and its
linkage to the east through a railroad network occurred in the period
of 1860–1897.
Moreover, due to its limited ability to collect taxes over this
period, the U.S. government used land sales to finance the devel-
opment of public goods. It also used land grants to incentivize the
construction of the railroads, and the railroad companies owned a
mile of land to each side of the tracks. As a result, to this day,
the Southern Pacific Transportation Company is one of the biggest
landowners in California. Adjacently, western land settlement was
fueled by railroad construction. In each municipality, the government
granted land to build schools and finance them through land sales.
The establishment of rural schools was one of the major achievements
of the U.S. Later, land grants were used to finance universities, and
every state had a land grant university system.
The major rural ideology of the U.S., based on the reality of the
North, was that family farms should be the main form of farming
and that agriculture should be a competitive sector consisting of
independent farmers instead of peasants. This ideology is called the
Jeffersonian Vision, named after the renowned Thomas Jefferson.
Within this ideology, the role of the government is to provide
education, information, and technologies. Nobel Laureate Theodore
Schultz, the father of the idea of human capital, emphasized the
difference between a peasant and a farmer (Schultz, 1975). Peasants
are traditional farmers who learn practices from their parents.
Traditional agriculture has been in equilibrium, where yields and
methods changed little, and therefore imitation was the best form
of acquiring practices. A farmer, however, lives in a dynamic envi-
ronment, requiring them to adjust to changes in economic conditions
or technology. After the start of the Industrial Revolution, and the
massive expansion of global trade, the importance of adapting to
changes increased. Schultz distinguished between two types of human
capital: worker ability and allocative ability. The former involves the
physical capacity to conduct tasks, while the latter involves business
and management decisions. He also used the term “the ability to deal
with disequilibrium” to describe allocative ability. Allocative ability
6 Agricultural Economics and Policy

requires literacy and numeracy, and rural schools were introduced to

train farmers with allocative ability.
Moreover, the thinkers behind the idea of family farms recognized
that individual farms could not develop new technologies to address
changing agricultural realities. Therefore, they later introduced land
grant universities with experiment stations that developed new agri-
cultural practices, and, later, Cooperative Extension, which trans-
ferred knowledge to farmers. The land grant colleges were established
through the Morrill Act of 1862 during the Civil War. These colleges
offered courses in agriculture and mechanical arts in addition to
classical studies. The U.S. Department of Agriculture was estab-
lished in 1862 as well, and one of its major roles was to provide
price information to farmers to ensure competitive and fair trade.
Agricultural experimentation stations were established by the Hatch
Act of 1887 and Cooperative Extension through the Smith–Lever Act
of 1914.
The idea of a competitive farm sector that is augmented by
collective actions has prevailed throughout the history of U.S.
agriculture. Farmers established the farm bureau, an organization
that aimed to represent them politically and have the market
muscle to obtain better terms of insurance and other products.
In a history where buyers tended to have monopolistic power,
there was a need for a countervailing power. Thus, agricultural
cooperatives that were established to negotiate terms of trade for
farmers helped balance the relationship by controlling supply. The
other developments in agriculture during this period include farm
mechanization, market expansion, and agribusiness development.
The process of farm mechanization started in the 1830s, reached
widespread proportions in the 1850s, and accelerated since 1860.
The introduction of the steamboat, the construction of the Erie
Canal, and the development of railroads in the early 1800s opened the
market to western farmers. The later development of the refrigerator
railroad car for shipping fresh meat, fruits, and vegetables also
widened the markets for farm producers and increased their income-
earning possibilities. Agribusiness was developed to produce farm
inputs, and to store, process, and distribute farm products. Market
 History of U.S. Agriculture: A Lesson in Development 7

centers such as Chicago, St. Louis, and New Orleans grew to handle
and distribute the growing agricultural surpluses.
The specialization of agricultural production according to their
comparative advantage by areas was largely completed in the period
of 1860–1900. The Northeast specialized in fruit, truck crops, and
dairy production. The Upper South specialized in tobacco production
and produced a surplus of corn. The Deep South continued to
specialize in the production and export of cotton even after the
abolition of slavery. The Midwest specialized in corn production,
and corn surplus was utilized to support a livestock-feeding industry.
Grain farming on the Great Plains was firmly established by 1890.
California agriculture developed rapidly in the 1850s, and fruit
and vegetable production developed quickly. The production areas
established during this period continued up to the present day.
Perhaps one of the reasons why U.S. citizens expect much from
the government, but do not like to pay taxes, is that for much of
its early history, the government acquired assets with little or no
payment and used or sold them to provide essential services. One
has to admit, though, that the investments made by the government
have been effective and yielded fruits. At the beginning of the 20th
century, the government realized that it now needed to raise taxes
to continue expanding the services it could provide. This led to the
establishment of various forms of taxation, including federal income
tax and the Internal Revenue Service. The tax revenue would allow
the government to establish safety nets that support agriculture.

4. The Intensification of Farming

Practices — 1900–Present
Throughout the 19th century, agriculture expanded westward. Pro-
duction per acre changed little, even though U.S. agriculture acquired
the capacity to adapt production systems to various conditions. The
U.S. is a large country, and adapting wheat, corn, cotton, and other
products to various regions required significant efforts and innovation
(Olmstead, 2008). But towards the end of the 19th century, it
became clear that expansion of U.S. agricultural production would
8 Agricultural Economics and Policy

21

19
Corn area and production relative to 1866
Corn production
17

15

13

11

5
Corn harvested area
3

1
1866 1886 1906 1926 1946 1966 1986 2006

Figure 1. Area of corn harvested and production, U.S. 1866–2022 (indexed to

1866=1).
Source: The U.S. Census Bureau.

require increasing yields per acre, or intensification. The 20th century

indeed has seen significant intensification as a result of breakthroughs
such as the introduction of nitrogen fertilizer, modern breeding, and
pest control. Some of these technologies were commercialized by
the private sector, but their use was adapted to various locations
by experimental stations and Cooperative Extension. The export
market has played an important role in the development of U.S.
agriculture. The level of agricultural exports reflected the demand
for U.S. farm products and was critical to the economic well-being
of U.S. farmers. While agricultural prices fluctuated — reflecting
changes in market conditions — they reached a peak during the
First World War. During the war, the U.S. was a food supplier for
much of the world, and agricultural land use peaked between 1918
and 1925.
As Figure 1 demonstrates, corn production in the U.S. increased
proportionally to acreage until the 1930s, and since then has risen
more than 5-fold while acreage harvested declined. As the U.S.
Department of Agriculture (USDA, 1969, Table 4) shows, crop
acreage peaked in 1929 at 359.2 million acres, declined significantly
 History of U.S. Agriculture: A Lesson in Development 9

20

18

16

14

12

10
$/bu.
8

0
1913
1915
1917
1919
1921
1923
1925
1927
1929
1931
1933
1935
1937
1939
1941
1943
1945
1947
1949
1951
1953
1955
1957
1959
1961
1963
1965
1967
1969
1971
1973
1975
1977
1979
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
2003
2005
2007
2009
2011
2013
2015
2017
2019
2021
Corn Soybeans Wheat

Figure 2. Major crop commodity prices in the United States, 1913–2022

($/bushel, in 1982 dollars).
Source: National Agricultural Statistics Service of the USDA.

during the recession, rebounded almost to its peak during the Second
World War, and then declined steadily to about 273 million acres in
1969. Since then, crop acreage has stabilized at around 310 million
acres. In total, agricultural land, including cropland, rangeland, and
woodland, is about 900 million acres. USDA (1969, Table 4) and
USDA (2012, Table 8) also show that the number of farms has
declined since 1935 when there were about 6.8 million farms in
the U.S., and by 2012 there were around 2.1 million farms. At the
same time, the size of the average farm increased and the productive
resources of American agriculture had become concentrated in fewer
and larger farms.
Figure 2 depicts the prices of major agricultural commodities
over the 1912–2022 period. Inflation-adjusted prices reached their
peak during the First World War (1914–1918), declined precipitously
during the Great Depression (1929–1932), peaked again during the
Second World War (1939–1945), declined afterward, but peaked
briefly in 1973–1974 due to both the 1972 United States — Soviet
Union Wheat Deal and the 1973–1974 Oil Embargo. In 1972, the
U.S. government sold the Soviet Union a large volume of grains
(about 10 million tons) at subsidized prices, not realizing that
10 Agricultural Economics and Policy

inventories were limited, and the net effect was a drastic increase
in prices. President Nixon imposed price controls on foods and
restricted exports in response. The government also established
a more systematic monitoring process of grain inventories. Prices
then declined until around 2005 and had risen as a result of the
introduction of biofuels. Over 2013–2022, crop prices fluctuated due
to supply, trade, and pandemic shocks.

5. Farm Problem and Agricultural Policies

The overall trends in the 20th century U.S. agriculture were declining
prices and a reduction in acreage. Declining prices, resulting in
declining farm income, were a dominating factor in U.S. agriculture
policy. Thus, many policies sought to stabilize prices, assure income,
and restrict acreage. Schultz (1964) and Cochrane (1979) developed
a framework to analyze the major features of agriculture around the
mid-20th century. They included:

(1) High rate of innovation.

(2) Many small farms.
(3) Homogenous product.
(4) Competitive markets.
(5) Inelastic demand.
(6) Randomness.
(7) Use of natural resources.
(8) Environmental and health side effects.
(9) Lack of mobility of inputs.
(10) Credit imperfection.

These features still apply today and encapsulate what people have
called the “farm problem.” This problem has several dimensions.
The first dimension is that commodity prices tend to decline
over time. The same may be true for the aggregate income of
agricultural commodities. Inelastic demand for agricultural com-
modities, competition, and technological change lead to reduced
farm prices and farm income. In particular, the adoption of new
supply-enhancing technologies by competitive industries benefited
consumers but reduced the overall income of the agricultural sector
 History of U.S. Agriculture: A Lesson in Development 11

due to inelastic demand. Two factors contribute to increased demand

for agricultural products and therefore agricultural income: One is
increased non-farm income domestically, but this effect is declining
because agricultural commodities have low-income elasticity in a
country like the U.S.; the second factor is increased global demand.
An increase in income in developing countries may result in increased
international demand for food, and that may lead to increased export
opportunities for U.S. farmers. Therefore, periods of low supply
overseas, for example during the wars, and the opening of new
markets like China and Russia tend to increase demand for and prices
of U.S. crops.
The second dimension is the instability of prices. Randomness
in supply and, sometimes, demand due to wars or changes in
weather or economic conditions lead to fluctuating agricultural
commodity prices. The third dimension is rural poverty. This is
caused by the fact that agricultural assets, including human capital,
are not mobile and cannot be easily reallocated to other uses. So,
a reduction in farm prices leads to a reduction in farm income
and farm assets, and sometimes causes significant bankruptcies and
crises. The fourth dimension is the environmental side-effects of
agriculture. Agricultural production may cause environmental and
resource degradation, such as polluting groundwater, depleting soil
resources, and harming wildlife. The great dust bowl of the 1930s was
a wake-up call for the need for land stewardship. The publication
of Silent Spring (Carson, 1962) raised awareness of the need for
pesticide management.
Several policies were developed to address these different dimen-
sions of the farm problem. Most of them were established by farm
bills, such as price support, income support, supply control, inventory
control, risk management, land conservation, environmental regu-
lation, credit subsidies, and export support programs. Some other
policies, such as biofuel policies and crop insurance policies, however,
were not established by farm bills, but were still closely related to
agriculture.
To address the problems of low income, the government estab-
lished a series of price support and supply control policies. In the
1950s, price support policies served to increase supply and resulted
12 Agricultural Economics and Policy

in a high level of inventory, and the excess supply reduced prices even
further. One solution to the excess supply problem was to send some
of the excesses as foreign aid to countries that had food shortages
but could not afford to import food. An unintended consequence was
low prices in these countries that impeded the growth of their own
agriculture sector. Recognition of this problem set a limit on food
aid, and now the aim is to use it when it is justified. Another policy
was deficiency payments, where the government allocated to farmers
a certain quantity target (called payment yield), and prices were
set based on the target. If the market price was below the target
price, farmers would receive a payment that is equal to the price
shortfall times the quantity target. This approach aimed to control
supply and support farmer income. But what occurred was increased
average prices and decline in risk, and thus supply increased. To deal
with this problem, governments established set-aside requirements,
where farmers were supposed to divert a certain percentage of their
land to conservation or fallowing to qualify for deficiency payments.
The establishment of some base yield and acreage that were enti-
tled to deficiency payments was aimed to protect income, rather
than price.
To address instabilities in agricultural prices, the government
developed inventory control policies, which can be viewed as a
break from supply control but still considered as direct intervention
policies. The government established a price floor and ceiling —
if the actual price was below the floor, the farmers were able to
keep their crops in inventory while receiving a loan at the floor level.
If the actual price went above the ceiling, the farmers were required
to sell and pay the loan. In many cases, the government covered
or subsidized the debt service on the loan. The idea behind this
scheme was to keep the price within a range. Of course, there were
periods when prices stayed low and inventory increased too much —
indeed, there was one year (1983) when the government established
a Payment-in-Kind program that paid farmers not to grow certain
crops and instead sell crops from inventory.
Another approach to deal with instability is crop insurance which
insures yields, and later, revenue. Farmers pay a premium and
 History of U.S. Agriculture: A Lesson in Development 13

are assured a minimum level of income based on yield or revenue

targets. The premium paid is based on several alternative levels of
coverage. The program is subsidized heavily, so the premiums paid
by farmers do not fully cover expected indemnity payments. The
government also covers the administrative costs of this program.
In addition to insurance, the government has disaster assistance
programs in cases of catastrophes. Most of the government programs
applied to major commodities, such as corn, wheat, and cotton.
There are also programs for peanuts and others for dairy. Specialty
crops, like vegetables, grown in specific states are not part of the
major USDA programs, even though they were recently covered
by crop insurance programs. But for these crops, growers may
organize and vote on product quality standards enforceable through
a marketing order. Because farmers are subject to credit constraints,
the government instituted credit products provided by the Farm
Credit Administration.
Agricultural policies include several environmental regulations.
The initial set-aside program has become the payment for ecosystem
services programs. The most notable is the Conservation Reserve
Program under which farmers are paid to modify their production
activities toward actions that provide ecological benefits. These ben-
efits can be the protection of wildlife habitat and the reduction of
soil erosion and wind erosion. Furthermore, the Environmental Pro-
tection Agency (EPA) was established in 1970 to protect people and
the environment from significant health risks, to sponsor and conduct
research, and to develop and enforce environmental regulations.
Quotas and standards are the predominant approach to controlling
agriculture-induced environmental and health risks. The Clean Air
Act (1970), Clean Water Act (1972), Federal Insecticide, Fungicide
& Rodenticide Act (FIFRA, 1947) and subsequent amendments are
enforced by the EPA to protect air and water as well as regulate
pesticide usage. The EPA has also issued rules to limit green-
house gas emissions from various sources to address climate change
issues.
One of the major issues of U.S. agriculture was labor shortages,
especially in states such as California and Florida, where fruits
14 Agricultural Economics and Policy

and vegetables are a major agricultural output. The government

developed programs to recruit inexpensive, international labor. One
such program was the Bracero Program (1942–1964), where farm
workers were recruited from Mexico to work in the U.S. at very
low wage rates. The program was abolished in 1964, and afterward,
farm workers unionized while migrants continued to enter the U.S.,
sometimes illegally, and work in farms at a higher rate than during
the Bracero Program period.1 Farmers were active in supporting
policies to reduce immigration restrictions and allow illegal workers
to become citizens. The 1986 Simpson–Mazzoli Act provided citizen-
ship to a large number of previously illegal immigrants, legalized
some illegal seasonal agriculture workers, and required only self-
reporting of worker status by employers. To comply with some of
the regulations, many farm owners used labor contractors to hire
their seasonal workers.

6. Conclusion: Changes in Perspectives on

Agriculture

Over time, the farm population decreased, and the average farm
income increased. Farmers themselves have gained in wealth com-
pared with the rest of the population (Gardner, 1992). In addition,
international trade barriers have been reduced, allowing farmers
to access more export markets, and farmers continue to develop
differentiated products that enable higher income. In addition, the
U.S. is part of the World Trade Organization, which aims to
reduce barriers to trade, including decreased subsidies for agriculture.
Deficiency payments were eliminated by the 1996 Farm Bill, but there
was increased emphasis on crop insurance as well as green policies.
These new trends resulted in a reduction of policies focused on
direct intervention strategies but an expansion of crop insurance
programs that became the new carrier of subsidies, as well as a
continuation of environmental policies. With growing awareness of

1
See Clemens et al. (2018) for a study regarding the impact of Bracero Program’s
termination on wage rates for domestic farm workers.
 History of U.S. Agriculture: A Lesson in Development 15

climate change, there is an effort to reduce agricultural greenhouse

gas emissions and, at the same time, to use agriculture to sequester
carbon. One of the challenges is to incorporate credits from carbon
sequestration in the various carbon reduction policies and markets.
The concern about climate change, combined with concern over
energy security, also led to the introduction of policies targeted
at increasing the production of biofuels. Agricultural policies thus
became meshed with energy policies. The introduction of biofuel
policies was part of the growing emphasis on establishing the
bioeconomy. Meanwhile, soil conservation has long been a concern
and has been well supported through a host of conservation programs
such as the Conservation Reserve Program. Moreover, technical
innovation and adoption are viewed as major means to address many
sustainable development issues in agriculture. Finally, to understand
agricultural policies better, one must consider political economy
issues. The remainder of this book will touch on many of these issues.

References
Carson, R. 1962. Silent Spring. Houghton Mifflin Harcourt.
Clemens, M.A., E.G. Lewis, and H.M. Postel. 2018. Immigration Restrictions as
Active Labor Market Policy: Evidence from the Mexican Bracero Exclusion.
American Economic Review 108(6): 1468–1487.
Cochrane, W.W. 1979. The Development of American Agriculture: A Historical
Analysis. University of Minnesota Press.
Gardner, B.L. 1992. Changing Economic Perspectives on the Farm Problem.
Journal of Economic Literature 30(1): 62–101.
Olmstead, A.L. and P.W. Rhode. 2008. Creating Abundance. Cambridge Books.
Schultz, T.W. 1964. Transforming Traditional Agriculture. Yale University Press,
New Haven, CT.
Schultz, T.W. 1975. The Value of the Ability to Deal with Disequilibria. Journal
of Economic Literature 13(3): 827–846.
The U.S. Department of Agriculture (USDA). 1969. Census of Agriculture, Part 1:
Farms: Number, Use of Land. Available at: https://agcensus.library.cornell.
edu/census parts/1969-farms-number-use-of-land-size-of-farm/ (accessed
January 2, 2023).
The U.S. Department of Agriculture (USDA). 2012. Census of Agriculture, Part 1:
Farms: Number, Use of Land. Available at: https://agcensus.library.cornell.
edu/wp-content/uploads/2012-United-States-st99 1 008 008.pdf (accessed
January 2, 2023).
This page intentionally left blank
 Chapter 2

Review of Production Economics

Production has long been a focus of agricultural economics. This is

because not only food security relies on a stable supply of agricultural
commodities but also agricultural production can impose consider-
able impacts on the environment. In this chapter, we first briefly
review some basic properties of production functions. Centered on
production, we then discuss economics of land-quality-augmenting
input application technologies (e.g., drip irrigation). In the last
section, we discuss aggregation issues associated with investments
and production.

1. Production Function and Its Parameters

Production function is defined by the maximum output that can
be produced with a given input combination. The basic element
is technology. The detail and accuracy of a production function
depend on its use. A production function is presented in more generic
terms in a general theoretical context than in specific empirical
applications.
Let y denote output and x denote input. The production function
is y = f (x), marginal product (MP) is fx = ∂f /∂x, and average

17
18 Agricultural Economics and Policy

f (x)

Stage I Stage II Stage III x

MP/AP

B
MP
C
D

AP

G
eB eC eD x
Stage I Stage II Stage III

Figure 1. Three stages of production.

product (AP) is y/x. Recall that

⎧
⎪
⎨ MP > AP > 0 at Stage I of production function
AP > MP ≥ 0 at Stage II of Production function
⎪
⎩
MP < 0 at Stage III of production function.

Figure 1 represents output as a function of input and introduces

the three stages of production. The second stage is the economic
region. This is the stage with positive but decreasing marginal prod-
uct or concave production function. A competitive profit-maximizing
firm is likely to operate at this stage of the production function.
 Review of Production Economics 19

Many mathematical specifications of production functions, such as

the Cobb–Douglas function (y = Axα , with 0 < α < 1), only
represent situations when all outcomes are at the economic regions.
Their use precludes identifying situations in which producers operate
at the third stage of production and have negative marginal product.
Quadratic production functions, y = a + bx − cx2 , allow outcomes at
the second and third regions of production functions but not at the
first. A simple and elegant production function which allows three
regions of production is not easy to construct, so we often favor
simplicity and use flawed production function specifications in many
analyses.
A basic issue raised in Figure 1 is that of economies of scale, the
relationship between average production cost and production scale.
It presents that at Stage I of production, there is increasing returns to
scale. This is because in this stage AP is increasing in input quantity,
and thus the average cost is decreasing in input quantity. However,
at the economic region (i.e., Stage II), there are constant or more
likely decreasing returns to scale.
Figure 2 presents the relationships between inputs in the pro-
duction process. The isoquants depicted in Graph A in the figure
represents the different input level combinations producing the same
level of output. Isoquants are useful to address issues, such as input
intensity and input substitutability. If x1 is capital and x2 is labor,
x1 /x2 measures capital intensity relative to labor. Production at
point A is capital intensive and at B is labor intensive.
Economists are also interested in assessing the ease of replacing
one input for another while maintaining fixed output. When produc-
tion function is of the Leontief type (i.e., with fixed proportion):
 
x1 x2
y = min , ,
a1 a2
the isoquant for each production level is L-shaped, as shown in Graph
B in Figure 2. For cost-minimizing firms, the isoquant is essentially
one point,
(x1 = a1 × y, x2 = a2 × y)
20 Agricultural Economics and Policy

Capital, x1

Y = Y1 > Y0

Y = Y0

0
Labor, x2
Graph A

Capital, x1
Y = Y1 > Y 0

Y = Y0

0
Labor, x2
Graph B

Capital, x1

Y = Y 1 > Y0

Y = Y0

0
Labor, x2
Graph C

Figure 2. Isoquants and factor intensity.

 Review of Production Economics 21

and input intensity is constant at x1 /x2 = a1 /a2 . If production

function is linear, y = a1 x1 + a2 x2 , the isoquant is a straight line (see
Graph C in Figure 2), x1 = y/a1 − a2 x2 /a1 , and there are infinite
substitution possibilities.
To allow quantification and comparison of production technolo-
gies, some key parameters of the production function are defined.
They are as follows: the input elasticity, ηi ; the scale elasticity, ε;
and the elasticity of substitution, σij . Let y = f (x1 , x2 , . . . , xn ) be a
production function, where y indicates output and xi the ith input.
Then the following quantities can be defined:
∂f
• Marginal productivity: fi = ∂xi .
∂f xi
• Input elasticity: ηi = ∂x xi
y = fi y .
n i
• Scale elasticity: ε = i=1 ηi .
• The elasticity of substitution between input i and j:
∂(xi /xj ) (xi /xj ) d ln(xi /xj )
σij = − / =− .
∂(fi /fj ) (fi /fj ) d ln(fi /fj )
The elasticity of substitution measures the percentage change in
xi /xj along an isoquant curve when the marginal rate of technical
substitution (MRTS) between xi and xj (i.e., fi /fj ) increases by one
percent. Intuitively, it is a measure of the ease of change in input
intensity. Note that if the production function is quasiconcave, then
σij ≥ 0, with σij = 0 implying fixed proportion production function
and constant input intensity (e.g., the Leontief production function).
At the opposite extreme, there is the linear production function
(y = a1 x1 + a2 x2 ), where σ = ∞ and input intensity easily changes.
For Cobb–Douglas production functions, y = Axα1 1 xα2 2 , where
α1 + α2 = 1 and α1 , α2 ≥ 0, we have marginal product fi = αxiiy ,
input elasticity ηi = αi , scale elasticity ε = α1 + α2 , and elasticity
of substitution σ = 1. The Cobb–Douglas production function is
limited because it has constant elasticity of substitution with value
at 1. Furthermore, they do not allow for negative marginal product.
Note that in the case of one input, ε = η, that is, input elasticity
is equal to scale elasticity. Thus, at the first stage of the production
22 Agricultural Economics and Policy

function, we have ε > 1, at the economic stage, 1 > ε > 0, and at

the third stage, ε < 0.

1.1. Production function under perfect competition

The parameters of the production function have special interpre-

tation under profit maximization and price taking. Under such
conditions, the firm’s optimization problem is
n
max P f (x1 , . . . , xn ) − xi W i ,
x1 ,...,xn
i=1

where Wi is the price of input i. The first-order condition for ith

input is

P fi − Wi = 0,

and it can be interpreted as the value of marginal product of input

i must equal its price. This condition can be expressed in terms of
input elasticity: It becomes
f i xi W i xi
Py − Wi xi = 0 ⇒ P yηi = Wi xi ⇒ ηi = , (1)
y Py
that is, the input elasticity equals the share of revenue spent on input

i. In the case of constant returns to scale, ηi = 1 and ηi = share
of input i in all expenditures. Under perfect competition,
f1 W1
= .
f2 W2

1.2. Duality and its implications

The cost and profit functions are key relationships for deriving
quantities demanded and supplied. The profit function is defined as
n
Π(P, W1 , . . . , Wn ) = max P y − xi W i ,
y,x1 ,...,xn
i=1

subject to y = f (x1 , . . . , xn ).
 Review of Production Economics 23

Hotelling’s lemma states that the output supply and input

demand are determined by

∂Π
= y ∗ (P, W1 , . . . , Wn ),
∂P
∂Π
− = x∗i (P, W1 , . . . , Wn ).
∂Wi

We can use duality to present quantities as functions of monetary

variables. This is particularly useful to estimate supply and demand
for agricultural commodities and agricultural input. Furthermore,
when data for output levels or input mix are not available, dual
relationships can be used to estimate them. For example, one may
have accounting data on the output of different firms and one may
have price. Using duality, one can estimate output and input. In
principle, one can use duality-based relationships to estimate even
production function parameters from monetary data (see Chambers
and Pope (1994) for an example).
One of the key challenges of applied economists is to estimate
production parameters. Sometimes it is easier to obtain prices,
expenditures or revenue data than quantity data. Duality and other
relationships derived under profit maximization provide a basis for
estimation of technology relationships without data on quantities.
For example, if we know a farmer’s revenue is 100, expenditure on
input 1 (e.g., labor) is 20, and expenditure on input 2 (e.g., fertilizer)
is 30, then based on Equation (1), the suggested labor elasticity is
0.2 and fertilizer elasticity is 0.3. If relative prices of labor increase
by 10% and labor expenditures become 21 and fertilizer expenditures
become 31, the implied elasticity of substitution under competition
can be derived as follows. First, under the initial condition,

W10 x01 20
0 0 = ,
W 2 x2 30

where Wi0 and x0i are the price and quantities under initial outcome.
Let z be the rate of change in x1 /x2 . That is, x11 /x12 = (1 + z)x01 /x02 .
Then, under the new outcome expenditures, the ratio W11 x11 /W21 x12
24 Agricultural Economics and Policy

can be written as
W11 x11 1.1W10 x01 2 21
1 1 = 0 0 (1 + z) = 1.1(1 + z) = .
W 2 x2 W 2 x2 3 31

We can solve out z = 63/(62 × 1.1) − 1 = −0.0762. Therefore, we

have the elasticity of substitution as σ = −(−0.0762/0.1) = 0.762.
There are studies on the estimation of technological parameters
from monetary data assuming profit maximization. However, it is
clear that the estimated relationships are not necessarily the true
technical parameters. Milton Friedman said that it is not clear that
decision makers are profit maximizers, but the data are such that it
seems “as if” decision makers are profit maximizers. Production func-
tion parameters that are estimated under the profit maximization
assumptions are, in essence, “as if” production function parameters.
One can separate between technological relationships and behavioral
relationships. Behavioral relationships incorporate technological and
behavioral assumptions. Production function parameters that are
estimated under duality have a strong behavioral component. Even
production theory recognizes that strict profit maximization is
unrealistic; behavior has to adjust to risk and uncertainty, and there
are new models of production behavior under uncertainty. Herbert
Simon introduces a notion of bounded rationality. He suggested
that the ability of humans to process and analyze data is limited.
Therefore, choices are not perfect and reflect this limited ability.
One of the challenges is to decipher the factor behind production
decisions and to understand what leads producers to make choices.
Choices under strict profit maximization may be different than under
risk aversion and limited analytic capacities. The same technological
relationships may result in different outcomes under different behav-
ioral assumptions. However, it is difficult to untangle the behavioral
and technological contributions to observed outcomes.
The previously discussed production functions are stylized into a
metaphorical salad: All the components are put together and mixed
instantaneously. Actual production relationships are more complex.
Time plays an important role in production and production includes
several stages. One of the challenges of production theory is to
 Review of Production Economics 25

introduce the dimensions of time in the production process. Antle

(1983) developed a model of a sequential production process, and
there are several other attempts to look at the different stages of
production in order to represent more realistic models of a production
process.
However, modeling is an act of abstractism. For some purposes,
we need a very simple representation of reality and in other uses we
need a more realistic representation. In many aggregate analysis,
a simple presentation of the production process associated with
the traditional production function is sufficient. As the analysis
become more aggregated, generic modeling is more relevant. For
micro analysis, one may need more detailed modeling that takes into
account specific biological and physical phenomena.
Before we proceed with explicit production modeling issues, we
will discuss another issue, the measurements and nature of inputs
used in the production process. Here we focus on capital, labor, land,
pesticides, and water. Each input has unique features that may be
essential in modeling behavior at the farm level.
Capital consists of all the equipment, structure, and machinery
used for production. It represents outcomes of previous production
activities that are embodied in some assets relating to present
production activities. Generally, capital is utilized with viable
inputs — labor, energy, and fertilizers — that are consumed by
the production process. Producers may purchase services of capital
goods or they may own capital assets that would be reflected
differently in their accounting documents. In each period, there is
a cost associated with the use of capital goods, which includes
the cost of physical depreciation as well as the periodical costs
for the resources that were used in the capital investment (e.g.,
interest costs). One difficulty in measuring labor comes from the
differences in quality between different individuals. Generally, there
can be different wage rates according to the quality of labor services
provided. Knowledge acquired through training and education in the
past is a determinant of productivity in the present. Compensation
for workers combines payment for the raw labor services as well as
a return for their human capital. Similar to labor, land is not a
26 Agricultural Economics and Policy

homogeneous input. Land quality varies depending on location and

physical characteristics. There are different mechanisms for payment
of land services including rental fee and sharecropping. Moreover,
quality of land may affect the effectiveness of new technologies.
Pesticides are damage control agents. Their productivity depends
on the environment, the pest situation, and the product. The value
of water depends on its use, quality, and location.
In the following section, we develop models to analyze problems
of water. The modeling will demonstrate how some of the basic
biological or physical properties of water affect the specifics of the
modeling of the production process, the nature of choices that are
applied, and the type of outcome that we observe.

2. The Economics of Land-Quality-Augmenting

Input Application Technology
In this section, we model how adoption of an input application
technology that augments land quality may affect the use of the
corresponding input. We focus specifically on drip irrigation tech-
nology and water use in the model, but the framework and insight
developed here can be applied to understanding the impact of many
other input application technologies on input uses, even including
the relationship between fuel efficiency enhancing technology and
gasoline use by automobiles.
Let y denote output per acre and e effective input per acre.
Applied input per acre is a. The output price is P and the water
input price is W . Let i be application technology indicator, where
i = 0 stands for the traditional technology and i = 1 for the modern
technology. Land quality is denoted by α ∈ (0, 1). We can view α as
a measure of input use efficiency of the land, which can be enhanced
by technologies. Suppose the production function is y = f (e), with
f  > 0 and f  < 0. Let hi (α) be the input efficiency function,
reflecting the fraction of input consumed by crop with technology
i and land quality α. Therefore, we have ei = hi (α)ai . For simplicity,
we assume h0 (α) = α. We further assume that 0 ≤ α ≤ h1 (α) ≤ 1,
hi > 0, and hi < 0. The cost of technology i per acre of land is ki ,
 Review of Production Economics 27

with k1 > k0 . Let δi be a technology selection indicator. If technology

i is chosen, then δi = 1; otherwise, δi = 0.
The optimization problem that a farmer faces when choosing the
technology is
1
max δi (P f (hi (α)ai ) − W ai − ki ) (2)
δi ,ai
i=0

subject to
1
δi ∈ {0, 1}, 0≤ δi ≤ 1, and ai ≥ 0.
i=0

The search for an optimal solution is conducted in two stages.

First, the optimal continuous choice is analyzed for each of the
alternative technologies:
πi = max P f (hi (α)ai ) − W ai − ki , (3)
ai

with the first-order condition:

P f  hi (α) = W, (4)
indicating that at the optimal water input level, the value of the
marginal product of input is equal to the price of water. Rearranging
Equation (4), we obtain P f  = W/hi (α), which states that at the
optimal water use level, the value of marginal product of effective
input is equal to the price of effective input.
Once the optimal quantity of input to be used under each
technology, ai , is found, the discrete choice problem is solved, by
choosing
δ1 = 1 if π1 > π0 and π1 > 0,
δ0 = 1 if π0 > π1 and π0 > 0,
δ1 = δ0 = 0 if π1 < 0 and π0 < 0.
The second-order condition of optimization problem (3) is
P fi h2i < 0. Total differentiation of (4) yields
P f  h2i dai + f  hi dP + [P f  hi hi ai + P f  hi ]dα − dW = 0. (5)
28 Agricultural Economics and Policy

Let Ψ(e) = f  (e)e/f (e) be the output elasticity of effective water,

φ(e) = −f  (e)e/f  (e) be the elasticity of marginal product (EMP) of
e, and η(α) = hi (α)α/hi (α) be the elasticity of input use efficiency
with respect to land quality.
From Equation (5), we can obtain
da∗i −f  hi ai f  ai
= 2 = − = > 0, (6)
dP P f  hi P f  e Pφ
da∗i 1 f ai
=  2 =  
=− < 0, (7)
dW P f hi P f hi f hi Wφ

da∗i [P f  hi hi ai + P f  hi ] ai hi f  hi

=− = − −
dα P f  h2i hi f  h2i
ai ηi (α) 1
=− 1− 0 if φi (hi (α)ai )  1, (8)
α φi (hi (α)ai )

dyi∗ dai f  hi ai yi f  e yi Ψ
= f  hi = = = > 0, (9)
dP dP Pφ f · Pφ Pφ
dyi∗ yi Ψ
=− < 0, (10)
dW Wφ
dy ∗ dai dai ai ηi
= f  hi (α) + f  ai hi (α) = f  hi (α) +
dα dα dα α
f  hi ai η yi Ψη
= = > 0. (11)
φα φα
Equations (6)–(11) have intuitive interpretations. Equations (6)
and (7) state that optimal input use under technology i increases
in output price and decreases in input price. Equations (9) and
(10) indicate the opposite is true regarding the optimal output.
Although an enhancement in land quality or input use efficiency
(i.e., an increase in α) will increase the optimal output (see Equation
(11)), its impact on input use depends on the elasticity of marginal
product of effective input per acre (i.e., ei = hi (α)ai ) (see Equation
(8)). Particularly, if the elasticity of marginal product of effective
input is equal to 1, then an increase in land quality or input use
 Review of Production Economics 29

efficiency will not affect the optimal input use. However, if the
elasticity is less than 1, then an increase in input use efficiency
will drive up input use. In other words, the input use outcomes
under an improved, more efficient technology could be opposite to
what is originally expected (e.g., using drip irrigation to save water).
This rebound effect exists commonly in conservation and resource-
saving practices. For example, the adoption of fuel-efficient cars
might actually increase gasoline consumption, instead of decreasing
it, because driving would be less expensive and people would drive
more and thus, perhaps, use more fuel.

2.1. Comparison of input use and output under

the two technologies

Technology switch from i = 0 to i = 1 is equivalent to land

quality improvement from α to h1 (α) (recall that for simplicity
we set h0 (α) = α). View optimal input use as a function of land
quality. Therefore, by the first-order Taylor expansion, we have
a1 ≈ a0 + ∂a∂α (h1 (α) − α). Define Δa ≡ a1 − a0 . Then, based on
0

Equation (8) and the assumption h0 (α) = α, it is readily checked

that
∂a0
Δa ≡ a1 − a0 ≈ (h1 (α) − α)
∂α
a0 1
=− 1− (h1 (α) − α)  0 when φ  1. (12)
α φ
Similarly, we have
∂y0 y0 Ψ
Δy ≡ y1 − y0 ≈ (h1 (α) − α) = (h1 (α) − α) > 0. (13)
∂α αφ
Thus, we can see that the adoption of modern technology always
increases output but saves water only when φ > 1. Intuitively, when
marginal product is sensitive to effective input use (i.e., φ > 1),
then an increase in effective input use will decrease the marginal
product significantly. The first-order condition in Equation (4) will
then ensure less water would be used in the optimal solution.
30 Agricultural Economics and Policy

Price
VMP of modern technology

W2

W1

VMP of tradition technology

0 Effective water, e

Figure 3. Technology adoption and input use.

Alternatively, based on Equation (13), we can say that when the

yield effect (i.e., Δy) is small (i.e., large φ), then Δa < 0; and that
when the yield effect is large (i.e., small φ), then Δa > 0. Figure 3
provides a visual presentation of the impact of the new technology on
output and water use. The two curves in this figure stand for the value
of marginal product (VMP) of water under the two technologies,
respectively. When water price is W1 , switching from the traditional
to modern technology would save water use. When water price is W2 ,
however, switching to the modern technology would increase water
use. Since the areas under the VMP curves approximate output (or
yield), we can see that the yield effect of the switching under price
W2 is larger than that under price W1 .
What do we know about the elasticity of marginal product
(EMP) of e, φ? Assuming that f (·) has three regions of production,
its marginal and average productivity (MP and AP ) are depicted in
the lower panel of Figure 1. The economic region (MP < AP and
MP > 0) is between C and G, and MP is negative to the right of G.
The MP reaches its peak at B, where f  (eb ) = 0. Hence, at point B,
the EMP is φ(eB ) = −f  (eB )(eB )/f  (eB ) = 0; at point G, the EMP
is +∞. Thus, between points B and G, the EMP increases from 0 to
∞. Assuming continuity, there is a point D with φ(e)  1 if e  eD
and if e < eG .
 Review of Production Economics 31

2.2. Implication for Cobb–Douglas production

function

The Cobb–Douglas production function is quite popular because of

its ease of use. We argue here that it is not very realistic to apply
it to micro-level studies. Suppose y = Aeψ0 , with (1 − ψ0 ) < 1.
In this case, the two elasticities of interest are constant: the output
elasticity of effective water, Ψ(e) = ψ0 , the EMP, φ(e) = 1 − ψ0 < 1.
As stated previously, Cobb–Douglas functions do not allow a region
with negative marginal product. Furthermore, consider the case with
i = 0, where h0 (α) = α, and thus e = αa: The first-order condition is
ψ0 P Aeψ0 −1 = W/α (see Equation (4)). Hence, it is readily to check
that the first-order condition implies W a/P y = ψ0 . However, the
share of water cost per acre is rarely constant in reality; it is likely to
increase with W . Furthermore, water prices vary radically in places
like California and water use per acre does not respond as drastically
as predicted by Cobb–Douglas. Supposing φ0 = 0.5 and WA = 10WB ,
we are unlikely to observe that a∗A = a∗B /100. Therefore, production
functions like the quadratic may be more realistic for depicting micro-
level behavior.

2.3. Quality and technology choices

In this subsection, we show that there are segments of lower quality

lands that adopt the new technology. Note first that
dπi ∂[P f (hi (α)ai ) − W ai − ki ]
=
dα ∂α
∂a∗
= (P f  hi − W ) i + P f  hi a∗i
∂α
ηi W ai ηi
= P f  hi ai = > 0,
α α
where the last equality holds because of Equation (4). Thus, profits
increase in land quality. Furthermore,

dΔπ d(π1 − π0 ) a∗ η1 − a∗0

≡ =W 1 .
dα dα α
32 Agricultural Economics and Policy

For α = 1, we have a∗1 (1) = a∗0 (1), πo (1) = π1 (1) + k1 − k0 . Since

0 ≤ α ≤ h1 (α) ≤ 1, we know that when evaluated at α = 1, then
η1 (1) < 1. Therefore,

dΔπ
(1) = W a∗1 (η1 (1) − 1) < 0.
dα
The modern technology is less profitable for α = 1, but the
profitability gaps decline as α becomes smaller, and at α = αs1 their
profits per acre are equal. There may be many feasible patterns of
technology adoption as functions of quality, but the highest quality
land is much less likely to adopt. The pattern we analyze is depicted
in Figure 4. For land with quality better than αs , profit under
the conventional technology is higher than that under the modern
technology (π0 > π1 > 0), suggesting that producers will stick to the
conventional technology (i.e., δ0 = 1). On the other hand, for land
with quality between αm s
1 and α , profit under the modern technology
is higher than that under the conventional technology (i.e., π1 > π0 ),

profit

π0

π1

α1m α0m αs α Land quality, α

Figure 4. Switching land quality.

 Review of Production Economics 33

and thus adoption of modern technology will occur (i.e., δ1 = 1).

Note that αm m
i is land quality such that πi (αi ) = 0. In other words,
it is the minimum land quality for the profit to be non-negative under
technology i. The value αs is switching land quality. At this quality,
π1 (αS ) = π0 (αS ). That is,

P f [h1 (αS )a∗1 (αS )] − W a∗1 (αS ) − k1 = P f [αS a∗0 (αS )] − W a∗0 (αS ) − k0 .
(14)
The introduction of the new technology will lead to production
at the extensive margin (i.e., land with quality between αm m
1 and α0 )
operating under the modern technology. Therefore, the profits of land
with different quality can be summarized as

π1 (α) for αm
1 <α<α
s
r=
π0 (α) for αS1 ≤ α ≤ 1.

The switching quality αS and marginal qualities are functions of

prices. Total differentiation equation (14) yields

[P f1 h1 − W ]da∗1 + P f1 h1 a∗1 dαS + f1 dP − a∗1 dW − dk1

= [P f0 h1 − W ]da∗0 + P f0 a∗0 dαS + f0 dP − a∗0 dW − dk0 .

Rearranging terms, we have

W [a∗1 η1 − a∗0 ] S
dα + (y1 − y0 )dP − (a∗1 − a∗0 )dW − dk1 + dk0 = 0.
αS
Therefore, we have

dαS (y1 − y0 )αS dαS (a∗1 − a∗0 )αS

=− ∗ ∗ > 0 and = >0
dP W [a1 η1 − a0 ] dW W [a∗1 η1 − a∗0 ]
if φ > 1 and η1 < 1.

Similarly, total differentiation of P f [h1 (αm ∗ m ∗ m

1 )a1 (α1 )]−W a1 (α1 )−
k1 = 0 yields

[P f1 h1 − W ]da∗1 + P f1 h1 a∗1 dαm ∗

1 + y1 dP − a1 dW − dk1 = 0,
34 Agricultural Economics and Policy

which implies
⎧ m
⎪
⎪ dα1 = − y1
⎪
⎨ dP <0
P f1 h1 a∗1
⎪
⎪
⎪ dαm
1 a∗1
⎩ = > 0.
dW P f1 h1 a∗1
From the above expressions, we can conclude that a higher output
price will trigger the existence of the rent-efficient firms (i.e.,
dαm1 /dP < 0). When φ > 1, it will trigger adoption of modern
technologies by firms around αs (i.e., dαS /dP > 0). Higher input
prices will trigger technology switching toward the modern one at
α = αs (i.e., dαS /dW > 0), and entry of producers with marginal
quality which will adopt the modern technology (i.e., dαm
1 /dW > 0).

2.4. Aggregation

Suppose the distribution of land quality is described by a probability

∞
distribution function g(α), so the total land area is A = 0 g(α)dα.

The area of land with quality in the range of α − 2 , α + Δα
Δα
2 is
g(α)Δα. Aggregate supply is
 αs  1
S
Y = y1 g(α)dα + y0 g(α)dα.
αm
1 αs

The marginal change in supply with respect to price is given by

 αS  1
S ∂y1 ∂y0
YP = g(α)dα + g(α)dα
αm
1
∂P αS ∂P

  ∂αS m ∂α1
m
+ g(αS ) y1 (αS ) − y0 (αS ) − y1 (αm
1 )g(α1 ) >0
∂P ∂P
(15)
and
 αS  1 m
∂y1 ∂y0 m ∂α1
S
YW = g(α)dα + g(α)dα − y(αm
1 )g(α1 )
αm
1
∂W αS ∂W ∂W
∂αS  
+ y1 (αS ) − y0 (αS ) g(αS )  0. (16)
∂W
 Review of Production Economics 35

For Equation (16), the first three items are negative, but the fourth
item (i.e., the switching effect of higher W on supply) may be
S undetermined.
positive, leaving the sign of YW

2.5. Impact of pollution tax

The model can be readily expanded to examine the impact of

some environmental policies, such as pollution tax, on agricultural
production. Recall that e = hi (α)ai is the effective input consumed
by the crop. Then Z = [1 − hi (α)]ai is the unused input that causes
pollution (e.g., irrigation water runoff or chemical runoff). Suppose
pollution tax rate is V , the maximization problem for technology i
becomes
max P f (hi (α)ai ) − W ai − [1 − hi (α)]ai V − ki (17)
ai

with first-order condition

 
1−hi (α)
P f  hi = W + V (1 − hi (α)) ⇒ Pf = W
hi +V hi (α) .

By following the approach in this section, one can derive the impact
of the tax on technology adoption, output, and resource uses. We
leave this to readers as an exercise.

3. Economic Analysis of Investments and

Production Aggregation
Modeling production processes is essential for developing realistic
policy analysis frameworks. In all of the modeling, one needs to
investigate the implications of the approach for policy purposes.
In this section, we study the putty-clay framework, which is a
general framework to view production choices and their outcomes.
Before beginning, we emphasize a few points. First, some decision
variables are discrete. For instance, decisions made about the nature
of technology to be adopted (e.g., drip vs. sprinkler irrigation or
biological vs. chemical pest control approaches) can be dichotomous
choices, assuming binary values of 0 and 1. Other decision variables,
such as the amount of water that should be applied, are continuous
36 Agricultural Economics and Policy

and observed in a total amount. Second, producers operate under

varying sets of circumstances that may result in different outcomes,
creating heterogeneity in production. The causes for variability may
be differences in environmental conditions, land quality, human cap-
ital, and physical capital. Third, decisions are often time-dependent:
Short-run choices entail much less flexibility but are easier to predict
than long-run choices. Fourth, aggregation is a challenge in both
short-run and long-run analysis. To obtain meaningful predictions
of production choices and market outcomes under heterogeneity,
meaningful aggregation procedures are essential.

3.1. The Cambridge controversy

The notion of production function is applied for different levels of

aggregation. We can speak about the production function of an
individual process (e.g., a production function of wheat in one field),
of producers (e.g., a production function of wheat producers with
several fields), of an industry producing the same product, of a
sector that includes several industries, and, finally, of the aggregate
economy.
Aggregation may require a redefinition of input and output,
especially for conceptual analysis, as one has to reduce the number
of variables to a bare minimum to illustrate some concept without
having an extremely complicated analysis. Even empirical analysis
may require reducing the dimensionality and aggregation. One ques-
tion is as follows: “Under what condition would aggregation become
meaningless and the results not useful?” The biggest controversy
has been related to economy-wide production functions. One of the
most important areas of research after Word War II was attempts
to understand the process of economic growth. Kuznets established
a national accounting data on output, capital, and aggregate labor.
Many researchers, most notably Robert Solow, developed a neoclas-
sical growth theory to analyze these data. The growth literature that
Solow developed was very important during the 1960s and early
1970s, and it spawned another body of literature that attempted
to explain the process of innovation. The earliest seminal article in
the literature on innovation and growth was Kenneth Arrow’s (1962)
 Review of Production Economics 37

writing on learning-by-doing. There has been a resurrection in the

mid-1980s from the works of Robert Lucas Jr. and, in particular,
Paul Romer, who introduced a new concept: endogenous growth.
Romer’s work has become an important element of macroeconomics,
but we return to our discussion of production and, in particular, the
Cambridge controversy that led to the putty-clay model which is our
subject of interest.
The Cambridge controversy was a debate between economists
in Cambridge, Massachusetts, headed by Robert Solow and Paul
Samuelson, proponents of the neoclassical production function and
neoclassical growth theory, and economists in Cambridge, England,
headed by Joan Robinson, Piero Sraffa, and Luigi Pasinetti. Neoclas-
sical growth theory assumes the existence of an aggregate production
function where national output is produced by aggregate labor and
aggregate capital stock. It also assumes that there is an endogenous
process of technological change that increases input productivity over
time. Solow estimated an aggregate model of economic growth of the
form

Yt = ALαt Ktβ eηt ,

where Yt , Lt , and Kt stand for aggregate output, aggregate labor,

and aggregate capital, respectively.
His model has had a good statistical fit, and η, the time coeffi-
cient, was found to be quite substantial, indicating the importance
of technological change. The model assumes that the economy has
a stock of capital, Kt , which is augmented by investment It , but
may decline due to depreciation. This approach suggests measuring
capital by dollar units and assumes that capital goods are malleable.
The malleability of capital seemed unreasonable to economists
in Cambridge, England. The English economists argued that there
is much specialization of capital goods — a tractor cannot print
books. Therefore, the notion of aggregate capital is meaningless, and
policies based on assumption of smooth substitution between capital
and labor are misleading.
In this macroeconomic debate on the formulation of produc-
tion, both groups had valid points. The basic idea of assessing
38 Agricultural Economics and Policy

aggregate productivity in the economy — taken by the Cambridge,

Massachusetts, scholars — was viable. The effort they started led
to important results, and growth theory is a very important area
of research. However, the England group was correct that higher
capital expenditures do not necessarily mean more flexibility in
production because capital goods are limited in their uses. One of the
important elements in Romer’s new model is the explicit recognition
of the role of specialized capital goods and the limited extent of
malleability that capital goods have. The Cambridge controversy
can be summarized succinctly as the argument about the magnitude
of the elasticity of substitution between capital and labor. The
neoclassical economists assume that the elasticity of substitution is
quite high and the English economists assume that it is very low
and relationships are converging to a fixed proportion production
function. The compromise within the Cambridge controversy was
presented in “putty-clay” models.

3.2. Putty-clay models

Putty-clay models were introduced by Johansen (1959) and Salter

(1960). They separated between micro and macro as well as ex ante
and ex post production functions. A micro production function is the
production function of an individual producer. A macro production
function is a production function of an industry. One challenge is
to develop aggregation procedures to move from micro to macro
relationships. The ex ante choices are the putty stage, before the
shape of the final machine is determined. Ex post choices are at
the clay stage where the equipment is well formed and limits the
flexibility of choices. The ex ante production function is used for
long-run choices before investment takes place where the capital
level is flexible. An ex post production function reflects choices when
capital outlay is completed and capital is less flexible. Putty-clay
models assume that, at the microlevel, ex ante production functions
are neoclassical and have positive elasticities between capital and
other inputs, but ex post functions have fixed proportions and
zero elasticity of substitution. Thus, the putty-clay models separate
 Review of Production Economics 39

between micro ex ante production function, micro ex post production

function, aggregate ex ante production function, and aggregate ex
post production function.

3.2.1. The Salter model

Salter (1960) introduced a graphical presentation that is useful for

explaining the putty-clay model. His model is dynamic, and he
looks at the determination of prices and investment at a given
period. At the start of the period, the industry has a distribution
of existing production units that were built in previous years. Every
year entrepreneurs make ex ante decisions about new capital. Here
we make some general assumptions about the trend in capital costs
and variable costs over time. Every year entrepreneurs determine
the cost of a new capital good, its production technology, and its
production capacity. Salter assumes that technology has constant
returns to scale, and the cost of variable inputs such as labor increases
over time relative to capital.1 Technological change and the relative
price of labor result in new technology with lower variable costs but
may have slightly higher annualized fixed costs.
Suppose we are at the beginning of period t. The industry inherits
capital that was built in previous periods. Let Ct−j be the productive
capacity of facilities that were built j years before t. We can refer to
these machines as vintage t−j, and Ct−j is the productive capacity of
vintage t − j. Productive capacity is the maximum output that these
machines can produce if they are utilized. Let Vt−j be the variable
input cost per unit of output of machines of vintage t − j. Thus, at
the beginning of the period, the industry has output supply of an
existing plant that is a step function such as the one depicted in
Figure 5.
If output price P is smaller than Vt−j , then the capacity of vintage
t − j would not be utilized. If output price is greater than Vt−j, then

1
The reason that capital becomes cheaper overtime is that technological change
results in improved machinery. Labor cost may decline only when population
growth is very drastic, but in most developed countries, capital cost has declined
relative to labor costs.
40 Agricultural Economics and Policy

Price, Pt

Vt-5
Pt-1 = ACt-1 A
Vt-4 Pt = ACt B
Vt-3
Vt-2
Vt-1

Ct-1 Ct-2 Ct-3 Ct-4 Ct-5

0 Ct Capacity, C

Figure 5. Total capacity and price in the Salter model.

the output capacity of vintage t − j will be utilized. Part or all of the

capacity of vintage t−j will be utilized if the price is equal to Vt−j . Let
ACt be the average cost per period (total cost divided by output) of a
machine of vintage t. AC includes both variable cost and annualized
fixed cost.2 New productive capacity is introduced in period t as long
as price is greater than average cost. Thus, in equilibrium, output
price has to be equal to average cost of the current vintage.
Figure 5 helps us to understand the determination of the
equilibrium of time t, assuming that the industry is facing negative
sloped demand curve D. Suppose that the equilibrium at period t − 1
was at point A. During period t − 1, the industry produced qt−1 units
of output using the productive capacity of vintage t − 1, t − 2, t − 3,
and t − 4. The productive capacity of vintage t − 5 was idle because
the variable cost of this vintage, Vt−5 , was higher than the price.
The price at period t − 1 is equal to the average cost of vintage t − 1,

2
The assumption of constant returns to scale allows us to present average cost
regardless of size.
 Review of Production Economics 41

which is ACt−1 . Now, suppose that the average cost of vintage t is

ACt , which is smaller than ACt−1 , and that ACt is between Vt−3 and
Vt−4 . The new output price Pt will be equal to ACt . The capacity of
vintage Ct−4 will not be utilized. The new capacity of vintage t, Ct ,
introduced at time t will be equal to Ct−4 plus the increase in quantity
demanded because of lower prices. This can be represented by a
shift to the right of the supply step function. The new equilibrium
is at point B in the figure. Thus, Salter’s analysis suggests that
old capital equipment continues to operate as long as revenues can
cover its variable cost. However, at a certain time, this capital will
be out of production because its variable costs are too high. In his
model, capital is not being destroyed, it is only becoming obsolete.
At the same time, new capital is introduced reflecting the fact that
there is a technological change that reduces average cost below the
previous prices. When a firm makes ex ante investment decisions, it
has to recognize that the economic life of capital is limited and it
has to compute the cost of capital accordingly. Furthermore, when
computing the cost of capital, it has to recognize that variable
costs may increase over time and may reduce both the economic
life of capital and its future earning capacity. The following section
addresses the investment choice taking into account changes in prices
over time and final economic life.
The Salter model provides the framework for long-term decisions
when investments in new capital is incorporated explicitly into the
analysis. In the shorter run, choices are limited to existing equipment.
Therefore, if one wants to know the immediate effect of changes in
policies, she may ignore the possibility of developing new equipment
but consider the impact given existing vintages.

3.2.2. A quantitative analysis of investments

An investment involves an initial outlay of capital and results in a

stream of benefits. To analyze an investment, one needs to know the
stream of costs and benefits over time. Let x0 , x1 , . . . , and xT denote
the net benefit from a project at period 0, . . . , T . When xt < 0, it
represents a cost. For example, x0 may be the initial investment, xt ,
42 Agricultural Economics and Policy

t = 1, . . . , T , are the returns. In some cases, there may be several

periods of negative outlay. The interest or discount rate, denoted
by r, is a fee for the use of $1.00 for one period. The interest rate
can provide a base for comparing an income stream  at different time
1
periods. A dollar earned next year is worth 1+r today and a dollar
 5
1
earned five years from now is worth 1+r dollars today. The net

present value (NPV) of an investment is N P V = Tt=0 (1+r) xt
t . When

time is continuous, one can use N P V = 0 e−rt xt dt.

T

A project is worthwhile if NPV > 0. One way to compare projects

is to compare their internal rate of return (IRR). It is defined by z
which solves
T  T
xt
=0 or e−zt xt dt = 0.
(1 + z)t 0
t=0

One can think of IRR as a return rate of the investment. If IRR is

larger than the actual interest rate (i.e., return of the investment is
larger than the opportunity cost of the investment), then the project
will be profitable. Suppose a project requires an investment of K to
be paid in T equal payments with an interest rate of r. The annual
payment required to break even will be Y , where
 T
r
e−rt Y dt = K or Y = K.
0 1 − e−rT
When a project is of infinite length, then it is simply Y = rK.
When capital equipment worth K dollars is used in the produc-
tion process, the capital expenditure during period t includes interest
cost rK (or rK/(1–e−rT ) if the total number of period is T ) and
depreciation (the loss of value because of utilization). If depreciation
is assumed to be proportional to the value of the stock, it can be
denoted as δK, with δ as a fixed depreciation coefficient. Investment
choices, purchases, and the use of capital goods are the results of
choices over time.
Consider a long-run micro model under which, once established,
the capital will be used for a long period. This involves the choice of
 Review of Production Economics 43

capital at the beginning period and the choice of labor at each period.
The information available is output prices, Pt , and labor prices, Wt ,
for t = 0, . . . , ∞. The production function is f (K, L), where K is the
capital and L is the labor. The objective function is
 ∞
max e−rt [Pt f (K, Lt ) − Wt Lt ] dt − K,
K,Lt 0

subject to

K ≥ 0 and Lt ≥ 0.

The ex post choice problem given K is

max Pt f (K, Lt ) − Wt Lt , for t = 1, . . . , T,

Lt

and the ex post decision rules are to choose Lt such that

∂f
Pt − Wt = 0, for all t,
∂L
and

Pt f (K, Lt ) − Wt Lt ≥ 0.

For every K, there is a function Lt (K) denoting labor use over

time. Once the optimal ex post decision rules are determined, the ex
ante choice is to choose K to
 ∞
 −rt 
max e Pt f (K, Lt (K)) − Wt Lt (K) dt − K.
K 0

If Pt grows faster than Wt , then Lt will be positive forever. However,

when the labor price grows faster than the price of output, there may
be a period T1 where maximum short-term profit is zero. Beyond
this period, no output is produced. T1 is therefore the economic life
of capital. That is, capital stops operation because of obsolescence
in period T1 and beyond. When prices fluctuate a lot, the firm may
shut down, then operate, then shut down, and so on.
Consider a simpler case where the ex post technology is a fixed
proportion. Then, at time zero, both capital and labor are determined
44 Agricultural Economics and Policy

and the ratios are followed thereafter. The objective function is

 ∞
max e−rt {Pt f (K, L) − Wt L} dt − K
K,L 0

subject to

Pt f (K, L)–Wt L > 0.

Consider the case when output and labor prices grow exponentially,
Pt = P0 eγt and Wt = W0 eθt . Suppose that the rate of growth of
output price is smaller than that of labor price, γ < θ, and further,
γ < θ < r. Assume that P0 f (K, L) > W0 L. In this case, production
occurs between t = 0 and t = T1 , and the optimization problem is
P0 f (K, L)   W L
0

max 1 − e−(r−γ)T1 − 1 − e−(r−θ)T1 −K. (18)
T1 ,K,L r−γ r−θ
The economic life of capital is determined by solving

P0 eγT1 f (K, L) = W0 eθT1 L.

When θ = γ, the dynamic problem becomes the classical static

problem. Once T1 is solved for, the ex ante optimal levels of K
and L are derived solving Equation (18). In this case, the dynamic
optimization looks like a standard static optimization with labor
and capital costs adjusted to deal with different growth rates and
economic lives of capital.

3.3. Aggregation from micro to macro

3.3.1. Short-run derivation in putty-clay models

Following the putty-clay assumption, each production unit has a

fixed-proportion, ex post production function. Production coefficients
vary among production units to reflect their vintages and other
variables that affect the ex ante choices. Note that each production
unit has only one unit of production capacity. Suppose we have
one input. Let y denote the productivity of a unit expressed as
output/input ratio, P the output price, and W the input price.
Production units (machines or plants) with y > W/P will operate
 Review of Production Economics 45

f (y)

0 K y

Figure 6. The pareto distribution.

with full capacity, while production units with y < W/P will be idle.
Let the density of the distribution of output capacity as a function
of output/input ratio be denoted by f (y), which means that the
output capacity of firms with y  − Δy  Δy
2 < y < y + 2 for a very

small Δy is f (y )Δy, and that total output capacity of the industry
∞
is C = 0 f (y)dy. The output supply of industry is given by
 ∞
Y (P, W ) = f (y)dy, (19)
W/P

and the input demand is given by

 ∞
f (y)
X(P, W ) = dy. (20)
W/P y
Suppose y has a Pareto distribution (see Figure 6). This distri-
bution is found to be useful describing income distribution and farm
size distribution:
Ay −(α+1) , for y ≥ K
f (y) =
0, for y < K,
46 Agricultural Economics and Policy

with α > 0 and K is the output/input ratio of the least efficient

machine. What is A? We know that total productive capacity is C,
and
 ∞
−(α+1) A −α ∞ A
C= Ay dy = y = . (21)
K −α K αK α

From (21), we have A = αCK α , hence,

αC  y −(α+1)
f (y) = .
K K
Under the Pareto distribution, productive capacity declines as
efficiency (measured by y) increases. Note that
∂f (y) y
(α + 1) = −
∂y f (y)
is the rate of decline in productive capacity associated with increased
efficiency. Higher α assures more skewed capacity distribution. The
Pareto distribution, like the log normal and exponential distribu-
tions, is approximated with the empirical distribution of income but
not necessarily distribution of output capacity.
From (19), the supply when capacity is distributed as a Pareto
function of output/input ratio is
 ∞  
−(α+1) A −α ∞ A P α
Y (P, W ) = Ay dy = − y = , (22)
W/P α W/P α W

and from (20), we have

 ∞ ∞
Ay −(α+1) A
X(P, W ) = dy = − y −(α+1)
W/P y (α + 1) W/P
 α+1
A P
= . (23)
α+1 W
From (23), we obtain
1
P α+1 α+1
= X .
W A
 Review of Production Economics 47

Substituting it to (22) yields

α
A α+1 α+1
Y = X
α A
or
α
Y = BX α+1 , (24)
where
α
A α+1 α+1
B= .
α A
Equation (24) provides what seems to be an aggregate production
function, but this is not a “technological” relationship. It is a hybrid
incorporating both technological (micro production function) and
behavioral (profit maximization) elements and relating aggregate
output and inputs under profit maximization.
The above analysis suggests that the input elasticity of the
aggregate production function, α/(α + 1), reflects the productive
capacity distribution and not a “technical” coefficient. Instead,
α/(α + 1) is an outcome of investment decisions of the past that
have resulted in the distribution of productive capacity. The Pareto
distribution used here is not realistic for cases where we have growing
industries and where there are relatively more capacity in newer
vintages than older ones. It is reasonable in describing declining
industries. When the input is land, it describes situations where
higher quality lands (with higher y) are growing scarce and the mode
of the productivity capacity distribution is at lower land qualities.
The reservation we have regarding the Pareto distribution raises
doubt on the appropriateness of aggregate Cobb–Douglas production
functions.

3.3.2. A simple model of aggregation with some variability

The putty-clay framework does not allow short-term flexibility.

However, existing plants may change variable inputs to affect output.
Thus, let us assume that production function is of constant returns
to scale, but plants vary in quality. Let q be a variable measuring the
48 Agricultural Economics and Policy

quality of the capacity unit. Higher q represents, say, higher human

capital of plant managers. Let production per capacity unit be

y = f (q, x),

where x is the variable input per capacity unit. Assume

fx > 0, fxx < 0, fqx > 0, fqxx < 0.

For example,

f (q, x) = q γ xβ , 1 > γ > 0, 1 > β > 0.

Let us also assume when a capacity unit operates, a fixed-cost C is

required. For a micro unit, the two choices are (i) whether to operate
or not and (ii) if operate, then how much input per capacity unit to
use.
Let P indicate the output price and W the input price. The
decision problem is

max P f (q, x) − W x − C.
x

The first-order condition is

P fx − W = 0.

The capacity unit operates if

P f (q, x) − W x − C > 0.

For the case f (q, x) = q γ xβ , at the optimal solution,

P βq γ xβ−1 = W, (25)

or the micro-level input demand is

1
P βq γ 1−β
x(P, W, q) = ,
W
 Review of Production Economics 49

and the micro-level supply is

β
γ P βq γ 1−β
y(P, W, q) = q .
W
In this case, x increases with quality and output price and declines
with input price. Profit per capacity unit is
β 1
P βq γ 1−β P βq γ 1−β
π(P, W, q, C) = P q γ
−W − C.
W W
Setting this expression equal to zero and solving for q, the marginal
producing quality, q m , can be found; it is a function of P , W , and C:
 1 1−β
W γ Cβ γ
m
q (P, W, C) = .
βP (1 − β)W
Let the distribution of capacity be denoted by g(q), where g(q)Δq
denotes the number of capacity unit between q − Δq Δq
2 and q + 2 . Let
 ∞
g(q)d(q) = M,
0

where M is the number of capacity units, then the aggregate supply is

 ∞
Y (P, W, q) = y(P, W, q)g(q)dq.
q m (P,W )

To find the supply slope, we need to differentiate Y (P, W, q). The

Leibnitz rule is used for differentiating integrals. For a function
 b(x)
F (x) = g(x, z)dz,
a(x)

the Leibnitz rule states that Fx (x) = bx g(x, b(x))–ax g(x, a(x)) +
b(x)
a(x) gx (x, z)dz. Using this rule, the slope of the supply curve is
 ∞
∂q m (P, W )
YP (P, W ) = yP (P, W, q)g(q)dq − y(P, W, q m )g(q m ),
∂P
   
q m
 
(1) (2)
50 Agricultural Economics and Policy

from which we can see that marginal change in supply includes

change in intensive margin (i.e., term (1)) and change in extensive
margin (i.e., term (2)). A lower price will reduce marginally output
of every operating production unit, and it will cause closure of some
borderline capacity units.

3.3.3. Determination of aggregate relationship in cases with

varying input quality

In this section, we use the irrigation model that was presented in

Section 2 as a base for aggregation. Just a recap of that model
is presented here. Suppose we have a constant returns to scale
technology, in which agricultural product is produced using land and
E, effective input. Given constant returns to scale, output and input
use can be expressed on a per-acre basis. Let y = f (e) be the yield
per acre, e be the effective input per acre (input used in production
process), a be the total applied input per acre, and finally, α be the
quality of land, a continuous variable from 0 to 1.
The amount of effective input is a function of the amount applied,
of land quality, and of the kind of technology: e = h(i, α)a, where i
is a technology index, assuming value 2 for modern technology and
value 1 for traditional ones. When no technology is used, i = 0.
h(i, α) = e/a is input-use efficiency. When input is water, it is
irrigation efficiency. Let α be an index of land quality with respect
to input use (e.g., water-holding capacity). Assume 1 ≥ h(2, α) ≥
h(1, α) = α > 0. Let us also assume that production requires fixed
cost per acre, Ci with C2 > C1 . Assuming profit maximization for
each α, there is an optimal a (applied water) and an optimal i
(technology) determined by solving
Π(i) = maxa P f (h(i, α)a) − W a − Ci , i = 1, 2
max
i Π(0) = 0,
where P is the output price, W is the input price, and Ci is the cost of
adopting technology i. The optimization is solved in two steps: First,
we solve for optimal a given i = 1, 2, and the first-order condition is
P fe h(i, α) = W,
 Review of Production Economics 51

or P fe = W/h(i, α), where W/h(i, a) is price of effective water. Then

we compute optimal i according to
⎧
⎪
⎨ 2 if Π(2) > Π(1) and Π(2) > 0
i = 1 if Π(1) > Π(2) and Π(1) > 0
⎪
⎩
0 if Π(1), Π(2) < 0.

When land quality is the highest (i.e., α = 1), a(2) = a(1) and
Π(1) > Π(2) because C2 > C1 . It can be easily shown that Πα (i) > 0,
for i = 1, 2 (i.e., profit increases with land quality), and that, for
high-quality lands, the marginal effects of quality (Πα (1) > Πα (2))
hold that loss in profit as land quality declines is larger under
traditional technology. This is reasonable since modern technology
tends to augment land quality, especially if we assume that the ratio
h(2,α)−h(1,α)
h(1,α) becomes bigger as α declines. This assumption is true
with respect to sprinkler or drip irrigation because their impact
on water-use efficiency compared to furrow irrigation is stronger in
locations with low water-holding capacity. This assumption suggests
the following technology-use pattern:

i = 0 for α ≤ αm ,
i = 2 for αm < q < αs ,
i = 1 for α > αs ,

where αm is the marginal quality under technology 2 and αs is

the switching quality. Recall that αm is determined by solving
Π(2, P, αm , a∗ (P, αm )) = 0 or

max{P f [h(2, αm )a] − W a − C2 } = 0, (26)

a

and that αs is determined by solving Π(1, αs ) = Π(2, αs ), or

max P f [h(2, αs )a]−W a−C2 = max P f [h(1, αs )a]−W a−C1 . (27)

a a

To see how changes in price will affect marginal or switching

quality, we need to perform a comparative static analysis based on
52 Agricultural Economics and Policy

(26) and (27). Differentiation of (26) with respect to α and P yields

dαm ΠP (2) + Πa (2)aP

=− .
dP Πα (2) + Πa (2)aα

Since first-order conditions are met at αm , Πa (2) = 0, and

dαm ΠP (2) y(2)

=− =− < 0. (28)
dP Πα (2) P fe hα (2)a(2)

Note that the second equality in (28) holds because P fe h = W based

α (i)
on the first-order condition. Define Ψ(i) ≡ hh(i) α as the elasticity
of irrigation efficiency with respect to land quality. Therefore, (28)
becomes

dαm αm y(2)
=− .
dP W Ψ(2)a(2)

Similarly, differentiation of (27) with respect to α and P yields

dαS ΠP (2) − ΠP (1)

=− .
dP Πα (2) − Πα (1)

Note that, since h(α, 1) = α, Ψ(1) = 1, we get

dαS (y(2) − y(1))αs

=− . (29)
dP W [Ψ(2)a(2) − a(1)]

It can be argued that Ψ(2) < 1, and we know that y(2) > y(1).
s
In most cases, a(1) > a(2); therefore, dα
dP > 0. Namely, increase in
output price will increase adoption by increasing the switching land
quality. Now, let us assume that g(α) is the land quality density
function; g(α)Δα denotes the amount of land with qualities between
α − Δα Δα
2 and α + 2 for a small Δα. Thus, aggregate output supply
and input demand are given by
 Review of Production Economics 53

 αs  1
Y [P, W, c1 , c2 ] = y(2, α)g(α)dα + y(1, α)g(α)dα, (30)
αm αs
 αs  1
A[P, W, c1 , c2 ] = a(2, α)g(α)dα + a(1, α)g(α)dα. (31)
αm αs
The Leibnitz rule is useful in analyzing the properties of the
aggregate supply response. Using this rule,
dY [P, W, c1 , c2 ] ∂αS ∂αm
= [y(2, αs ) − y(1, αs )] g(αs ) − g(αm )y(2, αm )
dP ∂P   ∂P  
term 1: + term 2: +
 αs  1
∂y(2, α) ∂y(1, α)
+ g(α)dα + g(α)dα.
αm ∂P αs ∂P
    
term 3: + term 4: +

(32)
The first two terms reflect changes in the extensive margin, and the
last two reflect changes in the intensive margin. All elements are
positive which means that supply curve slopes upward. The extensive
margins are especially important when there is a relatively large
amount of land having switching or marginal qualities. One can use
similar procedures to evaluate YW , AP , and AW .

4. Conclusion
In this chapter, we have discussed some basic properties of production
functions, emphasizing on duality and its implications. By using
drip irrigation as an example, we have analyzed the economics
of input-quality-augmenting technology. A key insight is that the
impact of such technology on input use depends on the elasticity
of marginal product with respect to effective input: If the elasticity
is less than one, then the adoption of this technology will increase
the optimal input use. This rebound effect has wide applications in
understanding conservation and resource-saving technologies. This
chapter has also covered basic economic analysis of investments
and production aggregation. Putty-clay models were introduced as a
54 Agricultural Economics and Policy

general framework for understanding production choices, technology

adoption, and investment decisions. The production functions and
related analyses introduced in this chapter establish the basis for
the analyses in many other chapters within this book, particularly
the chapters regarding risk and uncertainty, welfare analysis, as
well as technology innovation and adoption. Before we move onto
those chapters, let us first discuss a field that is closely related to
production economics: consumer demand.

References
Antle, J.M. 1983. Sequential Decision Making in Production Models. American
Journal of Agricultural Economics 65(2): 282–290.
Arrow, K.J. 1962. The Economic Implications of Learning by Doing. The Review
of Economic Studies 29(3): 155–173.
Chambers, R.G. and D.R. Pope. 1994. A Virtually Ideal Production System;
Specifying and Estimating the VIPS Model. American Journal of Agricultural
Economics, 76(1): 105–113.
Johansen, L. 1959. Substitution versus Fixed Production Coefficients in the
Theory of Economic Growth: A Synthesis. Econometrica 27(2): 157–176.
Salter, W.E.G. 1960. Productivity and Technological Change (2nd edition).
Cambridge: Cambridge University Press.
 Chapter 3

Consumer Demand

Consumer demand for a product reflects consumers’ evaluation of

the product and is thus critical for assessing consumer’s welfare
associated with the product. Moreover, understanding consumer
demand is critical for firms to set their optimal prices and quantities,
and to estimate the degree of substitution between differentiated
products. For market analysts, knowing consumer demand of a
product is an important component to evaluate the extent of market
power. In this chapter, we will first briefly review the key results
of the standard neoclassical demand theory, and then introduce
several demand estimation models that have been widely used in
the literature, including the Almost Ideal Demand System (AIDS)
model developed by Deaton and Muellbauer (1980a,b) and the Exact
Affine Stone Index (EASI) model developed by Lewbel and Pendakur
(2009). We then discuss the critiques of the standard theory and
corresponding models, and introduce demand theory and models
that are based on product characteristics, which originate from the
seminal work by Lancaster (1966).

1. Key Results of Standard Consumer

Demand Theory
Standard consumer demand theory analyzes consumer demand in
the space of products. Assume there are n products to be considered

55
56 Agricultural Economics and Policy

by a consumer, whose disposable income is denoted by I. Taking

product prices, pi , ∀i ∈ {1, . . . , n}, as given, the consumer maximizes
her utility by optimally choosing the quantity of each product,
qi , ∀i ∈ {1, . . . , n}, under her budget constraint. The consumer’s
decision problem can be written as

v(p, I) ≡ maxn U (q), s.t. p · q ≤ I, (1)

q∈R+

where p ≡ (p1 , . . . , pn ) and q ≡ (q1 , . . . , qn ) are respectively price

and quantity vectors of the n products, and U (·) is the consumer’s
utility function such that U  > 0 > U  . Here v(p, I) is defined as an
indirect utility function given product prices and consumer income.
The Lagrangian equation of problem (1) can be written as

max (U (q) + λ(I − p · q)), (2)

q∈Rn
+

where λ is the Lagrangian multiplier. The first order conditions of

the maximization problem in Equation (2) can be written as

∂U  (q)
− λpi = 0 for i ∈ {1, . . . , n}, and (3)
∂qi
I − p · q = 0. (4)

Here the Lagrangian multiplier λ can be interpreted as the marginal

utility of disposable income. Equation (3) states that in the optimal
solution, the marginal utility of product i (i.e. ∂U  (q)/∂qi ) is equal
to the marginal utility of income spent on one additional unit of this
product (i.e. λpi ). If ∂U  (q)/∂qi > λpi then the consumer should buy
more of product i; and if ∂U  (q)/∂qi < λpi then the consumer should
buy less.
By solving the n + 1 equations in (3) and (4) we can obtain
the optimal consumption levels of each product, denoted by q ∗ (p, I),
and the optimal Lagrangian multiplier, denoted by λ∗ . Here q ∗ (p, I)
is termed Marshallian demand, and it can also be thought of as
uncompensated demand because a rise in product price is not
compensated by any rise in the consumer’s nominal disposable
income. Based on Equation (1), the indirect utility function can be
 Consumer Demand 57

written as
v(p, I) = U (q ∗ ). (5)
One can readily check that ∂v(p, I)/∂pi ≤ 0, ∂v(p, I)/∂I > 0
and that v(p, I) is homogeneous of degree zero in (p, I). More-
over, once we obtain q ∗ (p, I), we can calculate demand elas-
ticity with respect to own price and income for product j as
ηj ≡ (dq ∗j (p, I)/dpj )pj /q ∗j (p, I) and ηjI ≡ (dq ∗j (p, I)/dI)I/q ∗j (p, I),
respectively. When ηj > 0 product j is called Giffen good. In contrast,
when ηj < 0, project j is called ordinary good. Particularly, when
ηj ∈ (−1, 0), we say the demand is inelastic; when ηj < −1, the
demand is elastic. Moreover, product j is called luxury (respectively,
normal and inferior) good if ηjI > 1 (respectively, ηjI ∈ (0, 1) and
ηjI < 0). One takeaway is that a Giffen good must be an inferior
good, but an inferior good is not necessarily a Giffen good.
An alternate way to model demand is to minimize expenditure
for a given utility level target. The optimization problem can be
written as
e(p, u) ≡ minn p · q, s.t. U (q) ≥ u, (6)
q∈R+

where e(p, u) is termed the expenditure function for a given price

vector p and utility level u. The Lagrangian for optimization
problem (6) is
min [p · q + η(u − U (q))], (7)
q∈Rn
+

where η is the Lagrangian multiplier, which reflects the increase in the

minimized expenditure if the fixed utility level u were to be increased
by one unit. Assuming a binding constraint, the first-order conditions
of problem (7) are
∂U  (q)
pi = η for i ∈ {1, . . . , n}, and (8)
∂qi
U (q) = u. (9)
By solving these n + 1 equations, we can obtain the optimal con-
sumption levels, q h (p, u), for each product given the price and target
58 Agricultural Economics and Policy

utility level. The solution q h (p, u) here is called Hicksian demand,

denoted by superscript h. In contrast to Marshallian demand,
Hicksian demand can be termed compensated demand because
when the price of a product increases, the consumer will behave
like her income is compensated to guarantee a minimum utility
level u. Once we obtain q ∗ (u, p), the expenditure function is simply
e(p, u) = p · q ∗ (u, p). One can readily check that ∂e(p, u)/∂pi ≥ 0,
∂e(p, u)/∂u > 0, and e(αp, u) = αe(p, u) (i.e., homogenous of degree
one in p).
Several important properties regarding the Marshallian demand
and Hicksian demand are discussed in almost every advanced micro-
economics textbook. Because they are critical for understanding the
demand models to be discussed further in the chapter, here we
provide a brief recap of these properties.

1.1. Roy’s identity and Shephard’s lemma

Based on the envelope theorem, we can obtain two important results

for the Marshallian demand and Hicksian demand: Roy’s identity
and Shephard’s lemma1 :
∂v(p, I)/∂pi
Roy’s identity: qi∗ (p, I) = − , (10)
∂v(p, I)/∂I
∂e(p, u)
Shephard’s lemma: qih (p, u) = . (11)
∂pi
Roy’s identity shows that the Marshallian demand can be
obtained from the indirect utility function. It states that the
Marshallian demand is the ratio between the marginal impact of
product price on the indirect utility and the marginal impact of
income on the indirect utility. The intuition of Roy’s identity can
be better illustrated if we rewrite Equation (10) as

−∂v(p, I)/∂pi = q ∗i (p, I) · ∂v(p, I)/∂I. (12)

1
Readers can find the proofs in advanced microeconomics textbooks such as Mas-
Colell et al. (1995) and Jehle and Reny (2001).
 Consumer Demand 59

Equation (12) indicates that at optimum, utility increase caused

by one-unit reduction in price pi (i.e., the left-hand side of the
equation) is equal to the expenditure freed up by the price reduction
(i.e., qi∗ (p, I)) times the marginal impact of expenditure on the
maximized utility (i.e., ∂v(p, I)/∂I).
Shephard’s lemma shows that the Hicksian demand can be read-
ily obtained from the expenditure function: The Hicksian demand for
product i is simply the partial derivative of the expenditure function
with respect to price pi . Because e(p, u) = p · q h (u, p), we have
∂e(p, u)/∂pi = qih (p, u) + p · (∂q h (p, u)/∂pi ). Note that the second
term on the right-hand side of this equation (i.e., p · (∂q h (p, u)/∂pi ))
is the impact of price change of product i on the minimized
expenditure through affecting demand for each product. Shephard’s
lemma states that p · (∂q h (p, u)/∂pi ) = 0, indicating that overall this
indirect expenditure effect (i.e., influencing expenditure by affecting
demand quantity) of price change of a product is 0 due to changes
in the expenditure on each product caused by demand quantity
changes canceling out each other in the optimum. Combined with
Young’s theorem, the Shephard’s lemma implies ∂qih (p, u)/∂pj =
∂qjh (p, u)/∂pi (i.e., symmetric substitution terms).

1.2. Slutsky equation

Before we discuss Slutsky equation, it is helpful to consider a few

relations between Marshallian demand and Hicksian demand as well
as between indirect utility function and expenditure function. These
relations are
⎧
⎪
⎪ q ∗ (p, I) = qh (p, v(p, I))
⎪
⎪
⎨q h (p, u) = q ∗ (p, e(p, u))
(13)
⎪
⎪ e(p, v(p, I)) = I
⎪
⎪
⎩v(p, e(p, u)) = u.

The first two equations in (13) show the duality between Marshallian
and Hicksian demand. The first equation indicates that at price
vector p, the Marshallian demand under budget constraint I is the
same as the Hicksian demand under the utility constraint that is the
60 Agricultural Economics and Policy

highest possible the consumer can reach at price p and income I, i.e.,
v(p, I). The last two equations in (13) show that given price p, the
minimum expenditure to achieve the maximized utility under income
I is I, and that given price p, the maximum utility the consumer can
reach under the minimum expenditure to reach utility level u is u.
The Slutsky equation can be written as
∂qi∗ (p, I) ∂qih (p, v(p, I)) ∂q ∗ (p, I)
= − qj∗ (p, I) i , ∀i, j ∈ {1, . . . , n}.
∂pj ∂pj ∂I
(14)
To better interpret the Slutsky equation, note that at optimum
qj∗ (p, I) = qjh (p, v(p, I)) = ∂e(p, v(p, I))/∂pj based on Equa-
tions (11) and (13). As a result, Equation (14) can be re-written as
∂qi∗ (p, I) ∂q h (p, v(p, I)) ∂e(p, v(p, I)) ∂qi∗ (p, I)
= i − ,
∂pj ∂pj ∂pj ∂I
∀i, j ∈ {1, . . . , n}, (15)
where the first term on the right-hand side of Equation (15) is the
substitution effect of price change and the second term is the income
effect caused by the price change. The price effect on Marshallian
demand is the sum of the substitution effect and the income effect.

2. Popular Demand Models Based on the

Standard Consumer Demand Theory
Based on the neoclassical demand theory illustrated above,
economists have developed some popular demand models for empir-
ical analyses. In this section we discuss three of them: (1) constant
elasticity of substitution demand model, (2) the Almost Ideal
Demand System (AIDS) model, and (3) the Exact Affine Stone Index
(EASI) model.2

2
Some other popular demand models in the product space include Linear
Expenditure System models, Rotterdam demand model, and translog demand
models. We refer readers to Stone (1954), Christensen et al. (1975), and Clements
and Gao (2015) for details about these models.
 Consumer Demand 61

2.1. Constant elasticity of substitution (CES)

demand model

The CES functions are popular in both consumer theory and produc-
tion theory due to their convenient properties such as monotonicity
and concavity. Moreover, Cobb–Douglas functions and Leontief
functions are special cases of the CES function family. A CES utility
function on n goods can take the following form:
 n 1/ρ
 ρ
U (q1 , . . . , qn ) = qi , (16)
i=1

where parameter ρ < 1 and ρ = 0. Under price vector p and income

I, a utility maximization problem like in optimization problem (1)
with the CES utility function can be solved as
1 ρ
q ∗i (p, I) = I · (pi ) ρ−1 · P 1−ρ , (17)
ρ
where P ≡ ( ni=1 piρ−1 )(ρ−1)/ρ can be viewed as a price index. Taking
natural logarithm of Equation (17), we obtain
1 ρ
ln q ∗i (p, I) = ln I + ln pi + ln P. (18)
ρ−1 1−ρ
If we have data for prices and quantity demanded of a group of
products over a period of time, then we can estimate a regression
model based on Equation (18),
ln q ∗it = β0 + β1 ln pit + β2 ln Pt + t , (19)
where t stands for time period. One might have noticed that the
price index P contains the unknown parameter ρ, and, therefore, the
values of Pt are unknown for the regression. In practice, researchers
can get around this issue by using time period dummies to replace
ln Pt .
The estimate of β1 can be interpreted as the demand elasticity
of product i with respect to its own price. The demand model offers
simplicity, but it imposes strict restrictions on price elasticity and
the elasticity of substitution between products. Specifically, within
this demand model, any product’s demand elasticity with respect to
62 Agricultural Economics and Policy

its own price is the same, and the elasticity of substitution between
any pair of products is the same as well. This is certainly not true
in reality. Therefore, some demand models that can relax these
restrictions are in order. We will discuss two such models for the
remainder of this section.

2.2. The AIDS model

Unlike the aforementioned CES demand model that is built on an

arbitrary utility function, the AIDS model starts with expenditure
function. In their seminal paper that introduces the AIDS model,
Deaton and Muellbauer (1980a) assume the expenditure function
takes the following form:

ln e(p, u) = (1 − u) ln a(p) + u ln b(p), (20)

where u ∈ [0, 1], with u = 0 indicating subsistence and u = 1 bliss.

Furthermore,
1 ∗
ln a(p) = a0 + Σk αk ln pk + Σk Σj γkj ln pk ln pj , (21)
2
ln b(p) = ln a(p) + β0 pβk k , (22)
k

where α, β, and γ are parameters.

Plugging Equations (21) and (22) into Equation (20), we obtain
the AIDS expenditure function
1 ∗
ln e(p, u) = α0 + Σk αk ln pk + Σk Σj γkj ln pk ln pj + uβ0 pβk k .
2
k
(23)

Based on Shephard’s lemma, we know that

∂ ln e(p, u) pi q i
= ≡ wi , (24)
∂ ln pi e(p, u)
where wi is expenditure share of good i among the group of studied
goods. Therefore, taking the derivative of Equation (23) with respect
 Consumer Demand 63

to ln pi , we obtain

wi = αi + Σj γij ln pj + βi uβ0 pβk k , (25)

k

where γij ≡ 1/2 · (γij∗ + γ ∗ ).

ji
As we discussed earlier, for a utility maximization consumer, the
indirect utility can be expressed as a function of prices and disposable
income, which is equal to the minimum expenditure needed to
obtain the maximized utility conditional on prices and income (see
Equation (13)). Therefore, based on Equation (23), we have

1 ∗
u = ln I − α0 + Σk αk ln pk + Σk Σj γkj ln pk ln pj β0 pβk k ,
2
k
(26)

where I is disposable income (or, equivalently, total expenditure) of

the consumer. Plugging Equation (26) into Equation (25), we have

wi = αi + Σj γij ln pj + βi ln (I/P ), (27)

where P is a price index such that

1
ln P = a0 + Σk αk ln pk + Σk Σj γkj ln pk ln pj . (28)3
2

The restrictions of the parameters in Equation (27) include

Σi αi = 1, Σi βi = 0, Σi γij = 0, Σj γij = 0, and γij = γji . (29)

With the restrictions in Equation (29), one can check that

Σi wi = 1, wi is homogeneous of degree zero in prices and total
expenditure, and that e(p, u) is homogeneous of degree one in p.

3 ∗
We obtain Equation (28) by using the fact that Σk Σj γkj = Σk Σj γkj .
64 Agricultural Economics and Policy

Based on wi = pi qi∗ (p, I)/I and Equation (27), the Marshallian

demand elasticity with respect to price can be written as
 
∂ ln qi∗ (p, I) ∂ ln wi ∂ ln P
ηij = = −δij + = −δij + γij − βi wi ,
∂ ln pj ∂ ln pj ∂ ln pj
(30)

where δij is the Kronecker delta equal to 1 if i = j and 0 otherwise.

If we have data for product prices and expenditure on each
product, then we can estimate Equation (27) and, thereby, price
elasticities in Equation (30). By comparing Equation (30) with
Equation (18), we can see that, unlike the CES demand models,
the own price elasticities across products are not necessarily the
same, indicating that the AIDS models are more flexible than the
CES models. For practical considerations of estimating the AIDS
demand system, we refer readers to Deaton and Muellbauer (1980a)
and Green and Alston (1990).

2.3. The EASI model

Although simple, relatively flexible, and widely applied, the AIDS

models nevertheless have some significant drawbacks. First, Engel
curves (expenditure on a good as a function of total disposable
income) derived from the AIDS models can only take quite restrictive
forms (e.g., linear-log form, see Deaton and Muellbauer (1980a)), and
Engel curves across all goods within the AIDS model assume the
same form. Second, the AIDS models cannot accommodate observed
and unobserved preference heterogeneity across consumers, while, in
reality, one would expect that the Engel curves for different products
may differ from one another (e.g., normal goods vs. inferior goods).
Moreover, studies have documented that preference heterogeneity is
prevalent in consumer choices and that it is important to consider the
heterogeneity in demand analysis (e.g., Beckert and Blundell 2008).
The EASI model proposed by Lewbel and Pendakur (2009) aims
to address the foregoing two issues. Similar to the AIDS models,
the EASI model also starts from an expenditure function, although
the expenditure function under the EASI model is quite different
 Consumer Demand 65

from that under the AIDS model. The tricks employed by the two
demand models to derive the Marshallian demand share equations
are also quite similar. Following Pendakur (2009), we first introduce
the simplest version of the expenditure function used by Lewbel
and Pendakur (2009), and then add on features to accommodate
preference heterogeneity. Continuing with the notation above, the
simplest expenditure function in the EASI models is

ln e(p, u) = u + Σi mi (u) ln pi , (31)

where mi (u) is a function of u such that Σi mi (u) = 1. By taking

derivative of this equation with respect to ln pi we obtain

wi (p, u) = mi (u). (32)

We make two observations from Equation (32) (the Hicksian expen-

diture share function). First, each good can have a completely
different expenditure function form than that of another good.4 This
feature will allow different Engel curves across all goods. Second, the
expenditure share function is unrelated to prices, which is a drawback
that needs to be overcome.
Similar to what we have done to derive the AIDS models, we can
obtain an expression of utility, u, based on Equations (31) and (32)
with the help of Equation (13),

u = ln I − Σi wi ln pi . (33)

Plugging Equation (33) into Equation (32), we obtain

wi = mi (ln I − Σi wi ln pi ), (34)

which is termed implicit Marshallian demand because the Marshal-

lian expenditure share, wi , appears on both sides of the equation.
However, all the variables in Equation (34) are now observables.

4
Note that in the AIDS models, the Hicksian expenditure share functions differ
across goods only by parameters αi and βi (see Equation (27)).
66 Agricultural Economics and Policy

Define implicit utility as

y ≡ ln I − Σi wi ln pi , (35)

which can be interpreted as disposable income I deflated by price

index Σi wi ln pi . For a given functional form of mi (·), (say, polyno-
mials), Equation (34) can be readily estimated by regressing wi on y
instrumented by ln I and ln pj , j ∈ {1, . . . , n}.
To add on unobserved preference heterogeneity, we simply
assume that the expenditure function takes the form

ln e(p, u, ) = u + Σi mi (u) ln pi + Σi i ln pi , (36)

where  = [1 , . . . , n ] represents unobserved preference characteris-

tics of consumers associated with good i ∈ {1, . . . , n}, and E() = 0.
Again, taking derivative of Equation (36) with respect to ln pi , we
obtain the Hicksian expenditure share function

wi (p, u, ) = mi (u) + i . (37)

Furthermore, by plugging mi (u) = wi (p, u, ) − i obtained from

Equation (37) into Equation (36), and rearranging, we again have
y = u = ln I − Σi wi ln pi as in Equation (35). Plugging this result
into Equation (37), we obtain the implicit Marshallian budget share
equation

wi = mi (y) + i . (38)

Again, these budget share equations can be estimated separately

by using instrumental variables as discussed above. Lewbel and
Pendakur (2009) term Equation (38) the Exact Stone Index (ESI)
demand because the nominal disposable income I is deflated exactly

by the Stone index defined by i pw i
i . One should note that even
though the demand models in Equation (38) can accommodate the
unobserved preference heterogeneity and allow for arbitrary form of
Engel curves for each good, prices do not directly enter the Hicksian
or Marshallian expenditure functions in Equations (37) and (38). To
address this issue, Lewbel and Pendakur (2009) employ the following
 Consumer Demand 67

expenditure function form

1
ln e(p, u, z, ) = u + Σi mi (u, z) ln pi + Σi Σj [αij (z) ln pi ln pj ]
2
+ Σi i ln pi , (39)

where vector z = [z1 , . . . , zT ] stands for observable demographics of

consumers. Taking the derivative of Equation (39) with respect to
ln pi , we obtain the Hicksian expenditure share function as

wi (p, u, z, ) = mi (u, z) + Σj αij (z) ln pj + i , (40)

where αij = αji ∀i, j ∈ {1, . . . , n}. Multiplying ln pi to both sides of

Equation (40) and sum across all goods, we obtain

Σi (wi ln pi ) = Σi (mi (u, z) ln pi ) + Σi Σj [αij (z) ln pi ln pj ]

+ Σi (i ln pi ). (41)

By solving out Σi (mi (u, z) ln pi ) from Equation (41) and plug it in

Equation (39), we obtain
1
y = u = ln I − Σi (wi ln pi ) + Σi Σj [αij (z) ln pi ln pj ]. (42)
2
Note that in Equation (42) the nominal disposable income I
is deflated by an affine transformation of Stone price index,
Σi (wi ln pi )− 12 Σi Σj [αij (z) ln pi ln pj ], given the name of EASI (Exact
Affine Stone Index) demand models. Plugging Equation (42) into
Equation (40), we then obtain the implicit Marshallian budget share

wi = mi (y, z) + Σj (αij (z) ln pj ) + i , (43)

where αij = αji ∀i, j ∈ {1, . . . , n}. Note that Equation (43)
remains the features that the Engel curves can have any shapes
for good i. Moreover, it now accommodates observable and non-
observable preference heterogeneity, as well as a linear price effect
on expenditure share. Assuming a functional form for mi (y, z), we
can estimate the EASI demand model in (43) by using instrumental
variables as discussed above.
68 Agricultural Economics and Policy

3. Critiques of Standard Demand Theory

and Models
Perhaps the strongest critiques of the standard demand theory and
models were from Kelvin Lancaster (1924–1999) and Gary Becker
(1930–2014).5 Lancaster (1966) criticized that although the standard
demand theory is elegant, it is also too general and vacuous. He
argued that the pursuit of generality resulted in a minimal set of
assumptions (i.e., utility maximization, income constraint, and quasi-
concave utility function), but a minimal set of results (i.e., Hicksian
demand curves are negatively sloped). Because of the pursuit of
generality, all goods in the standard demand theory are treated alike.
Anything that people want more of are goods. No attention is given
to the intrinsic properties of goods that separate bread from bicycles.
The standard demand theory assumes that preferences are given and
does not try to explain variation in preferences. Although the theory
can explain observed patterns in a static primitive economy, it cannot
deal with modern interesting issues such as prediction of demand
for new products and understanding the effects of product quality
differences.
Is Lancaster right? Almost, but his criticisms are somewhat
harsh. The notions of substitutes and complements and the notions
of separability were all attempts to enrich the power of standard
demand theory.6 These attempts, however, are weak. They do not
consider intrinsic properties, and therefore, they are not useful for
explaining product innovation and quality changes.
On the other hand, Michael and Becker (1973) argued that
the standard demand theories use income and prices as the main
explanatory variables for consumption patterns, and whatever is not

5
Becker was a student of Theodore W. Schultz (1902–1998). Schultz coined the
term “human capital” and started the literature on economics of education and
learning. He emphasizes the study of behavior outside the market and in situations
of disequilibrium. Another student of Schultz, Zvi Griliches (1930–1999), started
formal economic research on technology adoption. We will cover some of Griliches’
work in the chapter on technology adoption.
6
See Nevo (2011) for a brief discussion about separability.
 Consumer Demand 69

explained by income and prices is attributed to tastes. However,

income and price variations are found to explain only a small part of
the variation in family consumption patterns. Although taste has
an important role, there is no consideration of the formation of
taste in the standard demand theory. Moreover, a standard demand
theory explains consumption patterns of goods purchased in the
market, and that imposes an important limitation on phenomenon
investigation. In other words, applications of the standard demand
theory are restricted to market segments or monetary segments in the
economy, therefore, non-market activities are ignored. This limits
the usefulness of the theory to developed nations and reduces its
effectiveness in developing nations, in which markets are incomplete
or absent.
According to Becker, economics is a science that explains
behavioral choices with limited resources among competing needs.
These choices include both market and non-market decisions. Some
examples of non-market decisions include choices on family size,
lifestyle, occupation, political party, marriage, crime, migration,
suicides, and so on. Becker expanded the range of issues analyzed by
economists to include some of these non-market decisions. His works
are insightful but show the limitation of the notion of “economic
man.” A better understanding of choices and behavior may need
to incorporate more from psychology and sociology into economic
models.
Overall, the critiques from Lancaster and Becker revealed that
the standard demand theory ignored intrinsic properties of goods,
left too much explanation to “taste”, and cannot deal with important
phenomenon such as product innovation, product quality differences,
and non-market activities. Lancaster’s critiques originated an impor-
tant alternative demand theory that focuses on the characteristics
space, while Becker’s critiques were followed by family production
models, another alternative of the standard demand theory. Both
alternatives incorporate production activities into demand: a con-
sumption unit contains a production process that transforms goods
and other inputs into sources of utility. We now discuss these two
alternatives in the following two sections.
70 Agricultural Economics and Policy

4. Consumer Demand Based on Product

Characteristics
4.1. Lancaster’s approach

Unlike the standard demand theory discussed in Sections 1 and 2,

which assumes that consumers derive utility directly from goods,
Lancaster (1966) assumes that consumers derive utility directly
from the characteristics of goods, not from the goods themselves.
In Lancaster’s approach, consumers purchase goods from markets
as “inputs” for “activities” that convert goods to characteristics.
Assume that one unit of activity k will consume ajk units of good j.
Let xj denote the total quantity of good j consumed, and yk denote
the level of activity k. We then have

xj = Σk ajk yk . (44)

We further assume that one unit of activity k will produce bik units
of characteristics i. Therefore, the total amount of characteristics i
produced by all activities is

zi = Σk bik yk . (45)

In matrix format, Equations (44) and (45) can be written as x = Ay

and z = By, respectively. Suppose the dimensions of x, y, and z are
n × 1, m × 1, and r × 1, respectively. Then the size of matrices A
and B are n × m and r × m, respectively. B is called consumption
technology. In Lancaster’s model, consumers derive utility directly
from characteristics, which is reflected by utility function U (z). Based
on the foregoing model setup, the consumer’s optimization problem
can be written as

Max U (z), s.t. z = By, x = Ay, p x ≤ I, and x, y, z ≥ 0, (46)

where p is the price vector and I is disposable income.

Note that when m = n = r, we will have a one-to-one relationship
between goods and characteristics, i.e., z = BA−1 x and U (z) =
U (BA−1 x) = V (x). In this case, the consumer’s demand problem
 Consumer Demand 71

becomes the standard demand model, which shows that the standard
model can be a special case of Lancaster’s model.
Generally, however, we do not have m = n = r, so we have to
take a different path connecting goods to characteristics. Fortunately,
many such paths exist. For simplicity, let us assume that goods and
activities are identical. Thus, optimization problem in (46) can be
simplified as

max U (z), s.t. z = Bx, px ≤ I, and x, z ≥ 0. (47)

If r > n (i.e., the number of characteristics is greater than the

number of goods), then we have a primitive society. In this case,
not all consumption combinations are reachable. There is a relatively
small number of goods that cannot meet many needs, independent
of income constraints.
For advanced economies, we will have n > r (i.e., there are more
goods than characteristics). Consumers can attain any characteristic
combination in many ways. In this case, the consumer has to
make two types of choices. First, the efficiency choices to determine
the least costly way to produce a characteristic combination for a
given price vector p and characteristic vector z ∗ . This is a linear
programming decision problem that can be written as

min px, s.t. z ∗ = Bx and x ≥ 0. (48)

Repetition of this choice can yield a characteristic frontier such

that the cost of obtaining any z ∗ on the frontier is exactly I.
Figure 1 shows an example of such characteristic frontier when the
consumption technology involves two characteristics and four goods
(see curve E1 E2 E3 E4 in the figure). Note that problem (48) does not
involve any utility function, which indicates that the characteristic
frontiers are the same for all consumers with the same budget
constraint. Given the consumption technology B, the characteristic
frontiers will expand or shrink parallelly when budget increases or
decreases. It is also readily checked that the characteristic frontier is
concave.
After making the efficiency choices, the second type of
choices that the consumer will make is the private consumption
72 Agricultural Economics and Policy

Figure 1. An example of Lancaster (1966) model.

choice—a selection of consumption points on the characteristic

frontier. This choice depends on the consumer’s utility function and
is therefore subjective.
Let us continue with the 2-characteristic and 4-good example.
Suppose the relationship between the two characteristics and the
four goods are described as follows:

z1 = b11 x1 + b12 x2 + b13 x3 + b14 x4
(49)
z2 = b21 x1 + b22 x2 + b23 x3 + b24 x4 .
If the consumer spends all her income on good 1, then the amount
of good she can purchase is E1 ≡ I/p1 . In this case, the quantity of
characteristics 1 and 2 obtained will be such that z1 /z2 = b11 /b21 .
This relationship is depicted by ray OE1 in Figure 1. Similarly,
we have ray OE2 to OE4 in the figure. For example, point E2
indicates the maximum amount of good 2 that can be obtained by
the consumer, and ray OE2 has a slope of b12 /b22 , indicating the ratio
between z1 and z2 when only good 2 is available to the consumer.
Note that the shape of the curve E1 E2 E3 E4 indicates that the prices
of the four goods and the consumption technology B happen to make
 Consumer Demand 73

the combinations of goods 1 and 2, 2 and 3, and 3 and 4 efficient.

When prices or consumption technology changes, then the shape
of the curve E1 E2 E3 E4 may change as well. Moreover, a point on
segment E1 E2 indicates a linear combination of E1 and E2 (i.e.,
λE1 + (1 − λ)E2 where λ ∈ [0, 1]). One can readily check that all
points on the curve E1 E2 E3 E4 have the same cost.
Under the setup associated with Figure 1 we can study the
impact of a change in price. Suppose good 2 becomes cheaper. Then
the characteristic frontier moves from E1 E2 E3 E4 to E1 E2 E4 , and
the optimal choice changes from Q∗ to Q∗ in the figure, which is
determined by the characteristic frontiers and the indifference curves
(i.e., the dotted curves in the figure). Before the reduction in price
of good 2, good 3 was consumed; now good 3 is not in the optimal
solution and it is not used. As a result, the efficient substitution
effect eliminates the consumption of good 3 in our case and causes a
switch. Thus, negatively sloped demand is not only a result of concave
utility function but also of a certain consumption and production
technology set. It is not only justified by one’s taste but by all the
people’s consumption technology.
To summarize, one has to distinguish between efficiency substitu-
tion and private substitution. The former is an objective substitution
where a bundle of characteristics may be produced better by a
different set of goods after price changes. The latter is a subjective
substitution where a change in prices changes the characteristic
frontier and then there is a change in consumption within a given
technology reflecting utility and preference. These substitutions are
complementing one another, and they strengthen the notion of
negatively sloped demand.
We now explore more about a specific version of the Lancaster
model where m = n = r = N . In this case, both matrices A and B
are N × N matrix. Specifically, we have
⎛ ⎞⎛ ⎞−1 ⎛ ⎞
b11 ··· b1N a11 ··· a1N x1
⎜ . .. ⎟ ⎜ . .. ⎟ ⎜ . ⎟
z = BA x = ⎜
−1
⎝ .
. ..
. ⎟⎜
. ⎠⎝ ..
..
. ⎟ ⎜ ⎟
. ⎠ ⎝ .. ⎠, or
bN 1 ··· bN N aN 1 ··· aN N xN
zj = Σk Σl bjl a−1
lk xk , (50)
74 Agricultural Economics and Policy

where a−1
lk is the element in the lth row and kth column of matrix
−1
A , and j, k, l ∈ {1, . . . , N }. The optimal demand choice problem is

max U (BA−1 x), s.t. p x = I. (51)

x

It can be presented in Lagrangian form

L = max U (BA−1 x) + λ[I − p x], (52)

x

where λ is the shadow price of the income constraint. The first-order

conditions are
∂L ∂U ∂zk
= Σk − λpl = 0, l ∈ {1, . . . , N }, and (53)
∂xl ∂zk ∂xl
∂L
= I − p x = 0. (54)
∂λ
Based on Equation (50), we have (∂zk )/(∂xl ) = Σj bkj a−1
jl . Therefore,
Equation (53) can be re-written as
∂L ∂U
= Σk Σj bkj a−1
jl − λpl = 0, l ∈ {1, . . . , N }. (55)
∂xl ∂zk
Rearranging terms in Equation (55), we have
1 ∂U
Σk ×Σj bkj a−1
jl = pl . (56)
λ ∂zk
Terms in Equation (56) have an intuitive interpretation. Clearly,
the right-hand side of the equation is the cost of one more unit of
good l, price pl . The term on the left-hand side of the equation is
the marginal value to the consumer, represented in monetary form
(because it is divided by λ), caused by one unit increase in good l.
More specifically, the term (1/λ)∂U/(∂zk ) is the marginal value of
characteristic k and the term Σj bkj a−1
jl is the marginal productivity
of good l summed in producing characteristic k in all activities.
The model has originated a new branch of demand system models
that are different from the AIDS and EASI models discussed above.
Here, we introduce demand models in characteristic space, which are
a derivation of econometric models based on the Lancaster model.
 Consumer Demand 75

4.2. Demand models in the characteristic space

For simplicity, let us continue with the assumption that there is

a one-to-one relationship between goods and activities (i.e., A is
an identity matrix). Also, maintaining the notation above, we have
n goods and each good j ∈ {1, . . . , n} has r characteristics, labeled as
zj = (zj1 , . . . , zjr ) . One example is a local housing market where the
number of goods is the number of houses available in the local market
and the characteristics include the number of bedrooms, bathrooms,
stories, and so on. Let i be the index of a consumer. We assume
that there is a continuum of consumers so that the total number
of consumers sums up to 1. We further assume that consumer i’s
utility function on characteristics is Ui (z, , ei ), where  stands for
unobservable characteristics of goods and ei for consumer attributes.
Lastly, we assume that a consumer only purchases at most one good
(e.g., one house). Consumer i’s decision problem is then

max ui (Ii − Σj (dj pj ), ei ) + Σj (dj U i (zj , j , ei )),

di1 ,...,din

s.t. Ii − Σj (dj pj ) ≥ 0, dij ∈ {0, 1}, and Σj dj ≤ 1, (57)

where ui (·) indicates consumer i’s utility from income or all other
goods; zj is defined by equation (50); dj = 0 indicates that good j is
not purchased and dj = 1 indicates the opposite. Denote the optimal
solutions of problem (57) as {d∗i1 (p, z, Ii , ei ), . . . , d∗in (p, z, Ii , ei )}. If
d∗i1 = d∗i2 = · · · = d∗in = 0, then the consumer will not buy any of
the n goods. Aggregating individual demand for good j across all
consumers, we have the market demand for the good as

Qj (p, z) = d∗ij (p, z, Ii , ei )f (Ii , ei )dIi dei , (58)

where f (Ii , ei ) is the probability density function of individual income

and attributes. Here Qj (p, z) can also be viewed as market share
of good j because the total amount of consumers is normalized
to 1. Based on individual or market level of data about quantity
demanded, prices, expenditure, characteristics, as well as assumption
on the probability density function, we can estimate the marginal
76 Agricultural Economics and Policy

impact of characteristics on consumer demand.7 This model can be

used to forecast the demand for a new product that has different lev-
els of characteristics, but it fails if the new product’s characteristics
do not appear in any existing good.

5. Family Production Model

Michael and Becker (1973) emphasized on the time allocated to
activities processing goods at the household level. They distinguished
between goods and commodities (analogous to characteristics in
Lancaster’s model). The household spends time to convert goods into
commodities, and household utility is derived from commodities and
leisure. According to Michael and Becker (1973), “The consumer’s
demand for these market goods is a derived demand analogous to
the derived demand by a firm for any factor of production.”
Following the notation in the Lancaster model discussed above,
we again assume that the consumer’s utility function is U (z), where
z = (z1 , . . . , zr ) is a vector of commodities (or characteristics). For
simplicity, we assume there are r goods available, and each is used
to produce a unique commodity. The household allocates its time
to convert the r goods, x = (x1 , . . . , xr ), into commodities, with a
production function
zj = gj (xj , tj ; E), (59)
where gj (·) is the production function, tj is time used to produce
commodity zj , and E stands for the production environment. Note
that leisure decision is also covered in the model. For example, one
can view x1 as the amount of goods consumed during leisure time
(e.g., movie theater tickets) and t1 as the time spent on leisure.
Assume the time that the household supplies in the labor market
is tw , earning wage rate at w. Therefore, the total time constraint
of the household is tw + Σj tj = T , where T is the total amount
of time that the household has. Let V be the non-wage income of

7
We refer readers to Nevo (2011) for a detailed discussion about estimation for
demand models in characteristic space.
 Consumer Demand 77

the household. Therefore, the budget constraint of the household is

Σj pj xj = wtw + V . Merging the time and budget constraint, we
obtain S = wT + V = Σj (wtj + pj xj ). The household’s demand
problem can be written as

max U (g1 (x1 , t1 ; E), . . . , gr (xr , tr ; E)) s.t. Σj (wtj + pj xj ) = S,

x,t
(60)

whose Lagrangian is

L = U (g1 (x1 , t1 ; E), . . . , gr (xr , tr ; E)) − λ[Σj (wtj + pj xj ) − S]. (61)

The first order conditions include

∂L ∂U ∂gj
= − λpj = 0, ∀j ∈ {1, . . . , r}, and (62)
∂xj ∂zj ∂xj
∂L ∂U ∂gj
= − λw = 0, ∀j ∈ {1, . . . , r}. (63)
∂tj ∂zj ∂tj
These first order conditions imply that, at the optimum, the marginal
utility of a factor (good or time) is equal to its marginal cost (price
or wage rate).
Solution of the system provides the demands for goods, leisure,
and supply of labor in the market. Specifically, demand for goods are
functions of prices, income level, wage rate, and household production
environment. Total demand for a good (an input in commodity
production) is the sum of induced demands from all demand for
commodities. The demand for goods depends on the decomposition
of income between wage and non-wage income. Wage affects the
demand of goods which are substitutes or complements of labor in
production of commodities. The model can be expanded to include
several family members with different wage rates and to address labor
allocations within the family.
There are several differences between the models proposed by
Lancaster (1966) and by Michael and Becker (1973). First, the
sources of utility are treated and defined differently. On the one
hand, Lancaster’s (1966) characteristics (e.g., brightness, sweetness,
beauty) are abstract. On the other hand, Michael and Becker’s
78 Agricultural Economics and Policy

commodities are more concrete and physical, but they distinguish

between an omelet cooked at home and an omelet at a restaurant, or
a salad vs. just tomato and lettuce. Second, the models are different
in the nature of production technology. In Lancaster (1966), the
production process, with linear technology, includes three elements:
activities which combine goods to yield characteristics. In Michael
and Becker (1973), the production process assumes a neoclassical
production function in two stages: time plus goods produce com-
modities. Lancaster (1966) emphasizes intrinsic, atomistic elements
in goods and actual enjoyment of production, while Michael and
Becker (1973) emphasize the role of time in family production.

6. Conclusion
In this chapter, we have summarized some key results of the
standard demand theory and introduced a few demand system
models that are widely used in the literature. The AIDS model
is quite flexible and easy to estimate, however, its limitations lie
in that it puts restrictions on Engel curves that can be derived
from the estimates. Moreover, the AIDS model cannot accommodate
consumers’ preference heterogeneity. The EASI model addresses
these issues by starting from a different expenditure function than
that used by the AIDS model. Both the AIDS model and the EASI
model are models in the product space. These models suffer from
high dimensionality issue because there are numerous products in
the market, which requires estimates of large amount of parameters.
They also have difficulties to predict the demand for new products.
Demand models in the characteristic space originated from Lancaster
(1966) aim to address these issues. Another line of enrichment of the
demand theory comes from Becker who emphasizes on household
time allocation and household production. Overall, we have an array
of working models to choose from when estimating demand. However,
which model to choose will eventually be determined by the specific
research question and focus. For instance, if the research purpose is
to predict the demand of some new products, then demand models
in the characteristic space should be considered.
 Consumer Demand 79

References
Beckert, W. and R. Blundell. 2008. Heterogeneity and the Non-Parametric
Analysis of Consumer Choice: Conditions for Invertibility. The Review of
Economic Studies 75(4): 1069–1080.
Christensen, L.R., D.W. Jorgenson, and L.J. Lau. 1975. Transcendental Loga-
rithmic Utility Functions. American Economic Review 65(3): 367–383.
Clements, K.W. and G. Gao. 2015. The Rotterdam Demand Model Half a Century
On. Economic Modeling 49: 91–103.
Deaton, A. and J. Muellbauer. 1980a. An Almost Ideal Demand System. American
Economic Review 70(3): 312–326.
Deaton, A. and J. Muellbauer. 1980b. Economics and Consumer Behavior.
Cambridge University Press, Cambridge, UK.
Green, R. and J.M. Alston. 1990. Elasticities in AIDS Models. American Journal
of Agricultural Economics 72(2): 442–445.
Jehle, G.A. and P.J. Reny. 2001. Advance Microeconomic Theory (2nd edition).
Addison-Wesley, Boston, MA.
Lancaster, K.J. 1966. A New Approach to Consumer Theory. Journal of Political
Economy 74(2): 132–157.
Lewbel, A. and K. Pendakur. 2009. Tricks with Hicks: The EASI Demand System.
American Economic Review 99(3): 827–863.
Mas-Colell, A., M.D. Whinston, and J.R. Green. 1995. Microeconomic Theory.
Oxford University Press, Inc., Oxford.
Michael, R.T. and G.S. Becker. 1973. On the New Theory of Consumer Behavior.
The Swedish Journal of Economics 75(4): 378–396.
Nevo, A. 2011. Empirical Models of Consumer Behavior. Annual Review of
Economics 3: 51–75.
Pendakur, K. (2009), Chapter 7: EASI Made Easier, Slottje, D.J. (Ed.), Quanti-
fying Consumer Preferences (Contributions to Economic Analysis, Vol. 288),
Emerald Group Publishing Limited, Bingley, pp. 179–206. https://doi.org/
10.1108/S0573-8555(2009)0000288010.
Stone, R. 1954. Linear Expenditure Systems and Demand Analysis: An Appli-
cation to the Pattern of British Demand. The Economic Journal 64(255):
511–527.
This page intentionally left blank
 Chapter 4

Economic Analysis of Behavior under

Risk and Uncertainty

Risk and uncertainty are prevalent in agriculture. This chapter

presents economic modeling of decision-making under risk and uncer-
tainty. Choices under risk occur when the probability distribution of
the outcomes is objectively known to the decision maker. Choices
under uncertainty occur when no objective probability distribution
is given to the agent.
The expected utility model, introduced by von Neumann and
Morgenstern in The Theory of Games and Economic Behavior
(1944), represents the traditional approach to modeling behavior
under risk. It describes the relationship between an individual’s scale
of preferences for a set of acts and their associated consequences.
Given certain postulates about rational choice, von Neumann and
Morgenstern developed a set of axioms about the ordering, continu-
ity, and independence of individual choice and used this as a base to
derive the properties of the expected utility function, thus describing
the conditions under which an individual’s preferences under random
choices correspond to maximization under the expected utility model.
Friedman and Savage’s paper (1948) is perhaps the first where the
expected utility approach is applied to explaining economic behavior.
The utility function is defined on wealth, where diversification and
general risk aversion are explained by the function’s concavity.
Moreover, an S-shaped specification for the utility function explains

81
82 Agricultural Economics and Policy

why an individual may be risk-averse for some choices but risk-loving

for other choices.
Arrow (1971) and Pratt (1964) introduced measures of risk
aversion. They defined a measure of absolute risk aversion, RA =
−U  /U  , and a measure of relative risk aversion, RR = −wU  /U  ,
where U  and U  indicate, respectively, the second and first deriva-
tives of the von Neumann–Morgenstern utility function, and w is the
wealth. It was established that, under the expected utility hypothesis,
there exists a one-to-one relationship between preferences over
random income (or wealth) and the measures of risk aversion. Addi-
tionally, as income grows, one cares less about one “unit” of risk —
the measure of absolute risk aversion is declining but cares equally
about the risk involving a given share of his wealth — the measure of
relative risk aversion may be constant and perhaps even equal to 1.
The next step in the theory of decision-making under risk was
the development of models and concepts for measuring risk. The
first efforts in this direction used statistical indexes such as mean
and variance of the random outcome as arguments of the utility
functions. Recognizing that variance is not always a good measure
of risk, Hadar and Russell (1969) and Rothschild and Stiglitz (1970)
developed models and concepts that are useful for a more general
comparison of risky prospects. These approaches use probability
distributions and are independent of the decision maker’s utility
function. Concepts such as the mean-preserving spread and stochastic
dominance fall under this rubric and have been used in agricultural
economic research.
Although widely used in agricultural economics, expected utility
theory has been criticized for not being able to predict well people’s
behavior under risk. Prospect theory, developed by Kahneman and
Tversky (1979), has emerged as an alternative of expected utility
theory and has been recently applied in agricultural economic studies.
Different from expected utility theory, prospect theory differentiates
gains and losses from a reference point while considering people’s
biased reactions to large probabilities and small probabilities (i.e.,
under-weighting large probabilities and over-weighting small proba-
bilities). Moreover, assuming that decision makers know distributions
 Economic Analysis of Behavior under Risk and Uncertainty 83

of outcomes is, in many cases, unrealistic. Relaxing this assumption

involves modeling decision-making under uncertainty, where no
objective probability distribution is given.
In the following sections, we first discuss some of the main
approaches currently used to measure risk and risk aversion and
then introduce some models that are used to analyze decision-making
under risk and uncertainty.

1. Measures of Risk Aversion and Their

Interpretation
What does it mean to say that an individual is risk averse in the
context of expected utility? How can we measure people’s attitudes
toward risk? Arrow (1971) developed answers to these questions.
Working from the definition of a risk-averse individual as one who
“starting from a position of certainty, is unwilling to take a bet which
is actuarially fair”, Arrow derived a series of quantitative measures
of risk attitudes. The most important of these are the absolute and
relative risk-aversion coefficients, which can be derived from the
von Neumann–Morgenstern utility function. Arrow also hypothesized
that individuals would exhibit decreasing absolute risk aversion and
increasing relative risk aversion under most circumstances. These
hypotheses have critical implications for the empirical specification
of utility functions. Some of the most simply formulated utility
functions, however, do not exhibit the risk preference structure
postulated by Arrow.

1.1. Absolute risk aversion

Suppose U (w) is a von Neumann–Morgenstern utility function which

is bounded and twice differentiable, where w is the wealth. Let U 
be the marginal utility of wealth, and U  be the rate of change of
marginal utility with respect to wealth. Absolute risk aversion, Ra ,
is then defined as
U  (w)
RA = − .
U  (w)
84 Agricultural Economics and Policy

1.1.1. Absolute risk aversion with a discrete probability

distribution

Suppose a risk-averse individual is offered, for a small amount h, the

following gamble: win h with probability P or lose h with probability
1 − P . When will the individual be indifferent to the gamble? With
an initial wealth level of w0 , equalizing the expected utility of taking
the gamble to the utility achieved with no gamble, we obtain
P U (w0 + h) + (1 − P )U (w0 − h) = U (w0 ). (1)
Estimating a Taylor series approximation around w0 ,
1
U (w0 + h) ≈ U (w0 ) + U  (w0 )h + U  (w0 )h2 ,
2
1
U (w0 − h) ≈ U (w0 ) − U  (w0 )h + U  (w0 )h2 .
2
Substituting these approximations into (1), we get
P U (w0 + h) + (1 − P )U (w0 − h)
1
≈ U (w0 ) + (2P − 1)U  (w0 )h + U  (w0 )h2 .
2
Canceling like terms and rearranging terms, we obtain
1
(2P − 1)U  (w0 )h = − U  (w0 )h2
2
1
⇒ (2P − 1) = RA (w0 )h
2
1 1
⇒ P = + RA (w0 )h
2 4
dP 1
⇒ = RA (w0 ).
dh 4
Therefore, the coefficient of absolute risk aversion RA indicates
the odds of winning (i.e., P ) have to be affected to induce a risk-
averse individual to take a constant sum gamble. For a risk-averse
individual, this coefficient should be positive. Thus, the coefficient of
absolute risk aversion directly measures how much over fair odds an
individual requires before accepting a bet. As the amount gambled
 Economic Analysis of Behavior under Risk and Uncertainty 85

increases, a higher probability of winning is needed in order for an

individual to be indifferent between gambling and certainty.

1.1.2. Absolute risk aversion with a continuous probability

distribution

Consider a small gamble x with mean μ and variance σ 2 . If we express

this gamble in terms of the first two moments of the distribution and
use a Taylor series approximation around the mean, we get

EU (w0 + x) ≈ E U (w0 + μ) + U  (w0 + μ)(x − μ)

1  2
+ U (w0 + μ)(x − μ) . (2)
2
Defining a constant z as the certainty equivalent of x such that
EU (w0 + x) = U (w0 + z),
and taking the expectations operator through (2), we obtain
1
EU (w0 + x) = U (w0 + z) ≈ U (w0 + μ) + U  (w0 + μ)σ 2 . (3)
2
Since both μ and z are small, we can make the following
approximations:
U  (w0 ) ≈ U  (w0 + μ) ≈ U  (w0 + z)
and
U  (w0 ) ≈ U  (w0 + μ).
Therefore, based on Equation (3), we can have the following
approximations:
1
EU (w0 + x) ≈ U (w0 ) + U  (w0 )μ + U  (w0 )σ 2 ≈ U (w0 ) + U  (w0 )z.
2
Solving for z:
1
U  (w0 )μ + U  (w0 )σ 2 = U  (w0 )z,
2
1
z = μ − RA σ 2 .
2
86 Agricultural Economics and Policy

Finally, expressing the risk-aversion coefficient in terms of the

moments of the distribution:
μ−z
RA = 1 2 .
2σ

We conclude the following: (i) if μ > z, then RA > 0, indicating that

the individual is risk-averse; (ii) if μ = z, then RA = 0, indicating
risk neutral; and (iii) if μ < z, then RA < 0, indicating risk loving.

1.2. Relative risk aversion

Based on the notation above regarding utility function and wealth,

we further define relative risk aversion, RR , as
U 
RR = − w = RA (w) · w.
U
Relative risk aversion is a measure of risk proportional to the level of
wealth. It is the elasticity of marginal utility with respect to wealth
and can be thought of as the sensitivity of risk aversion to changes
in wealth. Consider a discrete probability distribution with a gamble
to win a fraction of t of wealth with probability P or lose it with
probability 1 − P . When an individual is indifferent to the gamble,
then we have

P U [w(1 + t)] + (1 − P )U [w(1 − t)] = U (w).

Following a similar procedure as with the absolute risk aversion

coefficient, we can derive the following equation:
1
U (w) + (2P − 1)U  (w)tw + U  (w)t2 w2 ≈ U (w).
2
Dividing both sides by U  (w)tw and canceling like terms, we obtain
1 U  1
2P − 1 ≈ − tw = RR t.
2 U 2
The higher the relative risk aversion coefficient, RR , the higher must
be the probability of winning for the individual to be indifferent, for
a given share of wealth at stake, t.
 Economic Analysis of Behavior under Risk and Uncertainty 87

1.3. Hypotheses about risk preferences

In his 1971 essay, Kenneth Arrow put forward two hypotheses

about the behavior of the measures of risk aversion. These are
decreasing absolute risk aversion (DARA) and increasing relative risk
aversion (IRRA). Decreasing absolute risk aversion implies that the
willingness of individuals to take small bets of fixed size increases
with wealth. Increasing relative risk aversion implies that, as wealth
increases, the proportion of wealth that the individual is willing to
risk declines. For example, Elon Musk will be more likely to accept
a $100 bet than a PhD student (assuming no obscenely rich PhD
student!), but he will be less likely to wager 10% of his wealth.
Going back to the derivations above in Section 1.1.1, note that
the assumption of decreasing absolute risk aversion is associated with
dP 1 ∂RA
= h < 0,
dw 4 ∂w
indicating that as wealth level increases, a lower winning probability
is needed for an individual to be indifferent between gambling and
certainty. From the derivation of the measure of relative risk aversion,
if we assume ∂R R ∂P
∂w > 0, then we have ∂w > 0. This implies that
for individuals with increasing relative risk aversion, as wealth level
increases, a higher winning probability is needed for an individual to
be indifferent between gambling and certainty.

1.4. The implications of risk preferences for

empirical specification of utility functions

Empirical applications of the expected utility framework may require

specification of the utility function. In order to achieve a reasonable
depiction of reality with a tractable form, it is desirable that the
specific utility function U (w), where w is the wealth (or income),
will have some of the following characteristics: (a) simplicity; (b)
positive and decreasing marginal utility (U  > 0 and U  < 0); (c)
decreasing (or at least non-increasing) absolute risk aversion (i.e.,
∂RA /∂w ≤ 0, where RA = −U  (w)/U  (w)); and (d) non-decreasing
relative risk aversion (∂RR /∂w ≥ 0, where RR = −wU  (w)/U  (w)),
88 Agricultural Economics and Policy

and if RR is constant, it is preferably near 1. A few specific cases of

utility functions are discussed as follows:
Quadratic Utility Functions: U (w) = a + bw − 12 cw2 , for w < b/c,
where a, b, and c are positive parameters. This utility function may
be objectionable because it implies increasing absolute risk aversion
(RA = c/(b − cw), and ∂R ∂w = cRA /(b − cw) > 0).
A

Cobb–Douglas Utility Functions: U (w) = Awα , 0 < α < 1.

This utility implies constant relative risk aversion and decreasing
absolute risk aversion with RR (w) = 1 − a and RA (w) = (1 − a)/w.
The deficiencies of this function include the following: (i) RR = 1
when α = 0, and (ii) the expectation of utility under this function,
E(Awα ), may result in complex expressions.
Logarithmic Utility Function: U (w) = ln w. Defined only for
w > 0, this function implies constant relative risk aversion with
RR (w) = 1 and decreasing absolute risk aversion. The expected value
of this utility, E(ln w), may be cumbersome in problems where w is
a linear function of decision variables.
Exponential Utility Functions: U (w) = 1 − e−rw . This utility
function implies constant absolute risk aversion with RA (w) = r.
It is easy to apply with distributions which can be defined by
their moment-generating functions. Moment-generating functions
are functions of the parameters of the distribution associated with
random variables. Specifically, for a random variable, z, its moment-
generating function can be written as Mz (t) = E(ezt ). Therefore, one
can see that for the exponential utility function, EU (w) = 1−E[e−rw ]
and E[e−rw ] is a moment-generating function. When w is a normally
distributed random variable with E(w) = μ and var(w) = σ 2 , the
moment-generating function to the second-order gives1
1 2
EU (w) = 1 − e−r[μ− 2 rσ ] . (4)

Since any solution that maximizes μ − (1/2)rσ 2 also maximizes

EU (w) in this case, when utility is exponential and the random
variable is normally distributed, maximization of a linear function

1
Chavas (2004, p. 39) presents a direct way to obtain Equation (4).
 Economic Analysis of Behavior under Risk and Uncertainty 89

of the mean and variance of income is equivalent to expected utility

maximization. That is,
 
1 2
arg max EU (w) = arg max μ − rσ .
X X 2

Let us see an example. Suppose w = w0 + P̄ Xε − C(X), where X

stands for output quantity; P̄ for average price; ε for random price
variability, ε ∼ N (1, σ 2 ); and C(X) for cost function with C  > 0
and C  > 0. Expected utility maximizing outcome in this case can
be obtained by solving
1
max P̄ X − C(X) − rσ 2 P̄ 2 X 2 .
X 2
The optimal solution is obtained by solving the first-order condition:

P̄ − C  (X) − rσ 2 P̄ 2 X = 0.

Suppose C(X) = cX. The first-order condition becomes

P̄ − c − rσ 2 P̄ 2 X = 0,

which gives
P̄ − c
X= .
rσ 2 P̄ 2
Readers may want to check if the supply (X as a function of P̄ )
can be negatively sloped in this case and think about the reasons.
Furthermore, readers may try to obtain the supply assuming that
price P follows a gamma distribution. We leave this as exercises for
readers. Hints can be found in Yassour et al. (1981).

1.5. Estimation of risk aversion coefficient

The Arrow–Pratt measures of risk aversion and conceptual models,

such as Sandmo (1971), have established a very rich theory of
decision-making under risk in agricultural production and resource
use. However, one of the challenges of applied economics and
agricultural economics research is to develop an empirical base to
90 Agricultural Economics and Policy

test the theory. This subsection identifies some of the problems and
alternative approaches to address them.
The most difficult problem in assessing empirical expected utility
models is the unobservable nature of an individual’s evaluation of
utility levels and the probabilities associated with them. Expected
utility models assume that choices under risk result from a mental
process where the utility of many wealth levels is assessed and mul-
tiplied with the right probabilities, resulting in the expected utility
of each prospect. Expected utilities are then compared to obtain the
optimal choice. Unfortunately, these evaluations of utility levels are
not observable, but, for standard economic analysis, we have to rely
on observable variables. These include choices, for example, to adopt
or not to adopt a technology, as well as characteristics of decision
makers, such as farm size, age, and education. Information about
the subjective probabilities of different outcomes is also not readily
available, which makes the estimation of decision makers’ parameters
even more difficult.
One approach for empirical estimation of expected utility is to
reduce the uncertainty of the researcher by conducting experiments
where the decision maker is presented with rewards and the probabil-
ities of rewards is specified. The decision maker has to make choices
between outcomes. In these cases, the researcher does not know the
utility functions but supposedly knows everything else associated
with the decision. Such experiments can be used to assess empirically
to what extent the expected utility model is realistic, and in the
case where it is a realistic description of reality, what the values of
key parameters are (i.e., the measures of absolute and relative risk
aversion).
In later sections, we discuss some of the experiments and tests
of expected utility hypothesis. In this subsection, we concentrate
on models that try to estimate risk aversion parameters and, in
particular, to test whether there is decreasing absolute risk aversion,
increasing relative risk aversion, and if relative risk aversion is
around one.
Binswanger (1980) used experiments where farmers in India were
given actual monetary rewards as part of different gambles, elicit-
ing risk-aversion parameters from his subjects’ individual choices.
 Economic Analysis of Behavior under Risk and Uncertainty 91

Since the income level of the individuals involved was relatively low,
with a reasonable amount of money, he was able to collect a large
set of data. He found substantial variability in the measures of risk
aversion between individuals, who demonstrate heterogeneity in risk
preferences. He also found that, for any given individual, the measure
of relative risk aversion did not change much with different gambles.
Other experimental studies, especially ones conducted in the
United States, were largely not done with actual rewards but instead
with hypothetical choices. These studies were mostly intended to
identify paradoxes in risk choice patterns that contradict expected
utility and not to elicit the parameters of the utility function.
Popular approaches to estimate the parameters of the utility
function were ones using programming models and econometric mod-
els. The following subsection describes how one can use econometric
models such as the Just–Zilberman model to elicit risk aversion
coefficients.

1.5.1. Estimation

Let us consider a simple model where a farmer allocates her total

land, L̄, between two crops, 1 and 2. Assume that the net returns
from crop i ∈ {1, 2} are πi = μi + i , where μi is a constant and
i is a random variable with mean at 0 and variance at σi2 . The
covariance between 1 and 2 is σ12 . Based on equation (7) in Just and
Zilberman (1983), the amount of land allocated for crop 1, L1 , can be
written as
μ1 − μ2 σ2 − σ12 A1
L1 = 2 + L̄ ≡ + A2 L̄,
(σ1 + σ22 − 2σ12 )φ(W̄ ) (σ12 + σ22 − 2σ12 ) φ(W̄ )
(5)
where the function φ(W̄ ) is a measure of absolute risk aversion as a
function of average wealth, W̄ , which is determined by the farmer’s
initial wealth and returns from land use.
This model, which is based on a Taylor series as an approximation
of the first-order condition, allows us to get a quantitative assessment
of the behavior of risk aversion measures. Since one cannot observe
the utility function changing with wealth for one individual, we
try to estimate how the measure of risk aversion is changing
92 Agricultural Economics and Policy

between individuals where we tried to estimate the measure of

risk aversion as a function of average wealth, W̄ , using our results
derived by our approximation. Several relationships which can be
derived using these models depend on the data available and the
degree of statistical sophistication.
Now, consider the simplest case. One assumes constant absolute
risk aversion and has only data on acre and land allocation with two
crops. In this case, the estimated models will be L1 = A1 + A2 L̄ + ε,
where ε is a random variable. Such a model can be estimated using a
simple linear regression to test some simple empirical hypotheses. For
example, if crop 1 returns have a higher mean and higher variance,
and the correlation of returns between the two crops is not large, one
can test hypotheses that A1 is positive and A2 is negative. If one ran
such models in different regions and obtained an A1 estimate for the
two regions, then one could test a hypothesis that a region where the
correlation between yield is larger has a larger A1 .
One can use this land allocation equation and incorporate it with
other elements that determine land allocation, for example, fixed
costs of different technologies as well as credit constraints. Marra and
Carlson (1990) have a nice application of this approach to allocate
assessment of adoption of double cropping in the United States.
If one assumes that φ varies with W̄ instead of being constant
and if a more sophisticated econometric model is used, then data
on L1 and L can lead to more insightful results. Furthermore, if
one assumes a constant ratio between farm size and expected wealth
W̄ ≈ αL̄, then risk aversion can be approximated and estimated as
a function of farm size. Suppose

φ(W̄ ) = C · W̄ −η ,

where η is the elasticity of absolute risk aversion with respect to

wealth,
∂φ W̄
η=−
∂ W̄ φ
and C is a scaling constant. In the case of constant absolute risk
aversion, η = 0. In the case of decreasing absolute risk aversion,
 Economic Analysis of Behavior under Risk and Uncertainty 93

however, η > 0. Let r = φw be a measure of relative risk aversion.

Note that if we define the elasticity of relative risk aversion as δ, it is
 
∂r W̄ ∂φ W̄
δ= = + 1 = 1 − η.
∂ W̄ r ∂ W̄ φ
We can see that in the case of constant relative risk aversion, η = 1,
and that in the case of increasing relative risk aversion, η < 1. Thus,
assuming decreasing absolute and increasing relative risk aversion
implies 0 < η < 1 (or 0 < δ < 1).
Moreover, when φ(W̄ ) = C W̄ −η and W̄ = αL̄, the absolute risk
aversion coefficient, φ, is φ = Cα−η L̄−η and the estimatable model
that corresponds to these assumption is

L1 = A3 /L̄−η + A2 L̄ = A3 L̄η + A2 L̄,

where A3 = A1 /(Cα−η ). An alternative formulation is

L1
= A2 + A3 L−δ .
L̄
Estimation of these models provides Â2 , Â3 , and δ̂ or η̂, which
allows testing when absolute risk aversion is decreasing in wealth
(approximated by size) and relative risk aversion is increasing in
wealth.
This formulation can be extended to identify other factors
affecting risk aversion. If socioeconomic data (denoted by S) are
available (e.g., age, gender, and education), one may replace C in
the expression of φ with g(S), a function of socioeconomic variable.
One plausible specification is φ(W̄ ) = g(S)W̄ −η . This specification
can lead to an estimatable relationship which is a function of farm
size and socioeconomic variables. More detailed data on profit and
wealth may allow all estimation of the Just and Zilberman (1983)
model with less approximation. Even then, one may need to use
bold assumptions. Data on the subjective values of mean at the
individual farm levels are not easily obtainable. Thus, one may
need to estimate these variables as well and introduce them to an
estimatable form. Similarly, average wealth of individual farmers
is needed to be computed from other accounting data. In essence,
94 Agricultural Economics and Policy

estimation of risk aversion coefficient from a simple specification

such as (5) requires much compromise and ingenuity. We also have
to recognize that the decision maker has the same problem of data
assembly as the researcher: None of the farmers know their μ1 , μ2 ,
and σ2 . A more complete model should recognize this.
Chalfant et al. (1990) developed an approach for when risk-
aversion parameters are utilized by farmers who recognize the uncer-
tainty of their estimates of the key profit distribution parameters.
The model is complex, but the optimal L1 depends not only on
the estimated means or variance of profits and the measures of risk
aversion but also on measures of the estimators’ reliability and the
moments of farmers’ profits.

2. Risk in Production

Risk occurs in several aspects of the production process. Various

institutional mechanisms for reducing risk have been developed.
Table 1 includes a partial list of these mechanisms and the type of
production risk they address. Many other institutions were developed
to address risks faced by firms. In this section, we focus on the
building blocks of modeling production risk.

2.1. Risk specification

Production risk is generally modeled through alternative specifica-

tions of the production function. Several examples of production
functions and their implications for risk analysis are discussed in

Table 1. Types of risks faced by producers and mechanisms to remedy.

Risks Remedies

Output price risk Future markets, forward contracts, crop insurance

Yield risk Crop insurance
Labor supply availability Mechanization, long-term labor contract
Input reliability Product warranty
Input price uncertainty Forward contacts
 Economic Analysis of Behavior under Risk and Uncertainty 95

this subsection. Let Y be output quantity, X input quantity, and ε

a random variable in each of the following models:

Model 1. Additive risk

Y = f (X) + ε, E(ε) = 0.

In this case, input use does not affect risk and the only type
of risk considered is output risk.
Model 2. Multiplicative risk

Y = f (X) · ε, E(ε) = 1.

In this case, any input that increases mean yield also increases
risks associated with yields.
Model 3. Linear risk (Just and Pope production function)

Y = f (X) + g(X)ε, E(ε) = 0.

Under this specification, impacts of inputs on yield mean and

risk can be differentiated. For example, some inputs may be
yield increasing (i.e., f  > 0) and risk reducing (i.e., g  < 0);
others may increase both yield and risk (i.e., f  > 0, g > 0).

2.2. Sandmo’s model

How will a firm behave when output price is a random variable

as opposed to when it is a constant? Sandmo (1971) used a
multiplicative risk specification to develop a model of a competitive
firm facing output price uncertainty. Specifically, price P is a random
variable with mean value at P̄ . Firms maximize expected utility with
cost function C(Y ), where C  > 0, C  > 0. The decision problem is

L = max EU (P Y − C(Y ) + w0 ),
Y

where w0 is initial wealth. The first-order condition is

∂L
= E{U  (P Y − C(Y ) + w0 )[P − C  (Y )]} = 0. (6)
∂Y
96 Agricultural Economics and Policy

By using the statistical theorem concerning the expected value

of the product of two random variables which states
E(XZ) = E(X)E(Z) + Cov(X, Z),
we can rewrite (6) as
Cov[U  (w), P − C  (Y )]
E[P − C  (Y )] + = 0, (7)
EU  (w)
where w = P Y − C(Y ) + w0 .
In order to determine the impact of risk on output, we need to
determine the signs of each element in this expression. The steps are
as follows:

(1) By definition, w = P Y −C(Y )+w0 and E(w) = P̄ Y −C(Y )+w0 .

Therefore, w − E(w) = (P − P̄ )Y and w = E(w) + (P − P̄ )Y.
(2) If P ≥ P̄ , then from step 1 we know that w ≥ E(w), and
therefore, that U  (w) ≤ U  [E(w)]. Similarly, if P < P̄ , then
w < E(w) and U  (w) > U  [E(w)].
(3) If we multiply (P − P̄ ) to both U  (w) and U  [E(w)], then, based
on step 2, we can conclude that (P − P̄ )U  (w) ≤ (P − P̄ )U  [E(w)]
whether or not P ≥ P̄ . Take the expectation of both sides, we
get
E((P − P̄ )U  (w)) ≤ U  [E(w)]E(P − P̄ ) = 0.
(4) Since Cov[U  (w), P − C  (Y )] = E[(P − P̄ )U  (w)], from Equation
(7) and the results in step 3, we have E(P − C  (Y )) ≥ 0, which
indicates C  (Y ) ≤ P̄ .

Based on the above steps, it is readily checked that under

risk neutrality, C  (Y ) = P̄ . With risk-averse behavior, however,
C  (Y ) ≤ P̄ . Since C  (Y ) > 0, we can conclude that risk-averse firms
will produce less than risk-neutral firms (see Figure 1). This finding
implies that price stabilization policies will lead to an increase in
output.
Suppose utility depends on wealth and w = P Y − C(Y ) + w0 .
What will be the impact of higher wealth on the optimal Y ?
 Economic Analysis of Behavior under Risk and Uncertainty 97

Figure 1. Production level in Sandmo’s model.

Note: Y ∗ indicates optimal production level under risk neutrality, and Ŷ indicates
optimal production level under risk aversion.

Based on the first-order condition in Equation (6), the second-order

condition is

∂2L
= E[U  (P Y − C(Y ) + w0 )(P − C  (Y ))2 ]
∂Y 2
− E[U  (P Y − C(Y ) + w0 )C  (Y )] < 0.

Total differentiation of the first-order condition with respect to w0

yields

dY E[U  (P Y − C(Y ) + w0 )(P − C  (Y ))]

=− ∂2L
. (8)
dw0 2 ∂Y

Let us first assume decreasing absolute risk aversion (DARA).

Let W̃ be the level of wealth associated with P̃ , and P̃ = C  (Y ).
98 Agricultural Economics and Policy

(w)
Therefore, if P < P̃, then − UU  (w) > RA (W̃ ), and if P > P̃, then
(w)
− UU  (w) < RA (W̃ ). As a result, we have

U  (w)
− [P − C  (Y )] < RA (W̃ )[P − C  (Y )]
U  (w)
⇒ −U  (w)[P − C  (Y )] < RA (W̃ )U  (w)[P − C  (Y )]
⇒ −E{U  (w)[P − C  (Y )]} < RA (W̃ )E{U  (w)[P − C  (Y )]} = 0
⇒ E{U  (w)[P − C  (Y )]} > 0.
The third line in the above four lines of equations holds because of the
first-order condition. In other words, when evaluated at the optimal
output level, E{U  (w)[P − C  (Y )]} = 0. Together with Equation
(8), E{U  (w)[P − C  (Y )]} > 0 indicates that under decreasing
absolute risk aversion, we have dY /dw0 > 0. This result states
that under DARA, more affluent risk-averse producers will provide
more output, and wealth has a positive effect on supply. Following
similar steps, readers can readily check that, under constant absolute
risk aversion (CARA), dY /dw0 = 0; under increasing absolute risk
aversion (IARA), dY /dw0 < 0.
Sandmo also shows that increasing relative risk aversion may
lead to a reduction in supply as income taxes increase. He was able
to partially show that increase in riskiness (measured by a mean-
preserving spread) reduces supply. His technique has been used to
obtain conceptual results in many problems with multiplicative risk.
We can also examine the impact of price risk on optimal output
level by following the same approach. Let P1 = γP + (1 − γ)P̄ , where
P̄ = E(P ). Note that here γ ∈ [0, 1] can be viewed as a measure of
price risk with γ = 0 indicating a fixed price, P̄ . Replacing P with
P1 in the firm’s profit maximization problem, we obtain
L = max EU ((γP + (1 − γ)P̄ )Y − C(Y ) + w0 ).
Y

∂γ |γ=1
∂Y
One can show that under DARA, we have < 0. To see this,
note that
 Economic Analysis of Behavior under Risk and Uncertainty 99

dY E[U  (w)(P − C  (Y ))(P − P̄ )Y + U  (w)(P − P̄ )]

=− ∂2L
dγ 2 ∂Y

E[U  (w)(P − C  (Y ))2 Y ] + Y (C  (Y ) − P̄ )×

E[U  (w)(P − C  (Y ))] + E[U  (w)(P − P̄ )]
=− ∂2L
.
∂Y 2

Under risk aversion, U  (w)(P − C  (Y ))2 Y < 0. In this subsection,

we also have shown that under risk aversion C  (Y ) − P̄ < 0 and
E[U  (w)(P − P̄ )] < 0 (see steps 1 to 4 above). Furthermore, we
have shown that under DARA, E[U  (w)(P − C  (Y ))] > 0. Based
on these results, we can conclude that dY /dγ < 0, indicating that
under DARA an increase in risk (measured by γ) will decrease the
optimal output. We can draw the same conclusion under CARA but
not under IARA.2
If DARA reflects the risk preferences of general decision makers,
then this subsection demonstrates that less output will be produced
under risk than under no risk; that an increase in risk aversion will
reduce output; and that optimal resource allocation by a risk-averse
firm requires that the value of the expected marginal product of a
resource exceeds its rental value (i.e., P̄ > C  (Y )). Moreover, under
DARA, reductions in fixed cost will increase output. Thus, financial
conditions may affect production decisions. Expected profit will be
the highest for firms which are closest to being risk neutral and have
the highest output.
Sandmo’s approach, while general, cannot be applied to situa-
tions with multiple and correlated risks. One may need to specify
utility functions in more detail to address these problems. The
model, focusing on price risk, does not allow for the analysis of
the differential impacts of an input on output and production risk.
Further, it is only possible to handle one random variable at a time.

2
Chavas (2004, Chapter 8) provides a detailed discussion on production under
risk.
100 Agricultural Economics and Policy

The advantage of the model is its generalized form, allowing for

the use of less restrictive utility forms, in contrast with specific
functional forms. Sandmo’s approach was extended by Feder (1977)
to model situations where multiplicative risk arises, and in particular,
the adoption of new technologies in less-developed countries. The
Sandmo approach was also used by Batra and Ullah (1974) for input
demand, Feder (1980) for technological adoption, Feder, Just, and
Schmitz (1980) for futures market behavior, and Chavas (1993) for
land allocation.

2.3. The mean–variance approach

The mean–variance approach was introduced by Tobin (1958) and

Markovitz (1959). It plays a key role in finance — being used as a base
for capital asset pricing. The basic idea
 of this model is that utility
from random prospects, EU (y) = U (y)f (y)dy, can be described
as a function of the moments of the distribution around a mean
outcome ȳ through a Taylor series expansion. Here, y is the random
returns from a prospect and f (y) is the probability density function
of y. Let ȳ denote the mean of y. Based on Taylor series expansion,
we have
∞
 (y − ȳ)2  n (y − ȳ)n

U (y) = U (ȳ) + U (ȳ)(y − ȳ) + U (ȳ) + U ,
2 n!
n=3

from which, taking the expectations of both sides, we have

∞
 U n (ȳ)
EU (y) = U (ȳ) + E(y − ȳ)n = g(M1 , M2 , M3 , . . .),
n!
n=2

where M1 , M2 , . . . denote the first and second moment, and so on.

If a distribution can be completely defined by n moments, then the
expected utility is a function of these moments. If a distribution
is defined by its first two moments, then the expected utility is a
function of the distribution’s mean and variance. For example, in
the case of financial assets, the price of any asset is determined
by its mean return and its variance with the market portfolio.
 Economic Analysis of Behavior under Risk and Uncertainty 101

Certain restrictive conditions on the utility functions and the

distribution of the random outcome variable are required in order
to be able to express expected utility as a function of the mean
and variance. These are (i) the utility function must be quadratic
or exponential in form and (ii) the outcome variable should be
normally distributed. Given these conditions, the expected utility
can be expressed as
σ2
EU (y) = αȳ + β ,
2
where σ 2 is the variance and α and β are parameters.
Freund (1956) proved the linearity of the expected utility func-
tion under the condition of normality and exponential utility. The
linear mean–variance approach has one big advantage: It is easy to
work with and it allows the consideration of behavior under risk
with a large number of random variables. It is used very frequently
(see Just and Zilberman (1983) for an early application and Miao
and Khanna (2017) for a recent application). However, the model
is objectionable on three grounds. First, quadratic utility implies
increasing absolute risk aversion. Second, exponential utility implies
constant absolute risk aversion. Third, normality of the outcome
variable may be unreasonable (e.g., crop yields have a negative
gamma distribution as Day (1965) has shown).
Despite these shortcomings, the linear mean–variance approach is
popular since it results in models useful for dynamic programming.
In the following section, some examples of how this approach has
been applied are discussed.

2.3.1. Applications of the mean–variance approach

The mean–variance approach can be useful in modeling a typical

farmer’s land allocation problem among M crops or uses. Let lm
denote total acreage for crop m = 1, . . . , M . Define L ≡ (l1 , . . . , lM ) .
Using the mean-variance approach, the farmer’s problem can be
stated as

max(U  − V  )L − rL ΣL,

102 Agricultural Economics and Policy

where U = (u1 , . . . , uM ) is the average revenue vector for all M

crops, V is the variable cost vector, r is a measure of risk aversion,
and Σ is the variance–covariance matrix of net returns per acre across
the M crops.
One way to apply the mean–variance approach is to construct an
efficiency locus of mean–variance (or standard deviation) tradeoffs.
This is done through a quadratic programming problem where
the land allocation that minimizes variance is computed to attain
expected profit from the land,

min L ΣL,
L

subject to a mean income constraint:

(U  − V  )L = Z̄,

where Z̄ is a mean income. Solving this minimization problem will

map each income level to a land-use choice which minimizes the
income variance. Then, the next step is to optimally choose a Z̄ and
corresponding variance to maximize the decision maker’s utility.

2.3.2. Mean–variance models

When the distribution of wealth (or profit) has two parameters,

expected utility can be expressed as function of the mean and
variance of wealth (or profit). With the negative exponential util-
ity function and normal distribution, utility maximization can be
expressed as a linear function of mean and variance of wealth (or
profit). These assumptions have been used extensively, especially
in agricultural economics and in finance. They are used primarily
in conceptual analysis when decision makers are affected by several
correlated random variables when more general frameworks, such as
Sandmo’s (1971), cannot be easily applied.
The linear mean–variance formulation has been extensively used
in modeling land allocation by farmers. Assume that a farmer has
L̄ acres of land which can be divided among n crops. The profit of
the ith crop per unit of land, πi , is normally distributed with mean
 Economic Analysis of Behavior under Risk and Uncertainty 103

μi , variance σi2 , and covariance cov(πi , πj ) = σij . The land allocation

problem is a constrained quadratic programming problem
⎡ ⎤
N
r 
N 
max Li μ i − ⎣ L2i σi2 + Li Lj σij ⎦ ,
Li 2
i=1 i=1 j=i

subject to

n
Li ≤ L̄,
i=1

where r is a measure of risk aversion. Let λ be the shadow price of

land. The Lagrangian formulation of this problem is
⎡ ⎤  
N
r ⎣ 2 2
N  
N
L = max Li μ i − L i σi + Li Lj σij ⎦ + λ L̄ − Li .
Li 2
i=1 i=1 j=i i=1

The optimality conditions, when there is an interior solution, are

⎡ ⎤
∂L 1 
= μi − r ⎣Li σi2 + Lj σij ⎦ − λ = 0, i = 1, . . . , N, (9)
∂Li 2
j=i

∂L  N
= L̄ − Li ≥ 0.
∂λ
i=1

At the optimal solution, λ equals the marginal contribution of

land to expected net benefits. For the ith crop, this marginal value
is equal to the mean of net  profit per  acres,  μi , minus 
marginal
2 1
contribution to risk cost M Vi ≡ r Li σi + 2 j=i Lj σij . Note
that the marginal contribution to overall risk depends on the inherent
risk of the crop and correlation of its profits with those from other
crops. A crop whose profits are negatively correlated to those of other
crops may reduce the overall cost of risk. Such a crop may have
substantial acreage even if it is less profitable on average than other
crops. By averaging all the first-order conditions, the marginal value
of land λ can be expressed as
λ = μ̄ − M V,
104 Agricultural Economics and Policy

where
1  1 
N N
μ̄ = μi and MV = M Vi .
N N
i=1 i=1

The first-order condition (9) can be rewritten as

μi − μ̄ − (M Vi − M V ) = 0.
A crop will be grown if it is either more profitable or less risky
than average. Lower risk may not reflect less variability but, rather,
negative correlation of profit with other crops.

2.3.3. Linking mean–variance and general EU models

There have been many attempts to generalize the linear mean–

variance framework. One such attempt is presented in Just and
Zilberman (1983). Consider the case when two crops are grown and
each has a constant return-to-scale technology. Suppose profits per
acre, Πi , is a random variable with mean μi . Land devoted to crop
i = 1, 2 is Li . When all land is used (i.e., L1 + L2 = L̄), the expected
utility problem becomes
max EU [Π1 L1 + Π2 (L̄ − L1 ) + w0 ], (10)
L1

where w0 is initial wealth. The first-order condition is

E{U  (w)(Π1 − Π2 )} = 0. (11)
Marginal utility at w can be approximated by
 
U  (w) = U  (w̄) + U  (w) L1 (Π1 − μ1 ) + (L̄ − L1 )(Π2 − μ2 ) , (12)
where w̄ = w0 +μ1 L1 +μ2 (L̄−L1 ). By introducing this approximation
to the first-order condition, (11) becomes

E U  (w̄)(Π1 − Π2 ) + U  (w̄)
  
× L1 (Π1 − μ1 ) + (L̄ − L1 )(Π2 − μ2 ) (Π1 − Π2 )
  
= U  (w̄) μ1 − μ2 − RA (w̄) L1 (σ12 − σ12 ) + (L̄ − L1 )(σ12 − σ22 )
= 0, (13)
 Economic Analysis of Behavior under Risk and Uncertainty 105

which suggests that

μ1 − μ2 σ22 − σ12
L1 = + L̄, (14)
V (Π1 − Π2 )RA (w̄) V (Π1 − Π2 )

where V (Π1 − Π2 ) = σ12 + σ22 − 2σ12 .

Under risk neutrality, a producer will specialize in the crop with
higher mean profit. With risk aversion, consideration of riskiness is
added to those of average profitability, and the weight of the expected
profit differential in determining L1 declines as V (Π1 −Π2 ) increases.
Suppose crop 1 has a higher profit and higher risk than crop 2. For
example, it maybe a modern export crop that is sensitive to weather
and economic conditions, while crop 2 may be a traditional crop.
It may be of interest to understand the impact of farm size on the
acreage of each crop. Differentiation of (14) yields

dL1 μ1 − μ2 dRA (w) w̄ dw̄ 1 σ22 − σ12

=− · · · · + .
dL̄ V (Π1 − Π2 )RA (w̄) dw̄ RA (w) dL̄ w̄ V (Π1 − Π2 )
(15)
Substituting (14) into the equation gives us
 2 
dL1 σ2 − σ12 dRA (w) w̄ dw̄ 1 σ22 − σ12
= L̄ − L1 · · · +
dL̄ V (Π1 − Π2 ) dw̄ RA (w) dL̄ w̄ V (Π1 − Π2 )
 
σ22 − σ12 dRA w̄ dw̄ 1
= 1 + L̄ · · ·
V (Π1 − Π2 ) dw̄ RA (w) dL̄ w̄
dRA (w) w̄ dw̄ 1
− L1 · · · . (16)
dw̄ RA (w) dL̄ w̄

Let ηR = − dR A W̄
dW̄ RA
be the elasticity of absolute risk aversion
with respect to wealth. Assume decreasing absolute risk aversion,
thus, ηR > 0. With this definition, the change in crop 1’s acreage
with respect to size can be presented as
 
dL1 σ22 − σ12 L̄ dw̄ L1 L̄ dw̄
= 1 − ηR + ηR .
dL̄ V (Π1 − Π2 ) w̄ dL̄ L̄ w̄ dL̄
As we discussed earlier in this chapter, decreasing absolute and
increasing relative risk aversion implies 0 < ηR < 1. It is reasonable
106 Agricultural Economics and Policy

to argue that (dw̄/dL̄) · (L̄/w̄) is also likely to be smaller than 1. If

we examine the case of constant absolute risk aversion where ηR = 0,
then
   2 
dL1 σ2 − σ12
sign = sign . (17)
dL̄ V (Π1 − Π2 )

Note that σ22 − σ12 = σ2 [σ2 − ρσ1 ], where ρ ≡ σ12 /(σ1 σ2 ) is the
correlation coefficient between the two crops’ profits and σi is the
standard deviation of crop i’s profits. Equation (17) suggests that, if
crop 2 is much less risky than crop 1 and the correlation between their
profit is high (so that σ2 < ρσ1 ), risk consideration will cause larger
farmers to grow less of the risky crop than the smaller farmers. When
0 < ηR < 1, the land share of the more risky technology declines
with farm size. Since risk costs increase more than proportionally
with farm size, larger formers are likely to grow relatively less of the
risky crops. The optimization problem (10) is not a constrained one,
but we have to realize that 0 ≤ L1 ≤ L̄. Therefore, letting the result
of this optimization problem be denoted as L∗1 , ultimately
⎧ ∗
⎪
⎨ L̄ if L1 > L̄
L1 = L∗1 if 0 ≤ L∗1 ≤ L̄
⎪
⎩
0 if L∗1 < 0.

Suppose μ1 > μ2 and ρ < σ2 /σ1 . One can readily check that if
σ1 is not very high, then L∗1 > L̄ may hold. This means that the
higher profits of crop 1 may lead to specialization of small farms if
its risk is not so high. Larger farms may grow both crops (note that
farm size affects RA (w̄)). However, when the profit correlation is not
sufficiently large and when 0 < ηR < 1, then the production of crop
1 will grow absolutely and its land share will decline with size.
Now, let us suppose μ1 > μ2 and ρ > σ2 /σ1 . In this case,
correlation is so large that, beyond a certain size, acreage of crop
1 declines with size. When μ2 > μ1 but σ12 < σ22 (i.e., when crop 1
is less risky but less profitable on average), it may not be grown by
small farmers but may be added to the portfolio of larger ones. It
may become the major crop of some very big operators.
 Economic Analysis of Behavior under Risk and Uncertainty 107

In this analysis, risk was the only reason for diversification,

and we ignored other constraints facing farmers. There are many
situations when other factors (e.g., labor or equipment scarcity)
cause diversification. Growers may grow several crops to spread
the harvesting season, thus overcoming labor or capital constraints.
Credit limitation may provide another reason. The high value crop
may require more credits and that may limit a farmer. Economists
have a tendency to attribute too much to risk considerations and to
ignore those other factors. In a more realistic analysis, one has to
study local conditions in detail to incorporate relevant constraints
before investigating land allocation.
The framework presented here can be extended to other choices.
It is a variation of financial portfolio analysis which is used to
investigate distribution of wealth among assets and analysis of
financial investments. Similarly, it applies to time allocation analysis
including land diversification between on-farm and off-farm activities
and migration decisions (time allocation between locations).

3. Measuring Risk: Mean-Preserving Spread

and Stochastic Dominance
Given a set of choices, which will a risk-averse individual prefer
and how will the degree of risk affect her choices? To address these
questions, economists have made many attempts to define a good
measurement for the riskiness of a prospect. Several general measures
were developed within the context of expected utility. Rothschild
and Stiglitz (1970) provided a methodology for the ranking of
prospects which have the same mean outcome but different levels of
risk. In addition, their method provides comparative statics results
describing the impact of risk on key parameters. Hadar and Russell
(1969) developed a method for ranking prospects with differing mean
outcomes. Both of these methods are set within the framework of
expected utility, ranking prospects derived from a von Neumann–
Morgenstern concave utility function to imply risk-averse behavior.
Moreover, the stochastic dominance rule is useful for the comparison
of risky prospects because it allows us to compare the risk associated
with each of two probability distributions and to determine which
108 Agricultural Economics and Policy

is preferable under an expected utility framework (Whitmore and

Findlay, 1978).
3.1. Mean-preserving spread

Rothschild and Stiglitz (1970) put forth four possible ways to

compare two prospects, namely, X and Y , with equal means. They
are as follows:
(1) For any X, Y with E(X) = E(Y ), if EU (X) ≥ EU (Y ) for every
U with U  > 0, and U  < 0, then Y is riskier than X. In other
words, if every risk-averse individual prefers X to Y , then Y is
riskier than X.
(2) If Y −d
→X + Z (i.e., Y is equivalent in distribution to X plus Z)
where Z is a random variable with E(Z|x) = 0, then Y is riskier
than X. In other words, if Y is the sum of X and another random
variable Z and the conditional expectation of Z for every X is
zero, then Y is riskier than X.
(3) If Y can be constructed from X through the use of a mean-
preserving spread (i.e., “spread” some probability weight from
the center to the tails), then Y is riskier than X.
(4) A conventional approach: Variance of X is smaller than that
of Y .
Rothschild and Stiglitz (1970) showed that the first three ways are
equivalent to each other in terms of ranking the riskiness of X and
Y but different from the fourth one. Regarding the fourth way, they
pointed out that it is possible that some risk-averse decision makers
may still prefer Y even if X and Y have the same mean and X has
smaller variance. A numerical example of how a mean-preserving
spread works may help shed some light. Given a random variable X,
we establish the following probability distribution:
x P (X = x)
1 1/8
2 1/8
3 1/2
4 1/8
5 1/8
 Economic Analysis of Behavior under Risk and Uncertainty 109

A new random variable Y is generated from X through the use

of a mean-preserving spread such that increasing the probability
weight assigned to 1 and 5 by 1/8, respectively, and decreasing the
probability weight assigned to 2 and 4 by 1/8, respectively. Y would
then have the following probability distribution:

y P (Y = y)
1 1/4
3 1/2
5 1/4

According to Rothschild and Stiglitz (1970), if and only if Y is

constructed by a sequence of a mean-preserving spread from X, then
EU (X) > EU (Y ) for any utility function U such that U  > 0 and
U  < 0. To illustrate the last point, consider the above example
where we moved from X to Y by a mean-preserving spread. In that
case,
 
1 1 1 1 1
EU (X) = U (1) + U (2) + U (3) + U (4) + U (5),
8 8 2 8 8
 
1 1 1
EU (Y ) = U (1) + U (3) + U (5),
4 2 4
1
EU (X) − EU (Y ) = [U (2) − U (1) + U (4) − U (5)],
8
which is positive due to a decreasing marginal utility under risk-
averse behavior (thus, U (2) − U (1) > U (5) − U (4)), and the
alternative X is preferred to Y .
The concept of mean-preserving spread is important because it
allows for an analysis of the marginal impact of risk. For example,
given a production system such that

Π = P Y − wX.

We can use the mean-preserving spread to manipulate P and create

a new variable with the same mean but different variance:

P1 = γP + Z,
110 Agricultural Economics and Policy

where γ ∈ [0, 1] is a constant and Z is a random variable with mean

E(Z) = (1 − γ)E(P ). We can then calculate dY ∗ /dγ to determine
the impact of price risk on the optimal output.

3.2. Stochastic dominance

Hadar and Russell (1969) developed the concept of stochastic

dominance which allows for the comparison of outcomes with
differing means. This method also allows for the ranking of uncertain
prospects without assuming a specific utility function. This approach
can be used for comparisons of both discrete and continuous risky
choices. However, its usefulness is limited since not all distributions
can be ordered through second-order stochastic dominance.

3.2.1. First-order stochastic dominance (FOSD)

Let X denote the value of a variable (e.g., wealth and income) which
can assume values in the range (−∞, ∞). Let the function FiX (x) be
defined recursively where
 x
X X
Fi (x) = Fi−1 (Z)dZ
−∞

and where F0X (x) is the probability density function of X.

Under this definition, F1X (x) is the cumulative distribution func-
tion (CDF) of X. Consider two random prospects, X and Y . There
is a first-order stochastic dominance of Y by X (i.e., X FOSD Y )
if
F1X (m) ≤ F1Y (m) for − ∞ < m < ∞,
with at least one point of strong inequality. Figure 2 illustrates FOSD
graphically. In other words, X is first-order stochastic dominant to
Y if the cumulative distribution of X is below that of Y for every m,
indicating that Prob(X ≥ m) > Prob(Y ≥ m) for every m.
It is easy to argue that if X FOSD Y , then every individual with
positive marginal utility will prefer X to Y . That is, if X and Y
represent income, then every individual will prefer X over Y since X
represents a higher probability of achieving a high income.
 Economic Analysis of Behavior under Risk and Uncertainty 111

CDF

1
FY(m) FX(m)

Figure 2. First-order stochastic dominance (FOSD).

Note: In this figure, F X (m) first-order stochastically dominates F Y (m).

3.2.2. Second-order stochastic dominance (SOSD)

SOSD is a weaker domination concept than FOSD. We define that

X second-order stochastically dominates Y if and only if
 m  m
F1X (z)dz ≤ F1Y (z)dz.
−∞ −∞

One can check that if X SOSD Y, then risk-averse individuals will

prefer X to Y and that if X FOSD Y, then X SOSD Y . SOSD
is useful when the decision makers are risk-averse. Figure 3 depicts
cumulative distributions under SOSD, from which we can see that
there is some cross-over between the CDFs of the two prospects, but
the area under F Y is greater than that under F X . The cumulative
functions in this case intersect twice, with F1Y (m) < F1X (m) in the
range between the intersection points.
Beyond FOSD and SOSD, stochastic dominance is much less
useful as a concept. It is unclear what choices a risk-averse individual
will make in the presence of third- or greater-order stochastic
112 Agricultural Economics and Policy

CDF

1
FY(m) FX(m)

Figure 3. Second-order stochastic dominance (FOSD).

Note: In this figure, F X (m) second-order stochastically dominates F Y (m).

dominance. Very restrictive assumptions about behavior are required

in order to derive results from higher orders of stochastic dominance.
When FOSD or SOSD is applicable, they are useful, but they are
not applicable in every situation. They only present a “partial”
ordering, ranking only some of the prospects while leaving others out.
In particular, stochastic dominance may not capture the tradeoffs
between risks and returns.

3.3. Variance as a measure of risk

Is variance a good measure of risk? Consider two variables with the

following probability distributions:
n+1 n−1
n with probability n
X= 1 1
n with probability n

n−1 n−1
n with probability n
Y = 2n−1 1
n with probability n.
 Economic Analysis of Behavior under Risk and Uncertainty 113

The means and variances of the two distributions will be as

follows:
n−1
E(X) = U (Y ) = 1 and V (X) = V (Y ) = .
n2
However, even though the means and variances are the same, many
risk-averse individuals will prefer Y to X. For this reason, variance
is not a good measure of risk, although it is often used as a proxy
because it is easy to calculate and the data necessary are usually
available.

3.4. Just and Pope production model

Just and Pope (1978) developed general models for analyzing cases
of production risk econometrically. Before their work, one had the
option of assuming either:

Additive risk: y = f (x) + ε, with E(ε) = 0 or

Multiplicative risk: y = f (x)ε, with E(ε) = 1.

Risk in the production function causes difficulties in the use of linear

programming estimation procedures since an investigation into the
properties of the random variables is required. Both the additive and
the multiplicative models are criticized heavily by Just and Pope.
The additive specification does not allow the uncertainty effect to be
correlated with the input mix, while the multiplicative specification
does not allow for inputs that have differing impacts on mean and
variance, such as fertilizer (yield increasing and risk decreasing).
To offer greater flexibility when describing stochastic technolog-
ical processes and related behavior, Just and Pope (1978) suggest

y = f1 (x) + f2 (x)ε, with E(ε) = 0.

By including two components in the production function, one relating

to output level and the other to output variability, the Just and Pope
production model allows for the differential impacts of an input on
output and risk. The shortcomings of this model are that it only
allows for the consideration of one random variable, and that it
114 Agricultural Economics and Policy

requires a large number of estimations, including the production

function, the risk element, and the behavioral element.

3.5. Exponential utility

Another approach to estimating behavior under risk calls for the use
of a utility function that is applicable to any distribution. Yassour
et al. (1981) adopted this approach and used an exponential utility
function in order to look at farm technology adoption decisions. This
utility function can be applied conveniently in conjunction with all
distributions which have moment-generating functions. The utility
function is written as
U (x) = −e−rx ,

and the expected utility function as

E[U (x)] = −E[e−rx ] = −M (−r),

where M is a moment-generating function and E[·] is the expectation
operator. Since the moment-generating function is a function of the
parameters, the utility function can be expressed in terms of this
function. One shortcoming of this approach is that it implies constant
rather than decreasing absolute risk aversion.
Saha (1993) proposed a new utility function, the expo-power
function, which allows for flexibility in the modeling of risk preference
structures. This form allows the data to reveal both the degree and
structure (i.e., increasing, constant, or decreasing) of risk aversion.
Use of this utility function means that no a priori assumptions about
risk preferences are necessary. The expo-power function is

U (w) = θ − exp{−βwα },
where

θ > 1, α = 0, β = 0, αβ > 0.
The properties of the expo-power utility function are as follows:
 α
(1) It is unique up to an affine transformation; (2) − UU  = 1−α+αβw
w

and − UU  w = 1 − α + αβwα ; (3) when α < 1 (= 1, or > 1), there is
 Economic Analysis of Behavior under Risk and Uncertainty 115

decreasing (constant, or increasing) absolute risk aversion; (4) when

β < 0 (= 0, or > 0), there is decreasing (constant, or increasing)
relative risk aversion; and finally, (5) the utility function is quasi-
concave for all w > 0.
The parameters, α and β, are the key determinants of the risk
preference structure. The effect of α on A(w) and R(w) — the
coefficients of absolute and relative risk aversion — depends on the
relative magnitude of w and the parameters.
In sum, expected utility has theoretical usefulness — such as in
Sandmo’s results and in the mean variance model — and practical
usefulness in applications through either the mean variance model
or the Just–Pope production function. Some of the shortcomings
of these models have been addressed by the work of Yassour et al.
(1981) and Saha (1993). The former allows for the consideration of
non-normal yield distributions in an easy-to-manipulate format. The
latter proposes a flexible utility form with no a priori assumptions
about risk preferences specified. A major problem is that, up to now,
only one variable (either yield or price) is considered random. Many
times, both are random. In this case, only the log-normal distribution
yields analytic results.

4. Safety Rules
The main appeal of the expected-utility approach to modeling
decisions under risk is that it is derived rigorously from a well-defined
and reasonable set of assumptions about preferences. One disadvan-
tage of this approach is the assumption that individuals know the
probabilities associated with each possible outcome of a prospect
and have based their decision upon this knowledge. In many cases,
the degree of information and computation required may make his
assumption unrealistic. An alternative approach for modeling choices
under risk and uncertainty is embodied by the various “safety rules”.
Safety rules correspond to expected utility in the same way
that classical statistics relate to Bayesian. Safety rules are simple,
reasonable, but somewhat arbitrary. They imply lexicographic rules
and an objective function that is linear in mean and variance,
116 Agricultural Economics and Policy

reflecting a “behavioristic” approach to modeling behavior. The

models are based upon simple decision criteria economists believe
people use in making day-to-day decisions. One common element of
these rules is that they reflect “satisfying” behavior, where people
make choices to meet some objective. Thus, under this approach,
people do not “maximize”, rather they aim to meet a target. A major
proponent of this approach was Herbert Simon, a Nobel Prize winner
in Economics. Simon introduced the notion of “bounded rationality”,
where people are constrained by high computation information and
competition costs, and therefore, they develop simple decision rules.
Simon saw these behavioristic rules as outcomes of optimization
subject to all computation and data costs and constraints, and his
views pose a challenge to economists to identify the constrained
optimization problems that have resulted in the behavioristic rules
whose existence is supported statistically.
The safety rules have multiple advantages. First, they can be
used with programming techniques. Second, they follow the logic of
conventional statistics, avoiding the Bayesian style of expected utility
theory. Third, they can provide a satisfying model of behavior in
many circumstances. Lastly, these models are useful both in positive
and normative analyses. Safety rules are frequently used by engineers
and regulators in constructing nuclear power plants and devising
earthquake regulations.
When economists identify persistent behavioral rules, they can be
used for prediction and analysis of appropriate policy. They can be
particularly relevant in setting government regulations. The safety
rules that we present here are intuitively appealing and are likely
to represent the behavior of at least some people. They correspond
in their structure to decision rules of “classical” statistics, and in
particular, they resemble the use of statistical significance in classical
“hypothesis testing.”

4.1. Roy’s minimum probability rule

Let π, denoting profit, be a random variable whose distribution

depends on a decision variable, X. For example, π = P X − C(X),
where P is a random variable. One safety rule is the Safety First
 Economic Analysis of Behavior under Risk and Uncertainty 117

approach introduced by Roy (1952). Under this approach, decision

makers will choose the decision variable to minimizing the probability
of π falling below a threshold level D. The objective function is

min P rob{π(X) ≤ D} or max P rob{π(X) ≥ D}.

X X

Under this approach, decision makers’ choices are dominated by the

desire to minimize the probability of falling under the threshold
level. This approach corresponds to a lexicographic utility model,
and is useful in modeling subsistence agriculture. If P ∼ N (P̄ , σ 2 ),
then Roy’s safety rule suggests setting the X that minimizes the
probability that a standard-normal random variable is smaller than
D−P̄ X+C(X)
σX . If the cumulative distribution of the standard normal
random variable is denoted by Φ(Z), then Roy’s rule is
 
∗ D − P̄ X + C(X)
X = arg min Φ .
X σX

4.2. Telser’s safety-first rule

An alternative approach is provided by Telser (1955). In his case, the

decision maker’s objective is to maximize expected profit subject to
a bound on the probability of profit falling below the threshold level
D. This model can be written as

max E[π(X)] s.t. Prob{π(x) ≤ D} ≤ α,

X

where α is a tolerance level of profit lower than the threshold.

For our previous example with a normal distribution, the Telser
rule is
 
D − P̄ X + C(X)
max P X − C(X) s.t. Prob Z ≤ ≤ α,
X σX

where Z ∼ N (0, 1) is a standardized normally distributed random

variable.
This rule has application in the development of many engineering
and health codes and is based on the concept of an onerous event in
a manner much like the use of critical levels in classical statistics.
118 Agricultural Economics and Policy

4.3. Kataoka’s safety-fixed rule

Kataoka (1963) developed the safety rule that maximizes a minimum

profit level (threshold level) where the random profit is higher
than this minimum profit with a probability of at least 1 − α.
Mathematically, the safety-fixed rule can be presented as

max D s.t. Prob{π(X) ≥ D} ≥ 1 − α.

X

This safety-fixed rule and the safety-first rule are inversely

related. The statistical significance level α is the parameter of the
safety fixed rule, and its objective is to find the profit distribution
which has the highest level of minimum profit where the cumulative
distribution value is α. The disaster level D is the parameter of Roy’s
safety-first rule, and its objective is to select the profit distribution
with the lowest level of cumulative distribution when profit is equal
to D. Figure 4 demonstrates the two rules graphically.
Suppose we have to choose among three activities. The cumula-
tive distributions of profits of the three activities are denoted by the

CDF CDF

1 1

Safety-First Rule Safety-Fixed Rule

Figure 4. Safety-first rule and safety-fixed rule.

 Economic Analysis of Behavior under Risk and Uncertainty 119

CDF

Figure 5. Choices among three activities based on safety-first and safety-fixed

rules.

Fi functions in Figure 5. Activity 1 is selected under the safety-fixed

rule when α is α1 and under the safety-first rule when D is D1 . When
the safety-fixed rule is not as restrictive and α is α2 , then activity 2
is selected. Activity 2 is also selected under the safety-first rule when
D is D2 .
Continuing with the normality example, we choose a line of
actions to maximize D subject to the constraint:
 
Π(x) − Π̄ D − Π̄
Prob < ≤ α.
σ σ
If the constraint is binding, this amounts to choose D such that
D − Π̄
= Zα ,
σ
where Zα is such that Prob(x < Zα ) = α under the standard normal
distribution. Thus, D = Π̄ + Zα σ, and maximizing D, in the case of
the normal distribution, amounts to maximize the sum of the mean
and a multiple of the standard deviation. If α is small (i.e., α < 0.5),
then Zα is negative and we have some form of risk aversion. We can
120 Agricultural Economics and Policy

see the link between this safety rule and the mean–variance utility
function.

4.4. Safety rules and expected utility

Safety rules can be expressed as special cases of the expected utility

framework. The safety-first rule can be expressed as the outcome
of the expected utility framework when the utility function is
lexicographic. The objective of the safety-first rule is to minimize the
probability of profits below the disaster level D, which is equivalent
to the expected utility maximization when the utility function is
1 when π ≥ D
U (π) =
0 when π < D.

With this utility function, the probability that profit is greater

than D is maximized under Roy’s minimum probability rule. The
main criticism of this minimum probability rule is that lexicographic
utility functions are unreasonable and individuals prefer making
more money than less once their profits are above the disaster level.
However, there are likely to be situations where the safety rules are
good approximations of reality, and their use is justified. Note that
one major flaw of the standard expected utility models is that they
put too much emphasis on income within utility functions, ignoring
other variables, such as social status (e.g., landlord vs. landless
peasant; hired vs. self–employed). If social status is lost when profits
are below a threshold level, and if such a status loss entails a drastic
reduction in welfare, then the use of safety first rule may provide a
good approximation of expected utility outcomes.

5. The Validity of the Expected Utility (EU)

Model
The attractive feature of expected utility (EU) theory is that it
is derived from a set of axioms about human behavior, creating a
reasonable, rational, and desirable basis for normative analysis. The
EU model has been widely embraced by economists for modeling
decisions under risk because of its ease of use, normative appeal
 Economic Analysis of Behavior under Risk and Uncertainty 121

Table 2. Four choice pairs.

Choice pair 1 Choice pair 2

X P (X) chosen X P (X) chosen
Option A 4,000 .80 20% 4,000 .20 65%
Option B 3,000 1 80% 3,000 .25 35%

Choice pair 3 Choice pair 4

X P (X) chosen X P (X) chosen
Option A −4,000 .80 92% −4,000 .20 42%
Option B −3,000 1 8% −3,000 .25 58%

and, arguably, for its reasonable accuracy in predicting behavior

under risk for many economic activities. However, the EU model
has repeatedly been shown to lack descriptive and predictive validity
in experimental settings. These empirical violations of EU have
given rise to the formulation of a large number of alternative
Generalized-EU models, even though not much is known about the
underlying reasons for the occurrence of EU violations. The following
experimental examples show some instances of where individuals
violate expected utility.

5.1. Kahneman and Tversky experiments

Kahneman and Tversky (1979) conducted experiments to test the

validity of the expected utility model. Individuals were asked to select
one of two options (Option A vs. Option B) from the following choice
pairs:
In Table 2, X stands for a random return under Option A or B,
P (X) is the probability of X taking a value under an option, and
“chosen” stands for the percentage of experiment participant chose
an option. For instance, in choice pair 1, Option A offers a $4,000
return at probability 0.8 (and, implicitly, a zero return at probability
0.2). Option B offers a sure $3,000 return. For choice pair 1, 20%
participants chose Option A and 80% chose Option B.
Assume that the utility function is U (·). Then the choice
outcome under pair 1 indicates that U (3000) > 0.8U (4000). Choice
122 Agricultural Economics and Policy

pair 2 indicates that .25U (3, 000) < .2U (4, 000), or U (3, 000) <
.8U (4, 000). We can see that the choices for the pairs 1 and 2
are inconsistent with each other. Similarly, choice pair 3 indicates
that .8U (−4, 000) > U (−3, 000), and choice pair 4 indicates that
.2U (−4, 000) < .25U (−3, 000), or .8U (−4, 000) < U (−3, 000).
Again, inconsistency arises between outcomes in choice pairs 3 and
4. Therefore, the validity or expected utility theory is cast into
doubt. Two major explanations for the apparent failure of the
expected utility model to correctly predict human behavior have
been advanced. First, people under-weigh outcomes that are probable
in comparison to certain outcomes, even when they both have the
same expected utility, which is known as the certainty effect. Second,
people are risk-averse when they gain and risk-loving when they lose;
known as the reflection effect.
Applied economists face a dilemma when choosing between
models of decision-making under risk. They must choose between
either (1) the EU model that has normative appeal but has been
shown to be systematically violated by behavior or (2) one of a
number of generalized models that lack normative appeal and allow
for some behavioral violations of EU. Here, we offer a brief discussion
of the EU model’s implications for behavior, the experimental
violations of EU, the approach of the generalized-EU models, and an
explanation for the occurrence of choices violating EU which offers
direction to applied economists for model selection in risky choice
environments.

5.2. Background: The EU model’s critical

implications for behavior

There are three main axioms in the EU framework. They are defined
over a binary relation where denotes weak preference, denotes
strong preference, and ∼ denotes indifference for preferences over
probability distributions p, q ∈ P that are defined over a common
(discrete or continuous) outcome vector x. The three axioms that
are necessary and sufficient for the EU representation u(·) over
preferences are as follows:
 Economic Analysis of Behavior under Risk and Uncertainty 123

Axiom O (Order): The preference ordering is complete (all

distributions p, q ∈ P are comparable via the ordering ) and
transitive (if p q and q r, then p r).
Axiom I (Independence): For all p, q, r ∈ P , and for all α ∈
(0, 1), if p q, then αp + (1 − α)r αq + (1 − α)r. This axiom holds
that preferences over probability distributions should only depend
on the portions of the distributions that differ (p and q), not on their
common elements (r). This independence of preference with respect
to r holds regardless of r and of the level of α that defines the linear
combination.
Axiom C (Continuity): For all p, q, r ∈ P with p q and
q r, there exist α, β ∈ (0, 1) such that
αp + (1 − α)r q and q βp + (1 − β)r.
This axiom gives a degree of continuity to the preferences.
Axioms O, I, and C can be shown to be necessary and sufficient
(Fishburn, 1983) for the existence of a function u(·) on the outcomes
x ∈ x that represents preferences through . In the discrete case
where p = {p1 , p2 , . . . , pn }, it gives the probabilities of occurrence for
x = {x1 , x2 , . . . , xn }:

n 
n
p q⇔ u(xi )pi ≥ u(xi )qi .
i=1 i=1

Most of the violations of EU hinge on an implication stemming

primarily from the Independence Axiom.
A useful diagram for viewing the implications for models of
behavior under risk was developed by Marschak (1950) and reintro-
duced by Machina (1982). Let us call it Machina–Marschak triangle
(see Figure 6). This triangle in two dimensions has boundaries from
0 to 1 (a simplex) consistent with rules of probability.
Probability distributions defining gambles over three (low,
medium, and high) discrete outcomes are represented by points inside
or on the boundaries of this triangle. The distribution’s probability
for the occurrence of the lowest outcome (xL ) is given on the
horizontal axis, the probability of occurrence for the highest outcome
(xH ) is given on the vertical axis, and the probability of the medium
124 Agricultural Economics and Policy

Indifference curve

Iso-expected value curve

0 1

Figure 6. Machina-Marschak triangle.

outcome (xM ) is given implicitly by 1 less the sum of the probabilities

for the high and low outcomes.
To further explain, points on the hypotenuse represent gambles
with no choice of occurrence for the middle outcome (the sum of
the probabilities for the low and the high outcomes is one), while
the point on the vertex of the triangle opposite to the hypotenuse
represents a gamble giving the middle outcome with certainty.
Preferences over gambles can be illustrated by indifference curves
within the triangle; the EU model holds that these indifference curves
must be parallel as in Figure 6. These indifference curves can be
compared with the iso-expected value curves in the figure, where
the gambles have the same expected values (note: not the expected
utility). The indifference curves shown in Figure 6 indicate that
the individual is risk loving, as increased risk (movements along an
indifference curve to the “Northeast”) require a lower expected value
 Economic Analysis of Behavior under Risk and Uncertainty 125

for indifference to hold. Individuals prefer movements toward the

“Northwest” of the triangle, as the probability of the highest outcome
increases while that of the lowest outcome decreases.

5.2.1. Violations of EU

There have been many advances in the economic analysis of decisions

under risk using the EU model. However, the majority of these papers
take the validity of the EU model as given. A serious challenge to
the use of EU was made by the Nobel Prize winner (in economics)
Maurice Allais soon after its introduction. His work and that of others
following him elicited choices between hypothetical risky alternatives
to show that EU lacked complete predictive, and hence descriptive,
validity.
Some well-known risky choice examples are given in a paper
by Kahneman and Tversky (1979) and that synthesizes work by
Allais and by others who have shown experimental violations of EU.
Kahneman and Tversky (1979) also present a model of choice which
strives for only descriptive (not normative) validity. Their paper
remains a standard in this subject of modeling choice under risk;
in particular, their experimental results have had a great deal of
influence on the literature. The first of Kahneman and Tversky’s
examples showing EU violations discussed here asks individuals to
select between two gambles as shown in Table 3. Note that gamble
A (respectively, C) is less risky than gamble B (respectively, D). On
the other hand, gamble A (respectively, C) has lower expected value
than does gamble B (respectively, D).
The EU model requires that the choice between A and B must be
compatible with the choice between C and D; i.e., if the more risky
alternative B is selected in the first choice, the more risky alternative
D must be selected in the second choice, and vice versa. One of
these choice patterns is required due to the Independence Axiom,
since the probability vectors {pC = (.75, .25, 0), pD = (.8, 0, .2)} over
the outcome vector x = ($0, $3000, $4000) defining alternatives C
and D, respectively, can be viewed as a linear combination of the
probability distributions {pA = (0, 1.0, 0), pB = (.2, 0, .8)} that define
126 Agricultural Economics and Policy

Table 3. Kahneman and Tversky’s experiments.

Choice 1: Select between Gambles A and B

Gamble A Gamble B
$3000 with probability 1.0 $4000 with probability .8,
$0 with probability .2
Choice 2: Select between Gambles C and D
Gamble C Gamble D
$3000 with probability .25, $4000 with probability .2
$0 with probability .75 $0 with probability .8
Choice 3: Select between Gambles E and F.
Gamble E Gamble F
$3000 with probability .9, $6000 with probability .45,
$ 0 with probability .1 $ 0 with probability .55
Choice 4: Select between Gambles G and H.
Gamble G Gamble H
$3000 with probability .002, $6000 with probability .001,
$ 0 with probability .998 $ 0 with probability .999

A and B, respectively. To see this, first let us define a distribution

carrying a certain outcome of $0 based on the outcome vector x =
($0, $3000, $4000) as (1.0, 0, 0). Then we have,

.25 · pA + .75 · (1.0, 0, 0) = (.75, .25, 0) = pC ,

.25 · pB + .75 · (1.0, 0, 0) = (.80, 0, .20) = pD .

In this case, the Independence Axiom implies that if pA pB

then pC pD . In their experiment using hypothetical payoff
outcomes, many (about 65%) of Kahneman and Tversky’s subjects
selected A over B in the first pair but selected D over C in the second
pair of choice, a choice pattern that violates EU. Figure 7 illustrates
this pattern of choice. This inconsistency between expected utility
theory predictions is termed “Certainty Effect” as Gamble A presents
a sure return.
The gamble pairs (A,B) and (C,D) are depicted in a Machina–
Marschak triangle (see Figure 7). One can readily check that lines
 Economic Analysis of Behavior under Risk and Uncertainty 127

Pn

0.8 B

Indifference curve

D
0.2
Indifference curve

A C
0 0.2 0.75 0.8 1 Pe

Figure 7. Gamble pairs (A, B) and (C, D) in a Machina-Marschak triangle.

AB and CD are parallel. These parallel lines are important for the
analysis of choice with respect to the EU model. If the majority
of subjects prefer gamble A to gamble B and prefer gamble D
to gamble C, then some indifference curves must be unparallel,
which contradicts the property under EU that indifference curves
are parallel.
Another well-known EU violation was found by Kahneman and
Tversky where they asked respondents to select between hypothetical
risky alternatives that have equal expected values (see choice pairs
3 and 4 in Table 3). The probabilities defining G and H can also
be written as a linear combination of the probabilities that define E
and F, with (1.0, 0, 0) defining a distribution giving a sure outcome
of $0 (we leave this as an exercise to readers). The EU model
requires consistency of choice: If E is selected in the first pair, G
must be selected in the second. On the other hand, if F is selected
128 Agricultural Economics and Policy

in the first choice, H must be selected in the second. However,

respondents also violated the EU model in their choices over these
two pairs, with most respondents selecting E over F and H over
G. The choice of the riskier H over G also violates second-degree
stochastic dominance predictions for risk averse decision makers as G
second-order stochastically dominates H. Note that the inconsistency
of expected utility predictions illustrated by choice pairs 3 and 4 is
termed “Common Ratio Effect” because the ratio of the winning
probabilities in Gambles E and G ((i.e., .9 to .002)) is the same as
that in Gambles F and H (i.e., .45 to .001).

5.2.2. The generalized-EU models

A number of models have been set forth as alternatives to EU in

light of the behavioral violations of EU such as those developed
by Kahneman and Tversky described above. These models weaken
the Independence Axiom of EU in order to allow for observed
behavioral violations. The upshot of these models is that preferences
are represented through both a function u(x) over the outcomes
and a nonlinear function g(s) for s ∈ {p, q} over the probability
distributions, giving the representation:

n 
n
p q⇔ u(xi )gi (p) ≥ u(xi )gi (q).
i=1 i=1

The function u(x) has similar structure as in the EU model;

the interesting part of these models is the function g(·). Quiggin
(1982) proposed a model where g(·) overweighted extremely small
probabilities when defined over either very low ($0 in the examples)
or relatively high ($4000, $6000 in the examples) outcomes. Later
work by Tversky and Kahneman (1992) incorporated Quiggin’s
structure of this function g(·) and developed the cumulative prospect
theory, which are discussed in the following section.
Another early, well known, and simple to illustrate form of the
generalized-EU models was developed by Machina (1982). In his
model, preferences locally correspond with EU in the same manner
as Taylor’s series approximations for non-stochastic functional forms.
 Economic Analysis of Behavior under Risk and Uncertainty 129

Machina further specified the behavior of the preferences in order

to allow for the empirical violations of EU through a curvature
change in the Machina–Marschak triangle known as “fanning out,”
where the indifference curves become steeper with increases in the
expected values of the gamble (movements toward the northwest)
(see graph (b) in Figure 5 of Machina 1982). This fanning out notion
has been used for other generalized-EU models but is neither strongly
motivated nor predictively accurate.

5.2.3. The similarity model: Alternative explanation

for the paradoxes

An appealing explanation of the patterns of choices showing incon-

sistencies with what EU proposes is that individuals evaluate risky
alternatives differently, dependent on the similarity of the alterna-
tives. This similarity has both objective and subjective connotations.
In risky choice, selection between the more similar alternatives would
likely be both (1) more difficult or (mentally) costly and (2) less
beneficial or important because the alternatives differ little in an
objective sense. As a result, comparisons between two sets of choice
pairs that differ considerably in their degree of similarity (such as
those used to show violations of EU) may give misleading implica-
tions about preferences in the nature of model misspecification, since
both preferences and perceptions are reflected in choices.
To illustrate the application of this similarity idea, consider the
pairs of risky alternatives in the certainty effect and the common
ratio effect examples discussed above respectively. In both of these
examples, one of the choice pairs (Gambles A and B in choice pair 1
as well as E and F in choice pair 3) is quite “dissimilar” as defined
by the Euclidean distance over the probability space:
 n 1
 2

d(p, q) = (pi − qi )2 .
i=1

In addition to distance differences, the pair A and B is qualitatively

different because A gives $3000 with certainty (i.e., no risk at all).
130 Agricultural Economics and Policy

Choices between these dissimilar pairs are compared with choices

between similar pairs (Gambles C and D in choice pair 2 as well as
G and H in choice pair 4). Kahneman and Tversky’s results can be
explained if individuals are more likely to select the riskier (Gambles
D and H) alternative when the alternatives are similar.
Work by Rubinstein (1988) and Leland (1994) has explored some
of the implications of various models of similarity on risky choice,
although the models suggested by these authors are quite limited in
their applications. Buschena and Zilberman (1999) have developed
and tested more general models for the similarity of risky choice
and have found considerable effects of similarity on both the pattern
of choice and on the occurrence of EU violations. The tests show
two primary results. First, as the risky choice pair becomes more
similar (the choice is less critical and the evaluation is more costly),
the riskier alternative is much more likely to be selected. Second,
violations of EU are much more likely to occur when the differences
in the dissimilarity between the two sets of risky choice pairs is large,
i.e., when there is a good deal of dichotomy between the dissimilarity
of the pairs.
The results of the similarity tests in Buschena and Zilberman
(1999) show some results that should be of significant interest to
general and applied economists, namely, the following:

(1) There is an operational and intuitively appealing explanation for

the occurrence of choice patterns that violate EU.
(2) There are a significant number of decisions over risky alternatives
for which the EU model is descriptively accurate. EU works well
for dissimilar choice pairs.
(3) Statistical analysis shows significant violations of the models
set forth as alternatives to EU (generalized-EU and others);
moreover, these violations were in the direction predicted by the
similarity model.

There remain a number of unanswered questions regarding the

effects of similarity on risky choice. Of particular note are those con-
cerning the effects of using real, rather than hypothetical, payoffs on
the influence of similarity on choice. Further questions of interest that
 Economic Analysis of Behavior under Risk and Uncertainty 131

are quite unexplored are the occurrence of EU violations themselves

and also the effects of similarity on the occurrence of violations, for
non-experimental choices (e.g., agricultural production decisions and
risky resource protection issues). There is some evidence (e.g., Bar–
Shira, 1992) that the EU model works well for modeling crop portfolio
choices.
Research modeling and testing the robustness of the findings of
experiments to risky decisions made in “everyday life” is a very fertile
one. This line of research, however, will likely be quite difficult to
carry out given data availability and calls for a good deal of creativity
in experimental design. Some of the recent findings to be discussed
in the following within this chapter indicate promise for models
of behavior incorporating factors reflective of more comprehensive
models of decision-making.

6. Prospect Theory
In recent years, prospect theory has attracted increasing attention
among agricultural economists, with applications for understanding
farmers’ decision-making under risk in various contexts. For instance,
Liu (2013) used prospect theory to explain Chinese farmers’ biotech-
nology adoption decisions, Babcock (2015) applied prospect theory
to understand US farmers’ crop insurance take-up decisions, and
Bocquého et al. (2015) and Anand et al. (2019) employed prospect
theory to study farmers’ adoption decision toward bioenergy crops
in France and the United States, respectively. Recently, Wilson and
Miao (in press) applied prospect theory to interpreting consumers’
food waste behavior. In this section we briefly discuss this theory.
Barberis (2013) provides an outstanding review of prospect theory,
including its applications in finance, insurance, behavioral economics,
labor supply, and industrial organization.
The original prospect theory was developed by Kahneman and
Tversky (1979) in their seminal paper “Prospect Theory: An Analysis
of Decision under Risk” published in Econometrica, where, as we
discussed in the previous section, the authors documented quite a few
examples that predictions from expected utility theory contradicted
132 Agricultural Economics and Policy

experimental results. Although focusing on prospects with only

no more than two non-zero outcomes, Kahneman and Tversky
established the key features of prospect theory: reference point, loss
aversion, and probability weighting. The reference point reflects the
phenomenon that people’s utility is largely determined by the change
in their incomes relative to a benchmark value (i.e., the reference
point). Incomes higher than the reference point are viewed as gains,
whereas incomes lower than the reference point are viewed as losses.
Loss aversion indicates that, for losses and gains with the same
absolute value (e.g., a loss of $100 and a gain of $100), people are
more sensitive to the former than the latter. In other words, the $100
loss would bring in much larger disutility to a decision maker than the
utility brought to her by the $100 gain. Probability weighting reflects
the idea that people tend to overweight small probabilities but
underweight large probabilities. Kahneman and Tversky (1979) argue
that probability weighting can help explain why people buy both
lotteries and insurance. Lotteries bring in gains with extremely small
probability and insurance covers low-chance losses. Then tendency
to overweight small probabilities makes both lotteries and insurance
appealing. A typical value function of prospect theory that reflects
the idea of reference point and loss aversion is depicted in Figure 8.
Probability weighting is depicted in Figure 9.
Let v(·) denote the value function for outcomes under prospect
theory. We normalize the reference to be 0. Suppose that a prospect
yields income x with probability p and income y with probability q,
where p + q = 1. The utility function defined on the prospect can be
written as
⎧
⎨ w(p)v(x) + w(q)v(y) x ≥ 0 ≥ y or y ≥ 0 ≥ x
U (x, p; y, q) =
⎩ v(y) + w(p)[v(x) − v(y)] x > y > 0 or x < y < 0,
(18)
where w(·) is a probability weighting function. In applications (e.g.,
Liu 2013), one can set
w(p) = exp[−(− ln p)α ] and (19)
 Economic Analysis of Behavior under Risk and Uncertainty 133

V(x)

x
Reference point

Figure 8. Value function of prospect theory.

⎧
1−σ
⎪
⎨x for x > 0
v(x) = 0 for x = 0 (20)
⎪
⎩
−λ(−x)1−σ for x < 0,
where α > 0 is a probability weighting measure, λ > 0 is the loss
aversion parameter, and σ ≤ 1 is the risk aversion parameter. When
α ∈ (0, 1), the probability weighting curve, w(p), is inverted S-shaped
(see Figure 9). When λ > 1 (respectively, λ = 1 or λ ∈ (0, 1)), the
decision maker is loss averse (respectively, loss neutral or loss loving).
Finally, when σ < 0 (respectively, σ = 0 or σ ∈ (0, 1)), the decision
maker is risk loving (respectively, risk neutral or risk-averse).
The original prospect theory was further modified by Tversky
and Kahneman (1992) to accommodate prospects with more
than two outcomes and to avoid some implausible predictions
(Barberis, 2013). The new version of prospect theory was called
134 Agricultural Economics and Policy

probability weighting

0 1 probability

Figure 9. Probability weighting function of prospect theory.

“cumulative prospect theory,” where probability weights were

assigned to cumulative probabilities instead of probabilities of
individual events. However, the feature of overweighting small
(cumulative) probabilities and underweighting large (cumulative)
probabilities remained in the new version. We now discuss cumulative
prospect theory.
Suppose a prospect has m + n possible outcomes, denoted by

(x−m , q−m ; x−m+1 , q−m+1 ; . . . ; x−1 , q−1 ; x1 , q1 ; . . . ; xn , qn ),

where m ≥ 1, n ≥ 1, xi ≤ xj ∀i ≤ j, and qk is the probability of xk

occurring, k ∈ {−m, . . . , −1, 1, . . . , n}. Here we still assume that the
reference point value is 0. Leaving the value function (20) unchanged,
the probability weighting function under cumulative prospect theory
is written as
⎧
⎪ φγ
⎨ w+ (φk ) = (φγ +(1−φk k )γ )1/γ for gains
k
(21)
⎪ δ
⎩ w− (φk ) = δ φk δ 1/δ for losses,
(φ +(1−φ ) ) k k
 Economic Analysis of Behavior under Risk and Uncertainty 135

where γ and δ are non-stochastic parameters, and φk is the cumu-

lative probability of outcome xk . This cumulative probability differs
from the usual cumulative distribution function (CDF). If xk > 0,
then φk = Pr{x ≥ xk } (i.e., probability of having a gain no less
than xk ). If xk < 0, then φk = Pr{x ≤ xk } (i.e., probability of
having a loss no less than xk ). One can check that w+ (φk ) and
w− (φk ) are both increasing in φk and with range [0, 1]. Particularly,
w+ (0) = w− (0) = 0 and w+ (1) = w− (1) = 1. The decision weight
assigned to outcome xk is defined as
⎧
⎪ +
⎪
⎪ w (qn ) if k = n
⎪
⎨ w+ (φ ) − w+ (φ ) if 1 ≤ k < n
k k+1
dk = (22)
⎪
⎪ w − (φ ) − w − (φ ) if − m < k ≤ −1
⎪
⎪
k k−1
⎩ −
w (q−m ) if k = −m.
Based on equations (20) and (22), the prospect’s expected value is

n
dk v(xk ). (23)
k=−m

The decision maker will choose prospect that generates the largest
expected value based on Equation (23).
A strand of literature pertains to using experimental meth-
ods to elicit participants’ risk and loss aversion parameters based
on prospect theory. In the realm of agricultural economics and
development economics, examples include Tanaka et al. (2010), Liu
(2013), and Wilson and Miao (2023). These studies typically present
three series of choice experiments between two options with actual
monetary payments to participants (see Table 2 in Liu (2013) for an
example). Within a series of choice experiments, researchers usually
fix one option and increase the attractiveness of the other option.
Here one choice experiment can be something like the following:

Option A Option B

win $20 with pr. 0.3 and $5 win $40 with pr. 0.1 and $3
with prob. 0.7 with prob. 0.9
136 Agricultural Economics and Policy

A series could include 7 to 14 choice experiments like this.

The researchers record the chosen option in each choice experiment
of each series. The switching point, the choice experiment where
the participant switches their choices between Options A and B
when compared with the immediate previous choice experiment
is of critical importance as it reveals much information about a
participant’s preferences. Based on the choices made, the researcher
can estimate a set of parameters (i.e., λ, σ, α) in Equations (19)
and (20) for each participant. It is worth noting that the average
values of λ, the loss aversion parameter, are around 2 across Tanaka
et al. (2010), Liu (2013), and Wilson and Miao (2023), indicating
that participants in their experiments are generally loss averse. Once
the individual parameters about loss aversion, risk aversion, and
probability weighting are obtained, one can associate participants’
risk and loss preferences to some economic variables (e.g., household
income, technology adoption decision, or willingness to pay for a food
item) of the participants. For details about the experiment design and
applications, we refer readers to Tanaka et al. (2010), Liu (2013), and
Wilson and Miao (2023).
Although being able to offer more reasonable predictions than
does expected utility theory for decision-making under risk, prospect
theory, including its modified version (cumulative prospect theory),
still have not replaced expected utility theory as a workhorse in
modeling people’s behaviors under risk. One obvious reason is
that, as one can see, the weighting function and decision weight
in Equations (21) and (22) are quite complex, which makes it
almost impossible to obtain analytical solutions for a model built
upon cumulative prospect theory. Anand et al. (2019) provide an
example of using numerical simulation approach to analyze US
farmers’ bioenergy crop adoption decisions under a prospect theory
framework. Another reason is that it is difficult for researchers to
determine the reference point for a prospect theory model (Barberis,
2013). A common practice is to select multiple reference point for
the model and examine the robustness of results to these reference
points. Babcock (2015) provides an excellent example in this regard.
 Economic Analysis of Behavior under Risk and Uncertainty 137

7. Decision-Making Under Ambiguity

When modeling decision-making under risk, we assume that decision
makers know the probability distribution of prospect outcomes. This
is a strong assumption and often unrealistic. For instance, it is
difficult for farmers to come up with a precise distribution of net
returns from adopting a new technology (e.g., advanced irrigation
system, drought-tolerant seeds, or a new harvester), perhaps because
the new technology in question may become obsolete at an unknown
probability. When we relax this assumption (i.e., known probabil-
ity distribution), we encounter decision-making under ambiguity.
Uncertainty can be viewed as a combination of risk and ambiguity
(Klibanoff et al., 2005; Barham et al., 2014). In what follows, we
discuss some models that have been used to depict decision-making
under ambiguity.

7.1. MaxMin, MaxMax, and α-MaxMin models

Let us start with the MaxMin model, which assumes that people are
pessimistic when viewing probabilities of prospect outcomes. In other
words, under this model, decision makers will always use the worst
possible probability distribution to evaluate the prospect (i.e., Min)
and then select the prospect with the highest minimum value (i.e.,
MaxMin). Suppose a decision maker is facing two options: Option
A that provides a sure return at $10, and Option B that provides
a $20 return with probability p and nothing with probability 1 − p.
However, the decision maker does not know the specific value of p.
She only knows the range of p, which is [p, p̄]. Let U (·) denote the
utility function of the decision maker. Under the MaxMin model, the
decision maker’s choice problem can be written as
!
max min {pU (20) + (1 − p)U (0)}, U (10) .
p∈[p,p̄]

The decision maker will choose Option B if and only if

minp∈[p,p̄] {pU (20) + (1 − p)U (0)} ≥ U (10).
138 Agricultural Economics and Policy

Due to the underlying pessimism of the MaxMin model, it

naturally reflects ambiguity aversion. In contrast, MaxMax assumes
optimistic decision-makers, who always use the best possible prob-
ability distribution when evaluating a prospect. Continuing with
the above example, the decision maker’s choice problem under the
MaxMax model can be written as
!
max max {pU (20) + (1 − p)U (0)}, U (10) .
p∈[p,p̄]

The decision maker will choose Option B if and only if

maxp∈[p,p̄] {pU (20) + (1 − p)U (0)} ≥ U (10). It is not surprising that
the MaxMax model predicts ambiguity seeking behaviors.
The MaxMin model and the MaxMax model are two extremes,
and none of them can accommodate both ambiguity averse and
ambiguity seeking behaviors. A combination of the two models,
termed α−MaxMin model, overcomes this limitation. Continuing
with the above example with Options A and B, we can write the
α−MaxMin model as

VαMaxMin (B) ≡ α · min {pU (20) + (1 − p)U (0)}

p∈[p,p̄]

+ (1 − α) · max {pU (20) + (1 − p)U (0)}, (24)

p∈[p,p̄]

where α ∈ [0, 1] can be viewed as a weight assigned to the MaxMin

model. Its value reflects the magnitude of ambiguity aversion of the
decision maker. One can readily check that when α = 1, then the
α-MaxMin model degenerates into the MaxMin model, and when
α = 0, then the α-MaxMin model degenerates into the MaxMax
model. Under the α-MaxMin model, the decision maker will choose
Option B if and only if VαMaxMin (B) ≥ U (10). Dimmock et al. (2015)
demonstrate that the α-MaxMin model can predict both ambiguity
averse and seeking behaviors.
 Economic Analysis of Behavior under Risk and Uncertainty 139

7.2. Impact of partial insurance under ambiguity:

A MaxMin model

By using a MaxMin model, Bryan (2019) shows that partial insur-

ance, whose indemnity payments are not directly related to a farmer’s
actual indemnity (e.g., rainfall index insurance), may hinder farmers’
adoption of new crop varieties. Here we briefly summarize the model.
Suppose farmers in a region originally plant a traditional crop,
from which they gain utility V T = α + A + i , where α is the average
utility from growing the traditional crop, A is disutility caused by
ambiguity aversion for ambiguity averse farmers, and i is a uniformly
distributed error term associated with each farmer i. Now, suppose
that the local government or an organization promotes a modern crop
variety. Their strategy is to offer rainfall index insurance to farmers
who adopt the modern variety. Note that the utility of growing the
traditional crop is not affected by the rainfall index insurance because
the insurance is unavailable to farmers who grow the traditional crop.
Consider a rainfall index insurance policy that generates a net
payment (i.e., indemnity payment minus premium) with value I when
rainfall is low (i.e., rainfall state RL ) and costs the policyholder
a premium with value P when rainfall is high (i.e., rainfall state
RH ). Assume that only two crop yield stats exist: low yield, yL , and
high yield, yH . Therefore, in total we have four states to consider in
the model, namely, (yH , RH ), (yH , RL ), (yL , RH ), and (yL , RL ). Let
λH and pH denote the probability of high yield and high rainfall,
respectively. The probability of high rainfall and low yield (i.e., state
(yL , RH )) is denoted by q. Therefore, we can readily check that the
probability of state (yH , RH ) is pH − q, (yH , RL ) is λH − pH + q,
and (yL , RL ) is 1 − λH − q. We assume that the policyholder knows
the exact values of λH and pH and that she is uncertain about the
value of q, just knowing that q ∈ [q, q̄]. Here the justification is that
farmers have good sense about the chance of having high yield or
high rainfall in a typical year; however, it is a bit difficult to have an
accurate estimate about the chance of having a low yield in a high
rainfall year. Also note that state (yL , RH ) is the worst state to the
140 Agricultural Economics and Policy

policyholder because, under this state, she suffers from low yield but
will not receive any indemnity payment.
If there is no rainfall index insurance at all, the expected utility
M
from growing the modern crop variety is simply VNoInsu ≡ λH U (yH )+
(1 − λH )U (yL ), from which we can see that q does not appear in the
expected utility. Farmers will adopt the modern crop variety if and
M
only if VNoInsu ≥ V T.
With the rainfall index insurance, for a given probability of state
(yL , RH ), q, and net payment under low rainfall state, I, the expected
utility from growing the modern variety is
M
VInsu (q, I) ≡ qU (yL − P ) + (1 − λH − q)U (yL + I)
+ (pH − q)U (yH − P ) + (λH − pH + q)U (yH + I).

Under the assumption that U  (·) < 0 < U  (·) and yL < yH , one can
M (q, I)/∂q < 0. This indicates that, with a
readily check that ∂VInsu
MaxMin preference, a decision maker’s expected utility from growing
M (q̄, I) and she will adopt the modern
the modern crop variety is VInsu
crop variety if and only if
M
VInsu (q̄, I) ≥ α + A + .

Bryan (2019) compares the impact of the rainfall index insurance

on an ambiguity-averse farmer’s incentive to adopt the modern crop
variety and that on an ambiguity neutral farmer’s. The ambiguity-
neutral farmer is assumed to have a belief of q at q SEU ∈ [q, q̄], and
hence her expected utility from growing the modern crop variety
M (q SEU , I). Clearly, because ∂V M (q, I)/∂q < 0, we have
is VInsu Insu
M (q̄, I) ≤ V M (q SEU , I). Therefore, one can draw the conclusion
VInsu Insu
that the impact of the rainfall index insurance on an ambiguity averse
farmer’s incentive to adopt the modern crop variety is smaller than
that on an ambiguity neutral farmer’s incentive (see Prediction 1 in
Bryan, 2019).
One can further explore whether a mandatory rainfall index
insurance will reduce ambiguity averse farmers’ incentive to adopt
the modern crop variety. Note that, unlike complete insurance,
partial insurance, even if actuarially fair, may make policyholders
 Economic Analysis of Behavior under Risk and Uncertainty 141

worse off if the basis risk is high (e.g., q is large in the above

example). This is because if the probability of high rainfall but low
yield is large, then the policyholders will often not get indemnity
payment when crop yield is low but still need to pay the insurance
premium. Consequently, the rainfall index insurance is actually make
the policyholder’s farming returns riskier than the returns without
insurance. Given the pessimistic nature of decision-makers under the
MaxMin model, an ambiguity averse farmer’s expected utility from
growing the modern crop variety in the presence of rainfall index
insurance may be lower than that from growing the modern variety in
the absence of the insurance. Therefore, the rainfall index insurance
may actually disincentivize the adoption of the new crop variety, even
though its intent is to increase the adoption.

7.3. Technology adoption under ambiguity:

An α-MaxMin model

As we have discussed in Section 7.1, the MaxMin model is incon-

sistent with ambiguity-seeking behaviors and the MaxMax model
cannot accommodate ambiguity averse behaviors. The α-MaxMin
model is a hybrid of the MaxMin and MaxMax models and can
predict both risk averse and risk-seeking behaviors (Dimmock et al.,
2015). In this section, we present a simple α-MaxMin model in the
context of technology adoption based on the model specification
illustrated in Dimmock et al. (2015).
Suppose a farmer is facing two technologies: a traditional tech-
nology and a new technology. For simplicity, we assume that the
traditional technology will provide a sure return at $1,000. The new
technology will provide a $2,000 return with probability p and a
$800 return with probability 1 − p. Further assume that the farmer’s
utility function is U (·) such that U  < 0 < U  . If probability p were
known, the expected utility theory would predict the farmer’s choice
by comparing U (1000) and pU (2000) + (1 − p)U (800). When p is
unknown to the farmer, however, the usual expected utility theory
framework discussed in early sections in this chapter that is used to
model decision-making under risk does not apply.
142 Agricultural Economics and Policy

The α-MaxMin model offers a useful framework when p is

unknown to the farmer. Define Cδ ≡ [(1 − δ)π, (1 − δ)π + δ] as the
prior probability distribution set for the farmer, where π ∈ [0, 1] is
the reference probability for having the $2000 return, and δ ∈ [0, 1] is
the perceived level of ambiguity (or, 1 − δ can be viewed as a degree
of confidence in π). Under the α-MaxMin model, a farmer with prior
probability distribution set Cδ will evaluate the new technology as

V (NewTech) = α · min {pU (2000) + (1 − p)U (800)}

p∈Cδ

+ (1 − α) · max {pU (2000) + (1 − p)U (800)} . (25)

p∈Cδ

The farmer will then compare V (NewTech) with U (1000) (the

utility from the traditional technology) and make technology adop-
tion decisions. One can readily check that, when δ = 0 (i.e., no
perceived ambiguity), the prior distribution set Cδ = π and the
α-MaxMin model in Equation (25) degenerates into an expected
utility theory model under risk: V (NewTech)|δ=0 = pU (2000) +
(1 − p)U (800).
One challenge of applying the α-MaxMin model is parameteriza-
tion (i.e., obtaining values for α, δ, and π). Dimmock et al. (2015)
present an experimental approach to elicit and estimate people’s
ambiguity aversion parameter (i.e., α) and ambiguity perception
parameter (i.e., δ). The approach has been applied by Mitra et al.
(2023) to studying the roles of date labels in determining food waste.
Based on a representative sample of U.S. population, Dimmock et al.
(2015) estimate that the values of α and δ are about 0.56 and
0.4, respectively. Certainly, the values of the two parameters would
vary across sub-samples of the population and specific prospects in
question. Particularly, similar to determining the reference point in
prospect theory, the reference probability value in the α-MaxMin
model highly depends on decision-makers’ experience and decision-
making context for each research project. There is no unique
“correct” value for π. Thus, we may want to examine the sensitivity
of α-MaxMin model results to the values of π.
 Economic Analysis of Behavior under Risk and Uncertainty 143

7.4. Technology adoption under ambiguity:

A different approach

Sometimes, the decision environment involves both risk and ambi-

guity. Barham et al. (2014) define this combination of risk and
ambiguity as “uncertainty” and present a model that can be used
to differentiate the roles of risk and ambiguity. In this section, we
present a simplified version of this model.
Let us continue with the previous example of technology adop-
tion. However, we will add a slight twist: Instead of assuming a prior
probability distribution set as under the α-MaxMin model, here we
assume that p, the probability of $2,000 occurring, is random and
has cumulative distribution function F (p). In addition to the von
Neumann–Morgenstern utility function U (·), the model introduces
ambiguity preference function h(·) with h > 0 > h .3 The model
evaluates the new technology as

W (NewTech) ≡ Ep [h(pU (2000) + (1 − p)U (800))] . (26)

By comparing the values of W (NewTech) and h(U (1000)), the model

can predict the decision maker’s choice between the new and old
technologies.
Following the idea of certainty equivalence, Barham et al. (2014)
define uncertainty premium and decompose it into risk premium and
ambiguity premium. First, one calculates the ex ante mean return of
new technology as M (NewTech) ≡ Ep [p · 2000 + (1 − p) · 800]. The
uncertainty premium is then define as a fixed amount of returns R
such that

W (NewTech) = h(U (M (NewTech) − R)), (27)

where W (NewTech) is defined in equation (26). Intuitively, the

uncertainty premium R can be viewed as the decision maker’s

3
Klibanoff et al. (2005) show that when the curvature of h(·) is large enough,
then the model reduces to the MaxMin model.
144 Agricultural Economics and Policy

willingness to pay to exchange the uncertain returns from the new

technology with the certain, mean payment M (NewTech).
Second, one identifies the ambiguity premium, Ra . Let μp be the
1
mean value of p, i.e., μp = 0 pdF (p). Then Ra is a fixed amount of
returns such that
W (NewTech) = h(μp U (2000 − Ra ) + (1 − μp )U (800 − Ra )), (28)
where W (NewTech) is defined in equation (26). Intuitively, the
ambiguity premium Ra can be understood as the decision maker’s
willingness to pay to eliminate the ambiguity of p and to fix the p
value at μp . The risk premium is defined as Rr ≡ R − Ra . Therefore,
Equation (27) can be rewritten as
W (NewTech) = h(U (M (NewTech) − Ra − Rr )), (29)
from which we can see that for prospects with the same ex ante mean
returns and risk premium, the ones with smaller ambiguity premium
(Ra ) provide the farmer with higher expected utility. Based on field
experiments and survey data collected in Minnesota and Wisconsin,
Barham et al. (2014) find that ambiguity aversion promotes the
adoption of genetically modified (GM) corn. They argue that GM
corn reduces return ambiguity due to their insect-resistance trait.
One can see that the model in Barham et al. (2014) interpret
ambiguity as “risk of risk,” assigning a (known) probability distri-
bution to the distribution of prospect returns. The MaxMin and
MaxMax models can be viewed as special cases of the Barham et al.
(2014) model, where the MaxMin model assigns probability 1 to
the worst probability distribution and the MaxMax model assigns
probability 1 to the best probability distribution.

8. Conclusion
Risk is crucial in agricultural production and marketing. Research on
risk is work in progress and it is critical for policy design (e.g., crop
insurance and investment incentives) and policy welfare analysis. In
this chapter, we have reviewed some common tools to measure and
model risks in agriculture, as well as to estimate risk parameters.
 Economic Analysis of Behavior under Risk and Uncertainty 145

These tools are important for developing risk management policies

and for future research work on risk. Traditional risk analysis based
on rational and standard neoclassical models fails to predict people’s
behaviors under risk and uncertainty in many occasions. New insights
from behavioral economics (e.g., loss aversion and decision-making
under ambiguity) should be incorporated when modeling policy and
developing analytical tools (Wuepper et al. 2023).

References
Anand, M., R. Miao, and M. Khanna. 2019. Adopting Bioenergy Crops:
Does Farmers’ Attitude toward Loss Matter? Agricultural Economics 50(4):
435–450.
Arrow, K. J. 1971. Essays in the Theory of Risk Bearing, Chapter 3 (pp. 90–133).
Chicago: Markham Publishing Company.
Babcock, B. 2015. Using Cumulative Prospect Theory to Explain Anomalous
Crop Insurance Coverage Choice. American Journal of Agricultural Eco-
nomics 97(5): 1371–1384.
Bar-Shira, Ziv. 1992. Nonparametric Test of the Expected Utility Hypothesis.
American Journal of Agricultural Economics 74(3): 523–533.
Barham, B. L., C. Jean-Paul, D. Fitz, Vanessa Rı́os Salas, and L. Schechter.
2014. The Roles of Risk and Ambiguity in Technology Adoption. Journal of
Economic Behavior & Organization 97: 204–218.
Barberis, N. C. 2013. Thirty Years of Prospect Theory in Economics: A Review
and Assessment. Journal of Economic Perspectives, 27(1): 173–96.
Batra, R. N. and A. Ullah. 1974. Competitive Firm and the Theory of Input
Demand under Price Uncertainty. Journal of Political Economy 82(3):
537–548.
Binswanger, H. P. 1980. Attitudes Toward Risk: Experimental Measurement in
Rural India. American Journal of Agricultural Economics 62(3): 395–407.
Bocquého, G., Jacquet, F., and Reynaud, A. 2015. Adoption of Perennial Crops
and Behavioral Risk Preferences. An Empirical Investigation among French
Farmers. Paper presented in the “Risk and Environment” Session at Journées
de Recherche en Sciences Sociales, SFER, Nancy, France. December 11–12.
Bryan, G. 2019. Ambiguity Aversion Decreases the Impact of Partial Insurance:
Evidence from African Farmers. Journal of the European Economic Associ-
ation 17(5): 1428–1469.
Buschena, D. E. and D. Zilberman. 1999. Testing the Effects of Similarity on
Risky Choice: Implications for Violations of Expected Utility. Theory and
Decision Volume 46(3): 253–280.
Chalfant, J. A., R. N. Collender, and S. Subramanian. 1990. The Mean and Vari-
ance of the Mean-Variance Decision Rule. American Journal of Agricultural
Economics 72(4): 966–974.
146 Agricultural Economics and Policy

Chavas, J. 1993. The Ricardian Rent and the Allocation of Land Under
Uncertainty. European Review of Agricultural Economics 20(4): 451–469.
Chavas, J. 2004. Risk Analysis in Theory and Practice. Elsevier Academic Press,
San Diego, California.
Day, R. H. 1965. Probability Distributions of Field Crop Yields. American Journal
of Agricultural Economics 47(3): 713–741.
Dimmock, S. G., R. Kouwenberg, O. S. Mitchell, and K. Peijnenburg. 2015.
Estimating Ambiguity Preferences and Perceptions in Multiple Prior Models:
Evidence from the Field. Journal of Risk and Uncertainty 51(3): 219–244.
Feder, G. 1977. The Impact of Uncertainty in a Class of Objective Functions.
Journal of Economic Theory 16(2): 504–512.
Feder, G. 1980. Farm Size, Risk Aversion and the Adoption of New Technology
under Uncertainty. Oxford Economic Papers 32(2): 263–283.
Feder, G., R. E. Just, and A. Schmitz. 1980. Futures Markets and the Theory
of the Firm under Price Uncertainty. The Quarterly Journal of Economics
94(2): 317–328.
Fishburn, P. C. 1983. Transitive Measurable Utility. Journal of Economic Theory
31(2): 293–317.
Freund, R. J. 1956. The Introduction of Risk into a Programming Model.
Econometrica 24(3): 253–263.
Friedman, M. and L. J. Savage. 1948. The Utility Analysis of Choices Involving
Risk. Journal of Political Economy 56(4): 279–304.
Hadar, J. and W. R. Russell. 1969. Rules for Ordering Uncertain Prospects. The
American Economic Review 59(1): 25–34.
Just, R. E. and D. Zilberman. 1983. Stochastic Structure, Farm Size and
Technology Adoption in Developing Agriculture. Oxford Economic Papers
35(2): 307–328.
Just, R. E. and R. D. Pope. 1978. Stochastic Specification of Production Functions
and Economic Implications. Journal of Econometrics 7(1): 67–86.
Kahneman, D. and A. Tversky. 1979. Prospect Theory: An Analysis of Decision
under Risk. Econometrica 47(2): 263–291.
Kataoka, S. 1963. A Stochastic Programming Model. Econometrica 31(1/2):
181–196.
Klibanoff, P., M. Marinacci, S. Mukerji. 2005. A Smooth Model of Decision
Making Under Ambiguity. Econometrica 73(6): 1849–1892.
Leland, J. W. 1994. Generalized Similarity Judgments: An Alternative Explana-
tion for Choice Anomalies. Journal of Risk and Uncertainty 9(2): 151–172.
Liu, E. 2013. Time to Change What to Sow: Risk Preferences and Technology
Adoption Decisions of Cotton Farmers in China. The Review of Economics
and Statistics, 95(4), 1386–1403.
Machina, M. J. 1982. “Expected Utility” Analysis Without the Independence
Axiom. Econometrica 50(2): 227–323.
Markowitz, H. 1959. Portfolio Section, New York: Wiley.
Marra, M. C. and G. A. Carlson. 1990. The Decision to Double Crop: An
Application of Expected Utility Theory Using Stein’s Theorem. American
Journal of Agricultural Economics 72(2): 337–345.
 Economic Analysis of Behavior under Risk and Uncertainty 147

Marschak, J. 1950. Rational Behavior, Uncertain Prospects, and Measurable

Utility. Econometrica 18(2): 111–141.
Miao, R. and M. Khanna. 2017. Costs of Meeting a Cellulosic Biofuel Mandate
with Perennial Energy Crops: Implications for Policy. Energy Economics 64:
321–334.
Mitra, P., R. Miao, and N. L. W. Wilson. 2023. Ambiguity, Date Labels, and Food
Waste. Working paper. Department of Agricultural Economics and Rural
Sociology, Auburn University.
Pratt, J. W. Risk Aversion in the Small and in the Large. Econometrica 32(1/2):
122–136.
Quiggin, J. 1982. A Theory of Anticipated Utility. Journal of Economic Behavior
and Organization 3(4): 323–343.
Rothschild, M. and J. E. Stiglitz. 1970. Increasing Risk: I. A Definition. Journal
of Economic Theory 2(3): 225–243.
Roy, A. D. 1952. Safety First and the Holding of Assets. Econometrica 20(3):
431–449.
Rubinstein, A. 1988. Similarity and Decision Making Under Risk: Is there a Utility
Theory Resolution to the Allais Paradox? Journal of Economic Theory 46(1):
145–153.
Saha, A. 1993. Expo-Power Utility: A ‘Flexible’ Form for Absolute and Relative
Risk Aversion. American Journal of Agricultural Economics 75(4): 905–913.
Sandmo, A. 1971. On the Theory of the Competitive Firm under Price Uncer-
tainty. American Economic Review 61(1): 65–73.
Tanaka, T., C. F. Camerer, and Q. Nguyen. 2010. Risk and Time Preferences:
Linking Experimental and Household Survey Data from Vietnam. American
Economic Review, 100(1): 557–571.
Telser, L. G. 1955. Safety First and Hedging. The Review of Economic Studies
23(1): 1–16.
Tobin, J. E. 1958. Liquidity Preference as Behaviour towards Risk. Review of
Economic Studies 25(2): 65–86.
Tversky, A. and D. Kahneman. 1992. Advances in Prospect Theory: Cumula-
tive Representation of Uncertainty. Journal of Risk and Uncertainty 5(4):
297–323.
von Neumann, J. and O. Morgenstern. 1944. Theory of Games and Economic
Behavior. Princeton University Press, Princeton, New Jersey.
Wuepper, D., S. Bukchin-Peles, D. Just, D. Zilberman. 2023. Behavioral Agri-
cultural Economics. Applied Economic Perspectives and Policy 45(4): 2094–
2105, https://doi.org/10.1002/aepp.13343.
Whitmore, G. A. and M. C. Findlay. 1978. Stochastic Dominance: An Approach
to Decision-Making under Risk. Lexington Books, D.C. Heath and Co.,
Lexington, MA.
Wilson, N. L. W. and R. Miao. in press. Food Waste, Date Labels, and Lost
Aversion: An Experimental Exploration. Applied Economic Perspectives and
Policy. doi: https://doi.org/10.1002/aepp.13507
Yassour, J., D. Zilberman, and G. C. Rausser. 1981. Optimal Choices among
Alternative Technologies with Stochastic Yield. American Journal of Agri-
cultural Economics 63(4): 718–723.
This page intentionally left blank
 Chapter 5

Welfare Analysis of Agricultural

Policies

Once an economic policy is implemented, it has the potential to

create a large impact on a society through affecting the decisions
of consumers and producers in the economy. One way to quantify
the policy impact is to conduct welfare analysis, whereby the policy
impact on consumers, producers, and government is examined,
and the efficiency of resource allocation is evaluated. The net
welfare impact of a policy in question is the sum of the welfare
impact on these three economic entities. Due to the prevalence
of agricultural policies (e.g., price support, acreage control, risk
management interventions, and input use regulations) in the United
States and many other countries, agricultural economics research has
a long tradition to conduct welfare analysis. In this chapter, we first
outline the basics of welfare analysis of some common agricultural
policies and then discuss a few advanced applications pertaining to
some major agricultural policies in the United States, such as price
support, input regulations, crop insurance, and biofuel policies. For a
comprehensive treatment of welfare analysis, we refer readers to Just
et al. (2004). Atkinson (2011) makes an appealing argument that
economists should resume their focus on welfare analysis in teaching
and research.

149
150 Agricultural Economics and Policy

1. Basic Welfare Analysis

Two groups of participants are essential for almost any markets of
goods: consumers and suppliers. We use the demand curve to measure
consumers’ willingness to pay (WTP) for a specific good (e.g., corn)
and the supply curve to measure suppliers’ willingness to sell (WTS)
of such a good. WTP of a consumer for a good is the maximum price
under which this consumer would buy, and WTS of a supplier is the
minimum price under which this supplier would sell. Imagine that we
have numerous consumers and suppliers in a market with a range of
WTP and WTS values. By plotting out these WTP and WTS values
we obtain the demand and supply curves of this market. Based on
these two curves we can construct consumer surplus and producer
surplus, two key concepts in welfare analysis. Consumer surplus is
defined as the area below the demand curve but above a price paid
by consumers (i.e., the aggregate of the difference between price paid
and WTP across all consumers), whereas producer surplus is defined
as the area above the supply curve but below the price received by the
producer (i.e., the aggregate of the difference between price received
and WTS across all suppliers).
Government may often intervene the market as another impor-
tant player. With various policies, government intervention alters
consumer and producer surplus, creating government revenue or
expenditure. The total social welfare is the sum of consumer surplus,
producer surplus, and government revenue (or negative expenditure).
Figure 1 provides an example of the definition of consumer
surplus and producer surplus. Without policy intervention in the
market, the demand curve and supply curves intersect at point
(Q∗ , P ∗ ) where the market clears. In this case, the consumer surplus
is area a + b + d and the producer surplus is area c + e + l. The
total social welfare is, therefore, a + b + c + d + e + l. Government
policies that causes the market to deviate from (Q∗ , P ∗ ) will create
social welfare losses (i.e., deadweight loss). Next, we will focus on
some major agricultural policies and analyze their welfare impact.
In developed countries, governments often implement policies
such as price support and supply controls. A price support sets a
 Welfare Analysis of Agricultural Policies 151

Supply

b i
d
e h
c
f g
m
l k j Demand
0

Figure 1. Welfare analysis of price support and quantitative restriction.

minimum price that is higher than the market equilibrium price so

that producers’ surplus can be increased. In contrast, supply controls
aim to increase commodity prices by reducing the supply, thereby
increasing producer surplus. In this section, we briefly discuss the
welfare analysis of these two types of policies. In Section 2 we analyze
crop insurance program’s welfare impact. In Section 3 we explore
welfare analysis of overlapped policies. Section 4 concludes.

1.1. Price support

The first price support programs in the United States were imple-
mented during the Great Depression with the creation of the
Commodity Credit Corporation (CCC). Although the specific price
support programs evolved over time (e.g., from loan support to
deficiency payment), the essence of these programs was to guarantee
a minimum price for covered commodities. Intuitively, price support
would increase the quantity supplied and decrease the quantity
demanded in the market, and, therefore, would initiate a supply
surplus in the market, which the federal government would purchase
in the early period of the price support programs (e.g., loan support).
However, due to the high costs of storage the government had to pay
for the surplus, later the government shifted to deficiency payments
wherein farmers sold their products at market prices but could get
152 Agricultural Economics and Policy

the price difference from the government if the market price is lower
than the support price (i.e., deficiency payment).
We now show an example about the welfare analysis of price
support. Suppose that Figure 1 depicts the supply and demand
in corn market. When there is no policy intervention, the market
equilibrium quantity and price are Q∗ and P ∗ , respectively. The
consumer surplus is area a + b + d, and the producer surplus is
c + e + l. The total social welfare is,

S 0 = a + b + c + d + e + l. (1)

Now the government introduces a price support program that sets

the price of corn at P which is higher than P ∗ . Due to the increased
price, the new quantity demanded QD is lower than the new quantity
supplied Qs (i.e., QS > QD ), and the consumer surplus shrinks
to area a. If the government purchases the overproduced quantity
QS − QD at price P , then the producer surplus would become area
b + c + d + e + i + l. The government outlay is area d + e + f + g + h +
i + j + k. Therefore, the total social welfare under the price support
program is

S 1 = a + b + c + d + e + i + l − (d + e + f + g + h + i + j + k)
= a + b + c + l − (f + g + h + j + k) . (2)

By comparing S 0 and S 1 , we can see the deadweight loss of the price

support policy is

S 0 − S 1 = d + e + f + g + h + j + k. (3)

Note that the deadweight loss would remain the same regardless
of whether the government purchases the overproduced quantity
(QS − QD ). To illustrate, when the government does not purchase
the overproduced quantity, the cost of overproduction falls on the
producer, whose surplus is area b + c + l − (f + g + h + j + k). The

total social welfare is S 1 = a + b + c + l − (f + g + h + j + k). By

comparing S 0 and S 1 we have

S 0 − S 1 = d + e + f + g + h + j + k, (4)
 Welfare Analysis of Agricultural Policies 153

which is the same as S 0 − S 1 . When the government purchases

overproduced quantity QS −QD , then the cost of overproduction falls
on taxpayers, but the net total social welfare change is the same.
Let us now consider how the overproduction is disposed. If the
government redistributes the purchased overproduction to consumers
for free, then consumer surplus will increase by d + e + f + g + j + k,
and the total social welfare change will be −h, the deadweight
loss. Under deficiency payment, a more realistic scenario could
be this: Knowing that the government will guarantee a price at
P , farmers will produce at quantity QS . The overproduction will
drive market price down to P̂ . Farmers sell their commodity at
price P̂ in the market, and get price difference P − P̂ from the
government. Therefore, with the deficiency payment, the consumer
surplus will be the area below the demand curve but above price
P̂ (i.e., area a + b + c + d + e + f + g + m), the producer
surplus will be the area above the supply curve but below price
P (i.e., area b + c + d + e + i + l), and government expenditure
will be (P − P̂ )QS = b + c + d + e + f + g + h + i + m. As
opposed to the surpluses without any policy interventions, under
deficiency payment consumer surplus increases by c + e + f + g + m,
and producer surplus increases by b + d + i. Considering government
expenditure, the change in total social welfare is still −h, the dead-
weight loss. This social welfare analysis under deficiency payment
will be used in the next section when we discuss welfare analysis of
overlapped policies.

1.2. Supply controls

Another policy instrument that can increase commodity price is

quantitative restriction. Let us continue with Figure 1. Instead of
setting a target price for corn, the government can set a maximum
quantity of corn that can be produced, e.g., QD . In this case, the
price of corn will be P . The consumer surplus and producer surplus
become area a and area b + c + l, respectively. Therefore, the total
social welfare under the quantity restriction is

S 2 = a + b + c + l. (5)
154 Agricultural Economics and Policy

By comparing S 0 and S 2 we have

S 0 − S 2 = d + e. (6)
However, in reality it is difficult to put an effective quantitative
restriction as it is often politically challenged by both consumers and
producers and motivates black market (think about Prohibition in
the U.S.). An alternative way to implement supply controls is for the
government to pay farmers to idle part of their land, which is often
called set-aside programs. A prominent example is the Conservation
Reserve Program in the United States, where the government pays an
annual rent to have farmers keep their land out of crop production
for 10–15 years. Figure 2 presents welfare effects of such set-aside
programs. After the program is implemented, the supply curve will
be shifted upward because for any given price, less land will be used
for production and therefore less quantity will be produced. The
consumer surplus will shift from area a + b + c to area a, with the net
loss equal to area b + c. The producer surplus will be changed from
area d + e to area b + e. Denote ΔG as the government outlay under
the program. Now, the net total social welfare loss is c + d + ΔG.
However, the set-aside program often generates environmental ben-
efits because keeping cropland out of production reduces fertilizer
and pesticide uses which are some major pollutants nowadays. From
this, with ΔE denoting environmental benefits from the set-aside

Supply

b c
e
d

Demand
0

Figure 2. Welfare impact of set aside programs.

 Welfare Analysis of Agricultural Policies 155

program, the net total social welfare loss caused by the set-aside
program is c + d + ΔG − ΔE.

2. Crop Insurance
Since the passage of the Federal Crop Insurance Act (FCIA) of 1980,
the federal crop insurance program has developed into a major pillar
of agricultural policy in U.S. agriculture. Neo-classical economics
predicts that a risk averse farmer would purchase crop insurance
if it is actuarially fair, but crop insurance take-up rate had been low,
even with a moderate premium subsidy rate (e.g., 30% of insurance
premium) as specified in the FCIA of 1980. Later legislations further
increased the take-up rate by significantly increasing premium
subsidy levels. Whether or not the government should subsidize crop
insurance has been debated by agricultural economists since 1980. We
refer readers to the U.S. Government Accountability Office (USGAO)
(2023), Coble and Barnett (2013), and Goodwin and Smith (2013)
for further discussions about premium subsidies.
In this section, we discuss the welfare impact of crop insurance
and of its associated policy (namely, premium subsidy), by following
a framework documented in Chetty and Finkelstein (2013). Assume
that there is a continuum of farmers that differ from each other in
terms of unobserved characteristics, . Let G() denote the cumula-
tive distribution function of . Assume that all farmers are facing one
type of crop insurance with premium p and are considering whether
or not they will purchase the insurance. Denote the expected utility
of farmer i with crop insurance by u1 (i , p), and without insurance
by u0 (i ). Farmer i will purchase the insurance if and only if
u1 (i , p) ≥ u0 (i ). Define π (i ) as the maximum
 willingness to pay of
1 0
farmer i for crop insurance, i.e., π (i ) ≡ max p : u (i , p) ≥ u (i ) .
The demand for crop insurance can be written as,
D(p) = ∫ 1 (π() ≥ p) dG(), (7)
where 1(·) is an indicator function.
Here we further assume that the cost of providing crop insurance
only includes the expected indemnity payment to farmers. In other
156 Agricultural Economics and Policy

words, there is no cost to administrate the insurance program (i.e.,

no loading factors). Denote the expected indemnity payment for
farmer i by c(i ). Therefore, the average cost and marginal cost of
the insurance program are

1
AC(p) = c()1 (π() ≥ p) dG() = E (c() | π() ≥ p), and
D(p)
(8)
MC(p) = E(c() | π() = p), (9)

where E(·) is the expectation operator. Chetty and Finkelstein (2013)

further assume that the marginal cost curve and the demand curve
intersect no more than once, which guarantees the existence and
uniqueness of the insurance market equilibrium.
Due to adverse selection, the marginal cost curve of providing
crop insurance is downward sloping: farmers who expect to have
larger indemnity payment (i.e., cost for insurance providers) will be
more willing to pay for the insurance. Furthermore, the downward
sloping marginal curve implies a downward sloping average cost
curve. Figure 3 depicts curves of demand, average cost, and marginal
cost associated with this insurance contract. Because the risk factor
i is unobservable to the insurer, the insurer cannot charge a
differentiated price based on the risk factor, they can only charge a

Demand

AC

MC
0

Figure 3. Adverse selection in crop insurance market.

Source: Adapted from Figure 2 in Chetty and Finkelstein (2013).
 Welfare Analysis of Agricultural Policies 157

uniform price to all buyers of this insurance contract. In the insurance

market, the competitive equilibrium is reached at (q̂, p̂), where the
average cost curve intersects the demand curve (see Figure 3). It is
readily checked that, assuming no loading factors, the insurer’s profit
is zero in the competitive equilibrium.
Following Chetty and Finkelstein (2013), we use certainty equiv-
alence to measure welfare. Recall that certainty equivalence of a
random return is a sure payment that equates the utility from this
sure payment with the expected utility from the random return.
Let e1 (i ) and e0 (i ) be the certainty equivalence of farmer i’s
returns with and without crop insurance, respectively. Specifically,
if we assume that farmers’ utility
 0 function
 is U (·), then we have
1 1 0
U (e (i )) = u (i , p) and U e (i ) = u (i ). The willingness to
pay for crop insurance is then π(i ) = e1 (i ) − e0 (i ). For risk
averse farmers, we expect that π(i ) > 0. According to Chetty and
Finkelstein (2013), the consumer surplus, producer surplus, and total
social welfare in the presence of crop insurance can be written as,

 1  
CS = e () − p 1 (π() ≥ p) + e0 ()1 (π() < p) dG(), (10)

PS = (p − c()) 1 (π() ≥ p) dG(), and (11)

T S = CS + P S

 1  
= e () − c() 1 (π() ≥ p) + e0 ()1 (π() < p) dG().
(12)

Notice that in the absence of crop insurance, farmers’ welfare is

simply ∫ e0 ()dG(). For farmers, the welfare increase due to crop
insurance is CS − ∫ e0 ()dG().
Moreover, based on Equation (12), to maximize total social
welfare, each farmer who has the willingness to pay for the insurance
larger than the insurance cost (i.e., expected indemnity payment)
should purchase the insurance. In Figure 3, the market equilibrium
is reached at (q̂, p̂), point D, where the marginal farmer’s willingness
to pay is equal to the average insurance cost. However, the efficient
equilibrium should be at point C (i.e., (q ∗ , p∗ )) where the marginal
158 Agricultural Economics and Policy

farmer’s willingness to pay is equal to the marginal cost of providing

the insurance. We can see that the total social welfare under crop
insurance outcome (q̂, p̂) and (q ∗ , p∗ ) are area ADEB and area ABC,
respectively. Clearly, the welfare loss due to adverse selection is area
CDE.
Policy instruments can be implemented to reduce this welfare
loss. One approach is to simply set insurance premium at p∗ via gov-
ernment regulation. Because p∗ is lower than p̂, setting the premium
at p∗ will increase the quantity demanded for the insurance to q ∗ ,
reaching the efficient equilibrium. In the case that the total farmer
population is smaller than q ∗ , the government can also mandate that
all farmers purchase crop insurance.1 Another policy instrument that
is most common in the case of crop insurance is premium subsidy.
Nowadays the U.S. federal government covers about 65% of the
crop insurance premium. Figure 3 shows that a premium subsidy
shifts insurance demand curve upward. If the shifted demand curve
intersects the average cost curve at point F , then the insurance
program can reach its efficient outcome (q ∗ , p∗ ). Because subsidizing
premium requires government expenditure, whether or not premium
subsidy will generate net welfare gains is an empirical question.
A recent working paper by Yu et al. (2023) examines the welfare
impact of premium subsidies in the context of U.S. crop insurance.
Its findings suggest that the current subsidy rate is higher than the
social optimal rate.

3. Welfare Analysis of Overlapped Policies

The welfare analyses above implicitly assume that the market is
in competitive equilibrium before the policy instrument in question
is implemented. However, in reality, it is often the case that one
policy instrument is implemented in the presence of some other

1
Note that if all farmers are risk averse and if there are no loading factors, the
marginal cost curve will be always below the demand curve. That is, the expected
indemnity payment is always smaller than the willingness to pay for the insurance.
In this case, (q*, p*), the intersect point of the marginal cost curve and the demand
curve, does not exist (see Figure 2 in Chetty and Finkelstein (2013)).
 Welfare Analysis of Agricultural Policies 159

existing policies. This is particularly true for U.S. agriculture because

various policy instruments had been introduced in the 20th century.
New instruments often have to take previous policies as “given”
and cannot do anything to those pre-existing policies. As a result,
Lichtenberg and Zilberman (1986) argue that policymakers are
often “policy takers”. For instance, although the EPA can regulate
pesticide uses, it has no legislative power on the deficiency payment
that may have affected pesticide uses in the first place. Therefore,
EPA’s regulation on pesticide use has to take place in the presence
of the deficiency payment, instead of a laissez-fair market. As to
be discussed in this section, ignoring the existence of other relevant
policies when conducting the welfare analysis of a policy instrument
may result in biased conclusions.

3.1. Production regulation in the presence

of deficiency payment

Lichtenberg and Zilberman (1986) analyze the potential biases of wel-

fare analysis when ignoring co-existing policies by using production
regulation in the presence of deficiency payment as an example. The
production regulation in question can be restrictions on pesticide
use, fertilizer use, or uses of genetically engineered seeds, which is
expected to reduce supply if implemented.
Figure 4 is used to illustrate the conceptual framework developed
in Lichtenberg and Zilberman (1986). In this figure, the demand
curve is denoted by the downward sloping line D, and the true supply
curve before the implementation of the production regulation is
labeled as S1 . Due to the existence of deficiency payment with target
price PT , without the production regulation, the quantity supplied
is Q1 , price paid by consumers is P1 , and the effective price received
by farmers is PT . Let CS 0 , P S 0 , and G0 denote consumer surplus,
producer surplus, and government expenditure without production
regulation. Then we have

CS 0 = ΔACP1 , P S 0 = ΔPT DG, and G0 = PT DCP1 . (13)

With production regulation, the supply curve will shift to S2 . Due

to the deficiency payment, the quantity supplied will become Q2 .
160 Agricultural Economics and Policy

Price

c
b

a d e

g h

i j l f
k

O Quantity

Figure 4. Welfare impacts of regulation in the presence of price support.

Source: Reproduced from Lichtenberg and Zilberman (1986). Permission obtained
from the authors and the American Economic Association.

Because the demand curve is unaffected, price paid by consumers

becomes P2 . The effective price received by farmers is still at PT . Let
CS 1 , P S 1 , and G1 denote consumer surplus, producer surplus, and
government expenditure with production regulation. Then we have
CS 1 = ΔABP2 , P S 1 = ΔPT EF, and G1 = PT EBP2 . (14)
By comparing Equations (13) and (14), we can obtain the
welfare changes of consumers and producers, as well as changes in
government expenditure. Specifically,
ΔCS = CS 1 − CS 0 = ΔABP2 − ΔACP1
= −(g + h + i + j + k + l), (15)
ΔP S = P S 1 − P S 0 = ΔPT EF − ΔPT DG = − (a + b + c) (16)
ΔG = G1 − G0 = PT EBP2 − PT DCP1
= −(c + e + f + g + h + i + j + k + l). (17)
 Welfare Analysis of Agricultural Policies 161

From Equation (17) we can see that a portion of government

expenditure reduction is represented by consumer surplus decrease,
−(g + h + i + j + k + l). Another portion is represented by the
loss in producer surplus, −c. These transfers do not affect total
social welfare, and the total welfare change caused by the production
regulation is therefore,

ΔT S = ΔCS + ΔP S − ΔG = e + f − a − b. (18)

Following Harberger (1971), Lichtenberg and Zilberman (1986)

decomposed e + f − a − b into two components. The first component
is −a, which is interpreted as the net social welfare loss caused by the
production regulation when deficiency payment does not exist. The
second component is e + f − b, which is the reduction in deadweight
loss when supply shifts from S1 to S2 while hold the deficiency
payment policy unchanged. Note that following the discussion in
Section 1.1 of this chapter, we know that the deadweight loss caused
by the deficiency payment is d + e + f when supply is S1 and is b + d
when supply is S2 . Therefore, the change in deadweight loss when
supply shifts from S1 to S2 is (d + e + f ) − (b + d) = e + f − b. In sum,
the net social welfare change caused by the production regulation in
the presence of deficiency payment is e + f − a − b.
The discussion above shows that, in the presence of price
support, PT , the production regulation has negative impact on
producer surplus. However, for consumers, due to their dual role as
consumers and taxpayers, the net effect of the production regulation
on them is positive (i.e., savings in government expenditure is
larger than consumer surplus reduction). Without price support, the
same production regulation is likely to have net negative impact
on consumers (i.e., reducing consumer surplus without savings in
government expenditure) but perhaps negligible or even positive
impact on producer surplus because of low elasticities of demand
for agricultural commodities. Therefore, Lichtenberg and Zilberman
(1986) conclude that price supports are “to strengthen producers’
opposition to and consumers’ support for regulation.”
What if price supports are ignored when conducting the welfare
analysis of production regulation? Lichtenberg and Zilberman (1986)
162 Agricultural Economics and Policy

illustrate that the consequence is a biased welfare estimate of the

regulation. They argue that if price supports are ignored, then the
supply curve will likely be estimated at Ŝ1 instead of at S1 because
the commodity is traded at quantity Q1 and price P1 in the market.
Suppose that the regulation will shift the supply curve from Ŝ1 to Ŝ2 ,
and that the production regulation in question has the same effect
on the supply curve with and without considering price support (i.e.,
Ŝ2 − Ŝ1 = S2 − S1 ). Figure 4 shows that when supply curve shifts
from Ŝ1 to Ŝ2 , consumer surplus will reduce by area i + j + k + l.
From Equation (15), we can see that the true consumer surplus
change is −(g + h + i + j + k + l). Therefore, ignoring price support
will underestimate consumer surplus reduction by g + h. Similarly,
assuming that all supply curves in Figure 4 are parallel, it will
underestimate producer surplus reduction by area i + j. Moreover,
ignoring price support overlooks savings in government expenditure
(i.e., c+e+f +g+h+i+j +k+l), and overestimates net social welfare
losses of production regulation by c + e + f + k + l when compared
with the approach without ignoring price support. When calibrated
using U.S. corn, cotton, and rice data, Lichtenberg and Zilberman
(1986) find that the overestimation can be as high as 30%–50%,
thus conclude that welfare analysis of production regulation ignoring
price support will “be biased against regulation” (Lichtenberg and
Zilberman, 1986).
Lichtenberg and Zilberman (1986) provide a classic example of
welfare analysis in the presence of an overlapped policy instrument.
In the next subsection we discuss one example that uses their
approach to analyze the welfare effect of biofuel tax credit.

3.2. Welfare effect of biofuel tax credit in the

presence of deficiency payment

The ethanol boom over the last two decades in the United States
gave rise to a great deal of research on the economics of ethanol and
biofuels in general. By using around one third of corn grain produced
in the U.S., ethanol production links the gasoline market with the
agricultural commodity market. Moreover, policy instruments sup-
porting biofuel development, such as Renewable Fuel Standard (RFS)
 Welfare Analysis of Agricultural Policies 163

originated from the Energy Policy Act of 2005 and RFS2 specified in
the Energy Independence and Security Act of 2007, may interact
with existing farm policies such as deficiency payment described
above. Numerous studies have examined the welfare impacts of these
biofuel policy instruments (e.g., de Gorter and Just, 2009a,b; de
Gorter and Just, 2010; Lapan and Moschini, 2012; Khanna et al.,
2016; and Moschini et al., 2017). In this subsection, we focus on de
Gorter and Just (2009a) as it offers an intuitive example examining
how biofuel tax credit may interact with deficiency payment when
studying the welfare impacts of the two.
The key ideas of de Gorter and Just (2009a) can be illustrated
through Figures 5–7. Figure 5 shows the impact of an ethanol tax
credit on the market equilibrium of the corn and fuel markets. In
panel (a) of Figure 5, line SC stands for U.S. corn supply and
line DN E stands for non-ethanol demand for corn (including both
domestic demand and international demand for U.S. corn). If there
is no ethanol production, the intersection of lines SC and DN E
determines corn price, PN E . In panel (b) of Figure 5, ethanol supply
curve is denoted by SE . The intercept of SE is assumed to be
at a level equivalent to PN E in panel (a).2 Furthermore, domestic
supply of gasoline is denoted by line Sd , and total fuel supply
(including domestic gasoline supply, imported gasoline, and any
ethanol) is represented by line SF . Without tax credit for ethanol,
the equilibrium fuel price is PG , determined by total fuel supply SF
and total fuel demand DF . We can see that PG is lower than the
intercept of SE , indicating that there is no ethanol production at PG
if there is no tax credit for ethanol, so PG is just the price for gasoline.
Now assume that the government introduces a tax credit, tc (51
cents per gallon as in de Gorter and Just (2009a)) for refiners or
blenders who mix ethanol with gasoline. In a competitive market, this
policy will increase ethanol price by the amount that is equal to the
tax credit (i.e., increasing the price to PG +tc ). This tax credit policy

shifts ethanol supply curve downward to SE . As a result, the total

2
Equation (1) in Gorter and Just (2009a) shows how to convert ethanol price
from dollar per gallon to dollar per bushel.
164 Agricultural Economics and Policy


fuel supply curve will be shifted downward to SF (note that there is a
 
kink on SF when fuel price is equal to the intercept of SE ). The new
 
equilibrium fuel price is PG , and ethanol price is PE = PG + tc . At

fuel price PG and ethanol tax credit tc , quantity of domestic gasoline
supplied is OG, imported gasoline is GH, and the quality of ethanol
produced is HJ (or, equivalently, OF).
The tax-credit-caused production of ethanol will increase the
total demand for U.S. corn and increase corn price. Note that the
non-ethanol demand curve for corn is not shifted by the tax credit.
Reflected in corn market illustrated in panel (a) of Figure 5, corn
price increases to PC , which is equal to ethanol price measured in
dollars per bushel. Due to the increased corn price, the quantity
demanded by non-ethanol uses shrinks from OC to OA, and the
quantity supplied of corn increases from OC to OB. The amount
AB is used for ethanol production. de Gorter and Just (2009a)
 
define water in tax credit as w = PN E − PGb , where PGb is fuel

$/bu ¢/gal
Sc SE

Sd
Pc = PE S′E SF
tc =
PEb tcb = [β/(1-δ)]tc
51¢
PNE /gal
w = water
PGb PG S′F
P' Gb P' G

DNE
DF

O O G H I J
A C B bushels F gallons
(a) Corn Market (b) Fuel Market

Figure 5. Equilibrium in corn and fuel markets under an ethanol tax credit.
Source: Reproduced from Figure 1 in de Gorter and Just (2009a). Permis-
sion obtained from the authors and the Agricultural and Applied Economics
Association.
 Welfare Analysis of Agricultural Policies 165

price measured in dollars per bushel considering the conversion rate

between corn and ethanol (see footnote 2). From panel (a) of Figure
5, one can see that within tax credit tcb , only the portion “above”
the water can increase corn price. If the water is “deep” enough (e.g.,
fuel price very low), the tax credit may not increase corn price at all.
Now let us see the welfare impact of the tax credit on fuel and
corn markets in the absence of deficiency payment. Panel (a) in
Figure 6 depicts the welfare changes in the fuel market. Note that
de Gorter and Just (2009a) assume that profits of ethanol producers
are zero and, therefore, do not include producer surplus of ethanol in

the discussion.3 The tax credit decreases fuel price from PG to PG ,
resulting in a gain in consumer surplus equal to area e+f +g+h+i+j.
Domestic gasoline producers and international gasoline producers
lose producer surplus of areas e + f and g + h + i, respectively. Area
i + j + k can be interpreted as transfer of funds from taxpayers to
fuel consumers through increased ethanol production. Overall, the
deadweight loss in the fuel market is area i + k, which has the same
size as area j.
Panel (b) in Figure 6 depicts the welfare impact of the tax credit
on corn market. Line Dd stands for domestic non-ethanol demand
for corn, and recall that, DN E stands for total non-ethanol demand
for corn (including international demand). As discussed above, the
tax credit will increase corn price from PN E to Pc . Non-ethanol
corn consumers lose consumer surplus at a magnitude represented
by area l + n + q + a. Corn producers gain producer surplus of area
l + n + q + a + b. Taxpayers’ fund transferred to corn producers
is represented by area a + b + c, where area a is the deadweight
cost of underconsumption of non-ethanol corn and area c is the
deadweight cost of overproduction of corn. Area b is transferred to
corn producers as producer surplus. What is unique in panel (b)
is that the authors identify a rectangular area, the hatched area,
that is pure deadweight loss. Note that taxpayers’ fund denoted by
this area is not transferred to any party in the corn or fuel market.

3
We thank David Just for pointing this out to us.
166 Agricultural Economics and Policy

¢/gal

Sd SF
SE

PG
f i k S′F
e h j
g
P'G

imports
DF

domestic ethanol

O G H J gallons
(a) Fuel Market

$/bu Sc

Dd

Pc =
PEb l n q a b c tcb = [β/(1-δ)]tc
PNE
w = water
PGb
PGb - P′Gb
P'Gb

DNE

O A C B bushels
(b) Corn Market

Figure 6. Welfare impact of an ethanol tax credit on fuel and corn markets.
Source: Reproduced from Figure 2 in de Gorter and Just (2009a). Permis-
sion obtained from the authors and the Agricultural and Applied Economics
Association.
 Welfare Analysis of Agricultural Policies 167

Even though the upper left corner of this hatched area overlaps with
part of producer surplus, one should note that this part of producer
surplus is already there before the implementation of tax credit.
Therefore, the upper left corner of this hatched area is not new
surplus, while it is an additional expense to taxpayers. Therefore,
the whole hatched rectangular is a deadweight loss. The numerical
simulation in de Gorter and Just (2009a) shows that the magnitude
of this rectangular deadweight loss can be as large as $2.3 billion,
much higher than the triangular deadweight loss (e.g., areas a and
c in panel (b) of Figure 6). Finally, the cross-hatched rectangular
area stands for taxpayers’ funds transferred to fuel consumers. It is
corresponding to area i + j + k in panel (a) of Figure 6, with areas i
and k are deadweight losses.
How does the ethanol tax credit interact with deficiency payment
for corn? In contrast to Lichtenberg and Zilberman (1986), in the
framework developed by de Gorter and Just (2009a), the deficiency
payment may be dormant, depending on the relative size of the defi-
ciency payment rate and corn price under the tax credit. de Gorter
and Just (2009a) specified all possible scenarios of market equilibrium
when both tax credit and deficiency payment are considered. Here we
only focus on one scenario that de Gorter and Just (2009a) discuss in
detail. Under this scenario, the deficiency payment rate, L, is greater
than corn price under tax credit only, PC , which is in turn greater
than non-ethanol corn price, PN E , which is in turn greater than
corn market price under deficiency payment only, PL , and in turn
PGb . That is, L > PC > PN E > PL > PGb . Figure 7 depicts the
relationship between these prices.
Under the setup in Figure 7, if the tax credit is the only
policy instrument in place, then the government expenditure is
tcb · (QC − CN E ), where tcb is the tax credit measured in dollar
per bushel, QC is corn production under tax credit, and CN E is
non-ethanol corn consumption. If the deficiency payment is the only
policy instrument, then the government expenditure is (L − PL ) ·
QL , where QL is corn production under deficiency payment only.
When we consider both tax credit, tcb , and deficiency payment, L,
168 Agricultural Economics and Policy

$/bu

Sc

Pc =
PEb g a b
PNE

h d
c tcb = [β/(1-δ)]tc

PL
e f wL
PGb
DNE

CNE QC QL bushels

Figure 7. Impact of co-existence of the tax credit and deficiency payment.

Source: Reproduced from Figure 3 in de Gorter and Just (2009a). Permis-
sion obtained from the authors and the Agricultural and Applied Economics
Association.

the government expenditure is,

(L − PC ) · QL + tcb (QL − CN E ). (19)

Clearly, the tax credit saves government expenditure on deficiency

payment by the amount of (PC − PL ) · QL , and the deficiency
payment increases government expenditure on tax credit by amount
tcb · (QL − QC ). However, the sizes of the two amounts are ambiguous
without further information. We therefore cannot determine whether
the total government expenditure is increased or decreased by the
co-existence of the two policy instruments when compared with the
expenditure under a single policy instrument.
Because QL = CN E + (QC − CN E ) + (QL − QC ), the government
expenditure savings in deficiency payment caused by the tax credit,
(PC − PL ) · QL , can be decomposed into three parts. They are:
(1) (PC − PL ) · CN E , represented by area g + h in Figure 7, which
can be interpreted as the consumer surplus reduction of non-ethanol
corn consumers; (2) (PC − PL ) · (QC − C N E ), represented by area
 Welfare Analysis of Agricultural Policies 169

a + c, which is part of tax credit cost that cancels the same amount
of deficiency payment; (3) (PC − PL ) · (QL − QC ), represented by
area b + d, which is another part of tax credit payment canceling
the same amount of deficiency payment. Note that areas e and f ,
parts of tax credit payment, do not offset any deficiency payment or
are transferred to any parties. Therefore, they are deadweight loss.
Specifically, area e is rectangular deadweight loss of tax credit in the
absence of deficiency payment, and area f is the deadweight loss of
tax credit caused by the existence of deficiency payment.

4. Conclusion
Welfare analysis is a key component in agricultural economics and
policy. In this chapter, we have provided a few examples of welfare
analysis on some major agricultural policies such as crop insurance,
deficiency payment, and biofuel policies. We emphasize considering
the interactions between policy instruments when conducting welfare
analysis. As crop insurance is gaining importance in the realm of
agricultural policy in the U.S., examining the interaction between
crop insurance and other policy instruments may yield meaningful
insights to understand the impact of these policies. Miao et al. (2016)
offer an example studying the interaction between crop insurance and
the Conservation Reserve Program (CRP), showing that were saved
insurance premium subsidy from enrolled CRP land to be considered,
then the CRP enrollment efficiency would be enhanced significantly.
Miao (2020) investigates how crop insurance may interact with
agricultural innovation in the context of climate change. Further
research is needed to understand how crop insurance may interact
with other agricultural policies related to renewable energy, organic
agriculture, rural development, and even nutrition.

References
Atkinson, A.B. 2011. The Restoration of Welfare Economics. American Economic
Review: Papers and Proceedings 101(3): 157–161.
Chetty, R. and A. Finkelstein. 2013. Social Insurance: Connecting Theory to
Data, Alan J. Auerbach, R.C., Feldstein, M., and E. Saez (Eds.), Handbook
of Public Economics, Vol. 5. Elsevier, Amsterdam.
170 Agricultural Economics and Policy

Coble, K.H. and B.J. Barnett. 2013. Why Do We Subsidize Crop Insurance?
American Journal of Agricultural Economics 95(2): 498–504.
de Gorter, H. and D.R. Just. 2009a. The Welfare Economics of a Biofuel Tax
Credit and the Interaction Effects with Price Contingent Farm Subsidies.
American Journal of Agricultural Economics 91(2): 477–478.
de Gorter, H. and D.R. Just. 2009b. The Economics of a Blend Mandate for
Biofuels. American Journal of Agricultural Economics 91(3): 738–750.
de Gorter, H. and D.R. Just. 2010. The Social Costs and Benefits of Biofuels:
The Intersection of Environmental, Energy and Agricultural Policy. Applied
Economic Perspectives and Policy 32(1): 4–32.
Goodwin, B.K. and V.H. Smith. 2013. What Harm Is Done by Subsidizing Crop
Insurance? American Journal of Agricultural Economics 95(2): 489–497.
Harberger, A.C. 1971. Three Basic Postulates for Applied Welfare Economics: An
Interpretive Essay. Journal of Economic Literature 9: 785–797.
Khanna, M., H.M. Nuñez, and D. Zilberman. 2016. Who Pays and Who Gains
from Fuel Policies in Brazil? Energy Economics 54: 133–143.
Just, R.E., D.L. Hueth, and A. Schmitz. 2004. The Welfare Economics of Public
Policy: A Practical Approach to Project and Policy Evaluation. Edward Elgar
Publishing, Inc. Northampton, MA.
Lapan, H. and G.C. Moschini. 2012. Second-Best Biofuel Policies and the Welfare
Effects of Quantity Mandates and Subsidies. Journal of Environmental
Economics and Management 63(2): 224–241.
Lichtenberg, E. and D. Zilberman. 1986. The Welfare Economics of Price Supports
in U.S. Agriculture. The American Economic Review 76(5): 1135–1141.
Miao, R., H. Feng, D.A. Hennessy, and X. Du. 2016. Assessing Cost-effectiveness
of the Conservation Reserve Program and Its Interaction with Crop Insurance
Subsidies. Land Economics 92(4): 593–617.
Miao, R. 2020. Climate, Insurance, and Innovation: The Case of Drought and
Innovations in Drought-Tolerant Traits in U.S. Agriculture. European Review
of Agricultural Economics 47(5): 1826–1860.
Moschini, G.C., H. Lapan, and H. Kim. 2017. The Renewable Fuel Standard in
Competitive Equilibrium: Market and Welfare Effects. American Journal of
Agricultural Economics 99(5): 1117–1142.
U.S. Government Accountability Office (USGAO). 2023. Farm Bill: Reducing
Crop Insurance Costs Could Fund Other Priorities. GAO-23-106228.
Yu, J., J.J. Wu, and R. Miao. 2023. Welfare Effects of the U.S. Federal
Crop Insurance Program: The Sufficient Statistics Approach. Working
Paper, Department of Agricultural Economics and Rural Sociology, Auburn
University.
 Chapter 6

The Political Economy of

Agricultural Policies

Economists do not run the world. In principle, there are at least two
mechanisms for resource allocation. One is the economic system (e.g.,
markets and prices) and the other is the political system. In some
cases, there is a hierarchy wherein the political system dominates the
economic system, but in most cases, the relationship is symbiotic.
So, economic analysis needs to incorporate political considerations
to realistically understand resource allocation.
There are several lines of analysis in political economy. The first
is a model by Black (1948), who developed the median voter theorem.
For every choice, we have a distribution of preferences, and the
evaluation of the median voter would result in an outcome where
a majority rule governs. It can be used to assess voters’ responses
to propositions, for example, whether to vote for or against GMOs.
Zilberman et al. (2014) address this issue and show how the parties
aim to affect the vote through expenditures and other activities.
A more sophisticated paper on the same topic is by Waterfield et al.
(2020), which distinguishes between willingness to vote, to ban, and
to pay for GMOs. It emphasizes the different considerations that
affect voting vs. buying. The model is elegant, and the two papers
are complementary.

171
172 Agricultural Economics and Policy

The median voter model is important for democracy or voting.

Still, in many cases, policymakers make regulatory choices, and
they have an objective function that considers their likelihood of
being re-elected and their gain from certain economic activities.
Becker (1983) had an interesting model. There are other models by
Besley and Coate (1997). Another important model is Grossman
and Helpman (1994), which started a big literature with a lot
of following. In all these models, a common element is that the
different groups have varying power. Groups with a lot of political
power can influence the political system. One approach to the
political system maximizes the sum of weighted consumer, producer,
and international consumer surplus. Because different groups have
different weights, the outcome may be biased towards a particular
group. An important author is Mancur Olson (1965), who wrote a
wonderful book on collective action that emphasizes the notion that
small, coherent groups with common objectives may have more power
than dispersed groups. That is the reason why grower groups are
successful in influencing agricultural policies. The book by Rausser
et al. (2011) uses cooperative game theory to reach this conclusion.
Groups build coalitions, and the relative power of different groups
depends on their ability to operate independently; groups that can
survive without the coalition may have more weight than others.
A strand of political economy studies focuses on agriculture
in particular. A strong paper is Rausser’s (1982) study on PESTs
(political economic-seeking transfers) vs. PERTs (political economic
resource transactions). A related model is presented in Rausser and
Foster (1990). Another nice application studying PESTs and PERTs
in agriculture is documented by Rausser (1992). The survey by
Swinnen and Vans der Zee (1993) is a good way to start studying
the evolution of agricultural policies from the perspective of political
economy. A more recent survey paper on the political economy of
agricultural policies is Anderson et al. (2013). Zilberman (1984)
examines the dynamics of agricultural policy, showing that techno-
logical change may lead to what we see today, which is a reduction in
the prices of agricultural commodities. Since farmers know they lose
 The Political Economy of Agricultural Policies 173

from innovation, they need to be subsidized to accept it. Zilberman

et al. (2014) discuss the politics of biotechnology and why political
consideration results in some of the industry’s challenges.
In the remaining part of this chapter, we will first speak about
political economy in general and then the political economy in
agriculture, emphasizing modeling that deals with resource allocation
in the political system and agricultural issues.

1. Basic Models in Political Economy

In this section, we will discuss three basic models in the literature on
political economy. They are the median voter model, pressure group
competition model, and a Nash bargaining model for various interest
groups.

1.1. Median voter theorem

As discussed in the introduction, the median voter theory predicts

that two competing political parties or candidates will tailor their
political proposals to appeal to the median voter and try to win their
vote. The theory was perhaps first postulated by Harold Hotelling
in his famous 1929 paper “Stability in Competition,” and then
developed and formalized by later studies such as Black (1948) and
Downs (1957). We present a simplified model with only two political
candidates here, but readers can refer to Black (1948) for a more
intricate model with n candidates.
Suppose there are an odd number of I voters.1 Let Vij denote
the voting behavior of individual i, with Vij = 1 if voter i votes for
candidate j and Vij = 0 if voter i does not vote for candidate j.
We further assume that voter i will select candidate j ∗ if and only
if her utility from selecting candidate j ∗ is greater than that from

1
When I is an even number, then we need a mechanism that can break a possible
tie. Black (1948) shows that the median voter theorem still holds in such a
case.
174 Agricultural Economics and Policy

Table 1. A simple numerical example for the median voter model.

Voter 1 2 3 4 (median) 5 6 7
ΔUi −53 −36 −7 6 8 22 30
Vote for Candidate 2 2 2 1 1 1 1

any other candidates (i.e., Vij ∗ = 1 if and only if Uij ∗ > Uij for all
j = j ∗ ). The winner, denoted by ĵ, will be the candidate who has the
 
most votes (i.e., Ii=1 Viĵ > Ii=1 Vij for all j = ĵ). For simplicity,
let us assume there are only two candidates, 1 and 2. Then, for voter
i, the difference in utility from choosing the two candidates is ΔUi =
Ui1 − Ui2 . We ascendingly order the voters according to the value of
this utility difference. It is readily checked that the winning candidate
will be the one whom the median voter selects. Table 1 presents
a simple numerical example, where the winner is Candidate 1, and
the median voter is pivotal for the winning candidate to obtain the
majority.
The median voter theorem is helpful for understanding why the
priorities of median voters shape campaigns and influence policies.
In many countries, where regional or tribal considerations affect
political alliances, policies and campaigning are often targeted to
switchable regions or groups. This is perhaps why the “swing states”
attract so much attention during presidential elections in the United
States.

1.2. Rent and rent-seeking

Rent, the amount paid for a factor of production above what it needs
to be economically viable, is one of the most important concepts in
economics. The economic rent is what people would consider above
“normal” profit. Textbook models suggest that at the equilibrium
firms operate with normal profit, but some agents will obtain rents
for unique capacities. For example, farmers operating at the margin
will make zero profit with prices equaling average total cost; other
farmers who have higher productivity, however, will gain additional
 The Political Economy of Agricultural Policies 175

surplus, i.e., rent. So, we can relate the notion of rent to that of
producer surplus, and one can even expand it and interpret consumer
surplus as the rent for the consumer who is not at the margin, where
the price paid is smaller than the marginal utility.
Rent-seeking involves activities that aim to increase one’s rent.
Numerous examples exist. For instance, domestic producers may
lobby to set or increase import tariffs so that they can enjoy protected
prices and markets. They may share the gains from tariffs with
politicians explicitly or implicitly. International groups such as the
World Trade Organization attempt to reduce this type of behavior.
Further, regulators can provide an exemption from environmental
laws, regulations, or zoning to some groups of producers under
their requests. For example, the mandates for advanced biofuels
established in the Energy Independence and Security Act of 2007
had been continuously waived. Another example is that the EPA
waived some small refineries’ obligation to blend a certain amount of
biofuels into gasoline via Small Refinery Exemptions.
Social welfare under perfect competition is typically higher
than that under monopoly. Monopoly causes deadweight loss with
producer surplus increasing but consumer supply decreasing (see
Figure 1). If a government provides a company with monopoly power,

Perfect Competition Monopolies

$ $

ATC
Consumer Consumer MC
Pm Surplus
MC Pm Surplus

Ppc Ppc DWL

Producer Producer
Surplus Demand Surplus Demand

MR MR
Qm Qpc Qm Qpc

Figure 1. Consumer surplus and producer surplus under perfect competition

and monopoly.
176 Agricultural Economics and Policy

say, through regulation, it creates rent, reducing social welfare but

generating profits for the monopolist. In many cases, the politician
gets a kickback, but they do have to take political accountability
into consideration. Suppose that the regulation moves output from
the competitive outcome, Qpc , to a lower regulation output, Qr . The
gain to the industry is the increased producer surplus: ΔP S(Qr ) =
P S(Qr ) − P S(Qpc ), where P S(Q) stands for producer surplus at
output Q. Note that ΔP S(Qr ) reaches its maximum when Qr = Qm ,
where Qm is the optimal output for the monopolist. Under the
regulation, consumers lose ΔCS(Qr ) = CS(Qpc ) − CS(Qr ). Assume
that the industry pays X dollars to the politician. The regulation,
owing to its welfare-reducing nature, may cause the politician to lose
re-election, say, at probability f (Qr ). Let us assume f (Qpc ) = 0 and
fQr (Qr ) = df /dQr ≤ 0. Finally, let Cp denote the politician’s gain
from staying in power (excluding any kickback money).
With the above notation, the optimization problem of the two
parties (i.e., the industry and the politician) can be written as

Max{Qr } N G = (1 − f (Qr )) ΔP S(Qr ) − f (Qr ) CP , (1)

subject to Qm ≤ Qr ≤ Qpc .
The first term on the right-hand side of Equation (1) is the
expected gain from the regulation (i.e., (1 − f (Qr )) ΔP S(Qr )) and
the second term is the expected loss from the regulation (i.e., loss
from losing power, −f (Qr )CP ). Suppose there is an interior solution.
Then the first-order condition that determines the optimal Qr is
dN G
= (1 − f (Qr ))ΔP S Qr − fQr (ΔP S(Qr ) + C P ) = 0. (2)
dQr
By rearranging Equation (2), we obtain

(1 − f (Qr )) (−ΔP S Qr ) = −fQr (ΔP S(Qr ) + C P ) . (3)

Note that (1 − f (Qr )) is the probability of staying in power, and

−ΔP S Qr is marginal gain in surplus with lower Qr . Equation (3)
states that the marginal expected gain in producer surplus from
reducing supply is equal to the expected economic loss associated
with the marginal increase in the probability of losing power due
 The Political Economy of Agricultural Policies 177

to restricted supply. Specifically, this expected economic loss is the

product of the marginal increase in the probability of losing power,
fQr , and the loss of both producer surplus and benefit from holding
power for the regulator, (ΔP S(Qr ) + C P ).
This model suggests that the producer’s gain from the regulation
will increase, for example, when demand is more inelastic or when
the vulnerability to losing power due to restricted supply is low.
Obviously, when there is a large potential to get an increased
producer surplus, then the restrictions on supply will be stronger.
On the other hand, the more that the politician gains from their
power (i.e., larger Cp ) or the more vulnerable they are (i.e., higher
f (Qr )), the less they engage in introducing cartels and monopolies.
This model suggests that manipulation of power to benefit the few is
less likely in a democracy where consumers are aware of the political
process and are able to mobilize their power against the artificial
restriction of supply.
Now let us see how the rent generated from the regulation might
be divided between the industry and the politician (i.e., “sharing
of spoils”). Let α ∈ [0, 1] be the relative power of the politician
to the industry when dividing the spoils. The probability of losing
power may be a function of the politician’s portion in the spoils
(denoted by X ). Specifically, the probability of the politician losing
position is (f (Qr , X)). Again, as before, the politician’s loss in case
of losing his position is denoted by CP . Therefore, the net gain for
the politician from the regulation is X − f (Qr , X) · CP . The industry
gains ΔP S(Qr ) as long as the politician stays in power, minus the
kickback money X to the politician with certainty. So, the net gain
of the industry (1 − f (Qr , X)) · ΔP S(Qr ) − X. The optimal Qr and
X can be determined by solving the following optimization problem:


Max{Qr ,X} α (X − f (Qr , X) · CP )
  
+ (1 − α) 1 − f (Qr , X) ΔP S(Qr ) − X (4)

subject to Qm ≤ Qr ≤ Qpc and 0 ≤ X ≤ ΔP S(Qr ).

178 Agricultural Economics and Policy

Assuming an interior solution, the first-order conditions for Qr and

X are:
−afQr Cp + (1 − a) (1 − f (Qr , X)) ΔP S Qr − fQr ΔP S(Qr ) = 0,
(5)
a(1 − fX Cp ) − (1 − a) (1 + fX ΔP S(Qr )) = 0, (6)

where fQr = ∂f (Qr , X)/∂Qr , fX = ∂f (Qr , X)/∂X, and ΔP S Qr =

∂ΔP S(Qr )/∂Qr .
Rearranging terms in Equation (5), we obtain

(1 − α) (1 − f (Qr , X)) ΔP SQr = fQr αCp +(1 − α) ΔP S(Qr ) . (7)

The left-hand side of Equation (7) is the weighted marginal gain

to the industry–politician coalition from reduction in quantity (i.e.,
expected increase in producers’ surplus). The right-hand side is
the expected marginal cost from a marginal reduction in regulated
output. A reduction in probability of winning will reduce the well-
being of both the politician and the industry. Equation (7) states
that at the optimal level of Qr , the weighted marginal gain of
Qr is equal to the weighted marginal cost. The greater is the
industry power (1 − α), the stricter is the regulation (i.e., lower
Qr values). For example, when a politician depends strongly on
domestic manufacturers, he may set regulations that limit import
opportunities. This was a key element in the export substitution
strategy in Latin America.
Similarly, by rearranging Equation (6) we obtain
α
(1 − fX Cp ) = 1 + f X ΔP S (Qr ) . (8)
1−α
The left-hand side of Equation (8) is the marginal gain to the
politician from increased portion of the spoils, weighted by its relative
political power a/(1 − a). Note that the marginal gain from increased
donation is income plus the expected extra political gain from higher
spending (remember that fX is negative; the more the politician
spends on elections, the less likely he or she is to lose). The right-
hand side of the equation is the marginal cost to the industry. It is
the direct expenditure (1 unit of X ) minus the gain due to reduction
 The Political Economy of Agricultural Policies 179

in expected political risk due to donation fX ΔP S(Qr ). When α

is higher, the politician has more political power, and, thus can
negotiate a larger contribution from industry for a given level of extra
surplus he or she provides the industry by regulation. Also, when
the marginal effect of campaign contribution is larger, the bigger the
donation (i.e., X ) will be.
From this simple “spoil-sharing” model we can see that factors
determining patterns of political payoffs may include the strength
of politician (a), the effectiveness of campaign spending (fX ), gain
to producers (ΔSP (Qr )), probability of getting caught (f (Qr , X)),
and price for getting caught (Cp ). One can infer that complexity
with lack of transparency leads to corruption, that lack of awareness
by losers allow for more exploitation, and that political awareness
matters. Numerous day-to-day examples, such as gifts to authority
figures, licensing, and regulations, can be explained using this model.
Rent-seeking behavior may be used to slow reforms or changes, such
as introduction of policies that limit new technologies or new export
practice.

1.3. Modeling competing interests

Different from earlier studies on political economy that focus on

voters, politicians, or political parties, Becker (1983) proposes a
theory postulating that political equilibrium is a result of competition
among interest groups. He argues that each interest group will invest
its resources (e.g., time, money, energy, and social connections) to
build up political pressure and maximize their incomes. Interest
groups vary in their resource endowment and in their capability
to generate political pressure. Therefore, the outcomes for interest
groups may differ in the political equilibrium. In this subsection, we
briefly introduce this model.
Suppose there are two interest groups, s and t. One can view
group s as the group that enjoys subsidy and view group t as the
group that is taxed. The numbers of members in groups s and t
are ns and nt , respectively. Within each group, the members are
identical. Each group will invest its resources to generate political
pressure to either increase its subsidy or reduce its tax. Assume that
180 Agricultural Economics and Policy

each member in group s invests as amount of resources and each

member in group t invests at . The political pressure produced by the
groups is, therefore, pj = pj (mj , nj ) where mj = aj nj , and j ∈ {s, t}.
The pressure from both groups will determine the payoffs for each
group because of the competition between the two groups. Moreover,
the aggregated payoffs of the two groups should be zero because the
subsidy for group s will be from the tax from group t. Specifically,
we have

ns G (Rs ) = I s (ps , pt , x), (9)

nt F (Rt ) = −I t (ps , pt , x), (10)
I s (ps , pt , x) + I t (ps , pt , x) ≡ 0, (11)

where Rs and Rt are subsidy rate and tax rate per member in groups s
and t, respectively; functions G(·) and F (·) capture the deadweight
losses from subsidy and tax, reflecting distortion effect of subsidy
and tax; and finally, I s (·) and I t (·) are “influence” functions that
translate the political pressure and other variables (labeled as x )
into subsidy amount or tax revenue.
For a member in group s or t, her overall payoff can be written as,

Zs = Zs0 + Rs − as , (12)
Zt = Zt0 − Rt − at , (13)

where Zs0 and Zt0 are initial payoffs prior to any political redistri-
bution. Suppose both groups are politically active, i.e., as > 0 and
at > 0. The payoff maximization conditions for the two groups are:
dRs dRt
= 1, and = −1. (14)
das dat
Based on Equations (9)–(11), the conditions in (14) are equivalent to

dRs 1 ∂I s ∂ps ∂ms Iss psm

= = = 1, (15)
das ns G ∂ps ∂ms ∂as G
dRt 1 ∂I t ∂pt ∂mt Its ptm
=− = = −1. (16)
dat ntF  ∂pt ∂mt ∂at F
 The Political Economy of Agricultural Policies 181

The second equation in expression (15) holds because ∂ms /∂as = ns ,

Iss ≡ ∂I s /∂ps , and psm ≡ ∂ps /∂ms . The second equation in expression
(16) holds for similar reasons. Becker (1983) shows that the sufficient
mm ≤ 0, pmm <
conditions for the optimization are Iss s < 0, I s > 0, ps t
tt
0, G > 0, and F  < 0. Assuming that each group, when it selects its
political investment level aj , takes the other group’s investment level
as given. Therefore, expression (15) can be viewed as a response
function of group s’s investment level, as , to group t’s investment
level, at . Similarly, expression (16) is group t’s response function to
group s’s investment level, as . The equilibrium is reached under a
pair of (a∗s , a∗t ) that satisfies both (15) and (16).
Figure 2 presents the response functions and equilibria, with the
initial equilibrium at point e0 . Suppose now group s becomes more
efficient in generating political pressure, say, due to their leadership
change. In other words, group s can respond with higher political
pressure for any given pressure generated by group t. This change
can be reflected as an upward shift of the response curve shifts from
s0 to s1 , causing the equilibrium to move from e0 to e1 . We can
see that when compared with the initial equilibrium e0 , the new

ps t0

t1

e2 s1
e1 s0

e0

pt

Figure 2. Political pressure response functions and equilibrium.

Note: This figure is adapted from Figure 1 in Becker (1983).
182 Agricultural Economics and Policy

equilibrium e1 implies higher political pressure from both groups

because the slopes of the response curves are positive. However, based
on Equation (15), the upward shift from s0 to s1 will cause both the
subsidy to group s and the tax on group t to increase. Becker (1983)
concludes that when a group is more efficient in exerting political
pressure, this group can either increase its subsidy or decrease its tax.
If group t also enhances its efficiency to generate political
pressure, then its response curve will move, say, to t1 . In this case, the
equilibrium will be settled at e2 . As a result, the pressure generated
by both groups would increase, but the equilibrium influence of
each group will be undetermined because it depends on the relative
changes in the influence of the two groups. Therefore, what matters
most for the political outcome of a group is not its absolute efficiency
in generating political pressure; it is the group’s relative efficiency
that will eventually determine the political outcome (Becker, 1983).
Becker further argued that, due to the free-riding problem, smaller
political groups (e.g., farmer groups) tend to have relatively higher
efficiency in generating political influence.

1.4. A game-theoretical framework

Another useful framework to analyze the interaction between politi-

cal groups is Nash–Harsanyi bargaining game, which models bargain-
ing and negotiations among groups and predicts the results therein.
The framework was developed under a series of seminal works by
Nash (1950, 1951, 1953) and Harsanyi (1962a,b, 1963). Rausser et al.
(2011, Chapters 2 and 3) provide a comprehensive review of this
framework. In this subsection, we briefly introduce a bargaining game
under which the reservation payoff (i.e., payoff when no agreement
is reached by political groups) of each group is exogenously given.
For the bargaining game with endogenously determined reservation
payoffs, we refer readers to Chapter 3 of Rausser et al. (2011).
We start with a model pertaining to two players, and then
expand this model to include n players. First, consider two groups,
1 and 2, that aim to maximize their individual payoffs through
 The Political Economy of Agricultural Policies 183

bargaining with the other group to reach an agreement (e.g., a

policy intervention). Suppose their reservation payoffs are u1 and
u2 , respectively. In other words, if no agreement is made between
the two groups, then group 1 will receive u1 and group 2 u2 . Let P
denote the payoff space of the two groups, which includes all vectors,
u ≡ (u1 , u2 ), of potential payoffs to groups 1 and 2. We assume that
P is compact and convex. Define H as a subset of P such that the
element in H is not (strictly or weakly) dominated by any element
in P . That is, for any element uH in set H, there does not exist an
element in P, u, such that u1 ≥ uH 1 and u2 ≥ u2 (i.e., H is the
H

efficiency frontier of P ). Define P ∗ as another subset of P such that

P ∗ ≡ {u ∈ P : u1 ≥ u1 , u2 ≥ u2 } and H ∗ as the efficiency frontier of
P ∗ . Furthermore, let u denote the optimal solutions of the bargaining
problem.
Based on the individual rationality axiom, it is readily checked
that no payoff outside of set P ∗ will be reached as an agreement
by the two groups. In other words, it must be true that u ∈ P ∗ .
Furthermore, the rationality assumption will imply that no Pareto
improvement can be made to optimal solutions u. That is, one cannot
find a payoff vector, u ∈ P ∗ , such that when compared with u, one
group is better off and the other group is not harmed, indicating that
u ∈ H ∗ . This axiom is termed Pareto optimality. Prior to Nash’s
game-theoretical framework of the bargaining problem, economic
theories could only refine the optimal solutions to H ∗ , which is
typically a set instead of a unique solution.
Based on the individual rationality axiom, the Pareto optimality
axiom, and three additional axioms, Nash (1950) showed that the
optimal solution to the bargaining problem is unique and can
be obtained by solving a straightforward optimization problem.
The three additional axioms are symmetry, linear invariance, and
independence of irrelevant alternatives. The symmetry axiom states
that P ∗ is symmetric: if a vector (a, b) ∈ P ∗ then (b, a) ∈ P ∗ .
The linear invariance axiom states that an order-preserving linear
transformation of a player’s utility function does not affect the
optimal solution of the bargaining problem. Finally, the axiom
184 Agricultural Economics and Policy

of independence of irrelevant alternatives states that removing

irrelevant alternative payoff vectors from set P does not affect the
optimal solution of the bargaining problem. Here the irrelevant
alternative payoff vectors include any vectors except the reservation
payoff vector and the optimal solution of the bargaining problem
based on the original payoff set P .
Based on the aforementioned five axioms, Nash (1950) showed
that the optimal solution of the two-player bargaining game can be
obtained by solving
max (u1 − u1 )(u2 − u2 ) . (17)
(u1 ,u2 )∈P ∗

This two-player bargaining game can be easily expanded to an

n-player game. Let u ≡ (u1 , . . . , un ) and u ≡ (u1 , . . . , u2 ) be the
reservation payoff and payoff of players (1, . . . , n), respectively. Also,
we expand sets P and P ∗ from two dimensions to n dimension
accordingly and label them as Pn and Pn∗ . The optimal solution of the
n-player bargaining game is the solution to the following optimization
problem:
max Πi∈{1,...,n} (ui − ui ). (18)
u∈Pn∗

The n-player bargaining game can be further generalized to

accommodate heterogeneous bargaining power across players (Britz
et al., 2010). Let αi denote the bargaining power (i.e., the weight)
of player i. Then, the bargaining game solution can be obtained by
solving:
max Πi∈{1,...,n} (ui − ui )αi . (19)
u∈Pn∗

In the next section we will see how this model is applied in agricul-
tural economics.

2. Applications in Agriculture

In this section, we will discuss the political economy of biotechnology,

biofuels, and water. We chose these three areas because their devel-
opment was controversial and often involved competition between
 The Political Economy of Agricultural Policies 185

various interest groups. Therefore, they serve as good examples of

applications of political economy in agriculture.

2.1. Economics of biotechnology regulations

Biotechnology in agriculture, such as genetically modified organisms

(GMOs), is perhaps the most controversial topic related to agricul-
ture nowadays. Various interest groups may hold starkly different
opinions toward biotechnology, generating political pressure from
different directions in order to shape the regulation of biotechnology.
Graff et al. (2009) provide a political-economy framework incorpo-
rating multiple competing interests to understand the formation of
biotechnology regulations in the United States, the European Union,
and developing countries. They model the agricultural biotechnology
regulator’s problem as follows:

Max (regulator’s welfare + A1 × consumer surplus

+ A2 × food retailers’ producer surplus
+ A3 × farmers’ producer surplus
+ A4 × major biotechnology suppliers’ producer surplus
+ A5 × new biotech innovators’ producer surplus
+ A6 × competing input suppliers’ producer surplus
+ A7 × academic institutions and scientists’ benefits
+ A8 × activist organizations’ benefits
+ A9 × environmental welfare), (20)

where the parameters A1 to A9 reflect the weight that the regulator

assigns to each interest group. Note that the weights are normalized
against the regulator’s welfare (i.e., the regulator’s welfare has weight
at 1 in Equation (20)). An interest group can influence its weight
by generating pressure on the regulator, by, for example, making
political contributions, lobbying, protesting, or voting. The regulator
maximizes the weighted sum of all interest groups’ welfare, including
his or her own.
186 Agricultural Economics and Policy

Consumer surplus in Equation (20) reflects the heterogeneous

taste, preference, and willingness to pay toward biotechnology.
Studies (e.g., Paarlberg, 2001; Wu, 2004) have shown that consumers
in developed countries do not benefit very much from biotechnology
like GMOs because the food expenditures of these consumers are
already relatively low. Furthermore, these consumers might be
concerned by the potential health risks imposed by GMOs. This
is perhaps why the majority of food consumers in the U.S. largely
ignore the welfare impact of GMOs and influence GMOs only through
individual purchase and voting.
Producer groups listed in Equation (20) have a heterogeneous
preference for GMOs.2 Food retailers may benefit or lose from the
production of GMOs depending on consumers’ response to GMOs.
Organic food stores may benefit from loose regulations of GMOs
because they can easily differentiate their products and charge a
price premium. Farmers may benefit from the increased yields of
genetically modified crops, but farmers who do not adopt GMOs
would be worse off when wide adoption increases crop production
and thus decreases crop prices. Although new biotech innovators will
almost surely benefit from a laissez-faire biotechnology regulation,
existing biotechnology suppliers may gain or lose from such regula-
tion. This is because, on the one hand, a laissez-faire biotechnology
regulation will make entry into the biotechnology sector much easier,
and therefore the incumbents will have more competition from new
entrants. On the other hand, under laissez-faire regulation, it will be
relatively easy for the incumbents to expand their own innovation
and product line. By definition, the competing input suppliers will
suffer from a laissez-faire biotechnology regulation.
Academic institutions and scientists play a unique role in
the political economy of biotechnology. First, their research out-
puts are often consulted in the decision-making process regarding

2
Even within one group, the preference need not be homogeneous. Therefore,
one can modify Equation (20) to divide one group into multiple sub-groups and
assign new weights to these sub-groups. For example, food retailers can be further
divided into large food stores (e.g., Kroger), organic food stores, and farmers’
markets.
 The Political Economy of Agricultural Policies 187

biotechnology regulation. For instance, studies on the health impact

of DDT (an insecticide) partially resulted in the ban of its use in the
U.S. in 1972. Therefore, stringent regulation may increase demand,
and hence financial support, for research from academic institutions
and scientists. Second, academic institutions and scientists may
gain or lose under biotechnology regulations in terms of research
grants, royalty income, and public esteem in the agricultural sector.
The relationship between biotechnology regulations and the gains
or losses of academic institutions is an interesting area for both
theoretical and empirical research.
Graff et al. (2009) noted that one should differentiate the welfare
of activist organizations and environmental welfare, even though
the latter is often the reason to continue the activities of activist
organizations. The welfare of activist organizations is determined,
at least partially, by their funding from grants and donations,
organization size, and reputation. However, environmental welfare
is something that accrues to the whole society, not just the activist
organizations.
Based on the discussion about these interest groups related to
biotechnology, Graff et al. (2009) conjecture that the following groups
are most likely to support a laissez-faire biotechnology regulation:
new biotech innovators, farmers who benefit from the increased yield
or reduced costs caused by biotechnology, and consumers who benefit
from reduced food prices but do not worry much about the potential
risk of biotechnology. Groups that support strict biotechnology
regulation may include consumers who are price-insensitive and
concerned by the potential risk of biotechnologies, brand-conscious
food retailers, competing input suppliers, activist organizations, and
regulators.
Based on the aforementioned model and discussion, Graff et al.
(2009) compared biotechnology regulations in the U.S. and those
in Europe. They pointed out that, although European consumers
could have reduced their expenditures on food had GMOs been
allowed in European markets, the cost of organizing consumer groups
to promote GMOs could be much higher than the reduced food
expenditures. It could also be the case that European consumers
were willing to pay a price premium to avoid the potential health
188 Agricultural Economics and Policy

and environmental risks of GMOs. Therefore, not much pressure

existed from the consumers’ side in Europe to promote the adoption
of GMOs. On the supply side, European farmers effectively benefited
from the ban on GMOs that served as a non-tariff trade barrier,
allowing the farmers to charge a price premium over GMO products.
Perhaps the most powerful opponents to GMOs in Europe are
the producers of conventional pest-control products for the clear
reasons that the adoption of GMOs will reduce the demand for their
products. Finally, activist organizations in Europe are strong.
In many developing countries, GMOs face stringent regulations
as well. Consumers and farmers perhaps would significantly benefit
from GMO adoption that could increase food supply and reduce pro-
duction costs. However, due to the large numbers of consumers and
farmers in developing countries, these two groups cannot efficiently
generate political pressure on regulators for loosening regulations on
GMOs. Producers of traditional agricultural inputs (e.g., pesticides
and herbicides) who would be harmed by GMO adoption can easily
form a group and generate political influence on regulators.
Stringent regulations on GMOs have significantly hindered the
development of “second-generation” biotechnology that is more
focused on the nutrient and environmental traits of crops, causing
large welfare losses to farmers and consumers. Recent breakthroughs
in gene-editing technologies, such as the clustered regularly inter-
spaced short palindromic repeats (CRISPR)/CRISPR-associated
protein (Cas) technologies, have the potential to accelerate innova-
tions in new crop varieties. However, the new technologies are facing
similar regulations and low social acceptance to those of GMOs in
Europe and many developing countries (Miao and Khanna, 2020).
Further research from scientists and economists on the benefits and
risks of these new technologies will assist societies in understanding
the true values of these technologies with more certainty.

2.2. Political economy of biofuels

In Chapter 5, we employed social welfare models to analyze the

welfare gains or losses of consumers and producers, as well as changes
in government outlays due to biofuel production and policy. In this
 The Political Economy of Agricultural Policies 189

subsection, however, we shift to the political economy and examine

some key factors that influenced biofuel policies in the U.S., Brazil,
and some other countries following Zilberman et al. (2014).
Similar to Equation (20), we develop a political economy model
of biofuel policies as follows:

Max(policymakers’ welfare
+ B1 × food and fuel consumers’ surplus
+ B2 × farmers’ producer surplus
+ B3 × environmentalists’ welfare
+ B4 × oil companies’ producer surplus
+ B5 × first-generation biofuel producers’ surplus
+ B6 × second-generation biofuel producers’ surplus
+ B7 × other energy producers’ surplus
+ B8 × automobile companies’ producer surplus
+ B9 × welfare of some other groups), (21)

where parameters B1 to B9 are the weight assigned to different

interest groups by policymakers.
During the early 2000s, U.S. policymakers were concerned with
energy independence, the trade deficit, and climate change. Develop-
ing biofuels can serve as means to mitigate these concerns. Obviously,
corn ethanol produced domestically will replace gasoline and thus
reduce crude oil imports, which will both enhance energy indepen-
dence and decrease the trade deficit. Although not policymakers’
top priority back then, mitigating climate change was also cited as a
reason for supporting biofuels in the United States. In Brazil, a major
driver of sugar cane ethanol in the 1970s was the concern of trade
deficit, which did not allow the country to import much crude oil.
Changes in food and fuel consumers’ surplus due to biofuels
vary significantly among countries. In developed countries such as
the U.S., the production of biofuels only had a mild impact (about
5%) on retail food prices (Harrison, 2009). Because only a small
190 Agricultural Economics and Policy

portion of household income is devoted to food in these countries,

this mild food price increase did not translate into large welfare loss
to consumers. In developing countries, however, since a large portion
of household income was spent on food, food price increases caused
large welfare losses to consumers (Wright, 2014). Fuel consumers
gained from biofuel because fuel price was lowered due to biofuel
production, but the gains mainly accrued to consumers in developed
countries because many consumers in developing countries do not
have cars (Zilberman et al., 2014).
Obviously, corn producers in the U.S. and sugarcane producers in
Brazil benefited from biofuel production. Thus, they supported bio-
fuel development. Environmentalists’ support for biofuel, however,
is significantly affected by research results on biofuel’s greenhouse
gas (GHG) emission reduction benefits. After the publication of two
Science articles in 2008 (i.e., Fargione et al., 2008; Searchinger et al.,
2008) which showed that biofuel production might actually increase
GHG emissions, environmentalists’ support for the corn-starch-based
biofuel largely shifted to disapprobation, becoming more inclined to
second-generation biofuels. Most oil companies and producers viewed
biofuel as a competitor and thus opposed it, although a few big
oil companies (e.g., BP and Shell) invested in the development of
biofuels.
Zilberman et al. (2014) noted that both first- and second-
generation biofuel producers benefited from biofuel production.
Particularly, for the first-generation biofuel producers who are also
corn growers, the benefits arose from both the rising corn price
and corn ethanol profits. Producers of other alternative energy, such
as solar and wind energy, viewed biofuel as a competitor because
they were competing for a limited amount of government support
and natural resources, such as land. Likewise, companies that make
electric vehicles may oppose biofuels while their flex fuel counterparts
would support biofuels. Finally, some other interest groups such as
airlines and the military welcomed the development of biofuels due
to concerns about stricter climate regulation, energy independence,
and the impossibility to power their aircraft using batteries in the
near future.
 The Political Economy of Agricultural Policies 191

2.3. Political economy of water

Mark Twain supposedly said, “Whiskey is for drinking, water is

for fighting.” Conflicts about water ownership and use have existed
forever and are subject to political debate and regulation. In
particular, the allocation of water rights, water transactions, and
water quality has been subject to policy debate and regulation.
Zilberman et al. (2023) argue that several factors affected the
evolution of water policy in the United States and other countries
over time. First, the relative abundance of water. Second, the
emphasis on supply expansion and development. Third, the capacity
of the government to invest in water projects. Fourth, the concern
for the environment. During early settlement in the U.S., the federal
government was relatively poor but had the power to distribute
rights to water and other resources. The government was interested
in the expansion of agriculture, which led to the introduction of
water rights, where individuals established rights to use water by
diverting them from larger bodies. With the prior appropriation
system prevalent in the West, priority in time set priority in use,
and there were restrictions on water trading. In the early days,
farmers established water districts to divert water for mining, urban
consumption, and agricultural activities. Starting at the beginning
of the 20th century, as the government’s financial capacity improved,
it invested in water projects for irrigation and hydroelectric power.
Some of these water projects did not pass benefit-cost analysis, but
the power of agriculture interest groups and real-estate development
led to their establishment. However, both the emergence of the
environmental movement and increased concern about government
spending led, in the 1970 and 1980s, to the establishment of principles
and guidelines for new water project developments that approximate
benefit-cost analysis, and they resulted in the decline of new water
projects. With constraints on supply expansion, irrigated agriculture
becomes more intensive, relying more heavily on sprinkler irrigation,
low-pressure center pivot irrigation, and drip irrigation.
As water scarcity increased, there was an increasing awareness
of the need to move from water rights to water trading, especially
192 Agricultural Economics and Policy

in California and Australia, where water rights owners were not

allowed to sell water. The survey by Schoengold and Zilberman
(2007) reviewed studies that suggest the transition from water rights
to water markets may lead to the adoption of new technologies,
increased agricultural productivity, and overall social welfare. How-
ever, the owner of water rights may lose from the transition to
water markets if they are not allowed to keep most of the revenue
from the sale of water. Therefore, the introduction of transferable
rights (so water rights owners keep most of the proceeds from the
sale) is more politically accepted by water rights owners than if the
state keeps significant parts of the proceeds. Some environmental
groups may oppose moving to water trading as it may increase
agricultural production and reduce wilderness. Furthermore, the
transition to water markets from water rights requires monitoring
and new equipment investment. Uncertainty about water rights is
another reason for the delay of the transition to markets, especially
when the state has a limited capacity to enforce reallocation in an
emergency. Partial trading was introduced in California in response
to drought situations, but trading remains limited and temporary due
to the state’s legal constraints and limited capacity to impose reform.
On the other hand, Australia has moved to a water trading system
in response to the Millennium drought (2001–2009) (Hanemann and
Young, 2020).
In much of the world, groundwater is the main source of
irrigation, and traditionally, landowners have been entitled to pump
the water under their land. That led to inefficiency because of the
“Tragedy of the Commons” and depletion of groundwater resources.
Government actors may have recognized a need to control over-
pumping, but farming communities may have used their political
influence to oppose it. In contrast, other politicians might have
opposed restrictions on groundwater pumping because of its short-
term impact on food prices or supplies. The result is the tendency to
over-pump, which may trigger a depletion of aquifers and a water and
farming crisis. In some cases, a political establishment will invest in a
water project to augment groundwater (if the government resources
and farming lobbies are strong). In other cases, farming activities
may decline, leading to poverty and migration.
 The Political Economy of Agricultural Policies 193

Water quality is another source of political debate. Increased

fertilizer use has led to a large accumulation of nitrates and other
chemicals in the groundwater. Furthermore, runoff from agricultural
land may lead to the transfer of chemicals to lakes and other bodies
of water, resulting in eutrophication; harmful algal blooms that lead
to fish kills and dead zones. An important example is the dead zone
in the Gulf of Mexico, where the alleviation requires fertilizer use
reduction in the Midwest and thus the cooperation of farmers. There
is a growing need for creative technological and politically feasible
solutions. Water quality contamination exists in many other parts
of the world, endangering human health and reducing clean water
availability. One solution is the development of affordable and sound
environmental technologies to cleanse the water and allow its use
for consumption, an expensive implementation but already in effect
in countries such as Israel and Spain (Simpson, 2018; Zarzo, 2021).
Further development of these technologies and policies that enhance
their adoption requires a significant public investment and may take
time (Zilberman et al., 2023).
The increased scarcity of water resources and increased techno-
logical sophistication and demand for water are pushing increased
reliance on water markets and trading that reduce inefficiency.
However, the pace of the change is affected by political and economic
situations that vary across locations (Saleth and Dinar, 2004).
Thus, seeking solutions that enhance efficiency while overcoming said
constraints offers a valuable line of research.

3. Conclusion
In this chapter, we have introduced a few popular political economy
models including the median voter model, the spoil-sharing model,
the competing interest model, and the Nash–Harsanyi bargaining
game. Their applications in agriculture were discussed with a focus on
biotechnology regulation, biofuel policies, as well as water rights and
trading. Political economy is an expansive research field and offers
a unique perspective to understanding policy evolution and resource
allocation. It can be applied to many areas in agricultural economics
and what we have discussed in this chapter are only a few, although
194 Agricultural Economics and Policy

important, applications. For instance, agricultural economists have

examined crop insurance and agricultural innovations from the
perspective of political economy (e.g., Robert, 2003; Alston and
Pardey, 2021; Chapter 7 in this book). We leave them to our readers
for further exploration.

References
Alston, J.M. and P.G. Pardey. 2021. The Economics of Agricultural Innovation.
Handbook of Agricultural Economics (Vol. 5, Chapter 75). Elsevier, Amster-
dam.
Anderson, K., G. Rausser, and J. Swinnen. 2013. Political Economy of Public
Policies: Insights from Distortions to Agricultural and Food Markets. Journal
of Economic Literature 51(2): 423–477.
Becker, G.S. 1983. A Theory of Competition among Pressure Groups for Political
Influence. The Quarterly Journal of Economics 98(3): 371–400.
Besley, T. and S. Coate. 1997. An Economic Model of Representative Democracy.
The Quarterly Journal of Economics 112(1): 85–114.
Black, D. 1948. On the Rationale of Group Decision-Making. Journal of Political
Economy 56(1): 23–34.
Britz, V., P.J.-J. Herings, and A. Predtetchinski. 2010. Non-Cooperative Support
for the Asymmetric Nash Bargaining Solution. Journal of Economic Theory
145(5): 1951–1967.
Downs, A. 1957. An Economic Theory of Political Action in a Democracy. Journal
of Political Economy 65(2): 135–150.
Graff, G.D., Gal Hochman, David Zilberman. 2009. The political economy of
agricultural biotechnology policies. AgBioForum 12(1): 34–46.
Grossman, G.M. and E. Helpman. 1994. Protection for Sale. The American
Economic Review 84(4): 833.
Fargione, J., J. Hill, D. Tilman, S. Polasky, and P. Hawthorne. 2008. Land
Clearing and the Biofuel Carbon Debt. Science 319(5867): 1235–1238.
Hanemann, M. and M. Young. 2020. Water Rights Reform and Water Marketing:
Australia vs the US West. Oxford Review of Economic Policy 36(1): 108–131.
Harrison, R.W. 2009. The Food versus Fuel Debate: Implications for Consumers.
Journal of Agricultural and Applied Economics 41(2): 493–500.
Harsanyi, J.C. 1962a. Measurement of Social Power, Opportunity Cost, and the
Theory of Two-Person Bargaining Games. Behavioral Science 7: 67–80.
Harsanyi, J.C. 1962b. Measurement of Social Power in N-Person Reciprocal Power
Situations. Behavioral Science 7: 81–91.
Harsanyi, J.C. 1963. A Simplified Bargaining Model for N-Person Cooperative
Game. International Economic Review 4: 194–220.
Hotelling, H. 1929. Stability in Competition. The Economic Journal 39(153):
41–57.
 The Political Economy of Agricultural Policies 195

Innes, R. 2003. Crop Insurance in a Political Economy: An Alternative Perspective

on Agricultural Policy. American Journal of Agricultural Economics 85(2):
318–335.
Nash, J.F. 1950. The Bargaining Problem. Econometrica 18: 155–162.
Nash, J.F. 1951. Non-Cooperative Games. Annals of Mathematics 54: 286–295.
Nash, J.F. 1953. Two-Person Cooperative Games. Econometrica 21: 128–140.
Olson, M. 1965. The Logic of Collective Action: Public Goods and the Theory of
Groups. Harvard University Press.
Paarlberg, R.L. (2001). The Politics of Precaution: Genetically Modified Crops in
Developing Countries. Baltimore, MD: Johns Hopkins University Press.
Rausser, G.C. 1982. Political Economic Markets: PERTs and PESTs in Food and
Agriculture. American Journal of Agricultural Economics (1982): 821–833.
Rausser, G.C. 1992. Predatory Versus Productive Government: The Case of US
Agricultural Policies. Journal of Economic Perspectives 6(3): 133–157.
Rausser, G.C. and W.E. Foster. 1990. Political Preference Functions and Public
Policy Reform. American Journal of Agricultural Economics 72(3): 641–652.
Rausser, G.C., J. Swinnen, and P. Zusman. 2011. Political Power and Economic
Policy: Theory, Analysis, and Empirical Applications. Cambridge University
Press, Cambridge.
Saleth, R.M. and A. Dinar. 2004. The Institutional Economics of Water: A Cross-
Country Analysis of Institutions and Performance. World Bank Publications.
Schoengold, K. and D. Zilberman. 2007. The Economics of Water, Irrigation, and
Development. Handbook of Agricultural Economics 3: 2933–2977.
Searchinger, T., R. Heimlich, R.A. Houghton, F. Dong, A. Elobeid, J. Fabiosa,
S. Tokgoz, D. Hayes, and T.-H. Yu. 2008. Use of U.S. Croplands for Biofuels
Increases Greenhouse Gases Through Emissions from Land-Use Change.
Science 319(5867): 1238–1240.
Simpson, K. 2018. What the world can learn from Israel’s water reuse pro-
grams. Medill Reports. Medill School of Journalism, Northwestern University.
October 18. Available at: https://news.medill.northwestern.edu/chicago/
what-the-world-can-learn-from-israels-water-reuse-programs/ (accessed Jan-
uary 1, 2024).
Swinnen, J. and F.A. van der Zee. 1993. The Political Economy of Agricultural
Policies: A Survey. European Review of Agricultural Economics 20(3):
261–290.
Waterfield, G., S. Kaplan, and D. Zilberman. 2020. Willingness to Pay Versus
Willingness to Vote: Consumer and Voter Avoidance of Genetically Modified
Foods. American Journal of Agricultural Economics 102(2): 505–524.
Wright, B. 2014. Global Biofuels: Key to the Puzzle of Grain Market Behavior.
The Journal of Economic Perspectives 28(1): 73–97.
Wu, F. (2004). Explaining public resistance to genetically modified corn: An
analysis of the distribution of benefits and risks. Risk Analysis 24(3): 715–
726.
Zarzo, D. 2021. Water Reuse in Spain. IDA Global Connections, Fall
2021 Issue. Available at: https://issuu.com/idadesal/docs/digital aaff ida
fall21/30 (accessed January 1, 2024).
196 Agricultural Economics and Policy

Zilberman, D. 1984. Technological Change, Government Policies, and Exhaustible

Resources in Agriculture. American Journal of Agricultural Economics 66(5):
634–640.
Zilberman, D., G. Hochman, S. Kaplan, and E. Kim. 2014. Political Economy of
Biofuel. Choices 29(1): 1–5.
Zilberman, D., A. Huang, L. Goldberg, and T. Reardon. 2023. The Evolution
of Symbiotic Innovation, Water, and Agricultural Supply Chains. Applied
Economic Perspectives and Policy 45(3): 1592–1603.
Zilberman, D., S. Kaplan, E. Kim, and G. Waterfield. 2014. Lessons from the
California GM Labelling Proposition on the State of Crop Biotechnology.
In Handbook on Agriculture, Biotechnology and Development. Edward Elgar
Publishing.
 Chapter 7

Agricultural Innovation

Innovation gave birth to new technologies that had changed agricul-

ture. From the self-scouring steel plow in the 1830s to automated
combines nowadays, and from synthetic nitrogen fertilizer in the
early 20th century to today’s nitrogen-fixing function of genetically
edited crops (Yan et al., 2022), new technologies have offered
agriculture numerous means to increase productivity and reduce
production externalities. In this chapter, we provide an overview of
the innovation literature in agricultural economics, discussing the
definition and measurement of technological change, a few key models
and their insights, as well as political economy and public policies
about agricultural innovation or innovation in general. We then
conclude with some interesting and possibly fruitful future research
directions in the realm of agricultural innovation.

1. Nature of Technological Change and R&D

The first set of questions we need to ask are: What is technological

change and how is it measured? In short, technological change reflects
the ability to get more output from the same amount of inputs. It
can be reflected by an increase in an index of quantity of output per
unit of input, or by a change in production function parameters.

197
198 Agricultural Economics and Policy

The first controversy we encounter in the economic literature

on technological change concerns how one accounts for changes in
input use when measuring technological change. Here we have two
schools of thought.1 First, Heady (1949) and Denison (1967) claim
that technological change accounts for all the growth in output that is
not accounted for by growth in physical input use. They measure an
exogenous shift of isoquants where input measures are unchanged.
They include changes in quality of input as part of technological
change and measure use of input in each single period. The second
school of thought, represented by Jorgenson and Griliches (1967),
claims that technological change reflects changes in production
functions that cannot be explained otherwise. Specifically, techno-
logical change reflects changes that cannot be explained in terms
of increased quality or quantity of inputs. Technological changes are
measured in input-output ratios when quality changes are considered.
This thought enlightens two manifestations of technological change:
increase in knowledge of the production process and improvement in
input quality. Accordingly, Peterson and Hayami (1977) define tech-
nological change as “[t]he phenomena of input quality improvements
or an increase in knowledge leading to increase in output per unit of
input” (p. 500).
This definition tries to generate a compromise in the embodied-
disembodied argument. For years, economists have asked: do techno-
logical changes embody themselves in new machinery (input changes)
or are they slow processes of change in production techniques
(management)? The disembodied hypothesis relied on evidence from
the airplane industry when the time required to assemble a plane
declined with the increase in the number of planes produced. People
developed the notion of a “learning curve” where output/labor =
f (time) or output/labor = f (output); function f is found to be
logarithmic. This relationship became the basis for Arrow’s (1962)
“learning by doing” theory.
The book by Salter (1960) is a cornerstone of an alternative
approach: the “Putty Clay” approach. Here, as we have seen in our

1
Godin (2015) documented the evolvement of the “technological change” concept
in the 20th century.
 Agricultural Innovation 199

production chapter, it is assumed that there is a choice of input-

output coefficients before a new technology is introduced and, once
an investment takes place, an input/output ratio is fixed. In other
words, before the new machinery is introduced, the production is in
the putty stage; after the new machinery is adopted, the production
has the “clay” and the input/output ratio is virtually constant. The
new technology is considered embodied in the machinery that need
to be bought. Technological change reflects itself in the putty stage
that every year the machinery is better. Note, however, that these two
theories — embodied versus disembodied — are more complementary
than contradictory. It is true that new machinery, embodying new
knowledge, is introduced constantly, but use efficiency improves with
experience, as the user acquires new managerial ability.
A second important issue relates to the direction of technological
change. With technological change, is the neoclassical production
function (the ex-ante one, a la Salter, 1960) changing in a neutral or
a biased manner? Technological change can be described as a shift
of the unitary isoquant toward the origin (one unit of output can
be produced with less inputs), but the shift can be “parallel” (that
is, the quantity of all inputs declines proportionally to their use) or
biased (some inputs reduce more than other).
Suppose the production function is f (k, l), where k and l stand
for capital and labor, respectively. There are three measures of
technological bias. All three measure what occurs to the ratio of
capital share to labor share (i.e., fk k/fl l, where fk is the marginal
product of capital k and fl is the marginal product of labor l) along
a certain line and can be interpreted as measures of what happens
to the ratio of overtime along that particular line. Hicks’ measure
focuses on bias along the fixed capital/labor ratio; Harrod’s measure
pays attention to bias along fixed capital/output ratio; and, finally,
Solow’s measure emphasizes bias along fixed labor/output ratio. For
instance, Hick’s measure of technological bias can be written as,
  ⎧
∂ ffkl kl ⎨ labor saving
|given k  0 ⇒ neutral (1)
∂t l ⎩
capital saving,

where t stands for time.

200 Agricultural Economics and Policy

Given these definitions, the question is: What is the nature of

technological change in the real world? According to Salter (1960),
technological changes are neutral. He states that manufacturers want
to minimize cost and do not care if this is due to labor or capital
savings. Therefore, they will encourage research possibilities that
reduce both capital and labor requirements. He agrees that observed
data indicate that capital/labor ratio is increasing over time, but he
explains it as the consequence of the increase in the relative price of
labor (the economy is moving along the iso-quants toward a more
intensive use of capital, but the shape of the isoquants remains
unchanged). Salter criticizes an alternative approach, the induced
innovation approach, which maintains that research and innovation
efforts that shift production functions are affected by relative prices
(more to follow on the induced innovation approach).
Hayami and Ruttan (1970) apply the induced innovation
approach to show that differences in endowment determine differ-
ences in relative prices which affect the direction of R&D in different
countries. Binswanger (1974) extended their work and developed an
empirical model for understanding technological bias. Atkinson and
Stiglitz (1969) have a theory following Joan Robinson’s ideas. Given
an initial isoquant, an initial technology, the R&D activity is close
to this technology and the ex-ante production function becomes less
and less elastic and sensitive as prices change. Thus, even in the
ex-ante production function, we may find regions of extreme rigidity.
The result will be that a country might be very efficient along one
k/l ratio, but less so along another.

2. Understanding Directions of Technical

Changes: Predominant Models
One focus of the innovation literature concerns the directions of
technical changes and the drivers of the directions. Are technologies
becoming more labor-augmenting or land-augmenting? What are
the underlying forces that determined the directions of technical
 Agricultural Innovation 201

changes? There are two major strands of literature in this realm. One
proposed the induced innovation hypothesis and the other introduced
the directed innovation hypothesis. In this section, we discuss them
briefly by focusing on two representative papers.

2.1. Induced innovation hypothesis:

Binswanger (1974)

Before we introduce this model, note that it is agreed that tech-

nological change includes many processes, but two are important:
innovation and diffusion. Both processes involve investment —
innovation is an investment in R&D and diffusion is an investment
in the products of R&D. Here, we analyze the determination of the
direction of an innovative effort. It is assumed that the innovator
is the user of the new technology — an unlikely assumption in
agriculture, where there is a separation between a developer of new
technology and its users. However, it is reasonable in other industries,
where a firm develops, in part or totally, its technologies. This model
is also relevant for a competitive economy.
The model is based on the Evenson and Kislev (1976) model
of applied research, where research is viewed as a search for new
products, input processes, and so on. Research outcome is a random
draw from a distribution. Basic research defines the distribution —
probability of success — and applied research uses the probability
rule and searching. To fix ideas, Evenson and Kislev (1976) model
the optimal number of trials in a series of agricultural experiments,
searching for new crop varieties with higher yields. Binswanger (1974)
assumes that R&D activities can be modeled using a similar frame-
work. The output of R&D activities are parameters of production
function and, as the R&D effort increases, one may expect to see
technologies with lower input-output ratios.
 LetL us assume that the firm’s production function is Y =
f K A , B , where A and B are measures of capital and labor pro-
ductivity and, as they decline, productivity increases. Now assume
that we start with initial parameters, A0 and B0 , and implement
202 Agricultural Economics and Policy

R&D efforts to develop new technology. The measure of success is

A0 − A1 B0 − B1
A∗ = , and B∗ = . (2)
A0 B0
Suppose there are two research lines, M and N. M is mainly
capital saving and N is labor-saving. Let n be the number of research
units (or simply, research input) under research line N, and m be
the number of research units under research line M. Let αN be the
impact of one unit of successful research type N on capital saving,
and αM be the impact of one unit of successful research type M on
capital saving. The impact of one unit of successful research on labor-
saving, β N and β M , is defined analogously. Finally, let μ(m) denote
the expected amount of successful research type M when input is m,
where μ > 0 > μ . Similarly, let μ(n) denote the expected amount
of successful research type N when input is n. The two research lines
are complements if αM > β M > 0 and β N > αN > 0. The two
research lines are substitutes if αM > 0 > β M and β N > 0 > αN .
The total expected effect of R&D are
A∗ = μ (n) αN + μ (m) αM = A∗ (m, n),
and B ∗ = μ (m) β M + μ (n) β N = B ∗ (m, n). (3)
Assume a Leontief production function, Y = min( K L
A , B ) and a fixed
output Y (imagine building a plant with fixed capacity). Suppose
that if there is no R&D, then K0 and L0 are needed to produce Y .
However, firms can conduct R&D to improve the efficiency of capital
and labor and perhaps reduce the need for these two inputs. Risk-
neutral firms will maximize the expected profits by choosing optimal
R&D investment m and n:
max[P Y − RK0 − W L0 + RK0 A∗ (m, n)
m,n

+ W L0 B ∗ (m, n) − mP M − nP N ], (4)
where P , R, W , P M , and P N are prices of final product, capital,
labor, research input for type M, and research input for type N,
respectively. Solving this optimization problem yields the optimal
R&D levels in the two research lines, m∗ and n∗ .
 Agricultural Innovation 203

By defining technological bias as

⎧
⎨capital saving
∗ ∗ ∗ ∗ ∗ ∗
Technological Bias = A (m , n )−B (m , n )  0 ⇒ neutral ,
⎩
labor saving
(5)
Binswanger (1974) found that bias and research scale depend on the
price and productivity of research line, cost of each factor, and the
output level. For instance, if research line M is more likely to be
successful (i.e., larger μ(m)) or to have larger impact on efficiency
(i.e., larger αM ), then technological change will be more likely to be
capital-saving. Likewise, a decrease in input price of research line M
will have the same impact on the direction of technological change.
Moreover, larger production level, Y , will call for more research.
Finally, factor price ratio, W/R, also plays a role in determining
the direction of bias. If labor becomes relatively more expensive (i.e.,
increasing W/R) then technological change tends to be labor-saving,
which is consistent with the induced innovation hypothesis.
The work by Binswanger (1974) is among a host of endeavors
that seek to establish the microeconomic foundation for the induced
innovation hypothesis. Nevertheless, the theoretical ground of the
model is limited. For instance, the production function is in a
restricted form, Leontief, and the model ignores risk and risk
aversion. To include risk and risk aversion, one can re-write the
optimization problem as,
K L
max EU P F , − W L − RK − mP M − nP N ,
m,n A B
s.t., A = A0 [1 − A∗ (m, n)] and B = B0 [1 − B ∗ (m, n)]. (6)

Another limitation of Binswanger’s (1974) model is that it ignores

market structure. In agriculture the farm input sector is independent
from the farming sector. The input producers are monopolists or
oligopolists and use their power to determine the quantity and quality
of research. The input producer has to take into account the realities
of the adoption process in their decision-making, and R&D decisions
204 Agricultural Economics and Policy

depend on adoption procedure parameters and are also determined

simultaneously with promotion choices. Moreover, the R&D process
in the Binswanger (1974) model is limited to biological and chemical
technological changes. In mechanical changes, there is a specific
dynamic aspect that is ignored. The research process can be then
described by an optimal control model.

2.2. Directed innovation hypothesis: Acemoglu (2002)

In what follows we briefly introduce the model in Acemoglu

(2002) that proposes the directed innovation hypothesis. Let y =
F (L, Z, A) denote the aggregate production function, where y
is output, L is labor, Z can be viewed as capital or land,
and A stands for a technology index. We further assume that
∂F/∂L > 0, ∂F/∂Z > 0, and ∂F/∂A > 0. Thus, larger A indicates
technological improvement. Acemoglu (2002) defines that, if the
production function is in the form of y = F (AL, Z) then a tech-
nological improvement is called L-augmenting; if the production
function has the form of y = F (L, AZ) then a technological improve-
ment is called Z-augmenting. Moreover, a technological change is
called L-biased if ∂((∂F/∂L)/(∂F/∂Z))/∂A > 0 and Z-biased if
∂((∂F/∂Z)/(∂F/∂L))/∂A > 0. Note that an L-augmenting tech-
nological improvement is not necessarily an L-biased improvement.
Consider the following constant elasticity of substitution (CES)
production function
σ−1
 σ
σ−1 σ−1
y = γ· (AL L) σ + (1 − γ) · (AZ Z) σ , (7)

where γ ∈ (0, 1) is a share parameter reflecting the importance of

a factor, AL and AZ are technology indices, and σ > 0 but σ = 1
is the elasticity of substitution parameter between factors L and Z.
When σ approaches to 1 (respectively, 0 and +∞), then the function
is Cobb–Douglas (respectively, Leontieff and linear). When σ > 1,
Acemoglu (2002) views the two factors as gross substitutes because
in this case the demand for one factor increases as the price of the
other factor increases. When σ < 1, the two factors are viewed as
gross complements. Clearly, AL is L-augmenting whereas AZ is Z-
augmenting.
 Agricultural Innovation 205

If we define Ω as the ratio of marginal product of Z to marginal

product of L, then it is readily checked that
σ−1
− σ1
∂y/∂Z 1−γ AZ σ Z
Ω≡ = . (8)
∂y/∂L γ AL L
From this equation we obtain
− σ1
∂Ω 1−γ σ−1 1 AZ Z
= · · · · . (9)
∂AZ γ σ AL AL L
Equation (9) shows that if σ > 1 (i.e., L and Z are gross substitutes)
then ∂Ω/∂AZ > 0 and thus AZ is Z-biased, and that if σ < 1 (i.e.,
L and Z are gross complements) then AZ is instead L-biased. Simi-
larly, if σ > 1 then AL is L-biased, and if σ < 1 then AL is Z-biased.
Note that an increase in AZ will increase the marginal product of
both Z and L. However, when σ < 1 (i.e., L and Z are gross com-
plements), an increase in AZ will increase the marginal product of L
by a larger amount than that of Z. When σ > 1 the opposite is true.
The model focuses on two sectors: an intermediate good sector
and an innovating sector.2 The former produces a labor-intensive
good, denoted by YL , and a capital-intensive good, denoted by YZ .
When producing YL , L-augmenting innovations are used, and when
producing YZ , Z-augmenting innovations are used. The innovating
sector consists of technology monopolists who supply technologies
that are used in the intermediate goods sector. Specifically, YL and
YZ are specified by
 NL
1
YL = xL (j)1−β dj Lβ and
1−β 0
 NZ
1
YZ = xZ (j)1−β dj Z β , (10)
1−β 0

2
To close the model, a consumer utility function with constant relative risk
aversion and an aggregate production function with constant elasticity of
  −1 −1
  
−1
substitution i.e., Y ≡ γYL  + (1 + γ) YZ  are included in the
model. In each period, the sum of consumption, investment, and R&D expenditure
is no higher than the aggregate production level.
206 Agricultural Economics and Policy

where β ∈ (0, 1) is a parameter; xL (j) is L-augmenting innovation

of type j and NL stands for the range of L-augmenting innovations;
and xZ (j) and NZ can be interpreted similarly for Z-augmenting
innovations.
Assuming competitive markets for intermediate goods, the opti-
mization problems of firms that produces YL or YZ are
 NL
max pL YL − wL L − κL (j) xL (j)dj and
L,{xL (j)} 0
 NZ
max pZ YZ − wZ Z − κZ (j) xZ (j)dj, (11)
Z,{xZ (j)} 0

where pL and pZ are prices of goods YL and YZ , respectively; wL and

wZ are prices of factors L and Z; and κL (j) and κZ (j) are prices of
innovations xL (j) and xZ (j), respectively. Taking these prices as well
as NL and NZ as given, the first-order conditions of the optimization
problems in (11) are:

1/β
pL
xL (j) = L and
κL (j)
1/β
pZ
xZ (j) = Z, (12)
κZ (j)
 NL
β
wL = pL xL (j)1−β dj Lβ−1 and
1−β 0
 NZ
β
wZ = pZ xZ (j)1−β dj Z β−1 . (13)
1−β 0

Equation (12) shows that the demand for innovations increases

in the output prices of the intermediate goods and factors employed
but decreases in innovation prices. Equation (13) simply shows the
marginal cost of a factor (L or Z ) is equal to its marginal benefit.
We now examine the innovating monopolists’ optimization prob-
lem. For simplification, we assume that the marginal cost of all
innovations is the same and denoted by ψ. Therefore, the profit
 Agricultural Innovation 207

maximization problem of supplying innovation xL (j) is

max {πL (j) = (κL (j) − ψ) xL (j)}. (14)

κL (j)

Based on xL (j) in Equation (12), the first-order condition in

optimization problem (14) implies that the profit-maximizing price
for the innovation is kL (j)∗ = ψ/(1 − β). For simplicity, Acemoglu
(2002) normalizes ψ to be 1 − β. Therefore, the optimal price
for L-augmenting innovations is 1. Similarly, the optimal price
for Z-augmenting innovations is also 1. The innovation demand
in Equation (12) and the optimal prices demonstrate that maxi-
mized instantaneous profits from supplying the L-augmenting and
Z-augmenting technologies are
1/β 1/β
πL = βpL L and πZ = βpZ Z. (15)

Consider a steady state and let the interest rate be r. Then

the innovating monopolists’ net present values from supplying the
innovations over the lifetime can be written as
1/β 1/β
βpL L βpZ Z
VL = and VZ = , (16)
r r
from which we obtain Acemoglu’s (2002) key message:
1  σ−1
− σ1
VZ pZ β Z 1−γ σ NZ Z σ
= · = , (17)
VL pL L γ NL L

where σ ≡ − ( − 1)(1 − β) is the derived elasticity of substitution

between factors L and Z.3 Note that the relative profitability from

 − 1
3 pZ ∂Y /∂YZ 1 − γ YZ 
The second equation in (17) holds because (a) = =
pL ∂Y /∂YL γ YL
1 (1−β)/β
in competitive market equilibrium; and (b) YL = p NL L and
1 − â L
1 (1−β)/β
YZ = p NZ Z obtained by substituting (12) into (10).
1−β Z
208 Agricultural Economics and Policy

Z-augmenting innovations and L-augmenting innovations, VZ /VL ,

determines innovators’ incentive to supply a certain type of inno-
vation. The larger the VZ /VL value, the greater incentive to provide
Z-augmenting innovation. The first equation in (17) shows that two
forces influence the relative profitability and thus the direction of
technical changes: the price effect, denoted by (pZ /pL )1/β , and
the market size effect, denoted by Z/L. Clearly, the price effect
will increase the incentive to supply innovations that produce the
more expensive intermediate goods, while the market size effect will
increase innovations that augment the more abundant factor. Note
that an exogenous increase in relative abundance of one factor has
both price and market size effects on the direction of innovation.
The price effect occurs because an increase in the abundance of
a factor decreases its relative price, and, therefore, decreases the
incentive to supply, whereas an increase in the abundance increases
the market size for innovation augmenting this factor, and, therefore,
increases the incentive to supply. If the price effect dominates the
market size effect, then an increase in the relative abundance of a
factor will decrease innovations augmenting this factor. However, if
the market size effect dominates the price effect, then the opposite
is true.
The second equation in (17) reveals conditions under which one
effect dominates the other. The conditions are strikingly straightfor-
ward: If σ < 1, then the price effect dominates the market size effect;
if σ > 1, then the opposite is true. The intuition is as follows. As
noted earlier, when σ < 1 then the two factors are gross complements.
In this case, an increase in the abundance of factor Z would increase
the demand for factor L, and thus its price, significantly. Therefore,
the relative price between Z and L will decrease so much that the
overall incentive to supply Z-augmenting innovations will decrease.
When σ > 1 then the two factors are gross substitutes, and thus
an increase in the abundance of Z will decrease the prices of both
Z and L. In this case, the relative prices of these two factors may
not change very much and therefore the price effect is moderate and
dominated by the market size effect.
 Agricultural Innovation 209

Interestingly, by using Equation (13) we can re-write the first

equation in (17) as
VZ wZ Z NZ
= · · , (18)
VL wL L NL
which shows that the induced innovation hypothesis proposed by
Hicks (1932) only partially captured the incentives to innovate by
emphasizing on the relative factor price, w Z /wL , while overlooking
the market size effect that is reflected by Z/L.

3. Politics, Policies, and Innovations

In addition to market forces, political forces or public policies can
also have a significant influence on innovations. In this section,
we discuss three important papers that examine the relationship
between innovation and political forces or public policies.

3.1. Political forces and innovation: de Gorter

and Zilberman (1990)

Although innovation was not directly mentioned, de Gorter and

Zilberman (1990) analyze the role of political influence and market
features on public expenditure, with various R&D projects as
motivating examples. Focusing on a closed economy under perfect
competition markets, they highlight the role of demand elasticity
and political influence on R&D investment. In this subsection, we
will briefly review their model.
Let U (q, Z) = u (q)+Z be the utility function of a representative
consumer, where q is a specific commodity (e.g., wheat) and Z
is a numeraire good. On the production side, let C(q, E) denote
the cost function of a representative producer who produces the
commodity, with E representing the public expenditure on R&D.
The funding source of the public expenditure can be consumers (via
tax), or a group of producers (via, e.g., producer associations), or
both. Suppose that α ∈ [0, 1] is the portion of public expenditure
210 Agricultural Economics and Policy

sponsored by consumers via tax. Obviously, if the public expenditure

is fully funded by consumers (respectively, producers) then α = 1
(respectively, α = 0). We assume that Cq > 0, Cqq > 0, CE < 0,
CEE > 0, and CqE < 0 to ensure that the cost function is convex in
both q and E, and that an increase in E will reduce the marginal
cost of q (i.e., shifting the supply curve rightward).
Let p be the price of the commodity and Y be the income of
the representative consumer. The consumer maximizes her utility by
choosing the consumption quantity:

max {u (q) + Y − pq − αE}. (19)

q

For the representative producer, the profit-maximizing problem can

be written as,

max {pq − C (q, E) − (1 − α)E}. (20)

q

Due to perfect competition, both consumers and producers are price

takers. We can thus derive the demand and supply relationships
based on problems (19) and (20) as follows: q D = D (p) ≡ u−1 q (p)
and q S = S (p, E) ≡ Cq−1 (p, E), where subscript q denotes derivative
with respect to q. Competitive market equilibrium will be reached
when q D = q S (or, equivalently, uq = Cq , marginal utility equal to
marginal cost), which determines the equilibrium price and quantity
as functions of expenditure: p∗ (E) and q ∗ (E).
By applying the implicit function theorem to the equilibrium
condition, uq (q) = Cq (q, E) = p∗ , one can obtain the impact
of public good expenditure on equilibrium quantity and price.
Specifically,

∗ CqE CqE η S η D CqE

qE = = and p∗E = , (21)
uqq − Cqq p(η S − η D ) 1 − η D /η S

where η S ≡ Sp · p/q = p/(Cqq · q) and η D ≡ Dp · p/q = p/(uqq · q)

are price elasticities of supply and demand, respectively. Because
CqE < 0, η S > 0, and η D < 0, one can readily check that qE ∗ >0
∗
and pE < 0. Equation (21) shows that the quantity and price effect
of R&D expenditure depends on the marginal cost effect of the
 Agricultural Innovation 211

said expenditure, as well as the supply and demand elasticities. For

instance, if the R&D expenditure has large impact on marginal cost
in terms of absolute value or if the demand is inelastic, then an
increase in R&D expenditure tends to have large price effect.
Plugging q ∗ (E) and p∗ (E) into the utility function and the profit
function, we obtain

U ∗ (E) ≡ u (q ∗ (E)) + Y − p∗ (E) q ∗ (E) − αE
(22)
π ∗ (E) ≡ p∗ (E) q ∗ (E) − C (q ∗ (E) , E) − (1 − α)E.

Differentiating equations in (22) with respect to E and using the

equilibrium condition Cq |q∗ = uq |q∗ = p∗ yields

UE∗ = −q ∗ (E) p∗E − α.
∗ = −C + q ∗ (E)p∗ − (1 − α).
(23)
πE E E

The equations in (23) show that a one-unit increase in R&D expen-

diture, E, will have three effects: (1) it will decrease the consumer’s
utility and the producer’s profit by α and 1−α, respectively; (2) it will
increase consumer’s utility by −q ∗ (E) p∗E due to the price effect, but
decrease the producer’s profit by the same amount (i.e., consumers
spend less on the same quantity of goods and accordingly, producers
receive less); and (3) it will reduce the producer’s production cost
by CE .
The total social welfare is W ∗ (E) = U ∗ (E) + π ∗ (E). Taking
derivative of W ∗ (E) with respect to E yields WE∗ = UE∗ + πE ∗ =

−CE − 1, which indicates that, assuming an interior optimal solution

for E, at the socially optimal R&D expenditure level, the marginal
cost of R&D expenditure (i.e., 1) should be equation to the marginal
benefit of the expenditure (i.e., −CE ). However, the socially optimal
R&D expenditure level (denoted by E ∗ ) may not be the same as the
expenditure level that is optimal to the producers (denoted by E1∗ )
or that to the consumers (denoted by E2∗ ). Note that E ∗ does not
depend on the value of α, but E1∗ and E2∗ do (see equations in (23)).
Depending on who funds the expenditure and who determines the
quantity of fund, there are four scenarios to consider: (a) consumers
solely fund the expenditure and determine its quantity; (b) producers
212 Agricultural Economics and Policy

solely fund the expenditure and determine its quantity; (c) consumer
solely fund the expenditure but producers determine the expenditure
quantity (government “captured” by producers); and (d) producers
solely fund the expenditure but consumers determine the quantity
(government “captured” by consumers). Assuming interior solutions,
under scenarios a) to c), the R&D expenditure is determined by
−q ∗ (E) p∗E − 1 = 0, −CE + q ∗ p∗E − 1 = 0, and −CE + q ∗ p∗E = 0,
respectively. Obviously, under scenario d) consumers would select
infinite amount of R&D expenditure (i.e., E2∗ = ∞). We can see the
optimal R&D expenditures will likely differ across the four different
scenarios, indicating that the political influence will affect R&D
expenditure.
Note that at the social optimum E ∗ , we have −CE = 1 and
thus UE∗ = −q ∗ (E) p∗E − α = −πE ∗ . Therefore, evaluated at E ∗ , if

−q ∗ (E) p∗E > α then UE∗ > 0 and πE ∗ < 0, indicating E ∗ < E ∗ . In
1 2
other words, if α is small or if the demand is more inelastic (i.e., price
effect of R&D is larger, see Equation (23)), then it is likely the case
that E2∗ > E ∗ > E1∗ . In plain English, this means that if consumers
only support a small portion of R &D or if the R&D will reduce price
considerably, then consumers will prefer R&D expenditure that is
larger than the social optimal expenditure and producers will prefer
an expenditure that is smaller than social optimum. In contrast, when
α is large or when the demand is more elastic, then the opposite is
true. de Gorter and Zilberman (1990) use these results to explain
why producers are more willing to fund R&D on specialty crops
(high demand elasticity) but less willing to fund R&D on major row
crops (low demand elasticity).

3.2. Political connection and innovation:

Akcigit et al. (2023)

Akcigit et al. (2023) explore how political connection can prevent

innovation. Assume that an incumbent firm’s production function is
y = l, where y is output and l labor input. The demand function
facing the incumbent is p = q β y −β , where p is output price and q
stands for product quality that is determined by innovation. Larger
 Agricultural Innovation 213

q indicates higher quality. Suppose that labor cost is w per unit and
that regulatory burdens increase the marginal cost of production by
a portion of τ . We also assume a monopolistic market. Therefore, the
profit maximization problem of an incumbent without any political
connection to remove the regulatory burdens is

π n = max{py − (1 + τ ) wl}, s.t. p = q β y −β and y = l. (24)

l

Solving this optimization problem we obtain the optimal labor

demand l∗ = q · [(1 − β)/((1 + τ ) w)]1/β , the optimal revenue R∗ =
q · [(1 − β)/((1 + τ ) w)](1−β)/β , the corresponding labor productivity
R∗ /l∗ = [(1 + τ ) w]/(1 − β), and the maximized profit π n∗ =
π· (1 + τ )−(1−β)/β q, where π ≡ β((1 − β)/w)(1−β)/β .
If the incumbent has political connections they can utilize them
to remove the regulatory burdens at a fixed cost wp , then the profit
maximization problem would be

π p = max{py − wl − wp }, s.t. p = q β y −β and y = l. (25)

l

Solving problem (25) we obtain the optimal labor demand ˆl =

q · [(1 − β)/w]1/β , the optimal revenue R̂ = q · [(1 − β)/w](1−β)/β ,
the corresponding labor productivity R̂/ˆl = w/(1 − β), and the
maximized profit π p = πq − wp .
One can readily check that ˆ l > l∗ , R̂ > R∗ , and R̂/ˆl < R∗ /l∗ ,
indicating that an incumbent, if transitioned from being politically
unconnected to connected, would hire more labor, generate higher
revenue, but operate at lower productivity level. By comparing π n∗
with π p , we can see that, if q > q̂ s ≡ wp /[π · (1 − (1 + τ )−(1−β)/β )],
then the incumbent will prefer to be politically connected.
Due to the specific form of the production function (i.e., y = l),
the productivity level here is equal to the monopolistic price. One
may conclude that political connection is welfare-enhancing because
now the price is lower and output is higher. This is only true
in a static scenario, where innovations and new market entrances
are ignored. However, in a dynamic framework where innovations
increase product quality and market entrants with new innovations
replace incumbent firms, political connections may prevent the
214 Agricultural Economics and Policy

entrance of new firms and thus dampen the dynamism of the market.
To see this, let us introduce entrances and exits into the model.
Imagine that a potential entrant works on innovation aiming to
improve the product quality, q, in the existing market. The success of
the innovation occurs at Poisson arrival rate δ. The size of the success,
λ, is a random variable with cumulative distribution function F (λ).
So, the product quality of the entrant with successful innovation is
q  = (1 + λ)q, and the demand is determined by p = (q  )β (y  )−β ,
where p and y  is the price and quantity of the new product. If the
entrant and the incumbent are in the same status regarding political
connections (i.e., both have connections or none have connections),
then the entrant will be able to replace the incumbent as long as λ >
0.4 In this case, because the success of innovation occurs following a
Poisson distribution with rate δ, the incumbent is replaced at rate δ.
When the incumbent is politically connected whereas the entrant
is not, then not every successful innovation with λ > 0 will cause the
incumbent to be replaced. Note that the lowest price that a politically
connected incumbent can charge is w, the marginal production
cost of the product without any regulatory burdens. However, the
lowest price that a politically unconnected entrant can charge is
(1+τ )w. Therefore, in order to have the price-adjusted quality of the
entrant’s product (i.e., (1 + λ)q/[(1 + τ ) w]) greater than that of the
incumbent’s product (i.e., q/w), the entrant’s innovation must be at
least able to improve the existing quality by a portion of λ∗ = τ .
Here we assume that the entrant and the incumbent will engage in a
quality-adjusted price competition until one of them exits, then the
market structure will return to be monopolistic.
To analyze the incumbent’s decisions on whether to establish
political connection, one more assumption is in order: with probabil-
ity α the entrant has political connection and with probability 1 − α
it has no political connection. If the incumbent firm chooses not to
be politically connected, then the net present value of its lifetime

4
For simplicity we assume that products of the entrant and the incumbent are
perfect substitutes after adjusting quality.
 Agricultural Innovation 215

profit, denoted by V0 (q), is

rV0 (q) = π · (1 + τ )−1 · q − δ · V0 (q), (26)
where r is the interest rate and π (1 + τ )−1 q is the instantaneous
profit after setting β = 1/2. Equation (26) states that the instan-
taneous rent of the incumbent is equal to the instantaneous profit
minus the expected loss caused by the incumbent being replaced by
an entrant in the next moment. Based on Equation (26) we can solve
V0 (q) out as
π · (1 + τ )−1 · q
V0 (q) = . (27)
r+δ
If the incumbent establishes political connection, the net present
value of its lifetime profit, denoted by V1 , is
rV1 (q) = πq − wp − δ [α + (1 − α) P r (λ > λ∗ ) ] V1 (q). (28)
The interpretation of Equation (28) is similar to that of Equation
(26). The instantaneous rent of the incumbent is equal to the instan-
taneous profit minus the expected loss from the incumbent being
replaced by an entrant. Note that the chance that the incumbent
is replaced by an entrant is δ (i.e., the chance that an innovation
arrives) multiplied by the sum of α (i.e., the chance that the entrant
is politically connected) and (1 − α) P r (λ > λ∗ ) (i.e., the chance
that the entrant is not politically connected but the innovation is
sufficiently large). Rearranging Equation (28), one can solve V1 (q) as
πq − wp
V1 (q) = . (29)
r + δ[α + (1 − α) P r (λ > λ∗ ) ]
The incumbent will establish the political connection if V1 (q) >
V0 (q), which is equivalent to
wp
q > q̂ d ≡ , (30)
r+δ 1
π(1 − r+δ · 1+τ )

where δ ≡ δ[α + (1 − α) P r (λ > λ∗ ) ]. Recall that in the static

framework the threshold quality above which the incumbent will
establish political connection is q̂ s = wp /[π · (1 − (1 + τ )−1 )] after
216 Agricultural Economics and Policy

we set β = 1/2. One can check that q̂ d < q̂ s , indicating that in a

dynamic framework the incumbent has more incentive to establish
political connection. By doing so, the incumbent makes new firm
entrance less likely to occur and thus it can survive longer. As a
result, the innovations that can be eventually used in production
occur at a lower rate. Specifically, as discussed earlier, when the
incumbent does not have political connection, the implemented
innovations occur at rate δ. When the incumbent has political
connection, however, the implemented innovations occur at rate
δ ≤ δ, where the equality holds when α = 1 or P r (λ > λ∗ ) = 1, or
both. The results show that if the entrant has political connection
with certainty, then the implemented innovations occur at the same
rate as that when neither the incumbent nor the entrant is politically
connected.

3.3. Innovation incentive systems: Wright (1983)

Public policies or institutional arrangements, such as patents, prizes,

and contracts, have been used as tools to incentivize and support
innovations. However, few studies have compared their efficiency
to promote innovations.5 Wright (1983) is perhaps the first study
that examines and compares the efficiency of these tools considering
asymmetric information. In this subsection, we briefly introduce the
model developed in the study.
Wright (1983) assumes that the analysis framework is static
with numerous homogeneous and risk-neutral inventors. The research
target is a process invention that, if successful, can generate benefit
B for society. For each inventor, the cost of generating μ units of
research activity is c(μ). Assuming perfect competition, for each
inventor the optimal research activity, denoted by μ∗ , will be achieved
at the minimum average cost. For simplicity, μ∗ is normalized to be
1, setting the number of firms in the industry equal to the number
of research activity unit in the market. Let m ≡ Σ μ be the total

5
Clancy and Moschini (2013) offer a comprehensive review regarding the innova-
tion incentives provided by patents, prizes, and research contracts.
 Agricultural Innovation 217

research activity in the society. The total cost of all research activity
is denoted by C (m) ≡ Σ c(μ), with C  (m) > 0 and C  (m) ≥ 0. Let
P (m) denote the probability that at least one firm out of m firms
will be successful.
Given the above assumptions, the socially optimal level of
research activity, denoted by m0 , will be determined by B ·P  (m0 ) =
C  (m0 ). That is, the social optimum is reached when the marginal
benefit of research (i.e., B · P  (m0 )) is equal to the marginal cost
(i.e., C  (m0 )). With complete information, the social optimal can be
reached by a research administrator offering a payment B · P  (m0 )
for each unit of research activity provided by the inventing firms.
Or, the same social optimum can be achieved by directly contracting
with inventing firms for m0 units of research activity under price
B · P  (m0 ). However, Wright (1983) shows that a patent or prize
policy that rewards the successful firms with full benefit of the
invention may not induce the socially optimal level of research
activity. The reason is explained as follows.
Due to the homogeneity assumption of firms, each firm has an
equal chance to be successful. If there are multiple successful firms,
then the award is shared equally among these firms. Therefore,
under the “full-benefit” patent or prize, the expected award for
a firm is B · P (m)/m per unit of research activity,6 which is the
marginal benefit of research activity for an individual firm. Because
P  (m) < 0, we have B · P (m)/m > B · P  (m). As a result, in the
equilibrium under the “full-benefit” patent or prize, the marginal
benefit and marginal cost of research activity, thus the quantity
of research activities, are higher than the corresponding socially
optimal levels. This result is known as the “common pool problem”
in the innovation literature following Dasgupta and Stiglitz (1980),
analogous to the inferior outcome from competitive fishing in a
common fishery described by Gordon (1954). One strategy to fix
this “common pool problem” under complete information is to just
offer a portion (say, γ) of the full benefit to the successful firms, such

6
This intuitive conclusion can also be readily shown by using the properties of
Binomial distribution.
218 Agricultural Economics and Policy

that γ · B · P (m)/m = B · P  (m). This can explain why patents have

limited, instead of infinite, life.
However, in reality, it is rare to have complete information. For
activities as complex as research, it is impossible for policymakers
to know the specific research costs or the benefits of a successful
research outcome, while the inventing firms may have much better
information on their costs and potential benefits. Wright (1983) goes
one step further from complete information, studying two scenarios
under which cost and, separately, benefit of invention are known to
the firms but unknown to policymakers. Here we discuss them briefly.
The first scenario involves random research costs with non-
random benefits. A heuristic approach is employed to approximate
welfare losses or gains from patents or prizes when compared to a
direct contract of research activities. Figure 1 depicts equilibrium
research activities when a prize is promised by policymakers under

Figure 1. Social welfare of prizes when cost information is asymmetric.

Note: The figure is a reprint of Figure 1 in Wright (1983). Copyright American
Economic Association; reproduced with permission of the author and of the
American Economic Review.
 Agricultural Innovation 219

this scenario. Note that Wright’s analysis is ex-ante, and the solutions
are optimal only in the sense of expectation. Assume that, from the
policymaker’s point of view, the marginal cost is (1 + θ)C   (m) and
 
(1− θ)C (m) with equal chance. As we discussed before, if full benefit
were to be awarded to successful inventing firms, the expected return
per unit of research activity would be B · P (m)/m, which is larger
than the marginal benefit of research activity, B · P  (m), resulting
in over-investment in research and welfare losses. As depicted in
Figure 1, the line B · P (m)/m crosses lines (1 + θ)C   (m) and
 
(1 − θ)C (m) at points Q and S, respectively, resulting in much
larger research activity than m0 , the ex-ante socially optimal research
level. To reach this level by using prize or patent, the policymaker
should only award a portion (i.e., Y0 ) of the full benefit to successful
inventing firms, depicted by line Y0 · B · P (m)/m in Figure 1. With
this prize or patent, the social welfare gains (or losses) of using prize
or patent when compared with contracting form m0 units of research
can be measured by Δw = 0.5 × [(αVEF − αAVD ) + (αFHG − αHJK )],
where α stands for “area.” Wright (1983) shows that under certain
assumptions, Δw can be approximated by a function of elasticity
of supply of research activity, elasticity of the average probability of
research success, and the elasticity of marginal probability of research
success.
Under the second scenario, the invention benefit, B, is assumed
to be random whereas the invention cost is assumed to be deter-
ministic. Here, the social optimal research activity under complete
information, m0 , can be reached by either direct contracting or
a discounted prize. The welfare gains and losses of awarding a
patent when compared with awarding a contract or prize to fix
the research activity at m0 can also be approximated by another
function of elasticities of research activity, average probability of
research success, and marginal probability of success. Wright (1983)
numerically shows that when success probability of research is high,
then contracts are preferred to prizes and patents. When success
probability is low but the elasticity of research supply is high,
patents would be preferred to prizes and contracts. Prizes will
220 Agricultural Economics and Policy

likely outperform patents and contracts when success probability is

moderate and the elasticity of research supply is small.

4. Agricultural Innovation and Climate Change

Agricultural innovation is expected to be a major tool for agriculture
to respond to climate change. A few studies have explored the impact
of climate change on agricultural innovation. Based on country-level
patent data over the world, Miao and Popp (2014) show that drought
promotes innovations of drought-tolerant seeds. Miao (2020) finds
similar results using patent data for drought-tolerant traits of 10
major crops in the United States, showing that drought-tolerant
innovations of these crops are quite sensitive to drought severity
measured by Palmer Drought Severity Index (PDSI): a one-unit
increase in PDSI will increase drought-tolerant trait patents by about
40%–60%.
A recent paper by Moscona and Sastry (2023) also investigates
the impact of climate change on agricultural innovation. However,
this paper pushes the research frontier one step further: after
quantifying the responsiveness of agricultural innovation to climate
change, they analyze how much the agricultural innovation mitigates
climate change’s impact on agriculture. Here we briefly describe this
study.
First, Moscona and Sastry (2023) develop a theoretical frame-
work to predict the direction of technological change under climate
change. Finding the prediction ambiguous, they proceed the analysis
with a novel empirical approach to examine the interaction between
climate change effect and agricultural innovation. Their theoretical
framework starts with a few assumptions. First, the production
function for farm i ∈ [0, 1] is
Yi = α−α (1 − α)−1 G(Ai , θ)α Ti1−α , (31)
where Yi is crop yield, α ∈ [0, 1] is a parameter reflecting the relative
importance of inputs, Ai can be viewed as local climate partially
governing farm i ’s productivity, θ measures the overall technological
advancement (or, technology quality), and finally, Ti is the quantity
 Agricultural Innovation 221

of technology adopted on farm i. Note that α−α (1 − α)−1 is included

for modeling convenience. It is further assumed that G(Ai , θ) > 0,
∂G/∂Ai > 0, ∂G/∂θ > 0, and ∂ 2 G/∂θ 2 < 0.7 With crop price, p,
and technology price, q, the optimal technology demand from farm i
1 1
is Ti∗ = α−1 p α q − α G(Ai , θ).
Moscona and Sastry (2023) assume that the innovator is a
monopolist who determines technology price, q, and overall technol-
ogy quality, θ. With marginal cost of producing the technology input
at 1 − α and the development cost at C(θ), where C  (θ) > 0 and
C  (θ) > 0, the innovator’s optimization problem can be written as,

1 1
−1 α −α
max (q − (1 − α)) α p q G (Ai , θ) dF (Ai ) − C (θ) , (32)
q,θ

where F (Ai ) is the cumulative distribution function of productivity

Ai . By solving the optimization problem in Equation  (32), we
∗ ∗
can obtain that q = 1 and θ = argmax p 1/α G (Ai , θ) dF
(Ai ) − C(θ) . See Online Appendix B of Moscona and Sastry (2023)
for detailed derivations.
Moscona and Sastry (2023) define climate substitutability of
technology as following: If G12 ≤ 0, then technological advances
are climate substitutes; if G12 ≥ 0, then climate complements.
Intuitively, if technology advances can decrease the marginal product
of climate, G1 , then the technology is substituting for climate. On
the other hand, if technological advances increase the marginal
product of climate, then the technology is complementing cli-
mate. Moscona and Sastry (2023) show that if crop price is fixed
(e.g., in a small open economy), then damaging climate change
will increase climate-substituting technology but decrease climate-
complementing technology. In a larger economy where crop price is

7
Note that ∂G/∂Ai > 0 could be a strong assumption. Many studies (e.g.,
Schlenker and Roberts, 2009) have shown that the impact of weather on crop yield
is non-linear: above a threshold value, an increase in temperature or precipitation
will harm crop yield. Therefore, a more realistic assumption could be ∂G/∂Ai > 0
when Ai is small whereas ∂G/∂Ai < 0 when Ai is large. Exploring how the
theoretical results in Moscona and Sastry (2023) would change under this more
realistic assumption could be an interesting exercise.
222 Agricultural Economics and Policy

endogenously determined, however, a damaging climate change will

still encourage climate-substituting technology, but its impact on
climate-complementing technology is ambiguous. The idea is that
damaging climate change reduces crop production and, thereby,
increases crop price. An increased crop price indicates increased
marginal product value of technology. This positive price effect on
technology innovation will further strengthen the increased demand
for climate-substituting technology, but will weaken or even reverse
the decreased demand for climate-complementing technology. There-
fore, the direction of changes in climate-complementing technology
is ambiguous.
To empirically determine the impact of climate change on
agricultural innovation, Moscona and Sastry (2023) harmonize the
U.S. Department of Agriculture Variety Name List, Plant Variety
Protection certificates, and U.S. patents on non-livestock agriculture
to construct various measures of agricultural innovation and conduct
econometric analysis. Climate stress is measured by the difference
between observed temperature and a crop’s upper threshold for
optimal growing temperature. Their econometric analyses, together
with numerous robustness checks, show that historical climate change
had incentivized agricultural innovation: Based on data for 69 crops,
climate change over 1950–2016 caused, on average, about 20%
increase in agricultural innovation.
After quantifying the impact of climate change on agricultural
innovation, Moscona and Sastry (2023) creatively move one step
further to quantify to what extent the climate-induced innovation
has mitigated the negative impact of climate change on agriculture.
Using land value as the dependent variable, the authors explore
how the exposure to agricultural innovation of one county may
mitigate the negative impact of climate change on land value, where
the exposure to innovation is measured by a county’s exposure to
extreme temperature, justified by the relationship between extreme
temperature and innovation discussed in the previous paragraph.
Moscona and Sastry (2023) find that in the past six decades,
agricultural innovation had mitigated the negative impact of climate
change by about 20%. Future agricultural innovation is estimated
 Agricultural Innovation 223

to reduce climate change impact on agriculture by about 13%,

indicating that agricultural innovation itself is far from enough to
eliminate the negative impact of climate change on U.S. agriculture.
Moscona and Sastry (2023) is the first study that attempts to
examine the extent to which agricultural innovation may mitigate
climate change’s impact on agriculture. It opens a new field of
research in agricultural economics and in the literature of climate
change adaptation. Numerous subsequent research inquiries can be
explored following Moscona and Sastry (2023). For instance, it
might be of interest to further explore the role of public research
(e.g., research in agricultural experiment stations) in mitigating
climate change impact, where Moscona and Sastry (2023) do not
find conclusive results, likely due to their rough measurement of
agricultural innovation. Second, understanding crop heterogeneity in
innovation’s impact on mitigating the climate effect can be of policy
relevance because it would be a prerequisite for optimal allocation
of research fund across crops. Furthermore, it is not well understood
how agricultural innovation affects the response to climate change on
a spatial level. As Evenson (1989) and Moscona and Sastry (2022)
point out, technologies developed based on local environment and
climate may hinder technological spillover, which may in turn inhibit
farmers from better adapting to climate change.

5. Conclusions
The literature on innovation is vast, and this chapter only covers a
few topics about agricultural innovation that the authors are most
interested in, leaving many important topics undiscussed, such as
returns to agricultural R&D, efficiency of various schemes managing
R&D funds, innovation and market power, as well as the relationship
between agricultural innovation and agricultural productivity. Read-
ers can find discussions of these topics in Huffman and Just (1994),
Fuglie et al. (1996), Moschini and Lapan (1997), Alston and Pardey
(2021), and Pardey and Alston (2021).
Innovation, from an idea to the final products or processes
that can be adopted by end users, involves complex activities,
224 Agricultural Economics and Policy

such as research, manufacture, and marketing. Its emergence and

success can be affected by numerous factors, such as existing knowl-
edge stock, breakthroughs in basic research, supply of researchers,
political forces, market structure, infrastructure of supply chains,
and consumers’ ever-changing tastes, to name a few. Sunding and
Zilberman (2001) divide the innovation process into five stages:
discovery, registration, development, production, and marketing.
A recent article by Zilberman et al. (2022) provides a comprehensive
discussion about innovation supply chain originating from innovative
ideas and culminating with final marketed products, where adoption
starts. We discuss adoption in the next chapter.

References
Acemoglu, D. 2002. Directed Technical Change. The Review of Economic Studies
69(4): 781–809.
Akcigit, U., S. Baslandze, and F. Lotti. 2023. Connecting to Power: Political
Connections, Innovation, and Firm Dynamics. Econometrica 91(2): 529–564.
Alston, J.M. and P.G. Pardey. 2021. The Economics of Agricultural Innovation.
Barrett, C.B. and D.R. Just (Eds.), Handbook of Agricultural Economics,
Vol. 5, Chapter 75, pp. 3895–3980.
Atkinson, A.B. and J.E. Stiglitz. 1969. A New View of Technological Change.
The Economic Journal 79(315): 573–578.
Arrow, K.J. 1962. The Economic Implications of Learning by Doing. The Review
of Economic Studies 29(3): 155–173.
Binswanger, H.P. 1974. A Microeconomic Approach to Induced Innovation.
Economic Journal 84(336): 940–958.
Clancy, M.S. and G.C. Moschini. 2013. Incentives for Innovation: Patents, Prizes,
and Research Contracts. Applied Economic Perspectives and Policy 35(2):
206–241.
Dasgupta, P. and J. Stiglitz. 1980. Uncertainty, Industrial Structure, and the
Speed of R&D. The Bell Journal of Economics 11(1): 1–28.
de Gorter, H. and D. Zilberman. 1990. On the Political Economy of Public
Goods Inputs in Agriculture. American Journal of Agricultural Economics
72: 131–137.
Denison, E.F. 1967. Why Growth Rates Differ: Postwar Experience in Nine
Western Countries. Brookings Institution, Washington, DC.
Evenson, R.E. 1989. Spillover Benefits of Agricultural Research: Evidence
from U.S. Experience. American Journal of Agricultural Economics 71(2):
447–452.
Evenson, R.E. and Y. Kislev. 1976. A Stochastic Model of Applied Research.
Journal of Political Economy 84(2): 265–282.
 Agricultural Innovation 225

Fuglie, K., N. Ballenger, K. Day, C. Klotz, M. Ollinger, J. Reilly, U. Vasavada,

and J. Yee. 1996. Agricultural Research and Development: Public and Private
Investments under Alternative Markets and Institutions. Washington, DC:
U.S. Department of Agriculture, Economic Research Service, Agricultural
Economic Report No. 735.
Godin, B. 2015. Technological Change: What do Technology and Change Stand
For? Working Paper. Project on the Intellectual History of Innovation,
Montréal.
Gordon, H.S. 1954. The Economic Theory of a Common-Property Resource: The
Fishery. Journal of Political Economy 62(2): 124–142.
Hayami, Y. and V.W. Ruttan. 1970. Factor Prices and Technical Change in
Agricultural Development: The United States and Japan, 1880–1960. Journal
of Political Economy 78(5): 1115–1141.
Heady, E.O. 1949. Basic Economic and Welfare Aspects of Farm Technological
Advance. Journal of Farm Economics 31(2): 293–316.
Hicks, J.R. 1932. The Theory of Wages. Macmillan, London, UK.
Huffman, W.E. and R.E. Just. 1994. Funding, Structure, and Management of
Public Agricultural Research in the United States. American Journal of
Agricultural Economics 76(4): 744–759.
Jorgenson, D.W. and Z. Griliches. 1967. The Explanation of Productivity Change.
Review of Economic Studies 34(3): 249–283.
Miao, Q. and D. Popp. 2014. Necessity as the Mother of Invention: Innovative
Responses to Natural Disasters. Journal of Environmental Economics and
Management 68(2): 280–295.
Miao, R. 2020. Climate, Insurance and Innovation: The Case of Drought and
Innovations in Drought-Tolerant Traits in US Agriculture. European Review
of Agricultural Economics 47(5): 1826–1860.
Moschini, G. and H. Lapan. 1997. Intellectual Property Rights and the Welfare
Effects of Agricultural R&D. American Journal of Agricultural Economics
79: 1229–1242.
Moscona, J. and K.A. Sastry. 2022. Inappropriate Technology: Evidence from
Global Agriculture. Working Paper, Harvard University.
Moscona, J. and K.A. Sastry. 2023. Does Directed Innovation Mitigate Climate
Damage? Evidence From U.S. Agriculture. The Quarterly Journal of Eco-
nomics 138(2): 637–701.
Pardey, P. and J. Alston. 2021. Unpacking the Agricultural Black Box: The Rise
and Fall of American Farm Productivity Growth. The Journal of Economic
History 81(1): 114–155.
Peterson, W. and Y. Hayami. 1977. Technical Change in Agriculture. Lee
Martin (Ed.), A Survey of Agricultural Economics Literature, pp. 497–540.
University of Minnesota Press, Minneapolis, MN.
Salter, W.E.G. 1960. Productivity and Technical Change. Cambridge University
Press, Cambridge, UK.
Schlenker, W. and M.J. Roberts. 2009. Nonlinear Temperature Effects Indicate
Severe Damages to US Crop Yields under Climate Change. Proceedings of
the National Academy of Sciences 106(37): 15594–15598.
226 Agricultural Economics and Policy

Sunding, D. and D. Zilberman. 2001. The Agricultural Innovation Process:

Research and Technology Adoption in a Changing Agricultural Sector, B.L.
Gardner and G.C. Rausser (Eds.), Handbook of Agricultural Economics,
Volume 1, Part A, Chapter 4, pp. 207–261.
Yan, D., H. Tajima, L.C. Cline, R.Y. Fong, J.I. Ottaviani, H.-Y. Shapiro,
E. Blumwald. 2022. Genetic Modification of Flavone Biosynthesis in Rice
Enhances Biofilm Formation of Soil Diazotrophic Bacteria and Biological
Nitrogen Fixation. Plant Biotechnology Journal 20(11): 2135–2148.
Wright, B.D. 1983. The Economics of Invention Incentives: Patents, Prizes, and
Research Contracts. American Economic Review 73(4): 691–707.
Zilberman, D., T. Reardon, J. Silver, L. Lu, and A. Heiman. 2022. From the
Laboratory to the Consumer: Innovation, Supply Chain, and Adoption With
Applications to Natural Resources. Proceedings of the National Academy of
Sciences 119(23): e2115880119.
 Chapter 8

Technology Adoption

Technological change is a multistage process, which includes both

technological innovation and adoption. Both processes require some
investment and sacrifice: in some industries, the firms apply their
own innovations, while in others, firms apply innovations from other
firms. At a society-wide level, a certain level of technological adop-
tion is necessary to make a dividend on innovation. Technological
innovations will not generate much real-world impact unless they
are adopted by society. Also, investments in innovations will not be
recouped without a certain level of adoption of the innovations. Thus,
social scientists, including economists, have long been concerned with
technological adoption, and a large literature has been devoted to
understanding the rationale for the uptake of a technology. This
literature distinguishes between two concepts: adoption, the decision
to use technology (measured on the individual level), and diffusion,
the aggregate of adoption measured as the proportion of users in
relation to potential users. The literature aims to understand who
adopts a new technology, why they adopt it, and how they adopt it.
This chapter provides an overview of various theories and models
on technology adoption, ranging from early models developed by
sociologists to recent models by economists. Earlier models developed
by sociologists in the 1960s–1970s mainly aimed to explain the
S -shaped diffusion curve. However, these models were criticized for

227
228 Agricultural Economics and Policy

ignoring the economic drivers of adopters. Economists incorporated

economic behavior such as profit maximization, risk consideration,
and even dynamics into decision models of technology adoption. In
recent years, advances in behavior economics enriched the literature
on technology adoption. We, therefore, include a brief review of the
roles of behavioral factors in affecting adoption.

1. Early Models
Early studies about technology diffusion can be at least traced
back to 1890, when Gabriel Tarde, a French sociologist, published
The Laws of Imitation. Emphasizing diffusion (since data on the
aggregate was readily available), these studies viewed societal-wide
uptake as time-consuming: it took about 40 years for a complete
adoption of the mechanical tractor and about three to five years
to complete the adoption of the tomato harvester. Other types
of technological adoption (e.g., modern fertilizers and new crop
varieties) also took time.
Ryan and Gross (1943) and Griliches (1957) found that the diffu-
sion of hybrid corn in the United States followed S-shaped diffusion
curves, depicted in Figure 1. At the “early adoption” stage, diffusion
is slow, but at the “takeoff” stage, the diffusion rate becomes very
high, although there is still a critical mass of non-adopters. As
diffusion reaches the “saturation” stage, the rate of diffusion becomes
quite low. This procedure was often modeled as a logistic function.
Let Pt be diffusion (i.e., adoption ratio) at time t. Then we have

k
Pt = , (1)
1 + e−(a+bt)

where k is the equilibrium ratio of diffusion, a is a measure of the

initial level of diffusion, and b is the measure of the rate of diffusion.1
Equation (1) implies that at each moment the log-ratio of adopters

1
Note that the maximum level of diffusion ratio, k, can be less than 1. Therefore,
k ≤ 1.
 Technology Adoption 229

Diffusion
rate

Saturation rate

Saturation

Takeoff

Early adoption

Time

Figure 1. Diffusion rate and time.

to non-adopters is a linear function of time:

Pt
log = a + bt, (2)
k − Pt
which can be estimated by using simple regressions in applied
analyses. Griliches (1957) estimated the three coefficients (k, a,
and b) to observe the effect of profitability on hybrid corn adoption in
Iowa. Comparing diffusion rates in different areas, he demonstrated
that diffusion was faster and more complete when the technology was
more profitable.

1.1. The imitation model

Rogers (1962), a sociologist, argued that imitation is a key driver

of technology diffusion: an individual who sees someone else using a
technology may wish to imitate this person. The basic mathematics
of imitation is similar to the mathematics of infectious diseases —
the imitator catches the new technology from the early adopter.
230 Agricultural Economics and Policy

Mansfield (1961) offered a formalization of imitation driving

diffusion: he considered an industry with n identical firms, with m(t)
as the number of firms that have adopted a certain technology by
time t and n − m(t) for those that had not. The rate of adoption
at time t among the non-adopters is defined by Δm(t)/(n − m(t)),
where Δm(t) ≡ m(t + 1) − m(t). Mansfield (1961) assumed that
 
Δm(t) m(t)
=f , π, c , (3)
n − m(t) n
where π and c stand for profitability and adoption cost of the new
technology, respectively. In other words, the adoption rate among
non-adopters is a function of the portion of firms that have already
adopted as well as the profitability and costs of the new technology.
By using Taylor expansion, and discarding all elements of order
greater than the second, Equation (3) can be approximated as
Δm(t) m(t) m(t)
≈ β1 + β2 + β3 π + β4 c + β5 π
n − m(t) n n
m(t)
+ β6 c + β7 πc + β8 π 2 + β9 c2 , (4)
n
where β’s are coefficients from the Taylor expansion.2
To solve for m(t), Mansfield moved to infinitesimal and got a
differential equation based on Equation (4),
 
dm(t) m(t)
= [n − m(t)] A + b , (5)
dt n
where A ≡ β1 +β3 π +β4 c+β7 πc+β8 π 2 +β9 c2 and b ≡ β2 +β5 π +β6 c.
Solving this differential equation, we obtain
m(t) eγ+(A+b)t − Ab
= , (6)
n 1 + eγ+(A+b)t
where γ is the integration constant. It is reasonable to assume that
when time goes back infinitely (i.e., t → −∞), the adoption will be

 2
2 m(t)
Mansfield (1961) assumed that the coefficient of n
is zero, claiming that
data support the assumption.
 Technology Adoption 231

zero (A = 0). Therefore, we obtain

m(t) 1
= −(γ+bt)
, (7)
n 1+e
which is the same as Equation (1) when k in Equation (1) is set to 1.
Mansfield predicts that the proportion of adopting firms will be
a symmetrical S-shaped curve over time. As the new technology
becomes more profitable, the rate of diffusion quickens and the cost
of the adoption process declines.
While Mansfield’s model is elegant, there are a few reasons why
it is unsatisfactory. First, it assumes identical firms and ignores
differences between firms due to size, wealth, and education (another
paper by Mansfield (1963) showed that these differences matter).
Second, it does not consider the dynamic, or the learning-by-doing,
aspect of adoption that reduces the cost of innovation. Third,
the model lacks basic economic mechanisms of decision-making. In
particular, it evades questions of utility, focusing solely on imitation
as the driver of uptake.

1.2. The threshold model

The threshold model, introduced by David (1975) and presented

in Sunding and Zilberman (2001), distinguishes between micro-
level behavior, heterogeneity among decision-makers, and dynamic
processes. The micro-level decision maker decides whether to adopt
(in the case of indivisible technologies) and the level of (in the
case of divisible technologies) by maximizing profit, expected utility,
or some other objectives. But individuals are different: at a given
period, some people adopt and others do not, divided by a threshold,
which determines the rate of diffusion. Over time, dynamic processes
of learning and technological improvement that reduce the cost of
technology change the rate of diffusion.3

3
Of course, sometimes alternative technologies appear and technologies are dis-
adopted, while in other cases technologies are abandoned because people have bad
experience with them. The term “threshold” comes from the fact that in some
models, for example purchasing luxury cars, the consumers are divided according
232 Agricultural Economics and Policy

In real life, there are many sources of heterogeneity, and iden-

tifying the key parameters that determine the set of adopters is a
major statistical challenge. With the availability of data on individual
behavior as well as the statistical tools of logit and probit, researchers
can apply the threshold model to estimate the parameters that
will affect the likelihood of adoption by different individuals at
every moment. For simplicity, in this chapter, we consider a farming
industry in which farm sizes differ. Let l denote farm size and assume
its density function is g(l). Therefore, the total number of farms and
∞
total land acreage in the farming industry are N ≡ 0 g(l)dl and
∞
L ≡ 0 lg(l)dl, respectively. Suppose that farmers are facing a new
technology which, at time t, costs Ft to be established on a farm
and generates Δπt more profits per acre than does the conventional
technology. If farmers are attempting to maximize their profits, then
farmers with farm size such that l · Δπt ≥ Ft will adopt the new
technology. Define lt∗ ≡ Ft /Δπt as the threshold farm size above
which farmers will adopt the new technology. The share of farms
that have adopted the new technology by time t is
∞
l∗ g(l)dl
Yt = t . (8)
N
By using Leibniz’s formula, from Equation (8) we obtain

∂Yt g (l∗ ) ∂lt∗

=− t . (9)
∂t N ∂t
∂l∗
It is reasonable to assume that ∂tt < 0 because the fixed establish-
ment cost of the new technology is expected to be decreasing over
time and the profitability advantage of the new technology is likely to
increase due to the increased efficiency of using the new technology
over time (i.e., “learning by using”). Moreover, because g (lt∗ ) is
always positive, we can conclude that ∂Yt /∂t > 0. Differentiate

to income and at each moment there is a threshold level of income above which
people are assumed to adopt the technology. But this threshold level declines over
time.
 Technology Adoption 233

Equation (9) with respect to t, we have

   
∂ 2 Yt 1 ∂g (lt∗ ) ∂lt∗ 2 ∂ 2 l∗
=− + g (lt∗ ) 2t . (10)
∂t 2 N ∂lt∗ ∂t ∂t

In Equation (10), the curvature of the diffusion curve depends on

(1) the slope of the density function at the threshold acreage (i.e.,
∂g(lt∗ )/(∂lt∗ )), (2) the relationship between the threshold acreage and
time (i.e., ∂lt∗ /∂t), (3) the value of density function at the threshold
acreage (i.e., g(lt∗ )), and (4) the curvature of lt∗ as a function of t.
As long as ∂ 2 Yt /∂t2 is positive when t is small and negative when t
becomes larger, the diffusion curve will assume an S-shaped curve.
For instance, if the adoption threshold farm size, lt∗ , decreases
over time at a constant rate (i.e., ∂ 2 lt∗ /∂t2 = 0); if the farm size
distribution is unimodal, as in Figure 2, then (∂ 2 Yt )/(∂t2 ) will be
small and positive when t is small. As t increases, (∂ 2 Yt )/(∂t2 ) first

Figure 2. Unimodal farm size distribution.

234 Agricultural Economics and Policy

increases and then decreases to 0 and then negative. In this case, the
diffusion curve is S-shaped.
The diffusion in cases of divisible technologies can be computed
in similar ways. For instance, with a new seed variety, farmers may
experiment with it on a small portion of their plot. With learning-
by-doing, the uncertainty of the technology may decline. Thus, the
land share of the technology, as well as the number of adopters, will
increase, resulting in S-shaped behavior of diffusion as well.

2. Static Expected Utility Models

The imitation model and threshold model discussed above do not
consider any risks associated with new technologies. However, risks
are prevalent, particularly for new technologies, because potential
adopters have little experience with them. In this section, we discuss
the static expected utility model concerning technology adoption
developed by Just and Zilberman (1983). Note that we discussed
this model in Chapter 4 but mainly to focus on diversification. In
this section, however, we will focus on the insight into technology
adoption derived from the Just and Zilberman (1983) model.
Consider a farmer who is considering (1) whether or not to adopt
a new crop variety on their farm, and (2) if they adopt, then how
much of their land will be devoted to the new crop variety.4 Suppose
the profit per acre of the conventional crop is π0 = m0 + 0 , where
m0 is a constant but 0 is a random variable with E(0 ) = 0 and
V ar(0 ) = σ02 . The profit per acre of the new crop variety is π1 =
m1 (f ) + v1 (f )1 , where f is the cost per acre incurred for planting
the new crop variety, m1 (f ) and v1 (f ) are deterministic functions of
f , and 1 is a random variable with E(1 ) = 0 and V ar(1 ) = σ12 .
Moreover, if the farmer adopts the new crop variety, then a fixed cost,

4
This is a quite general decision problem regarding technology adoption, partic-
ularly in agriculture, and the framework illustrated below developed by Just and
Zilberman (1983) has been used in many other studies such as Miao and Khanna
(2017a,b) and Majeed et al. (2023). As of February 2025, the paper has been
cited more than 600 times according to Google Scholar, and the citation per year
of this paper is still trending up after 42 years since its publication.
 Technology Adoption 235

k, would be incurred regardless of the portion of land devoted to the

new variety. Suppose the farm size is L and the farmer fully utilizes
their land. The land devoted to the conventional crop variety and the
new variety is denoted by L0 and L1 , respectively. The technology
adoption decision problem can be written as
max E{U (pL L + π0 L0 + I · (π1 (f )L1 − k))} (11)
I∈{0,1}
L0 ,L1 ,f

s.t. L0 + I · L1 = L, and L0 , L1 , f ≥ 0,
where E is the expectation operator; I is an adoption index, with
I = 0 indicating non-adoption and I = 1 adoption; and pL is end-of-
season land price.
To solve the maximization problem in Equation (11), one can
compare the expected utility under two scenarios. The first one is
the non-adoption scenario where I = 0. Under this scenario, the
farmer’s expected utility is
V0 = E{U (pL L + π0 L)}. (12)
The second is the adoption scenario under which the farmer chooses
the size of L1 (i.e., acreage devoted to the new crop variety) to
maximize their expected utility:
V1 = max E{U (pL L + π0 L − L1 + π1 (f )L1 − k)}. (13)
L1 ∈(0,L],f >0

In Section 2.3.3 of Chapter 4, we discussed the solutions to problem

(13) and the comparative static analyses in detail. Here, we expand
upon the discussion in that section to suggest that the input
accompanying the new crop variety increases the variance of new
variety profits (i.e., ∂v1 /∂f > 0). If the relative risk aversion is
increasing, then larger farmers tend to use less of such input than
do small farmers. However, if the input decreases the variance and
if the relative risk aversion is increasing, then larger farmers tend to
use more of such input.
The farmer will make the optimal choices by comparing V0 and
V1 . If V0 > V1 then the farmer will not adopt the new crop variety.
If V0 ≤ V1 , then the farmer will adopt and plant the new variety on
236 Agricultural Economics and Policy

L∗1 acres of their land, where L∗1 is the optimal solution to problem
(13). That is,
0 if V0 > V1
I∗ = (14)
1 if V0 ≤ V1 .
As Just and Zilberman (1983) point out, it is difficult to compare
V0 and V1 without knowing the farmer’s risk preference and the
impact of input f on the risk of the new crop variety. However,
because farm size has direct risk implication (i.e., the variance of
total profit increases as farm size increases), under conditions of
a positive fixed establishment cost for a new crop adoption (i.e.,
k > 0) and increasing relative risk aversion matched with non-
increasing absolute risk aversion, Just and Zimmerman outline three
possibilities. The first one is that no farms will adopt the new
crop variety due to its high establishment cost. The second one is
that farms smaller than a threshold will not adopt due to the high
establishment cost, but farms larger than the threshold will adopt.
The last one pertains to two farm size thresholds that divide farms
into three groups. Farmers with a farm size smaller than the lower
threshold will not adopt because the fixed establishment cost cannot
be justified by the profits of the new crop; farmers with a farm size
larger than the lower threshold but smaller than the higher threshold
will adopt because they can justify the fixed cost and the disutility
from the increased risk is not yet too large. Finally, farmers with
farm size larger than the higher threshold will not adopt because the
disutility from the increased risk is too large.5

3. Dynamic Models
Although the earlier adoption models and the static expected utility
models capture some important aspects of technology adoption, such
as the S-shaped diffusion curve and risky returns of new technology,

5
Note that the last case only occurs when the covariance between the conventional
crop profits and new crop profits is larger than the variance of conventional crop
profits.
 Technology Adoption 237

they miss some other important features of technology adoption.

First, both the costs and benefits of technology adoption are
dynamic. The new technology itself is likely to become less expensive
as its producers get more efficient in producing the new technology
through, say, “learning-by-doing.” Compared to the $10,000 word
processor which Jimmy Carter bought in 1981 to write his memoir
(Bird, 2021), a word processor available today is far less expensive
and more powerful. Moreover, through “learning-by-using,” adopters
can enlarge the benefits derived from the new technology over time.
Second, technology improvement is unpredictable. A potential user
looking to purchase a technological innovation may consider whether
the future price of the good or even whether another improvement
is in the pipeline. Sometimes delaying adoption may be a better
decision than adopting today. In this section, we will introduce
several models that explicitly consider the aforementioned factors
in technology adoption.

3.1. Learning and adoption of new technologies

In this subsection, we will explicitly consider the “learning by doing”

and “learning by using” effects on technology adoption. Consider a
farmer facing a new technology that can increase their instantaneous
profit by Δπ(τ − t) at any time τ ≥ t if the technology is adopted
−t)
at time t. Note that “learning-by-using” implies ∂Δπ(τ ∂τ > 0 and
∂Δπ(τ −t)
∂t < 0. In other words, the longer the adopter has used the
technology, the larger the benefit she can obtain from it.
It is also reasonable to assume that the rate of “learning-by-
2 2
using” is decreasing over time (i.e., ∂ Δπ(τ∂τ 2
−t)
< 0 and ∂ Δπ(τ
∂t2
−t)
> 0).
Let K(t) be the one-time adoption cost of the new technology at
time t. We assume that the adoption cost decreases over time in a
decreasing rate, i.e., K  (t) < 0 < K  (t). Set the current time to be 0.
The farmer’s decision problem is to decide the optimal time t ≥ 0
to adopt the new technology, considering the effects of “learning-by-
using” and “learning-by-doing” on the benefits and costs of adoption.
Let the discount rate be r, the farmer’s optimization problem can be
238 Agricultural Economics and Policy

written as
 ∞ 
−rt −r(τ −t)
max e Δπ(τ − t)e dτ − K(t) . (15)
t t

Assuming an interior solution, then the first-order condition of the

maximization problem in (15) is
∞
∂Δπ(τ − t) −r(τ −t) ∂K(t)
−Δπ (0) + e dτ + rK(t) − = 0. (16)
t ∂t ∂t
Equation (16) illustrates the gains and losses from delaying the
adoption per one marginal moment from the optimal adoption
time, t∗ . Specifically, the first term, −Δπ (0), indicates the direct
profit loss from delaying the adoption at t∗ . The second term,
∞ ∂Δπ(τ −t) −r(τ −t)
t ∂t e dτ , is the loss of total profits over the techno-
logical lifetime caused by the reduced “learning-by-using” due to
the delay (note that here we assume the lifespan of the technology is
infinite). The third term, rK(t), is the saved interest for the one-time
adoption cost due to the delay. Finally, the fourth term, −∂K(t)/∂t,
is the saved adoption cost due to the delay. In the optimal solution,
the sum of these four terms is 0.
It is readily checked that an increase in interest rate, r, will reduce
the loss from decreased “learning-by-using” (i.e., the second term
in Equation (16)), and increase the saved interest for the adoption
cost (i.e., the third term in Equation (16)). Therefore, higher interest
rates tend to delay new technology adoption. Moreover, if production
technology is constant returns to scale, then both the profit increase
from the new technology and the “learning-by-using” effect will
be proportional to farm size, so large farms would adopt the new
technology earlier than small farms.

3.2. A deterministic dynamic model

In addition to “learning-by-doing” and “learning-by-using,” drivers

of the changes in benefits and costs of technology adoption
include variations in the environment and resource availability. For
instance, climate change has increased the frequency of droughts and,
 Technology Adoption 239

therefore, spurred the adoption of drought-tolerant seeds. The deple-

tion of an aquifer may increase the adoption of more efficient
irrigation technology, such as drip irrigation. Shah et al. (1995)
developed a deterministic dynamic model to illustrate the adoption of
advanced irrigation technology considering the decreasing availability
of irrigation water. Although focused on irrigation, the model in
Shah et al. (1995) can be readily transferred to understanding tech-
nology adoption under several changing resource or environmental
constraints. In this subsection we briefly discuss this model.
Suppose farmers are facing two irrigation technologies: a conven-
tional technology (e.g., furrow irrigation) and a modern technology
(e.g., drip irrigation). The modern technology can increase water
use efficiency, but at a higher establishment cost. Assuming constant
returns to scale production function, we denote the output per acre
as y = f (e), where e is effective water use per acre and is determined
by e = hi (α)a, as we discussed in the production chapter. Here α is
land quality, a is actual water use, and hi (α) reflects how efficient
the land with quality α, together with the irrigation technology
(i = 1 indicating the conventional technology, and i = 2 the modern
technology), will use the actual water applied to it. We assume that
h1 (α) = α < h2 (α) ≤ 1 and h2 (α) > 0 > h2 (α), which implies
that modern technology improves the water use efficiency of any
land when compared with the conventional technology and that the
improvement is decreasing as land quality increases.
Let the range of α be [αL , 1], where αL is the lower bound of land
quality. Furthermore, let g(α) be the density function  1of land quality
α. By normalizing total land area to be 1, we have αL g (α) dα = 1.
Further, let δit (α) denote the share of land with quality α that adopts
technology i ∈ {1, 2} at time t. Then we have 0 ≤ i δit (α) ≤ 1 for
any t and α. Therefore, we have
 
1 
0≤ δit (α)g(α) dα ≤ 1. (17)
aL i

Finally, let ki denote the per-acre cost of using technology i

at time t. Then, we write out the expressions for the total output
240 Agricultural Economics and Policy

(Yt ), total water used (At ), and total technology cost (Kt ) at time t.
They are:
 
1 
Yt = δit (α)f (hi (α)ait ) g(α)dα, (18)
aL i
 
1 
At = δit (α)ait g(α)dα, (19)
aL i
 
1 
Kt = δit (α)kit g(α)dα. (20)
aL i

Denote the stock of available groundwater at time t by St . Excluding

any recharge of the groundwater, we have

Ṡt = −At . (21)

Let U (Yt ) be the instantaneous benefit from consuming total output

Yt , and c(St ) be the cost of extracting one additional unit of
groundwater when the water stock is St at time t. Therefore, the
total social welfare at time t is U (Yt ) − c(St )At − Kt . The social
planner’s goal is to maximize net social welfare over time 0 to T
by optimally choosing land share allocated to each technology (i.e.,
δit ) and water use under each technology (i.e., ait ). Denoting discount
rate by r, the social planner’s optimization problem can be written as
T
max e−rt [U (Yt ) − c(St )At − Kt ]dt, (22)
δit ,ait 0

s.t. (17) to (21), ST ≥ 0, and given S0 .

The current Hamiltonian value is

H = U (Yt ) − c(St )At − Kt − λt At − pt Yt + wt At + γt Kt

1 
+ δit [pt f (hi (α)ait ) − wt ait − γt kt ]
αL i
  

+ μt (α) 1 − δit (α) g(α)dα , (23)
i
 Technology Adoption 241

where λt is the shadow cost of groundwater stock at time t, and

pt , wt , γt , and μt (α) are the shadow prices of the output, water
applied, aggregate technology cost, and land capacity constraints.
The maximization of the current value Hamiltonian requires
∂H
λ̇t = rλt − = rλt + At c (St ), (24)
∂St
∂H
= U  (Yt ) − pt = 0, (25)
∂Yt
∂H
= −c(St ) − λt + wt = 0, (26)
∂At
∂H
= γt − 1 = 0, and (27)
∂Kt
 
 
max δit [pt f (hi (α)ait ) − wt ait − γt kt ] + μt (α) 1 − δit (α) .
δit ,ait
i i
(28)
Equation (24) indicates that the change in the shadow price of the
stock at time t, λ̇t , is the interest of the shadow price, rλt , minus
the extraction cost increase caused by the marginal depletion of the
groundwater stock (note that c (St ) < 0). Equation (25) states that
the shadow price of total output is equal to the marginal benefit of
the output. Equation (26) states that the shadow price of applied
water is equal to the extraction cost of one additional unit of water
plus the shadow price of groundwater stock. Equation (27) implies
that the shadow price of technology cost is one. Finally, Equation
(28) requires that optimal δit and ait should be chosen for all kinds
of land at the micro-level.
Obtaining the optimal solutions from expressions (24)–(28) is
quite technical and we refer readers to Shah et al. (1995) and Caswell
and Zilberman (1986) for details. Here we focus on the insights of
technology adoption derived from the model. Shah et al. (1995) show
that under the conditions that h1 (α) = α ≤ h2 (α) ≤ 1, h(1) = 1, and
h2 (α) > 0 > h2 (α), there is a land quality threshold value, denoted
by αst , above which the modern technology will not be adopted. This
is because the gains from modern technology are smaller for high-
quality land than for low-quality land, and the smaller gains cannot
242 Agricultural Economics and Policy

justify the adoption cost. They also show that there is another land
quality threshold, denoted by αm t , below which the land will be
left idle (i.e., neither the conventional technology nor the modern
technology will be adopted on the land). Therefore, at time t, the
modern technology is adopted on land with quality over [αm s
t , αt ],
and the conventional technology is adopted on land with quality
higher than αst . The dynamics of αm s
t and αt determine the diffusion
of modern technology over time.
Shah et al. (1995) show that if i ≡ −f  (e)e/f  (e) > 1 and
ηi ≡ hi (α)α/hi (α) ∈ [0, 1], then ∂αst /∂pt > 0 and ∂αm t /∂pt < 0,
which indicates that an increase in output price will increase the
adoption of modern technology. From equation (26), we obtain ẇt =
c (St )Ṡt +λ̇t . Based on equations (21) and (24), we have ẇt = rλt ≥ 0,
indicating that water price is always non-decreasing. Shah et al.
(1995) show that if i > 1 and ηi ∈ [0, 1], then the increase in water
price will increase both αm s
t and αt , leaving the diffusion of modern
technology determined by the distribution of land quality. If the
land quality distribution is unimodal, then the modern technology
diffusion may be S-shaped. Figure 3 provides an example of the

Figure 3. Unimodal land quality distribution.

 Technology Adoption 243

relationship between modern technology diffusion and land quality

distribution. In this example, we assume that αm t is affected little
by the increasing water price over time, whereas αst increases by
the same amount each period. In this case, the rate of adoption of
modern technology will increase initially but then decrease, creating
an S -shaped diffusion curve.

3.3. Real option models

Technology adoption, like many other investments, involves various

degrees of irreversibility. For instance, time spent on studying,
purchasing, and learning how to use the new technology cannot be
recovered. Newly purchased equipment, if to be re-sold, often endures
significant discounts. Moreover, benefit and cost uncertainties are
prevalent among new technologies. The benefits of drought-tolerant
varieties depend on the weather of the growing season because
these varieties may perform better than traditional varieties during
dry years but worse during wet years. The cost of new machinery
may fluctuate considerably over time due to trade frictions, wars,
or disease outbreaks (think about human history over 2018–2023).
Real option models consider both irreversibility and uncertainty
associated with technology adoption, shedding new insights that are
missing in the aforementioned models in this chapter. Real option
models view technology adoption as an option (right, but not an
obligation to adopt), which can be exercised at any time chosen by
the adopter. As new information in future periods may mitigate the
uncertainty related to the benefits and costs of the new technology,
under certain conditions delaying adoption may be the optimal
choice. We here use a simple two-period model to illustrate this point.
Suppose that a farmer is considering adopting a new technology
(e.g., an irrigation system). Let Ri denote the revenue generated by
the new technology in period i ∈ {1, 2}. Suppose R1 is deterministic
and R2 is random. With equal chance, R2 can take values of 100 and
200. The farmer will be certain about the value of R2 at the beginning
of period 2. For simplicity, we assume that the adoption cost, K,
is constant over the two periods, with a fixed value at K = 150.
We also ignore the discount rates. Now, the question is, should the
244 Agricultural Economics and Policy

farmer adopt the new technology in period 1 or delay the decision to

period 2?
If the farmer adopts in period 1, then the expected net present
value (NPV) of adoption is

N P V0 = R1 −K +E (R2 ) = R1 −150+0.5×(100 + 200) = R1 . (29)

The standard NPV approach would recommend adoption in period

one as long as R1 > 0. However, what if the farmer waits till period 2
and based on their decision on new information in that period? Doing
so, she would forgo the revenue in period 1 (i.e., R1 ) but she avoids
the possibility of losing money in period 2. This is because, if she
adopts in period 1, then, in period 2, she has a 50% chance of
receiving a low return, 100. If R1 is smaller than 50, then receiving
100 in period 2 indicates that the total revenue in the two periods
cannot cover the adoption cost. Overall, if she delays the adoption
decision to period 2 after the new information about period 2 revenue
is revealed, she will adopt when the revenue is 200 and not adopt
when the revenue is 100. In sum, the expected NPV of delaying the
decision to period 2 is

N P V1 = E [max (R2 − K, 0)] = 0.5 × 0 + 0.5 × (200 − 150) = 25.

(30)
By comparing N P V0 and N P V1 , we can conclude that, if R1 ≤ 25,
then the farmer should delay the adoption decision to period 2. If
R1 > 25, then the farmer should adopt in period 1. We can see that
the real option approach recommends a higher period 1 revenue (25
instead of 0) to trigger the adoption than does the standard NPV
approach. This is because the standard NPV approach ignores the
value of the “option” that the farmer owns to adopt in period 2,
which should be viewed as part of the opportunity cost of adopting
the technology in period 1. Therefore, we can define the option value
(the value of flexibility) as max[N P V1 − N P V0 , 0]. In this case, the
option value is max[25 − R1 , 0].
The two-period example above illustrates a powerful framework
to model adoption decisions or other decisions under irreversibility
and uncertainty. Two-period real option models have the essence
 Technology Adoption 245

of real option models (i.e., new information revealed in a future

period and the decision-maker has the option to delay an action
to benefit from the new information) and are very much tractable.
As a result, they are widely used in the literature. For instance,
Arrow and Fisher (1974) used a two-period model to study whether
an area should be developed or preserved. Moledina et al. (2003)
examined firms’ strategic abatement behaviors under emission tax
and permit using a two-period real option framework. Miao et al.
(2012) studied investment in cellulosic biofuel technologies under a
waivable mandate by using a two-period real option model. Finally,
Miao et al. (2022) employed a two-period real option model to study
whether or not to convert a parcel of land into a new use considering
uncertainties of returns from alternatives.
An extension of the above two-period example into a multi-
period framework is straightforward. Let r denote the discount rate
and Δπ (st ) the profit difference between the new technology and
traditional technology, where st is a random variable. The farmer’s
decision problem is to optimally choose the adoption time, T , to
maximize the expected benefits from adoption. The problem can be
written as
∞  
Δπ (st ) K
max Est t + , (31)
T ∈{0,1,...,∞}
t=T
(1 + r) (1 + r)T

where Est is the expectation operator with respect to st . The solution

to problem (31) involves identifying an adoption policy under which
whenever the realization of a random variable st reaches a threshold
value, then adoption will occur. The properties of the stochastic
process {st } determine the optimal adoption time. A general insight
from the solutions to problem (31) is that, when the variance of
st increases the expected optimal adoption time will be larger.
Intuitively, when future returns are becoming more uncertain and
adoption involves irreversible costs, then delaying the action may be
a rational choice. Dixit and Pindyck (1994) provide a comprehensive
set of treatments on real option models like the one in (31) and its
many other variations. Recent applications of the real option models
include Ye et al.’s (2022) study on the impact of renewable energy
246 Agricultural Economics and Policy

policies on renewable energy technology development and adoption,

focusing on the increased risk of technology obsolescence caused by
the policies.

4. Beyond Profit and Utility Maximization

The preceding models (except the imitation models) in this chapter
are based on neoclassical economic theory, where decision rules
are derived from maximizing expected profits or utility. In the
past four decades, however, developments in behavioral economics,
experimental economics, and many other fields in economics have
significantly enriched our understanding of people’s decision-making
associated with technology adoption. A large literature has focused
on understanding people’s loss, ambiguity, time, and social prefer-
ences and their implications on technology adoption. In this section,
we provide a brief review along these lines of research.
An emerging body of research employs prospect theory to
understand farmers’ technology adoption behavior.6 Liu (2013)
conducted field experiments to examine Chinese cotton farmers’
adoption decisions regarding Bt cotton, finding that higher loss
aversion delays the adoption of Bt cotton, but overweighting small
probabilities accelerates the adoption. Bocquého et al. (2015), in
contrast, find that both loss aversion and overweighting small
probabilities decrease bioenergy crop adoption of farmers based on
a survey and experiments conducted in France. Also focusing on
bioenergy crop adoption but using a numerical simulation approach,
Anand et al. (2019) find that an increase in loss aversion will decrease
miscanthus adoption but will increase switchgrass adoption.
Studies on technology adoption based on expected utility the-
ory or prospect theory assume that farmers know the probability
distribution of the returns of a new technology, which is a strong
assumption because the returns are determined by numerous fac-
tors including the uncertain technical characteristics of the new

6
See Chapter 4 for an introduction of prospect theory.
 Technology Adoption 247

technology, market environment of output, and individual adopter’s

capability of managing the technology (Chavas and Nauges, 2020).
A few recent studies in agricultural technology adoption have
considered the impact of ambiguity aversion on adoption. Barham
et al. (2014) conducted field experiments with U.S. Midwestern
farmers and found that farmers with higher ambiguity aversion tend
to adopt genetically modified (GM) maize earlier, due to GM maize
reducing the ambiguity of pest-damage levels. In contrast, Ward and
Singh (2015) do not find that ambiguity aversion has an obvious
impact on the adoption of drought-tolerant rice in rural India. Based
on a MaxMin expected utility model, Bryan (2019) shows that the
ambiguity created by rainfall index insurance decreases the demand
for such insurance and the incentivizing impact of the insurance
on technology adoption among ambiguity-averse farmers. Chapter 4
describes an example employing the α-MaxMin model to understand
people’s adoption decisions under ambiguity.
Because the benefits of technology adoption can be recouped over
years of the lifespan of the technology, how farmers discount returns
in the future is critical for estimating the overall benefit of adoption.
Studies (e.g., Duquette et al., 2012) have shown that farmers’
discount rate can be as high as 40% (i.e., $1,000 one year from now
is only as good as about $700 right now). Standard intertemporal
decision-making models have been assuming a constant discount rate
since the seminal work by Samuelson (1937) that predicted time-
consistent behaviors. Although convenient, these models encounter
difficulties in explaining some “anomalies” in people’s intertemporal
decisions, such as preference reversals (Thaler, 2015; Ch. 11). Exist-
ing studies (see Frederick et al. (2002) for a comprehensive review)
in experimental economics literature have shown that hyperbolic
discounting (the discount rate becomes smaller as time departs
further away from present) can better fit people’s observed discount
rates over time and better explain intertemporal behaviors. Although
Tanaka et al. (2010) and Liebenehm and Waibel (2014) estimate
farmers’ time preference parameters based on hyperbolic discounting
models, few studies have examined the role of hyperbolic discounting
248 Agricultural Economics and Policy

in agricultural technology adoption. Some early works in this realm

include Duflo et al. (2011) and Clot et al. (2017). The former
study found that a small and time-limited price discount of fertilizer
after harvest, when farmers have relatively abundant cash, increases
fertilizer use among farmers with hyperbolic discounting by inducing
a commitment to using fertilizer. Clot et al. (2017) find that farmers
with higher risk aversion are more impatient, and less likely to
enroll in long-term conservation programs in which program failure
is possible.
Because technology adoption is a complicated process that
involves farmers’ judgment on various potential effects of a new
technology on farm operation, studies have shown that farmers’
human capital (education and on-the-job learning) affects technology
adoption via different channels (Huffman, 2020; Foster and Rosen-
zweig, 2010). Huffman (1977) shows empirical evidence that farmers
with higher human capital can respond to technical and economic
changes more efficiently. A natural inference from this conclusion is
that farmers with higher human capital will be more likely to succeed
in harvesting the benefits of new technology and to adopt (Huffman,
2001; Abdulai and Huffman, 2014; Pannell and Zilberman, 2020).
Farmers with higher education levels are often more patient and
therefore are more willing to make investments to increase future
returns (Liebenehm and Waibel, 2014). Moreover, human capital
is likely to be positively correlated with cognitive ability, which
accelerates adoption (Barham et al., 2018; Dessart et al., 2019).
A factor that is closely related to human capital in the realm
of technology adoption is learning, which has been a focus of many
studies in the adoption literature (see Foster and Rosenzweig (2010)
for a comprehensive review). Learning mitigates the uncertainty of
the new technology and increases the adopters’ ability to better use
the technology (Foster and Rosenzweig, 1996). Therefore, educating
farmers on the new technology, either through public extension
services or peer farmers, may accelerate adoption (Genius et al.,
2014). Due to the positive externality of the learning conducted by
early adopters, Foster and Rosenzweig (1995) suggest that subsidies
for early adoption could be used to enhance adoption.
 Technology Adoption 249

In addition to loss, ambiguity, and time preferences, the behav-

ioral economics literature also considers the roles of social
preferences, such as peer effects, social status, and social networks
on technology adoption. Dessart et al. (2019) and Streletskaya
et al. (2020) provide excellent reviews on behavioral economics and
technology adoption. An interesting discussion about behavioral
economics associated with agricultural production can be found in
Wuepper et al. (2023). Based upon randomized controlled trials,
many experimental economics studies have examined factors that
constrain agricultural technology adoption. Bridle et al. (2020)
provide a comprehensive review of recent experimental findings of
these studies, with a focus on farmers in Sub-Saharan Africa and
South Asia, outlining four obstacles to technology adoption: credit
constraint, risk exposure, information scarcity, and limited access to
input and output markets. The authors suggest that risk mitigation
and enhanced access to information are critical for encouraging
adoption.

5. Technology Adoption and Marketing

Based on the work of Kalish (1985), Rogers (2003), and Zilber-

man et al. (2012), we posit that individual decision-makers go
through four stages in their adoption choice. First, the awareness
of the new product may be influenced by observing other people
and is consistent with the imitation model. Second, assessment
that involves experimentation via producer-offered demonstrations
or experimentation with products by friends and family. Third, the
adoption decision that may be based on constrained optimization,
but the information used for this decision is affected by the judgment
of others and social norms. Behavioral economics provides some
new insights into the way people make choices (Thaler, 2015). One
important notion is loss aversion suggesting that people emphasize
avoiding losses over achieving wins (see Chapter 4 for details).
Therefore, they need various mechanisms of reassurance about the
performance and the value of a new technology in evaluating it.
Mechanisms like demonstration that allow for testing the technology
250 Agricultural Economics and Policy

before purchase, money-back guarantees, warranties, or insurance of

performance that protect against failure are important in deciding
to adopt a new technology.
Frequently, people may decide to rent a technology (e.g., a
car) in order to experiment with it to reduce the perceptions
of risk and increase the likelihood of adoption (Zilberman et al.,
2012). Credit support can also reduce the fixed cost of adoption.
Furthermore, Kahneman (2011) distinguished between important,
large-scale technologies that require deliberation and less important
decisions that can be made ad hoc. While buying a new car requires
some deliberation and rational decision-making, the decisions about
complementary components are often made ad hoc. The fourth and
final stage is re-evaluation to determine whether to purchase the
product in the future.
Potential buyers are heterogeneous in terms of human capital,
income, education, and preferences, and different buyers will adopt at
different times. Another source of heterogeneity is self-efficacy, which
is a person’s perception of their own ability to influence the outcome
of events (Bandura, 1994), separating innovators and imitators (Bass,
1969). Innovators are individuals who “decide to adopt an innovation
independently of the decisions of other individuals in a social system”
while imitators are adopters who “are influenced in the timing of
adoption by the pressures of the social system” (Bass, 1969).
Over time, dynamic processes will change the key parameters
and expand the range of adopters. These include learning about the
technology (with experience, people better understand the parame-
ters of the technology and their risk concern declines), learning by
doing (i.e., a manufacturer improves the productivity of product
production and it becomes cheaper), learning by using (the user
benefits from experience with technology), and network externality
(when the benefit of the technology depends on the number of users,
for example, the telephone or internet). Therefore, more people find
the technology attractive over time.
New products may be introduced in different forms. Lu et al.
(2016) suggest that when it comes to an individual product with
 Technology Adoption 251

high fixed costs, early periods are characterized by few purchases

for their own use, and some people are likelier to purchase the
product to rent to individuals that are less able to buy or utilize the
product. For example, large farmers will buy a combine and other
farmers will contract services. Over time, as the price of the product
declines, rental rates may decline, and ownership may spread. The
first company that sold sprinklers in California was called “Rain for
Rent.” Over time, farmers purchased their own irrigation equipment.
Rental agreements are important both to share the fixed costs and
to allow farmers to test the technology. IBM leased mainframe
computers to buyers and only a few very big organizations owned
computers. Now, as the price of computers has gone down, most
Americans own their own computers. Furthermore, as the product
is established, variations of a product allow for different segments
of users to own it. Ownership has two advantages: (i) reduced
transaction costs (you have it when you need it) and (ii) pride of
ownership. Potential buyers compare these gains with the extra costs
of ownership, such as maintenance.
In addition to improvements in the technology that reduce
cost, technological improvement results in increased functionality
and efficiency. Once companies develop supply chains to market a
technology, they need new models and features to resell it. The
shorter the life of a technology, the more important is continuous
innovation by companies to survive. These improvements in tech-
nology further advance the diffusion processes. One mechanism to
enhance a technology is to provide a virtual ecosystem where others
may contribute complementary products, like apps. For example, the
diffusion of computers has been enhanced with the introduction of
editing and business software, games, and online videos (Putler and
Zilberman, 1988). Furthermore, suppliers of the technology aim to
improve upon it to produce new versions, gaining from variations
as consumers are more willing to pay for specific features. When
an existing version of a product becomes obsolete, they are sold to
new groups of buyers (e.g., low-income buyers) and the high-income
premier segments purchase the newer versions. Furthermore, some
252 Agricultural Economics and Policy

producers will develop new brands and supply chains to emphasize

their unique value proposition, as Apple did with moving from cell
phones to smartphones.
In many cases with new products, however, producers may
be uncertain that people will purchase their products. Having an
intermediary that develops a supply chain to transmit a product
between the producer and buyer thus carrying much of the risk on
both ends, may facilitate the introduction and development of the
new product. This is quite frequent, such as Costco develops a new
supply chain to buy quinoa from producers in the Andes.
Phillips (2011) introduced the notion of relentless innovation,
connecting to the experience of companies like Nokia and Blackberry
that they must innovate or die. Another example is the shift
away from driving as people, particularly younger generations,
choose alternatives, such as ride-sharing or Uber/Lyft (addition-
ally facilitated by being more adept with other technologies, like
smartphones). Thus, car companies have more incentive to develop
autonomous vehicles (AVs).
In the case of AV, new innovations will include improvements
to the performance of the car as well as tailoring the car to reach
different market niches. Thus, the automobile industry is likely to
have several major producers and an ecosystem of small firms that
produce the “apps” that complement the AV. Nobel Laureate Gary
Becker (1965) and the seminal work of Rosen (1974) introduced
the notion of household production functions and hedonic pricing,
respectively. In this case, consumers would pay for features and
components of technologies. The price of a product embodies the
price of the value of its components. But, when an AV is introduced,
people will have the convenience of riding their own car while
also performing other activities, including entertainment, resting, or
working. Because the AV will provide more amenities than traditional
cars—transport in addition to the environment for business, living,
and entertainment—one may expect more diverse sets of products
in AV than in current vehicles. So, while we look at the AV
as a mechanism to reduce dependence on the personal car, the
literature suggests that it will actually enhance the range of functions
 Technology Adoption 253

performed by a vehicle, and especially personal vehicles. In the long-

run, as prices decrease, it may even result in increased car ownership
rates.

6. Conclusion
Technology adoption is a key step for farmers to enhance their
productivity, competitiveness, and profitability. In this chapter, we
have summarized major economic models explaining technology
adoption, as well as technology adoption drivers beyond profit
and utility maximization. We believe that technology adoption will
continue to be an active research area in agricultural economics due
to the continuous emergence of new technologies and the urgent need
for new technologies to address challenges such as climate change and
food security.
Two major related issues that affect adoption are regulation and
acceptance. For example, governments regulate new medicines as well
as the use of chemicals in agriculture. One of the major reasons
for product regulation is safety. The literature on risk perception
shows that people are more concerned with the unknown than
the known, and are more concerned about the risk they cannot
control (moving by an autonomous vehicle) than the risk they control
(driving a car). As a result, certification from a trusted government
may lead to consumers or retailers to start adopting a technology.
The introduction of new technologies (e.g., AVs) will be gradual and
will involve extra cautionary regulations, until concerns about risks
are sufficiently reduced. We must keep in mind that the earliest
cars had people running before them to warn against accidents.
However, the regulatory procedures have an inherent benefit-risk
tradeoff, often reflecting political and economic considerations. The
often urgent need for new medicines, such HIV/AIDS treatments,
leads to expedient processes, and one could expect a similar outcome
with biotechnology in medicine compared to that in agriculture.
In the case of acceptance, consumers may oppose a new product
for environmental or other reasons, and may fight politically to
prevent its introduction, even when it is deemed sound to use it.
254 Agricultural Economics and Policy

The introduction of a new technology that has multiple social

ramifications and is regulated entails a political economic battle
involving multiple constituents (Herring and Paarlberg, 2016). They
include groups that are affected negatively by the technology or have
an alternative solution to the problem that the technology addresses.
Returning to biotechnology in agriculture, actors include chemical
companies or groups supporting ecological agriculture and organic
associations. At the same time, however, biotechnology in medicine
was more readily accepted as it was perceived to address immediate
needs and “life or death” situations. In the case of AVs, taxi drivers,
and even Uber or Lyft drivers may try to impede their introduction
while senior living facilities, safety agencies, and others will be more
supportive. Overall, people and organizations that oppose the use of
new technologies may use the political system to block its use, fully
or partially. If they are powerful enough, they may be successful, as
in the case of GMOs in Europe. Or they may be successful in delaying
introduction, as in the case of some GMO traits in the U.S.

References
Abdulai, A. and W.E. Huffman. 2014. The Adoption and Impact of Soil and Water
Conservation Technology: An Endogenous Switching Regression Application.
Land Economics 90(1): 26–43.
Anand, M., R. Miao, and M. Khanna. 2019. Adopting Bioenergy Crops:
Does Farmers’ Attitude toward Loss Matter? Agricultural Economics 50(4):
435–450.
Arrow, K.J. and A.C. Fisher. 1974. Environmental preservation, uncertainty, and
irreversibility. Quarterly Journal of Economics 88, 312–319.
Bandura, A. 1994. Self-Efficacy. John Wiley & Sons, Inc. New York, NY.
Barham, B.L., J.P. Chavas, D. Fitz, V. Rı́os Salas, and L. Schechter. 2014. The
Roles of Risk and Ambiguity in Technology Adoption. Journal of Economic
Behavior and Organization 97: 204–218.
Barham, B.L., J.P. Chavas, D. Fitz, and L. Schechter. 2018. Receptiveness to
Advice, Cognitive Ability and Technology Adoption. Journal of Economic
Behavior and Organization 149: 239–268.
Bass, F.M. 1969. A New Product Growth for Model Consumer Durables.
Management Science 15(5): 215–227.
Becker, G.S. 1965. A Theory of the Allocation of Time. The Economic Journal
493–517.
Bird, K. 2021. The Outlier: The Unfinished Presidency of Jimmy Carter. Penguin
Random House LLC, New York, NY.
 Technology Adoption 255

Bocquého, G., F. Jacquet, and A. Reynaud. 2015. Adoption of Perennial Crops

and Behavioral Risk Preferences. An Empirical Investigation among French
Farmers. Paper presented in the “Risk and Environment” Session at Journées
de Recherche en Sciences Sociales, SFER, Nancy, France, December 11–12.
Bridle, L., J. Magruder, C. McIntosh, and T. Suri. 2020. Experimental insights
on the constraints to agricultural technology adoption. Working Paper.
Bryan, G. 2019. Ambiguity Aversion Decreases the Impact of Partial Insurance:
Evidence from African Farmers. Journal of the European Economic Associ-
ation 17(5): 1428–1469.
Caswell, M.F. and D. Zilberman. 1986. The Effects of Well Depth and Land
Quality on the Choice of Irrigation Technology. American Journal of
Agricultural Economics 68(4): 798–811.
Chavas, J.P. and C. Nauges. 2020. Uncertainty, Learning, and Technology
Adoption in Agriculture. Applied Economic Perspectives and Policy 42(1):
42–53.
Clot, S., C.Y. Stanton, and M. Willinger. 2017. Are Impatient Farmers More
Risk-Averse? Evidence from a Lab-in-the-Field Experiment in Rural Uganda.
Applied Economics 49(2): 156–169.
David, P.A. 1975. Technical Choice Innovation and Economic Growth: Essays
on American and British Experience in the Nineteenth Century. Cambridge
University Press, Cambridge, NY.
Dessart, F.J., J. Barreiro-Hurlé, R., and Van Bavel. 2019. Behavioural Factors
Affecting the Adoption of Sustainable Farming Practices: A Policy-Oriented
Review. European Review of Agricultural Economics 46, 417–471.
Dixit, A. and R. Pindyck. 1994. Investment under Uncertainty. Princeton
University Press, Princeton, NJ.
Duflo, E., M. Kremer, and J. Robinson. 2011. Nudging Farmers to Use Fertilizer:
Theory and Experimental Evidence from Kenya. American Economic Review
101(6): 2350–2390.
Duquette, E., N. Higgins, and J. Horowitz. 2012. Farmer Discount Rates:
Experimental Evidence. American Journal of Agricultural Economics 94(2):
451–456.
Foster A.D. and M.R. Rosenzweig. 1995. Learning by Doing and Learning from
Others: Human Capital and Technical Change in Agriculture. Journal of
Political Economy 103(6): 1176–1209.
Foster A.D. and M.R. Rosenzweig. 1996. Technical Change and Human Capital
Returns and Investments: Evidence from the Green Revolution. American
Economic Review 86: 931–953.
Foster A.D. and M.R. Rosenzweig. 2010. Microeconomics of Technology Adoption.
Annual Review of Economics 2: 395–424.
Frederick, S., G. Loewenstein, and T. O’Donoghue. 2002. Time Discounting and
Time Preference: A Critical Review. Journal of Economic Literature 40(2):
351–401.
Genius, M., P. Koundouri, C. Nauges, and V. Tzouvelekas. 2014. Information
Transmission in Irrigation Technology Adoption and Diffusion: Social Learn-
ing, Extension Services, and Spatial Effects. American Journal of Agricultural
Economics 96(1): 328–344.
256 Agricultural Economics and Policy

Griliches, Zvi. 1957. Hybrid Corn: An Exploration in the Economics of Techno-

logical Change. Econometrica 25(4): 501–522.
Herring, R. and R. Paarlberg. 2016. The Political Economy of Biotechnology.
Annual Review of Resource Economics 8: 397–416.
Huffman, W.E. 1977. Allocative Efficiency: The Role of Human Capital. Quarterly
Journal of Economics 91: 59–79.
Huffman, W.E. 2001. Human Capital: Education and Agriculture. B.L. Gardner
and G.C. Rausser (Eds.), Handbook of Agricultural Economics, Vol. IA,
pp. 333–381. Amsterdam, Netherlands: Elsevier Science/North-Holland.
Huffman, W.E. 2020. Human Capital and Adoption of Innovations: Policy
Implications. Applied Economic Perspectives and Policy 42(1): 92–99.
Just, R.E. and D. Zilberman. 1983. Stochastic Structure, Farm Size and
Technology Adoption in Developing Agriculture. Oxford Economic Papers
35(2): 307–328.
Kahneman, D. 2011. Thinking, Fast and Slow. Farrar, Straus and Giroux, New
York, NY.
Kalish, S. 1985. A New Product Adoption Model with Price, Advertising, and
Uncertainty. Management Science 31(12): 1569–1585.
Liebenehm, S. and H. Waibel. 2014. Simultaneous Estimation of Risk and Time
Preferences among Small-Scale Cattle Farmers in West Africa. American
Journal of Agricultural Economics 96(5): 1420–1438.
Liu E. 2013. Time to Change What to Sow: Risk Preferences and Technology
Adoption Decisions of Cotton Farmers in China. The Review of Economics
and Statistics 95(4): 1386–1403.
Lu, L., T. Reardon, and D. Zilberman. 2016. Supply Chain Design and Adoption
of Indivisible Technology. American Journal of Agricultural Economics 98(5):
1419–1431.
Majeed, F., M. Khanna, R. Miao, E. Blanc-Betes, T. Hudiburg, and E.H. DeLucia.
2023. Carbon Credit Markets Can Reduce the Riskiness of Bioenergy Crop
Production. Journal of the Agricultural and Applied Economics Association
2(2): 181–197.
Mansfield, E. 1961. Technical Change and the Rate of Imitation. Econometrica
29(4): 741–766.
Mansfield, E. 1963. The Speed of Response of Firms to New Techniques. The
Quarterly Journal of Economics 77(2): 290–311.
Miao, R., D.A. Hennessy, and B.A. Babcock. 2012. Investment in Cellulosic
Biofuel Refineries: Do Waivable Biofuel Mandates Matter? American Journal
of Agricultural Economics 94(3): 750–762.
Miao, R. and M. Khanna. 2017a. Effectiveness of the Biomass Crop Assistance
Program: Roles of Behavioral Factors, Credit Constraint, and Program
Design. Applied Economic Perspectives and Policy 39(4): 584–608.
Miao, R. and M. Khanna. 2017b. Costs of Meeting a Cellulosic Biofuel Mandate
with Perennial Energy Crops: Implications for Policy. Energy Economics 64:
321–334.
 Technology Adoption 257

Miao, R., D.A. Hennessy, and H. Feng. 2022. Grassland Easement Evaluation and
Acquisition with Uncertain Conversion and Conservation Returns. Canadian
Journal of Agricultural Economics 70(1): 41–61.
Moledina, A.A., J.S. Coggins, S. Polasky, and C. Costello. 2003. Dynamic
Environmental Policy with Strategic Firms: Prices Versus Quantities. Journal
of Environmental Economics and Management 45(2): 356–376.
Pannell, D. and D. Zilberman. 2020. Understanding Adoption of Innovations
and Behavior Change to Improve Agricultural Policy. Applied Economic
Perspectives and Policy 42(1): 3–7.
Phillips, J. 2011. Relentless Innovation: What Works, What Doesn’t and What
That Means For Your Business. McGraw Hill Professional.
Putler, D.S. and D. Zilberman. 1988. Computer Use in Agriculture: Evidence
from Tulare County, California. American Journal of Agricultural Economics
70(4): 790–802.
Rogers, E.M. 1962. Diffusion of Innovations. The Free Press, New York, NY.
Rogers, E.M. 2003. Diffusion of Innovations (5th ed.). The Free Press, New York,
NY.
Rosen, S. 1974. Hedonic Prices and Implicit Markets: Product Differentiation in
Pure Competition. Journal of Political Economy 82(1): 34–55.
Ryan, B. and N.C. Gross. 1943. The Diffusion of Hybrid Seed Corn in Two Iowa
Communities. Rural Sociology 8(1): 15–24.
Samuelson, P.A. 1937. A Note on Measurement of Utility. The Review of
Economic Studies 4(2): 155–161.
Shah, F.A., D. Zilberman, and U. Chakravorty. 1995. Technology Adoption in the
Presence of an Exhaustible Resource: The Case of Groundwater Extraction.
American Journal of Agricultural Economics 77(2): 291–299.
Streletskaya, N.A., S.D. Bell, M. Kecinski, T. Li, S. Banerjee, L.H. Palm-Forster,
and D. Pannell. 2020. Agricultural Adoption and Behavioral Economics:
Bridging the Gap. Applied Economic Perspectives and Policy 42(1): 54–66.
Sunding, D. and D. Zilberman. 2001. The Agricultural Innovation Process:
Research and Technology Adoption in a Changing Agricultural Sector, B.L.
Gardner and G.C. Rausser (Eds.), Handbook of Agricultural Economics,
Volume 1, Part A, Chapter 4, pp. 207–261.
Tanaka, T., C.F. Camerer, and Q. Nguyen. 2010. Risk and Time Preferences:
Linking Experimental and Household Survey Data from Vietnam. American
Economic Review 100(1): 557–571.
Tarde, G. 1890. Les Lois de l’imitation. Paris: Alcan. English edition: The Laws
of Imitation, translated by Parsons, E.C., The Mershon Company Press,
Rahway, NJ.
Thaler, R.H. 2015. Misbehaving: The Making of Behavioral Economics. WW
Norton & Company.
Ward, P.S. and V. Singh. 2015. Using Field Experiments to Elicit Risk and
Ambiguity Preferences: Behavioural Factors and the Adoption of New
Agricultural Technologies in Rural India. The Journal of Development
Studies 51(6): 707–724.
258 Agricultural Economics and Policy

Wuepper, D., S. Bukchin-Peles, D. Just, and D. Zilberman. 2023. Behavioral

Agricultural Economics. Applied Economic Perspectives and Policy 45(4):
2094–2105.
Ye, F., N. Paulson, and M. Khanna. 2022. Are Renewable Energy Policies Effec-
tive to Promote Technological Change? The Role of Induced Technological
Risk. Journal of Environmental Economics and Management 114: 102665.
Zilberman, D., J. Zhao, and A. Heiman. 2012. Adoption versus Adaptation, with
Emphasis on Climate Change. Annual Review of Resource Economics 4(1):
27–53.
 Chapter 9

Agricultural Supply Chain

Microeconomics seeks to understand the market for goods (e.g.,

bread, meat, oil), which tend to be studied in isolation with rather
simple, and few, linkages. However, consumers frequently do not
consume a raw good (e.g., wheat grain or a cow), but rather
a processed good (e.g., a sandwich or hamburger), the outcomes
of complex supply chains that innovators and producers have to
take into account when developing their products. Economists have
not paid much attention to supply chains (at least before the
COVID-19 pandemic), and when they do, they often take supply
chains as “given.” But the reality is that every time an innovation is
introduced, whoever controls the innovation, thinks about the design
of a supply chain to implement it. This chapter emphasizes the role of
supply chains as mechanisms to implement innovations in the context
of agriculture, studying the design and performance of supply chains.
We will do so by using many examples such as Tyson Foods, beer, and
biofuels. We can apply the concept of the supply chain to just about
any product or service (even education) to understand its elements,
linkages, purposes, and external influences.

259
260 Agricultural Economics and Policy

1. Product Characteristics and Hedonics

1.1. Demand theory and supply chain

Demand theory is a major element of economics. Its pursuit of

generality resulted in a minimal set of assumptions and results. In
particular,

(1) It assumes utility maximization subject to income constraint

and the main result is that (compensated) demand is negatively
sloped.
(2) All goods are treated equal without much attention to the
intrinsic properties of goods (the only valuable distinction is
between goods that are substitutes versus complements and
between inferior, normal, and luxury goods).
(3) Preferences are considered as given.
(4) The number of goods is constant without new innovations.
(5) Consumption decisions are made within markets without atten-
tion to household-level choices.
(6) The household labor associated with consumption is not
considered.

The traditional demand theory has been criticized by Nobel

Laureates Gary Becker and Kelvin Lancaster to be elegant, general,
and vacuous, but not very useful in addressing some of the most inter-
esting and pressing issues (see Chapter 3 in this book). For example,
it is not useful for assessing the demand for new products because it
has limited capacity to identify sources of product improvement and
product differentiation. But much of the modern economy is based
on improving existing products. Lancaster (1966) emphasized more
on the processing of raw goods rather than consumable goods, while
Becker (1965) introduced the notion of household production function
where family labor is used in conjunction with purchased goods to
produce the final consumables with the desired characteristics for
the household. We will develop a framework based on their work and
relate it to our work on supply chains.
 Agricultural Supply Chain 261

Labor at Factory (LF) Capital of Factory (KF)

Feedstock (X)

Labor in Household (LH) Capital in Household (KH)

Differentiated Product (Y )

Consumables (C)

Figure 1. A supply chain from feedstock to consumables.

Like Becker and Lancaster, we will assume that there is a

sequence in the production system, and this sequence is a supply
chain. The starting point is a commodity, like cattle, that we
call feedstock and is denoted by X. Feedstock is processed in a
factory, with inputs like labor (LF ) and capital (KF ), to produce
a differentiated product (Y ), say, meat. Within the household, the
product is further processed with family inputs including labor (LH )
and capital (KH ) to produce a consumable (C), like a steak. This
basic model is presented in Figure 1. We can formalize it by denoting
the production function of the differentiated product at the factory
as Y = f (KF , LF , X) and the production function of the consumable
as C = g(KH , LH , Y ). Of course, this simple, one-dimensional model
can be expanded to have vectors of each variable. For example,
households may buy many types of differentiated products, using the
labor of different family members and different types of equipment
(i.e., capital), to produce different types of consumables.
262 Agricultural Economics and Policy

Now let us assume that the household has initial income denoted
by I and time T that can be used for household work LH or work
outside the home that earns wage W (in dollars). Let us also assume
that the household operates in a market economy and the price of Y
is P and the cost of using one unit of household capital is r.1
Thus, the net income after expenditures on Y and KH is: IN =
I − P Y − rKH + W · (T − LH ). The household derives utility from the
consumption of C and net income. The household utility is denoted
by U (C, IN ). To simplify, plug C = g(KH , LH , Y ) and IN = I −
P Y − rKH + W · (T − LH ) in the utility function. The household
determines their purchase of Y , their amount of KH and LH . The
optimization problem is,
 
Max U g (KH, LH , Y ) , I − P Y − rKH + W · (T − LH ) . (1)
{Y,LH ,KH }

The first-order conditions are derived using the chain rule:

dU (C, IN )) dg(KH , LH , Y ) dU (C, IN )
· =P· . (2)
dC dY dIN
dU (C,IN ))
dC
Let λ = dU (C,IN ) denote the monetary value of consuming one unit of
dIN
C at the optimal solution. Condition (2) implies that the optimal Y
is determined when the value of its marginal product λ · dg(KHdY, LH ,Y )
is equal to the price P . Similarly, one can derive the first-order
conditions that denote the amount of household labor and capital
allocated to producing C.
dU (C, IN )) dg(KH , LH , Y ) dU (C, IN )
· = · W. (3)
dC dLH dIN

dU (C, IN )) dg(KH , LH , Y ) dU (C, IN )

· = · r. (4)
dC dKH dIN

1
For simplicity we avoid using a dynamic model on decisions about investments
in durables that require multi-period analysis. Instead, we assume that the
household pays a fee for each unit of equipment they use.
 Agricultural Supply Chain 263

Conditions (3) and (4) suggest that the optimal level of household
labor and capital allocated to producing C is where the value
of their marginal products are equal to W and r, respectively.
Condition (2) provides the foundation for the consumer demand for
the differentiated commodity, Y , which will depend on the initial
income, I, the cost of capital, r, and the cost of labor, W . Similarly,
conditions (3) and (4) provide the basis to estimate how much
household time and capital will be supplied to produce C, given the
prices P , W , and r.
Our analysis is highly simplified. Certainly, there is heterogeneity
among households where each household has its own demand func-
tion, and these demand functions are combined to produce aggregate
demand. But the basic principle is that choices at the household level
determine how much consumables that each household will derive
and the demand for differentiated products. The analysis can also be
expanded at the factory level to determine the optimal supply of Y
given the prices of feedstock, capital, and labor at the factory. The
interaction of the demand and the supply for Y will determine its
price.

1.2. Product characteristics and hedonics

Now let us add another level of complexity based on a classic paper

by Rosen (1974). Assume that Y , the differentiated product, can
have different levels of quality, q. In this case, we assume that utility
is now also a function of q and it becomes U (C, IN , q).2 Every
family must then determine both the quantity and the quality that
they consume. In a competitive economy, the family faces a hedonic
pricing formula, where price as a function of quality: P (q). Hedonic
pricing is a method of analysis used to assess the marginal value of
a characteristic.
Generally, we expect that the price of the commodity will improve
with its quality. In this case, the optimization problem of the

2
For simplicity, we do not consider the more general case where production
function of C is affected by q, such that C = g (KH, LH , Y, q).
264 Agricultural Economics and Policy

household becomes,

Max U (g (KH, LH , Y ) , I − P (q)Y − rKH + W · (T − LH ), q).

{Y,LH ,KH ,q}
(5)
In this case condition (2) becomes,
dU (C, IN , q)) dg(KH , LH , Y ) dU (C, IN , q)
· = · P (q). (6)
dC dY dIN
And conditions (3) and (4) will change accordingly. But more
importantly, we have another first-order condition:
dU (C, IN , q)) dU (C, IN , q) dP
= · . (7)
dq dIN dq
dU (C,IN, q))
dq
Now let μ(q) = dU (C,IN, q )
be the marginal monetary value of a unit
dIN
of quality to the household. The optimal quality of Y purchased by
a given household will be at the point where the quality’s marginal
monetary value is equal to the marginal increase in price caused by
the increase in quality, namely, μ(q) = dPdq . The marginal monetary
value of a unit of quality is an element in hedonic pricing. From this
analysis, one can derive demand for both the quantity and quality of
the product, which will be dependent on the hedonic pricing formula
P (q) as well as r, I, T , and W .
Rosen’s analysis suggests that higher-income people purchase
higher-quality products and that the marginal price of quality is
increasing. Rosen’s analysis also considered firms’ behavior. When
producers are heterogeneous, different producers will elect to produce
products of different quality. Market forces will lead to an equilibrium
that determines P (q), and for each level of quality, the quantity of
demand equals the quantity of supply.
Many products have different product characteristics, which cap-
ture different elements of quality. For example, foods have calories,
vitamins, and protein, as well as taste and appearance indicators.
Some elements of quality are continuous variables (e.g., calories or
protein content in a food item) and others are discrete variables
(e.g., a food product can be either organic or not). In addition, some
 Agricultural Supply Chain 265

House price
with view

no view

40

10
House size

Figure 2. Relationship between house size and house price considering ocean
view.

elements of quality are objective (e.g., vitamin content) while others

are subjective (e.g., taste or beliefs). Each product characteristic has
a hedonic price. Economists have developed mathematical formulas
to estimate the price of a good as a function of its product
characteristics. The most common use of hedonic pricing is in the
valuation of houses. A simple model may assume that the price of a
house in a given region of San Francisco (Ph ) depends on its size S
and whether or not it has an ocean view (V = 1 if it has an ocean
view and 0 otherwise). A simple linear formula is:
Ph = αS + βV,
where α is the hedonic price of the square foot and β is the shadow
price of an ocean view. Figure 2 shows that if one has data on house
prices, the two coefficients can be estimated. Of course, actual studies
estimate more complex formulas with different hedonic prices for
different room types, the safety of neighborhood, quality of yard,
and so on.

1.3. Hedonics, innovation, and supply chain

Nowadays, because food demand is inelastic and increased supply

reduces prices, a big challenge for agricultural firms and regions is
not necessarily to produce more food, but rather to increase the
266 Agricultural Economics and Policy

value added.3 Increasing value-added and quality of products is key

to many industries. Innovation that improves product quality tends
to change the hedonic price equation, and therefore the elements of
the equation are changing over time. For example, drip irrigation
allowed farmers to grow fruits and vegetables on lower-quality land
(e.g., steep hills or sandy soil) where they were not grown before.
After the introduction of drip irrigation, the price of lower-quality
land increased, and because of the increased food supply, output
prices declined, and the value of higher-quality land declined (Caswell
and Zilberman, 1986). Moreover, the food system over the last
100 years has gone through a value-added revolution, where more
processed food requires fewer labor requirements at the household
level (Levenstein, 2003). Changes in food availability and in demand
are shifting the hedonic prices of different food characteristics,
including nutrients, taste, and convenience.
The way that changes in quality are introduced depends on the
capacity of the innovating organization to implement their innovation
through the design of supply chain and marketing activities. One
example is the transition from early cell phones to smartphones,
which resulted in new features of phones and a new valuation
of their attributes. This transition was also associated with the
emergence of Apple and Samsung and the decline of Nokia. To
better understand these transitions, we need to understand product-
changing innovation processes and associated supply chains.

2. Innovation and Supply Chains

Innovations are the foundation of changes in technologies and
institutions. There is a large literature on the economics of innovation
(Sunding and Zilberman, 2001) that provides classifications of inno-
vations and explains their generation, implementation, and impacts.

3
The main agricultural problem in the U.S. and developed world has been
oversupply that led to subsidies to reduce acreage. At the same time, the global
population is growing, and we need to increase food production. So, agribusinesses
and governments work both to increase food security as well as the quality and
value-added of food.
 Agricultural Supply Chain 267

While most of the literature takes markets and supply chains as

given when considering innovation implementation, we stipulate that
implementation of innovations often involves creating or redesigning
supply chains. Moreover, supply chains are not rigid entities, but
rather flexible structures that both respond to and affect technologies
and policies. In this section, we will first discuss innovation and
innovation processes, and then link them to supply chains. Obviously,
this is only an introduction. Both topics, together with the linkages
between them, will get more coverage as we move forward in this
chapter.
Innovations are new ways of doing things. Schumpeter (1934),
the great Austrian economist who taught at Harvard, recognized the
importance of innovation and technological change. He emphasized
that economic systems were constantly evolving and that innovations
were the key driving forces that led to the creation of new firms and
markets and the destruction of old ones (which he called “creative
destruction”). Innovating firms may acquire market power, but they
often dissipate as new innovations and competitors emerge. These
processes lead to increased availability of products and decreased
costs. Schumpeter’s ideas, in many ways, contrast with the tradi-
tional emphasis in microeconomics on equilibrium and competition.
Many of the modeling tools of microeconomics can be used to
illustrate and expand Schumpeterian ideas. Our analysis here is based
on some of his concepts linking explicitly innovation processes with
supply chains and markets that innovations induce or affect.
There are many ways to categorize innovations. One major
distinction is between embodied and disembodied innovations (see
Chapter 7 in this book). Embodied innovations are reflected by
new products or services that can be sold, whereas disembodied
innovations are mostly ideas and concepts. The owner of an embodied
innovation can capture much of the benefit by selling it in the
market, but it is much more difficult to capture the financial
benefit from disembodied innovations. Therefore, the government
may need to support the creation and adoption of disembodied
innovations. A more detailed classification distinguishes between
product innovation (e.g., tractor, automobile), process innovation
268 Agricultural Economics and Policy

(e.g., internal combustion engine), institutional innovation (e.g.,

banks), managerial innovation (e.g., double ledger accounting), and
resource use innovation (e.g., growing flowers in Kenya). Classifica-
tion can go even further. For example, innovations that address pest
control problems can be classified as mechanical (e.g., using a hoe),
chemical (e.g., pesticides), biological (e.g., exploiting predator-prey
dynamics), genetic (e.g., new seed varieties), managerial (e.g., crop
rotation), and informational (e.g., monitoring pests and weather).
Different classification systems serve different purposes and there is
no clear hierarchy.
Most innovations start from a concept, and the innovation pro-
cess translates the concept into something that can be implemented,
which can be a product, practice, or institution. Once this outcome
is established, then we have supply chains that commercialize it. An
example of a product innovation is the personal computer. Steve
Jobs and Steve Wozniak thought about a user-friendly personal
computer.4 It resulted in the Apple I and II products. The supply
chain of these products evolved from selling through a few stores
(Apple I) to developing a national distribution system (Apple II).
An example of a process innovation is pre-packaged fresh salads. It
was originated by James R. Lugg, an agricultural scientist. He led
a research team that, by mixing some gases appropriately, created
a method to preserve salad greens for up to two weeks. They
obtained a patent and commercialized it through a company called
Fresh Express. In both cases, the innovations were embodied in
products, and therefore were carried by the private sector. On the
other hand, the notion of integrated pest management (IPM) was
introduced by UC Berkeley entomologist A.E. Michelbacher in 1939
and was further refined by researchers at the university in the late
1950s. It emphasized the selection of appropriate pest treatment
based on monitoring of pest population. It was a disembodied
management practice and, at least in the beginning, it was promoted,
taught, and implemented by the public sector through university and

4
See http://www.computerhistory.org/atchm/steve-jobs/.
 Agricultural Supply Chain 269

Research

Discovery

Patenting

Development

Production

Marketing

Adoption

Figure 3. The innovation process.

government extension services. Since it has been proven profitable

and its implementation requires specific skills, a new private sector
of professionals and firms has emerged providing IPM services.
The transition of innovation from a concept to an implementable
outcome (product or service) may follow multiple paths. Most
innovations in the past were introduced by practitioners who refined
them through trial and error. The wheel and the use of fire for cooking
are two examples of this path. Even now many important innovations
arise from practitioners. But in the past century, academic research
has been the source of many important innovations, especially in
chemistry, biology, medicine, engineering, and computer science.
The transition from an idea to a product is frequently called the
“Innovation Process”, and it is depicted in Figure 3. In the case of,
say, a medical innovation, researchers at the university come up with
a discovery based on their research. The discovery leads to a new
concept, an idea to solve a problem. The concept is then experi-
mented with in the lab. After the concept is established, the scientist
may seek a patent and publish the results. But to be ready for
commercialization, the innovation must go through the development
process, which includes upscaling and testing (both to make sure the
product performs well and meets regulatory requirements). Once the
product is ready, then commercialization starts with production and
marketing that, hopefully, leads to consumer adoption.
270 Agricultural Economics and Policy

This schematic analysis is far from perfect. There are a lot of

feedback mechanisms within the system. The development process
may discover bugs and flaws that prompt new research. The produc-
tion process is gradual and responds both to consumer feedback and
may require new developments. But this overall framework is useful
for analysis of the institutional elements associated with innovation,
which we will emphasize later. For example, while research may occur
at the university, development may occur at start-ups and production
at major companies. The transfer of technology from universities
to the private sector is a major policy challenge. New institutions,
like offices of technology transfer at universities, have emerged to
facilitate this transfer. As we will discuss later, university researchers
are also part of start-ups and the interaction of universities with
industries to form the education–industry complex is a major factor
that explains the strengths of California versus other regions.
Although useful, this schema ignores the basic issue that we
emphasize, which is the creation and modification of supply chains.
Supply chains are the organizational arrangements responsible for
the creation and movement of a product from its raw materials to
final customers. Once an innovation is working and generates new
products or services, then its introduction to the market requires
establishing or modifying supply chains. For example, when Apple
introduced iPhone, it needed to develop new markets for its final
product and for its components. It also needed to decide on how to
distribute and market iPhones. For instance, Apple needed to decide
what and how much to produce in-house and where to source inputs.
Its decisions were constrained by the preferences of consumers and
by the availability and capability of suppliers. The design of supply
chains to implement this innovation is an economic problem that we
will investigate in this chapter.

3. Modeling Supply Chains

In traditional agricultural systems, most of the production was done

on a farm. A farmer would have grown food to feed her family and
their livestock, harvested and processed the grain, fruit, vegetables,
 Agricultural Supply Chain 271

and meat, and then sold them. With innovations came specialization
and the increase in activities occurring beyond the farm gate. Today
a farm unit that specializes in growing crops is likely to obtain inputs
such as seeds, fertilizers, pesticides, and irrigation water from input
suppliers and then sell its crops to a packing house, which processes
and sells the product to retailers for purchase by end consumers.
Supply chain covers the various stages that the final product goes
through.
Every product we consume — our car, phone, movies, the water
we drink, and medicine — has a supply chain. Every supply chain
has several segments, and each segment often has subsegments.
Understanding supply chains and how they work is challenging, but
crucial to a good analysis of the economy and its operation. As we
mentioned earlier, supply chains change with new innovations and
policies. New modes of transportation (e.g., train), discoveries (e.g.,
relativity, DNA), and political events lead to changes in what is
produced and where, and the various linkages between producers
and consumers. To understand the complexity of supply chains we
must simplify how we represent them for analytical purposes.

3.1. Two models of supply chains

We will start with a simple two-segment supply chain for producing a

processed agricultural product. The first segment of this supply chain
is producing an agricultural product, called feedstock, and the second
segment is processing the final product. You may think of fields that
produce sugar cane and a refinery that processes them to produce
sugar or ethanol; orchards that grow fruits and a packinghouse where
the fruits are sorted, stored, and prepared for shipment; and cows
that produce milk and dairy and farms that process and ship them as
various products. Figure 4 presents a schema of such a supply chain.
Note that the supply chain does not include the final consumers,
who represent the demand for the final product. The company
that owns the processing facility may be vertically integrated and
produce the feedstock as well, it may purchase the feedstock from
independent farmers, for example, through contract or via a market,
272 Agricultural Economics and Policy

Production of In house
feedstocks by production of
farmers feedstocks

Processing and
production of
final products

Final consumers

Figure 4. Two-segment supply chain.

or it may combine some in-house production of the feedstock with

some purchase from suppliers outside the farm.
The two-segment supply chain is useful to analyze the supply
chain choices associated with the implementation of new innovations.
When an entrepreneur comes up with an innovation, say to produce
flowers in Kenya for the European market, they need to have a source
for the flowers and a processing facility that will collect, sort, and ship
them. They also have to decide the parameters for processing and
how to split production in-house versus purchasing from others. If
the entrepreneur aims to maximize profits and has limited credit and
other resources, the choice of the capacity and distribution of effort
between internal and external sources is a constrained optimization
problem. Establishing a flower exporting facility is a long-term
investment requiring multiple years to pay off with uncertainty
about future conditions. The entrepreneur attempts to maximize the
expected discounted net present value of profits given credit or other
resource constraints, as well as technologies they face.
We will solve these types of problems in the following sections,
but note that the expected profits in each period include the
revenue from sales of the final product (depending on the quantity
produced and the demand for the product) minus the expenses
(cost of processing, in-house production, and feedstock purchases).
 Agricultural Supply Chain 273

We need to distinguish between the fixed costs of investing in

facilities and permanent equipment versus the variable costs of
production. The limited capacity to borrow and supervise may force
the entrepreneur to rely on either feedstock produced by others, or
a smaller processing facility with a significant in-house production
facility. When we look at the decision problem of the entrepreneur
as one occurring over time, we need to recognize that because
of technological change and learning, the planned activities often
change. Furthermore, the entrepreneur may have market power in
the final product or feedstock market that could affect the prices
they receive for output or the amount paid for feedstock. These
considerations will affect the choices made for final output and the
allocation between the feedstock purchased from outside suppliers
and that produced in-house.
The decision of the entrepreneur also depends on government
policies. For example, the initial investment may increase in the
presence of programs that provide cheap credit to infrastructure
investment in developing countries. In other situations, a foreign
entrepreneur may have limited capacity to control land for feedstock
production, which may limit the capacity to produce in-house and
thus increase reliance on external providers.
While the two-segment supply chain is easy to demonstrate, a
three-segment system (Figure 5) might be more appropriate for agri-
culture. The three-segment supply chain has been used generically
in economic analysis (Reardon and Timmer, 2012) where there is a
distinction between the upstream (farms), midstream (processing),
and downstream (wholesalers and retailers). For example, Tyson
Foods, a large livestock processor, relies mostly on contractors to
produce its feedstock (poultry, chicken) but frequently supplies them
with inputs like grains and genetic materials and provides guidance
for production. In other cases, farm cooperatives may collectively
purchase the input and sell the output of the farmers. The three-
segment supply chain has also been applied to natural resource man-
agement. For example, Chakravorty et al. (2009) depicted the supply
chain of water in agriculture with the water source (upstream),
aqueduct (midstream), and distribution (downstream).
274 Agricultural Economics and Policy

Farm
input
suppliers

In house External
Policies feedstock feedstock Technology
production production

Processing
and
production
of
final
product

Figure 5. Three-segment supply chain.

Let us return to the example of Tyson Foods because it can

be used to illustrate how innovations lead to the redesign of supply
chains. According to Schaefer (2014), the company started in the
1930s when the founders realized that there was a large demand for
poultry in New York that could be supplied by farmers in Arkansas.
They thus started as a trucking company, buying poultry from farms
in Arkansas and selling it in New York (a two-segment supply chain).
Then they had another innovative idea to take advantage of their
trucks and economies of scale (regarding purchasing) by providing
their producers with feed, thus converting to a three-segment supply
chain. Later, they realized that the new genetics improved the quality
of chickens, so they decided to invest in R&D and start providing
poultry producers with genetic materials. This meant that they
added another element to the first segment of the three-segment
supply chain, which is producing genetic material.
 Agricultural Supply Chain 275

Further, it may be more useful, in many cases, to consider a four-

segment supply chain, which includes farm inputs, farms, processing,
and distribution. Each segment of the system can be broken down
into subsegments, which means the detailed supply chain is much
more complex and has many segments. For example, the input
supply segment may include many elements — fertilizer, irrigation,
and machinery. The supply chain of fertilizer may include feedstock
(manure), processing to refined product (fertilizer), and distribution
(shipping). This suggests that supply chains are nested systems, and
the degree of detail that we consider depends on the problem at hand.
Our analysis has thus far deemphasized one aspect of the supply
chain: marketing. Marketing is part of the distribution, but in
many cases, there is a marketing chain nested inside the supply
chain. A finished product may be purchased by wholesalers who
then sell it to retailers, who in turn sell it to the general public.
Companies may have their own salespeople and may use different
strategies to promote sales at different stages of the supply chain.
In some cases, they may go to trade shows and then establish
contacts with potential buyers. There are many marketing tools
like advertisements, demonstrations, samplings, and warranties. The
introduction of the internet also affects the way sales are promoted.
Some of these aspects will be discussed further in this chapter.

3.2. Innovation-induced supply chains

One of the major concepts that we emphasize in this chapter is

innovation-induced supply chains. New innovations are likely to spur
the creation of supply chains to implement these innovations. But
these supply chains are often partial elements of larger supply chains.
For example, the introduction of GMOs required a two-segment
supply chain, involving the generation and insertion of traits into
seeds, which then entered an existing supply chain. Monsanto had
facilities to conduct research to identify, refine, and test traits. They
decided to buy some seed companies to sell GMO seeds directly
to farmers. Then, they sold other seed companies, like Pioneer, the
276 Agricultural Economics and Policy

right to insert the patented traits in their own seeds. Corn ethanol
is another example of an innovation that required a three-segment
supply chain. The first segment is the production of feedstock (corn)
on farms; the second is processing that produces multiple outputs,
ethanol and other byproducts (such as dried distillers’ grain (DDG)
used as animal feed); and the third segment is the distribution of
some of the outputs, such as ethanol and DDG.
First, let us consider an innovation that leads to further dif-
ferentiation of a product. For example, a new product trait that
affects quality. Returning to Tyson Foods, around 1990 they decided
to, instead of just selling whole chickens, extend their processing to
provide a wide variety of products (whole chickens, chicken parts, and
even pre-cooked and processed chicken products). In this case, the
production system has to be expanded and become more complex.
A distribution section with its own segments can be added and
divided to accommodate the range of subproducts. Similarly, after
introducing pre-packaged salads, Fresh Express, driven by “relentless
innovation,” introduced packages with salad dressing of different
flavors. This again adds complexity to supply chains and the design
of the supply chain must determine how to allocate production to dif-
ferent product features based on demand, relying on hedonic pricing.
In the case of process innovation that can be adopted by different
products, the supply chain design must recognize the sequence
of products and regions in which technology will be introduced.
A process innovation, in many cases, does not fit immediately into an
existing production system, and thus the production system needs
to be modified to accommodate the new process innovation. The
supply chain design must consider that the suppliers, in many cases,
assist the end consumer in adapting the process to their needs. When
IBM introduced mainframe computers, a product allowing users to
manage various computations, they also provided sales engineers who
helped to customize the design of the system to the buyers’ needs
and provided training to their staff. In the case of drip irrigation,
especially when it was introduced, drip manufacturers were providing
drip system designs to adapt to the specific conditions of the
 Agricultural Supply Chain 277

buyers. Furthermore, the introduction of drip irrigation to some crops

required changing agronomical practices. In California, extension
specialists modified the production of tomatoes for processing to
allow for the use of drip irrigation (Taylor et al., 2014). In these
cases, the supply chain may also have an adaptation segment with
contributions from agencies like Cooperative Extension.
The design of the supply chain is also affected by changes in
processing or shipping technologies. The introduction of the railroad
allowed for the expansion of the grain supply chain so that grains
could be grown in the Midwest and shipped to the coast. The
introduction of cold storage in transport allowed for the expansion
of processed foods across regions and borders. For example, most
seafood consumed in the U.S. comes from Asia (Belton et al.,
2015). The flower industry grew and became less localized with
the introduction of cold storage in air freight. It allowed regions
that have advantages in the production of flowers, like California,
Florida, Israel, and Kenya, to become major producers of flowers
internationally.
Similarly, policy affects supply chains. Introduction of policies
that allowed breweries to sell beer on-site led to the proliferation
of microbreweries in the US (Swinnen, 2011). Improved credit
conditions, such as reduced interest rates through monetary policy,
may lead to the introduction of new supply chains to implement
innovations that become profitable, or increase the output volume
of supply chains by reducing the cost of operation. The successful
adoption of drip irrigation in California relative to other regions in
developing countries can be explained by government investment in
research and extension services that collaborate with manufacturers
to adapt technologies to local conditions (Taylor et al., 2014).

3.3. Symbiotic supply chains

Zilberman et al. (2022) emphasize the symbiotic relationship between

the innovation supply chain and product supply chain. The former
focuses on institutions undertaking innovation as a research and
278 Agricultural Economics and Policy

Innovation supply chain Product supply chain

Predicting
Research Designing
Development Upscaling demand Implementation
discovery supply chain Consumers
adoption

Feedback between Consumers

supply chains feedback

Technology

Policy

Figure 6. Symbiotic supply chains.

development process that leads to new products; and the latter

centers on companies starting with innovative products and mar-
keting them. Zilberman et al. (2022) combine the two supply chains,
analyzing a sequence of actions from labs to consumers. They argue
that the two supply chains affect each other and that, concurrently,
they are influenced by economic and policy considerations.
Figure 6 presents the symbiotic relationship between the two
supply chains, through which the transition from innovative ideas to
final marketed products occurs. First, in the innovation supply chain,
basic concepts are refined for commercial uses via developing, testing,
and upscaling. Second, the product supply chain produces and
markets the product. On the one hand, the output of the innovation
supply chain, determining the characteristics of the product, will
affect the formation of the product supply chain. On the other
hand, feedback from each segment of the product supply chain can
be incorporated in the innovation supply chain to further improve
the product or to invent new products. Moreover, the two supply
chains are influenced by public policies on R&D, education, trade,
and intellectual property rights protection, among many others.
So far, we have illustrated how supply chains can be presented as
a collection of activities that follow a logical sequence to produce
final outputs from raw materials through intermediary outputs.
 Agricultural Supply Chain 279

The design of a supply chain can be presented as an optimization of

expected discounted profits subject to constraints. New innovations
may lead to the establishment of new or redesigned supply chains,
and similarly, government policies can affect the design of new and
existing supply chains. The next two sections will introduce tools
that allow the planner to design supply chains.

4. Designing Supply Chains — Static Analysis

As we argued, the introduction of new innovations is leading to
the creation and modification of supply chains. In this section, we
will present and solve the supply chain design issues mathematically.
Like many economic problems, supply chain design is a constrained
optimization problem. The design problem varies according to the
nature of the innovation and the constraints facing the entrepreneurs
who aim to implement the technology. For every design problem,
we need to identify the objective function, decision variables, and
constraints. In reality, every problem is unique, but we will attempt
to investigate the impact of one category of innovation presented
within a simple supply chain and analyze its implications. We will
aim to develop some general lessons, but there is room for future
research to analyze the impact of different types of innovation on
different supply chain structures.
Here we consider a supply chain for a product processing
agricultural feedstock. We design a two-segment supply chain that
corresponds to Figure 4 in the previous section. The innovation may
be a new product altogether, or a new location to produce a product.
The optimization problem that we will solve is similar to the one
presented by Du et al. (2016). We will start with a basic model, and
then expand it. The main choices that a firm or entrepreneur who
applies the innovation (referred to as the enterprise) are the total
quantity of production and the amount of feedstock that will be
produced in-house versus purchased through markets or contracts
with third parties. Obviously, this model is abstract and ignores
key variable inputs and equipment, but the role of a model is to
emphasize key issues.
280 Agricultural Economics and Policy

Let Xh denote the quantity of feedstock produced in-house and

Xm as that purchased from a third party. The total feedstock is
denoted by

X = Xh + Xm , (8)

where Xh ≥ 0 and Xm ≥ 0. Let the final output be denoted by Q,

and the production function is

Q = f (X) = f (Xh + Xm ). (9)

We assume that marginal productivity of feedstock, denoted by

MP(X) = ∂f (x)/∂X, is positive and non-increasing in X. The cost of
production includes cost of processing the feedstock, Cp (X), the cost
of producing the feedstock in-house, Ch (Xh ), and the expenditure
of purchasing feedstock from third parties, Em (Xm ). The cost of
producing the purchased feedstock from the third party is Cm (Xm ),
but we assume that these suppliers are competitive and will charge
the marginal cost of Xm per unit supplied. This marginal cost
is denoted by MCm (Xm ) = dCm (Xm )/dXm . So, the outlay of
expenditure on purchased feedstock is Em (Xm ) = Xm MC (Xm ). The
total cost of the firm that considers introducing the innovation is
therefore

T C = C p (X) + Ch (Xh ) + Xm MCm (Xm ). (10)

Now assume that the price of Q is P and the enterprise is facing

an inverse demand P = D −1 (Q). The enterprise’s revenue is thus
f (Xh + Xm )D −1 (Q). To simplify our analysis we will determine
the profit-maximizing Xh and Xm and will use (9) to determine
the optimal output. With the aforementioned notation, we write the
supply chain design optimization problem as

L = Max f (Xh +X m ) D −1 (f (Xh +X m )) − Cp (Xh +X m )
{Xh ,Xm }

− Ch (Xh ) − Xm MCm (Xm ) (11)

s.t. Xm ≥ 0 and Xh ≥ 0.
 Agricultural Supply Chain 281

For convenience, define enterprise revenue R as f (Xh +X m )

D −1 (f (Xh +X m )) and enterprise cost C as [Cp (Xh +X m ) +
Ch (Xh ) + Xm MCm (Xm )]. Thus Equation (11) is equal to the
maximization of profits subject to the non-negativity constraint.
First, assume that we have an interior solution. Using the chain rule,
the differentiation of the first-order condition with respect to Xh
yields
 
dL ∂f (X) −1 ∂D −1 (Q) ∂Cp (X) ∂Ch (Xh )
= D (Q) + Q − − = 0.
dXh ∂X ∂Q ∂X ∂Xh
(12)
We note the following:
∂f (X)
Marginal productivity of input: MP (X) = ;
∂X
∂D −1 (Q)
Marginal revenue of output: MR (Q) = D −1 (Q) + Q ;
∂Q
∂Cp (X)
Marginal cost of processing MCP (X) = ; and,
∂X
∂Ch (Xh )
Marginal cost of producing in-house MCh (Xh ) = .
∂Xh
Using these definitions, the first-order condition can be re-
written as

MR(Q)MP(X) = MCP (X) + MCh (Xh ). (13)

Condition (13) states that at the optimal level of Xh , the marginal

contribution of a unit of Xh to the enterprise revenue, which is
its marginal productivity (i.e., MP(X)) times the marginal revenue
produced by each unit of output (i.e., MR(Q)) is equal to the sum
of marginal cost of processing and producing one unit of feedstock
in house.
The first order condition of Xm is:
 
dL ∂f (X) −1 ∂D −1 (Q)
= D (Q) + Q
dXm ∂X ∂Q
∂Cp (X) ∂MCm (Xm )
− − MCm (Xm ) − Xm = 0. (14)
∂X ∂Xm
282 Agricultural Economics and Policy

Now recall that expenditure on purchased inputs is Em (Xm ). Let the

marginal expenditure on purchased inputs M E (Xm ) = MCm (Xm )+
∂MCm (Xm )
∂Xm Xm . With this interpretation, the first-order condition
(14) can be rearranged to yield

MR(Q)MP(X) = MCP (X) + ME (Xm ). (15)

Condition (15) states that at the optimal level of Xm , the marginal

contribution of a unit of Xm to the firm revenue is equal to the
sum of marginal cost of processing and the marginal expenditure
on purchased inputs. By comparing conditions (13) and (15) we note
that

MCh (Xh ) = ME (Xm ) ≥ MCm (Xm ). (16)

The inequality in expression (16) holds because ME(Xm ) −

m (Xm )
MCm (Xm ) = ∂MC∂X m
Xm ≥ 0. Expression (16) shows at the
optimal solution, the marginal cost of in-house production is equal
to the marginal expenditure of purchasing from others and may be
greater than the marginal cost of the third-party production. This
reflects the monopsonistic power of the enterprise, meaning that at
the optimal solution the enterprise may produce more in-house and
purchase less from external sources to reduce the price of purchased
inputs.
Furthermore, the enterprise is a monopoly in the output market
and a monopsony in the purchased input market. This means that it
will produce less output overall than would occur under competition,
and it will use fewer inputs overall. Much of the burden of the
reduction of inputs will be borne by the input suppliers, reflecting
market power in both markets. The reduction in output produced
and input purchased by the enterprise compared to a competitive
environment will result in extra profit because of higher prices and
lower purchased input cost. This extra profit is the compensation
for the entrepreneurial effort. From a traditional welfare economics
perspective, this outcome is inefficient, but entrepreneurs or firms
may not engage in implementing an innovation in the absence of this
extra profit.
 Agricultural Supply Chain 283

Thus far we assume an interior solution. But there may be cases

of corner solutions as well. For example, the internal cost of producing
inputs may be very high due to a lack of internal expertise in the
production of the input, or because the government may restrict
activities or access to resources (e.g., land and water) needed to
produce the feedstock. In this case, all the production will be carried
out externally (i.e., Xh = 0) and the optimal outcome will occur at
the Xm where

MR(Q)MP(Xm ) = MCP (Xm ) + ME(Xm ). (17)

In this case, the marginal revenue associated with increased pur-

chased input is equal to the marginal cost of processing it and
the marginal expenditure of purchasing it. The enterprise will be
a monopoly in the output market and a monopsony in the input
market. If production in-house was prevented due to regulations, the
production would be lower than if in-house production were allowed
and thus regulations would reduce overall welfare. Further, purchases
from other sources are higher when in-house production is banned,
thus outside suppliers may support such a ban even though it reduces
overall welfare.
The enterprise behaves in this case as a middleman, which is a
monopoly in its output market and a monopsony in its input market.
This is a very desirable position to be from a profitability perspective,
as the middleman maximizes profits from its activities in both
markets. But the middleman’s presence also results in the largest
reduction in overall welfare compared to other arrangements —
all result in higher output and purchased input use levels, lower
output price, and higher purchased input price. Apple’s situation
in the cellphone market largely resembles a middleman solution, and
indeed it is the richest company in the world at the time of writing
the book.
In some cases, purchasing from others may not be feasible (i.e.,
Xm = 0). This may occur when the innovation introduces both a
new feedstock and process in a new region where there are no local
farmers conducting such activities. For example, the introduction
of sugarcane for ethanol in regions of mostly abandoned land or
284 Agricultural Economics and Policy

extensively used rangeland (e.g., in Brazil). It may also occur with the
introduction of the production and processing of winter vegetables in
a desert region where part of the innovation is to support production
with a new source of water. In these cases, the optimal outcome will
occur at the Xh where

MR(Q)MP(Xh ) = MCP (Xh ) + MCh (Xh ). (18)

In this case, the enterprise has a monopoly only in the output

market. The overall production will be larger than if some feedstock
productions were feasible under similar cost structures by local
producers because the enterprise will use their monopsony power
to reduce the purchases of inputs from these suppliers.
It is useful to assess how sensitive the outcomes of the enterprise
behavior to changes in key parameters are. The extra gains from the
markets’ power associated with the innovation are greater when the
price elasticities of final output demand and purchased input supply
are smaller. Over time, these monopoly gains may not last, as other
firms may enter the markets associated with the innovation, which is
consistent with the Schumpeterian view that monopoly power tends
to vanish as new entrants are aiming to capture extra rents associated
with market power.
Frequently, the processing process is utilizing only part of the
feedstock, and the rest is gone to waste. In this case, the final output
is a function of “effective feedstock.” When the innovation has higher
feedstock use efficiency (higher percentage of feedstock utilized in
production) it will result in a greater quantity of the final output
and lower output price. It may reduce feedstock use (especially when
the gain in overall output is modest), but in some cases the increase
in input use efficiency will result in a large increase in output and
increase in feedstock use, increasing the price of purchased feedstock.
This is consistent with the analysis of Khanna and Zilberman (1997)
on the impact of the adoption of resource-conserving technologies.
When not all feedstock is utilized, uses for the residue are
introduced over time. These innovations expand the supply chain
because the processing of the feedstock is expanded, and the new
 Agricultural Supply Chain 285

Production of In house
feed stocks by production of
farmers feedstocks

Processing
and
production
of final
products

Consumers of Consumer of
output 1 output 2

Figure 7. Expanded supply chain.

activity generates new outputs. In this case, the supply chain is

producing two products with two different sets of consumers (see
Figure 7). A more detailed analysis should consider a third segment
of the supply chain (namely distribution of the two outputs). But
in this case, the enterprise must decide whether it is profitable to
add the additional processing of the feedstock to produce a second
output. If the new processing is added, then the revenue stream
from the feedstock is increasing, as well as the cost of processing.
Under plausible conditions, this extra demand will lead to a larger
volume of feedstock production. Two examples of such a situation
are the use of bagasse (the residue of sugarcane) to produce ethanol
and the use of dried distillers’ grain (the residue from corn ethanol
production) as animal feed. With these added options, optimal
feedstock production is likely to increase.
Two other considerations that are important to address are credit
and risk. The implementation of innovations requires investment —
the enterprise may finance it by relying on some combination of its
internal resources, issuing stock, receiving government support, or
seeking financial loans. The outcome of an investment is uncertain
and financial plans are, at best, a good estimate due to significant
future uncertainty. Furthermore, credit availability by banks may be
limited.
286 Agricultural Economics and Policy

4.1. Credit

Both the processing activities and the in-house production requires

an investment and are especially vulnerable to credit constraints.
Rigorous analysis of the credit constraints requires a more detailed
analysis, but we can adjust our model by modifying the conditions
of the model (11) by adding a constraint to yield a modified opti-
mization problem as follows:

L = Max f (Xh +X m ) D −1 (f (Xh +X m ))
{Xh ,Xm }

− Cp (Xh +X m ) − Ch (Xh ) − Xm MCm (Xm ) , (19)

s.t. βp C p (Xh +X m ) + βh C h (Xh ) ≤ g(W0 α (R − C)) and Xm ,

Xh ≥ 0,
where βp is the proportion of in-house costs associated with the
upfront investment in processing, βh is the proportion of in-house
costs associated with the upfront investment in input production, W0
is the net worth of the enterprise, and α is the credit coefficient which
translates this profitability potential to creditworthiness. Credit is
given based on the value of asset generated by the enterprise,
depending on its estimated profitability.
We assume that the credit constraint limits the upfront invest-
ment made by the enterprise and our formulation translates it to a
constraint based on cost.5 We will not solve this optimization here,
but adding these constraints may change the outcomes because the
extra cost of credit impacts the level of investment in processing
facilities and in-house production. One expects that companies
that are constrained by credit are likely to reduce their in-house
production and may utilize much of their credit to finance processing
facilities while relying on purchases from others for feedstock. One
reason why a major wine company like Gallo invested mostly in their
processing capacity was that they were not able to finance much of
the land needed to produce grapes (i.e., their feedstock).

5
We assume that external suppliers can obtain their own credit and therefore the
enterprise is not burdened by financing them.
 Agricultural Supply Chain 287

There are situations where external suppliers may not be able

to obtain their own credit, most often occurring in developing
countries where credit institutions and insurance schemes are limited.
In these situations, the enterprise may need to finance, at least
partially, external feedstock producers, which is likely to limit the
size of the feedstock processing but may ultimately increase in-house
production.

4.2. Risk

The analysis thus far assumed full certainty. In reality, implemen-

tation of innovation is subject to significant risk (Du et al., 2016).
The enterprise faces several types of risk. The four prominent forms
are processing cost risk, output price risk, external supply risk, and
in-house production risk. When refining technology is relatively new,
the cost of the refining facility is subject to random shocks. Each
form of risk presents its own set of considerations.
If the enterprise is risk-averse, and the marginal utility from the
profit is declining, higher processing cost risk may result in a lower
activity level. This suggests that as facilities become more reliable,
the production and processing of feedstock may increase. An increase
in output price risk will also tend to reduce the size of all activities.
The introduction of future markets or price insurance may increase
the likelihood of implementing innovation and increase the size of the
production or processing activities. An increase in the riskiness of
external suppliers may reduce overall activities and increase reliance
on in-house production. Similarly, an increase in the riskiness of
in-house production will reduce overall activities and increase the
reliance on external suppliers. In cases of storable feedstock (e.g., corn
for ethanol), one way to deal with supply risk is through storage. But
this option is not available to people who rely on refining sugarcane
for ethanol because the storability of sugarcane is low.

5. Designing Supply Chains — Dynamic Analysis

Implementing innovation is a lengthy process and the analysis of
the supply chain designed to implement this innovation is better to
be done within a dynamic framework. Such a framework needs to
288 Agricultural Economics and Policy

recognize the changes in the forces that shape the economics of the
demand for and cost of the outputs resulting from the innovation
and accommodate these changes in the design of the supply chain.
In this section, we will first discuss some dynamic forces affecting the
implementation of the technology and how to model them within a
decision-making framework. Then we will analyze the implications
of the dynamic consideration in the supply chain design.

5.1. Economic factor in dynamic analysis of supply

chain design

Consider the same enterprise we discussed in the last section

(developing a new product based on refining agricultural feedstock),
concentrating on the case where they purchase all feedstock from
others. In doing so, we will emphasize the impact of dynamics on
the processing facility. There are several key features that affect the
dynamic analysis in this context.
First is the stock of capital used in the processing facility. This
capital stock includes buildings and equipment, and its definition
may expand to include knowledge (human capital). This capital
stock is increased by investments and may decline by depreciation.
In mathematical representation, we will assume a planning horizon
of T + 1 years and the time indicator, t, ranges from t = 0 to
t = T . We assume that the enterprise makes an initial investment
denoted by I0 and the investment in year t is denoted by It . The
investments contribute to the accumulation of capital, and let Kt be
the capital stock at the beginning of period t. Capital may evolve
over time according to the equation of motion
Kt = Kt−1 (1 − δ) + It−1 , (20)
where δ is the depreciation rate. For instance, if δ = 0.05, then capital
declines in value by 5% each year and to maintain capital stock,
the enterprise must make an investment in capital each period of
It = δKt . The equation of motion implies that investment made in
period t − 1 will bear fruit starting at period t.
Second, the output of the processing process depends on the
feedstock input at time t, Xt , and the capital stock of the firm, Kt .
 Agricultural Supply Chain 289

Another argument of the production function is time, as producers

tend to improve productivity over time due to learning-by-doing. At
each period, starting from period 1, the processing facility is refining
Xt units of feedstock. The production function is

Qt = f (Xt , Kt , t). (21)

We assume that the marginal productivity of both capital and

feedstock in producing the final output in period t is nonnegative
∂f ∂f ∂2f ∂2f
and decreasing. Namely, ∂X ≥ 0, ∂K ≥ 0, ∂X 2 < 0, and ∂K 2 < 0. An

increase in capital leads to an increase in the marginal productivity of

∂2f
the feedstock (i.e., ∂X∂K > 0). Given other inputs, the productivity
of the processing is increasing at a decreasing rate over time (i.e.,
∂f ∂2f
∂t ≥ 0, ∂t2 ≤0).
Third, we assume that the processing cost at period t, Cp (Xt , Kt )
depends on feedstock and capital. The costs of processing decline
with the use of capital but tend to increase with the volume of
feedstock processed. We assume positive and increasing marginal
∂C ∂2C
processing costs of feedstock (i.e., ∂Xp > 0; ∂X 2p ≥0). More capital
∂C
reduces processing cost at a declining rate (i.e., ∂Kp ≤ 0 and
∂ 2 Cp
∂K 2
≥ 0). Increased capital reduces the marginal costs of processing
∂2C
feedstock (i.e., ∂K∂X p
≤ 0).
Fourth, we assume that marginal costs of the feedstock depend
on feedstock input and time, MCm (Xt , t). The marginal cost of
∂MC
processing is increasing with the volume of feedstock (i.e., ∂X p ≥ 0)
and the marginal cost declines over time due to technological change
(i.e., ∂MC
∂t
m
≤ 0), indicating that over time the expenditure on
purchased input may decline for a given level of feedstock.
Fifth, the willingness to pay for the final output is likely to grow
over time due to population growth, increase in income (if it is not an
inferior good), and learning-by-using on behalf of users that increases
the value of the technology to these users. Some factors, such as the
emergence of competing products and increased efficiency in the use
of the product, may reduce the demand over time. We assume that
inverse demand Pt = D −1 (Qt , t) is increasing with time and declining
with quantity.
290 Agricultural Economics and Policy

With these assumptions, we can write the dynamic optimization

problem of the enterprise. It must determine, at each period, the
quantity to produce, feedstock to purchase, and the amount of
investment to make. These decisions are subject to the equation
of motion of capital in (20), the production function in (21), and
the non-negativity of Xt , It , and Kt . We assume that the discount
rate is r and that the entrepreneur maximizes the total net present
value obtained from the enterprise. The optimization problem can
be written as

T
1  −1
L = Max t f (Xt , K t , t) D (f (Xt , K t , t))
{It ,X t } (1 + r)
t=1

− Cp (Xt, Kt ) − Xt MCm (Xt , t) − It − I0 , (22)
s.t.
Kt = K(t−1) (1 − δ) + It−1 for t = 0 . . . T − 1 (the equation of
motion of capital),
Xt , It , Kt ≥ 0 (non-negativity constraints),
K0 = 0 (no initial capital).
In our formulation of the optimization problem the temporal
profit of the enterprise at each period from 1 to T is equal to
the revenue minus processing costs, purchased input costs, and
investment costs. These profits are discounted to compute the net
present value, and then the initial investment is subtracted. As
problem in (22) is a quite standard dynamic optimization problem,
we forego solving it here but will present some features of the
solution.
First, the formula to determine the amount of feedstock used at
each period (Xt ) is identical to the level determined from the static
model. Namely, at the optimal level of Xt , the marginal contribution
of feedstock to firm revenue, MR(Qt )MPx (Xt ), is equal to the sum of
marginal contribution of feedstock to the processing cost, MCP x (Xt ),
and the marginal outlay on purchased inputs ME(Xt ). Note that
because now the production and cost functions have more arguments
than do their counterparts under the static scenario, we modified the
notation so that MPx (Xt ) is the marginal productivity of feedstock.
 Agricultural Supply Chain 291

Similarly, MCP x (Xt ) is the marginal processing cost with respect to

feedstock. We note that both productivity and marginal costs depend
on other variables (capital and time), but for simplicity they are not
included in the arguments.
MR(Qt )MPx (Xt ) = MCP x (Xt ) + ME(Xt ). (23)
Second, the temporal marginal benefits of capital at each period,
denoted by MBK (Kt ) is equal to its marginal contribution to the
firm’s revenue, MR(Qt )MPK (Kt ), plus the marginal reduction in
processing costs due to increased capital, − MCP K ( K t ).
MBK (Kt ) = MR(Qt )MPK (Kt ) − MCP K ( K t ). (24)
Third, the temporal marginal benefit of investment at any moment,
MBI (It ), is the net present value of marginal contribution of the
capital goods generated from this investment throughout the life of
the project, considering discounting and depreciation. In particular,

T
(1 − δ)j−t−1
MBI (It ) = MBK (Kj ) . (25)
j=t+1
(1 + r)j−t

Thus, the temporal marginal benefit of investment in period 0 will

start at period 1 and will continue till time T , but this benefit will
decline because of discounting and depreciation. The greater the r
and δ the smaller the future benefits of an investment. This analysis
suggests that the volume of investment in period t is such that the
temporal marginal benefit of the investment is equal to 1 monetary
unit, which is the temporal marginal cost of investment.

T
(1 − δ)j−t−1
MBI (It ) = MBK (Kj ) = 1. (26)
j=t+1 (1 + r)j−t

5.2. Implications of the dynamic considerations

Our analysis suggests that in every period the enterprise operates as

a middleman in the output and feedstock market. The extent that
it gains monopoly or monopsony rents depends on its market power
at each period. If over time more competitors enter — so that the
292 Agricultural Economics and Policy

innovation is losing its uniqueness — the above-normal profits of the

enterprise decline.
The volume of operation of the enterprise increases over time
when learning-by-doing increases the productivity of processing the
feedstock or the supply of feedstock is increasing and becomes
cheaper, or the demand for the final product is increasing. For
example, the scale of new forms of biofuel depends on (i) the extent
of increased productivity of conversion of feedstock to fuel, (ii) the
increase of the productivity of the feedstock sector, and (iii) the
increased demand either because of the increased price of fossil fuel
or increased compensation for greenhouse gas emission sequestration
(Khanna and Crago, 2012).
Since the enterprise starts with no capital, it must invest in the
early period to build capital stock. Smaller interest or depreciation
rates will increase the size of the initial investment. The investment
is likely to grow if the demand for the final product increases. Capital
goods that increase feedstock use efficiency are more valuable when
the demand for the final product is high or when feedstock costs
are high. If the interest rate and the depreciation are high and the
demand is expected to grow over time, the investment may be spread
over more years.
The dynamic analysis suggests that under plausible conditions,
the introduction of a supply chain to implement an innovation of a
new product or production in a new place may require a significant
initial investment to overcome the initial condition (K0 = 0). This is
consistent with the evidence in the literature on the need to overcome
threshold investments when establishing supply chains of foods, for
example, new supermarket facilities and distribution chains (Reardon
and Timmer, 2012). The optimal initial investment for implementing
innovation is larger, and the net present value is larger as well when
the investor faces a lower interest rate. So, foreign investors that
obtain their capital in developed countries with lower interest rates
may have an advantage in introducing new production systems in
developing countries. This is in addition to the added knowledge
they may have. That may be one of the reasons that foreign direct
investment played a crucial role in the development of the food sector
 Agricultural Supply Chain 293

in developing countries (Reardon et al., 2009). As the interest rate in

China and other developing countries has declined, their dependence
on finance from foreign and hedge funds has decreased.
Learning is another important feature of our analysis. In cases
with large potential gain from learning-by-doing, the enterprise may
not invest much in the early years, but instead, wait so that learning
reduces the cost of processing and feedstock. An enterprise may
implement innovation at a slower pace if the early demand for
the final product is low, but they expect the demand to grow
substantially over time. Of course, they can affect this growth
through marketing. If the learning increases, it may reduce the
volume of investment at late periods as the marginal gain from
capital declines. This is especially the case when demand for the
final product is not growing much.
The dynamic analysis thus far ignores credit and risk. The ability
to obtain credit increases as the performance and profitability of
the enterprise is more apparent. Thus, in the early years, credit
constraints may restrict the volume of the enterprise and limit its
ability to take advantage of its potential. Promising but resource-
limited enterprises that control an innovation may be taken over
by established companies or hedge funds with sufficient resources
to invest. This is demonstrated by the several innovative but small
agricultural biotechnology or seed firms that have been taken over
by major companies such as Monsanto and Pioneer.
Our dynamic analysis allows for learning by the enterprise. There
may be a significant degree of uncertainty about the processing
technology and the enterprise may make limited investment in
capacity aiming to master the technology better instead. Learning
reduces the production risk the enterprise faces. Especially when the
enterprise is risk averse, learning may lead to increased operation
volume over time. Note that learning may reveal flaws that can either
be corrected or may lead to a decline or closure of operation. Waiting
for the results of learning may lead to outcomes where only after
the kinks are gone may the large-scale investment in the technology
start. For example, franchising of food technology may occur only
after satisfactory operational procedures have been established. Ray
294 Agricultural Economics and Policy

Kroc, the founder of McDonald’s, spent a few years perfecting the

operation of the company in its two first stores before large-scale
franchising (Boas and Chain, 1976).
Our dynamic analysis did not consider the possibilities of in-
house production. However, there may be situations when external
suppliers are either unavailable or inefficient, and the enterprise
must then establish feedstock activities, either by owning them
outright or by financing and guiding contractors. In other cases,
if feedstock suppliers are not able to keep pace with the growth
of demand, the processor may either initiate its own production
operation or investment in expanding existing suppliers or seeking
out new suppliers.
Thus far we discussed the optimization problem and analyzed
the properties of the solution assuming that the investment is being
made. But a key condition for making the investment is that the
maximum total net present value of the optimization problem in
(22) is positive, and most often exceeds an established threshold. This
suggests that enterprises that face lower interest rates are more likely
to implement new innovations and establish the associated supply
chains. Innovations that result in outcomes that are facing higher
demand and technologies that have low rates of depreciation are more
likely to be implemented. Implementation is most likely to occur
when affordable and reliable sources of feedstock are available.
The internal rate of return (IRR) is used to compare a set of
investments. It is the value of r at which the solution to the net
present value optimization problem in (22) is equal to 0. There may
be several solutions to this problem. Therefore, to find a realistic
IRR, it is worthwhile to repeat the optimization problem for different
levels of r, starting from 0%, until an r that yields NPV = 0 is found.

6. Marketing
One of the major challenges of business is to market its products.
Marketing is the interaction between a business and its client to aim
to enhance goodwill and induce sales or income. Firms use marketing
tools (e.g., advertising, salespersons, money-back guarantees), and
 Agricultural Supply Chain 295

one can think about something parallel to a production function

(maybe a revenue or income function) that will be a function of
marketing tools. So, in essence, a firm may determine an optimal
combination of marketing tools in a profit-maximizing way, similar
to how a farmer uses inputs to maximize profits. A key to designing
a marketing strategy is to understand consumer behavior — one
element is adoption behavior, which is the pattern of decisions
associated with using a new product or service. Another important
aspect is that consumer choices are influenced by risks, including the
risk of reliability of the product as well as the fit of the product to
consumer needs, and, therefore, some marketing tools aim to reduce
these risks.
The literature on adoption can provide a lot of insights into the
design of marketing tools that enhance adoption. So, understand-
ing individual behaviors, heterogeneity, and dynamics of diffusion
processes is important in designing marketing tools. One cannot
underestimate the importance of understanding demographics and
recognizing the notion of “different strokes for different folks.” People
vary in their tastes, income, ability, and interest. Therefore, they
vary in their assessment of the usefulness and value of a product.
Marketing strategies determine the way in which a product or media
text is sold to a target audience, the techniques used to attract and
persuade consumers, and the process of planning and executing the
conception, pricing, promotion, and distribution of ideas, goods, and
services to satisfy customers.
The challenge of marketers is to target the right population in
their choice of location of an enterprise (or social network and media
for online activities), and the pricing and advertising strategies used.
While in Chapter 8 we spoke about the shift of the threshold of
adoption among different income levels, marketers think about the
shift of the frontier of adoption among different segments of the
population.
A key element associated with marketing is risk and uncertainty.
If consumers had perfect and complete information, they would make
the right choices by themselves. Marketing is used to address issues
of asymmetric information and other transaction costs. Buying a new
296 Agricultural Economics and Policy

product or service can be difficult. One must deal with risks, credit,
transportation, logistics, and product specifications. Marketers can
help in this regard. While basic economic models of behavior under
uncertainty deal mostly with price and quantity risks, marketing
deals with these risks and more. When a new product is introduced,
a potential buyer does not know what it does (performance), how
well it works (reliability), and if it will work for that buyer (fit).
Marketing thus emphasizes issues of performance risk, reliability risk,
and fit risk. Marketing also reduces transaction and transition costs.
The purchase of a product may require adjustments by the buyer
and the user. A new machine requires training and reassignment of
labor. Learning how to use a product takes time and is often costly,
at least initially. Additionally, finance for a new product may not be
easily available, and if the buyer needs to obtain credit by themselves
it takes time and effort (and sometimes can lead to buyer remorse).
Therefore, marketers aim to reduce transition costs, provide training,
assist with credit and system design, and, finally, offer sampling
coupons to reduce the cost of initial trials.

6.1. Marketing tools

From a consumer’s perspective, we can divide the adoption procedure

between being completely unaware of a product and eventually
being a consumer of it into five stages: awareness, information
search, evaluating alternatives, purchasing decision, and product
feedback. In the awareness stage, consumers become aware of the
product, which may prompt a further information search related to
the product (i.e., the information search stage). During the stage
of evaluating alternatives, consumers compare and evaluate the
product against potential substitutes, and then enter the purchasing
decision stage. Lastly, consumers might provide sellers some feedback
regarding their experience with the product.
In each stage, some marketing tools can be used to facilitate the
adoption of the product. Table 1 provides a list of marketing tools
and identifies their use and their place in the adoption process. In
what follows of this subsection, we will describe these marketing tools
in detail.
 Agricultural Supply Chain 297

Table 1. Some marketing tools under each adoption stage.

Adoption stages Marketing tools

Awareness Advertisements, demonstrations, salespeople, trade

shows
Information search Advertisement, demonstrations, salespeople, trade
shows
Evaluating alternatives Demonstrations, product training, salespeople, trade
shows
Purchasing decision Money-back guarantees, warranties, renting and
leasing, providing credit, salespeople
feedback Salespeople, trade shows

First, advertisements are the most widely used marketing tool,

meant to increase awareness and trigger sales. There are different
advertisement tools, and they evolve with technology. In the past,
people advertised their products by singing or shouting at markets.
They wrote and printed their advertisements on bulletin boards,
store walls, and other surfaces. Advertisements became a major
source of support to newspapers and then radio and TV. In
the 21st century, pop-ups in search engines and the internet are
major modes of advertisements. Advertisers may not know how
effective their advertisements are, but they will pay more for well-
targeted advertisements to direct their message to the appropriate
demographic group.
One type of advertisement is announcements that market a new
product or improvement. These are advertisements that announce a
new product is born, or an existing product is now improved. These
advertisements, on TV, in newspapers, or internet, are challenged to
provide several details in limited space. They introduce to the world
a new product or service with few details, hoping to raise consumers’
curiosity so that they will dig further. Apple’s introduction of the
Mac was a fantastic commercial at the 1984 Super Bowl that
raised buyers’ curiosity about it. Introducing buyers to a new type
of product or idea sometimes requires a long narrative or video
trailer, which leads to the development of infomercials. As behavioral
economics emphasizes, it is important to frame the presentation and
298 Agricultural Economics and Policy

give the consumer a reference point. Therefore, sometimes innovators

who introduce a new product may compare it (favorably) to existing
products. To add credibility, there may be a celebrity endorsement,
often connected to the product, such as Michael Jordan with Nike.
Second, there are reminders. Kahneman (2011) distinguished
between “slow” (deliberate) and “fast” (intuitive) thinking. Decision
on day-to-day low-cost items is fast. When a consumer desires a soft
drink, she will think “fast”: If the first drink that comes to her mind
is Coke, she will buy it rather than Pepsi. The role of advertisements,
in this case, is to remind you about a product so you think about it
rather than its competitors.
Third, there are timely advertisements. They provide critical,
timely information that can lead to sales. A discount might spark
interest among consumers who would otherwise find a product too
expensive. Clearance sales are especially effective, as it puts a time
limit on when the consumer can obtain the product.
Fourth, brands give a manufacturer identity and reduce uncer-
tainty about quality and performance. A key element of the brand is
familiarity as well as image and star appeal. Brands are important
when it comes to reliability risk. When purchasing a product that
is supposed to last for a long time, having a reliable brand is a
major advantage. Brands are also very important when it comes to
goods that are visible (e.g., clothes, cars). They indicate taste and
ability to pay. Brands are assets for a company and their value can
be estimated, in principle, by their contribution to increased demand
and the market power they generate.
Fit uncertainty is a major barrier to adoption and demonstrations
may address it. Demonstrations enable consumers to try the product
before purchasing, and thus reduce fit uncertainty. All the shirts in
a department store complete their primary purpose, but one should
try one on before they walk out with a shirt that is too small or
large (they also should make sure their significant other will like it).
We can model fit risk quite simply: the outcome of consuming a
product is a random variable. The benefit is B when the product
fits with probability q. When the product does not fit, and cannot
be used, the buyer has negative utility, L, representing the lost cost.
 Agricultural Supply Chain 299

The maximum the individual will be willing to pay for the product
with risky fit is P0 = Bq − L(1 − q). If the individual takes a
demonstration, for example, a test drive, assumed to provide full
information, they will now be willing to pay B, in case of fit, and
0 otherwise. So, the seller can charge more for the product with a
demonstration. He may sell less, but the customers will be happy,
enhancing long-run growth of the business; and if the demonstration
is not expensive, it is worthwhile for the seller. If the lost cost is
high, or the demonstration is time-consuming, it may be worthwhile
for retailers to pay people to take a demonstration because otherwise,
many people may not buy the product.
There are different types of demonstration — some include trying
on clothes, driving a car, and others include watching how a machine
works or trying a software program. Because of the complexity
of supply chains, final buyers may not buy from producers but
from retailers. Producers must market their products to wholesalers
and retailers. One frequent tool is trade shows, which can provide
opportunities for side-by-side demonstrations and comparisons. It
also facilitates interaction between sellers and buyers, as well as
between users and potential adopters.
Another tool to deal with fit risk is the money-back guarantee
(MBG). Frequently, in the case of experience goods, where the
assessment of value is difficult in advance, the buyer may be worried
about lack of fit and losses associated with it. Money-back guarantees
allow a buyer to return the product within a limited time frame
(and sometimes with a penalty). The net benefit from purchasing a
product priced at P under MBG is B − P with probability q and
−RC with probability 1 − q, where RC is the return cost including
any possible penalty. Thus, the expected value of the purchase is
(B − P ) q − RC (1 − q). The highest price one will be willing to pay
to a product purchased with MBG is Pmbg = [Bq − RC (1 − q)]/q
which is greater than Bq − L(1 − q), consumers’ largest willingness
to pay when there is no MBG.6

6
Here we assume that the loss from owning an unfit product is larger than the
return cost (i.e., L > RC).
300 Agricultural Economics and Policy

The gain from money-back guarantees is greater when the fit risk

(i.e., 1 − q) is smaller, the return cost is smaller, and the loss aversion
is larger. One of the main advantages of a big retail store, like The
Gap, is that the buyer can return their product at any location,
which lowers the return cost. Once the product is returned, it might
be readded to the rack or sold to “seconds” stores (e.g., outlet stores).
MBGs are not appropriate for every product. The offering of
MBG may lead to a moral hazard, where people will buy a product
they need for one event (e.g., a wedding gown), use it, and then
return it. So MBG is appropriate with products that are used for a
long time and require some experience for assessing their value and
gaining from their use.
In addition to its role in reducing fit risk, MBG has been seen
as a mechanism to signal quality. If the suppliers are confident that
their products are good and that consumers will keep their products,
they are more likely to offer the return option. This role is especially
important in industries where most firms do not offer MBG, or when
there are differences in the length of the return period. MBG has been
used in agriculture for a long time. McCormick, who introduced the
first reapers, used MBG to promote his product.
Demonstration and MBG serve a similar purpose. Demonstration
provides presale tests and MBG allows reversing a decision. Some-
times they will be substitutes (if you have a good demonstration,
you do not need an MBG), so a firm can decide which to use. In
other cases, if the demonstration cannot remove all the fit risks, then
they are complementary. The demonstration can screen and eliminate
potential buyers that the product clearly does not fit and the MBG
provides an option protecting buyers against misfits.
In cases of divisible products (e.g., seeds), providing free samples
is a good way to reduce fit risk. When consumers are not familiar with
a product or have a negative prejudice against it, a free sample may
allow the testing that will lead to purchase. In some way sampling
is a special case of a demonstration. Sampling, like a demonstration,
may misfire and eliminate a potential buyer who had a bad sample
and does not buy the product. However, overall, it is worthwhile to
keep satisfied customers.
 Agricultural Supply Chain 301

While MBG and demonstrations are mechanisms to address

mostly fit risks, warranties are crucial in addressing performance
risks. This is true especially for durable equipment (e.g., combines)
which must be available during critical time periods. Warranty
programs not only repair the products, but provide substitute
products during the repair period. A warranty is a form of insurance
and is also a signal of quality. The quality of the warranty and the
MBG are important contributors to establishing brand quality.
There are other means to deal with fit risk. For instance,
consumers may lease or rent a product before buying it. Moreover,
secondhand markets reduce the need for MBGs. So, if someone finds
that the product does not fit their needs then they may resell it,
either in a market or online marketplace, like e-Bay, which reduces
transaction costs even more. But prices in secondhand markets
are significantly lower than the price of new products; because of
asymmetric information, the buyer does not know the quality of
the product. Thus, even in secondhand markets, demonstration and
MBG could still be used to fetch a better price.
Rental and leasing have other purposes as well, especially their
ability to address insufficient scale or credit. If the per period cost
of owning the product, say, a combine, is denoted by I, and the
gain per acre is π, then only farms of size L ≥ I/π should buy the
product from a profitability perspective, while smaller firms can rent
it, incurring a much lower per-period cost. In cases of durable goods,
firms with insufficient credit may lease them. Leasing may enable
a firm to overcome concerns about obsolescence, in that they avoid
getting stuck with an older model when the new, improved one is
available. However, buyers benefit from outright owning a product
because they will be able to make the adjustments they see fit.
Marketing tools aim to reduce barriers to sales. Credit availability
is one constraint and sellers frequently provide, or help attain,
credit. Sellers of machinery and equipment provide credit to reduce
transaction costs, and reduce the likelihood of buyer’s remorse. One
popular mechanism car dealers provide to reduce transaction costs
is to purchase the older cars of potential new car buyers. People
who do not want to take the time to sell their own car can take
302 Agricultural Economics and Policy

a loss and use the older car to finance the purchase of a new
car. Furthermore, car dealers establish mechanic shops to provide
customers with maintenance and repair.
Another important mechanism to reduce transaction costs is
product training. Adopting new products requires changes in busi-
ness practices as well as modification of existing arrangements along
the supply chain. When a company purchases software, employees
need to be trained on how to use, maintain, and repair it. Part
of the warranty is addressing major repairs, but when it comes to
daily operations, the sellers provide training. One of the key elements
of franchising is that the franchisor trains the franchisee and their
employees (e.g., McDonald’s Hamburger University).
Salespeople are important for marketing in many circumstances.
They are expensive but much more versatile. They are crucial to
the wholesalers because they can approach potential retailers as
well as build the relationships and the capabilities that will enable
them to market the products to the public. Salespeople approach
potential buyers, who are likely to be opinion makers, early in the
diffusion process to introduce them to the product. They provide
demonstration product support and assist with credit and early use.
When it comes to complex products, like irrigation systems
and computers, companies may employ sales engineers who design
the product to meet buyers’ needs. These engineers serve as both
salespeople and trainers. In some cases, potential buyers put a bid for
a new product or service and the salespeople are the ones involved
in preparing the bid and executing much of the contract. As the
product gets cheaper and more widespread, the role of the salesperson
declines. They were much more significant during the mainframe
computer era than now during the personal computer era.
Price is perhaps the most important marketing tool. A firm can
affect the speed of diffusion by modifying prices. Prices serve as
signals of values and can be modified in response to existing and
expected market conditions and consumer choices. Prices can be
manipulated to allow for the learning of consumer preferences. For
example, different price schemes may be used in different but similar
markets to experiment with pricing schemes for products of different
 Agricultural Supply Chain 303

quality levels and different levels of service. Prices change over time
and across locations. Early in the adoption process, a company
may provide discounts to induce early adopters. In some cases, it
may provide discounts to adopters who increase its value through
their association with the product. For example, a car company
may sell cars to Google or Apple at a discount because of the
association (Ford car with a Google logo on it may be perceived as an
implicit endorsement). Similarly, car companies may provide volume
discounts to finalize a sale because large sales have lower transaction
costs. Frequently, buyers may collaborate and form cooperatives to
purchase in bulk. Companies may have fixed prices on relatively low-
cost items sold on the floor, but when it comes to higher-priced items
(cars or TVs), salespeople often offer a discount, increasing the value
of the transaction from the buyer’s perspective.

6.2. Choice of marketing tools

The choice of marketing tools is an economic choice like the choice

of other inputs. The firm has an expected revenue function that
is equal to sales times price. It also has costs, which include the
cost of purchasing the products and of marketing. Revenues are
a function of advertisements, demonstrations, and other marketing
tools. For example, advertisements may increase the number of
people attracted to a product and their willingness to pay for the
product. If there is uncertainty regarding fit, demonstrations may
reduce this uncertainty. It may reduce the number of potential
buyers, but the willingness to pay among the buyers who remain
is higher. MBG can reduce further the fit risk and thus increase
willingness to pay for the product. Having a good brand and product
support will increase the willingness to pay for the product. A firm
needs to have an estimate of how each factor affects revenue.
With this information, the firm seeks to maximize expected
revenue minus the cost of the marketing tool that it uses (e.g.,
demonstration, advertisement, money-back guarantee). The optimal
level of advertisement is where the expected marginal revenue is
equal to the marginal cost of advertisement. Similarly, the optimal
304 Agricultural Economics and Policy

level of demonstration is where the marginal revenue of providing

demonstrations is equal to the marginal cost of demonstration.
To optimize the use of marketing tools, firms must be able to
quantify the performance of the marketing tools they use. This is a
major challenge. First, they need to track the impact of different
marketing tools, which is difficult. For example, firms have little
knowledge of the effectiveness of their advertisements. Subsequently,
they may conduct surveys of consumers’ awareness of their campaign,
but still, these measurement tools are imperfect. Furthermore, they
must respond to the competition and sometimes they must protect
their product from “attacks” by competitors. They have to be
aware of alternative products, and the strategies these companies
employ. When Apple introduces a new product or service, Samsung,
LG, and others need to respond. So, if a competitor offers a new
warranty program, each competitor must decide how to respond.
Managing marketing strategy is an ongoing process that relies on
basic economic principles.

7. Conclusion

When conducting an economic analysis of a product, conventional

microeconomics often focuses on the supply and demand for this
single product and ignores the supply chain that makes the product
available in the first place. This chapter has expanded this view
by emphasizing on the entire supply chain of a product. Due to
the special role of innovations in a supply chain, we stressed the
intertwined relationship between innovations and supply chain. Two-,
three-, and four-segment supply chains have been discussed in this
chapter. Moreover, it discussed supply chain designs from both static
and dynamic perspectives, employing the approach of optimization
under constraints. Finally, the chapter carefully described several
marketing tools that can facilitate consumers’ adoption of a new
product.
Supply chain designs have attracted increasing attention from
agricultural economists since the outbreak of the COVID-19 pan-
demic which substantially disrupted the food supply chain around the
 Agricultural Supply Chain 305

world. Designing robust and resilient food supply chains has critical
implications for farm income, food safety and security, as well as
public health (Taylor et al., 2020). Given the scant attention that
conventional microeconomics has paid to supply chains as we noted
above, economic analysis on supply chain designs and evaluation may
prove to be a fertile area for future research, in both theoretical and
empirical realms.

References
Becker, G.S. 1965. A Theory of the Allocation of Time. The Economic Journal,
493–517.
Belton, B., A. Hein, K. Htoo, L.S. Kham, U. Nischan, T. Reardon, and
D. Boughton. 2015. Aquaculture in Transition: Value Chain Transformation,
Fish and Food Security in Myanmar (No. 230981). Michigan State University,
Department of Agricultural, Food, and Resource Economics.
Boas, M. and S. Chain. 1976. Big Mac: The Unauthorized Story of McDonald’s.
E.P. Dutton & Co., Boston.
Caswell, M.F. and D. Zilberman. 1986. The Effects of Well Depth and Land
Quality on the Choice of Irrigation Technology. American Journal of
Agricultural Economics 68(4): 798–811.
Chakravorty, U., E. Hochman, C. Umetsu, and D. Zilberman. (2009). Water
Allocation under Distribution Losses: Comparing Alternative Institutions.
Journal of Economic Dynamics and Control 33(2): 463–476.
Du, X., L. Lu, T. Reardon, and D. Zilberman. 2016. The Economics of
Agricultural Supply Chain Design: A Portfolio Selection Approach. American
Journal of Agricultural Economics 98(5): 1377–1388.
Kahneman, D. 2011. Thinking, Fast and Slow. Macmillan, London.
Khanna, M. and C.L. Crago. 2012. Measuring Indirect Land Use Change with
Biofuels: Implications for Policy. Annual Review of Resource Economics 4(1):
161–184.
Khanna M. and D. Zilberman. 1997. Incentives, Precision Technology and
Environmental Protection. Ecological Economics 23(1): 25–43.
Lancaster, K.J. 1966. A New Approach to Consumer Theory. The Journal of
Political Economy 74(2): 132–157.
Levenstein, H.A. 2003. Revolution at the Table: The Transformation of the
American Diet (Vol. 7). University of California Press.
Reardon, T., C.B. Barrett, J.A. Berdegué, and J.F.M. Swinnen. 2009. Agrifood
Industry Transformation and Small Farmers in Developing Countries. World
Development 37(11): 1717–1727.
Reardon, T. and C.P. Timmer. 2012. The Economics of the Food System
Revolution. Annual Review of Resource Economics 4(1): 225–264.
306 Agricultural Economics and Policy

Rosen, S. 1974. Hedonic Prices and Implicit Markets: Product Differentiation in

Pure Competition. Journal of Political Economy 82(1): 34–55.
Schaefer, K.A. 2014. The Meat Racket: The Secret Takeover of America’s Food
Business. American Journal of Agricultural Economics 96(5): 1507–1508.
Schumpeter, J.A. 1934. The Theory of Economic Development: An Inquiry
into Profits, Capital, Credit, Interest, and the Business Cycle (Vol. 55).
Transaction Publishers, New Jersey.
Sunding, D. and D. Zilberman. 2001. The Agricultural Innovation Process:
Research and Technology Adoption in a Changing Agricultural Sector.
Handbook of Agricultural Economics 1: 207–261.
Swinnen, J.F. (Ed.). 2011. The Economics of Beer. OUP, Oxford.
Taylor, C.A., C. Boulos, and D. Almond. 2020. Livestock Plants and COVID-19
Transmission. Proceedings of the National Academy of Sciences 117(50):
31706–31715.
Taylor, R., D. Parker, and D. Zilberman. (2014). Contribution of University of
California Cooperative Extension to drip irrigation. ARE Update 18(2): 5–8.
Zilberman, D., T. Reardon, J. Silver, and A. Heiman. 2022. From the Lab to
the Consumer: Innovation, Supply Chain, and Adoption with Applications
to Natural Resources. Proceedings of National Academy of Sciences 119(23):
e2115880119.
 Chapter 10

Health Risks, Food Security,

and Food Safety

Food is a major contributor to health, and it is important to develop

food and agricultural policies with an emphasis on health. Many food
and agricultural policies aim to enhance consumers’ nutritional
status and address the challenges of food security, food safety, and
environmental health. In this chapter, we will develop a framework
to address human health risks associated with food. The framework
could help us understand challenges such as hunger, safe food, and
some of the environmental side effects of agriculture. Much of the
chapter will review some of the literature on food security and food
safety globally and in the United States, followed by a conclusion.

1. Analyzing Health Risks

While the economic literature on risk emphasizes income risks, a
larger body of literature addresses health risks. A key element in this
literature is the risk generation function (Lichtenberg and Zilberman,
1988; Zilberman and Jin, 2016), which determines the probability
that a member of the population will die, get sick, or will be in
undesirable health status (e.g., obesity) during a period. For example,

307
308 Agricultural Economics and Policy

a risk generation function may provide estimates of probabilities that

members of a certain population will have an undesirable nutritional
status (e.g., stunting) or are subject to the risk of morbidity or
mortality due to the food situation. Regarding food safety, we can
consider the probability of mortality due to poisoning or other food
quality issues. Regarding environmental health, we can consider
population decrease or even extinction probability to a species, be it
a bird or other wildlife, caused by activities of food or agriculture.
The risk generation function reflects multidisciplinary models,
and it is frequently used in epidemiological, toxicological, and
engineering studies. Health risk is a product of multiple parameters
that can be affected by individual and policy choices. Chances of
death due to malnutrition, for example, are a function of several
factors relating to an individual’s access to food (Zilberman and Jin,
2016). Each of the factors that contribute to risk may be a random
variable that represents variability: one aspect is heterogeneity, or
how individuals vary in their capacity to withstand food shortages;
another is randomness in weather; and a third concern is how a
policymaker acts given the uncertainty in food supply.
Our model is inspired by the risk assessment framework of
Wilson and Crouch (1987), and Robson et al. (2022). We assume
a population with I groups, where i = 1, . . . , I is a group indicator.
The size of the population of group i is denoted by Ni . Let Ri =
fi (X, Bi ) be the probability of a deleterious outcome (e.g., mortality
or morbidity due to a pandemic, lack of food, or poisoning) to a
member of group i during a specific period. It is affected by policy
variable X and behavioral choices of the group denoted by Bi . The
risk generation is an outcome of multiple processes, including the
occurrence of a shock (i.e., emergence of an epidemic), exposure, and
vulnerability. Policy responses may reduce the shock, for instance,
by spraying or other treatments to slow the spread of the epidemic,
by protective clothing and other mechanisms to reduce exposure,
or by medical treatment, be it vaccine or pre-infection treatment
to reduce vulnerability. Policymakers may need to determine the
optimal level of intervention to minimize the expected cost. Here we
 Health Risks, Food Security, and Food Safety 309

use pandemic as an example, where the policymakers must make a

choice between reducing mortality risk and reducing economic cost.
Let us denote the value of statistical life (VSL) lost by Vi , which
may vary by groups reflecting differences in age or other variables
(Viscusi and Aldy, 2003). This concept applies to small risks and
regional differences reflecting differences in endowment, but it does
not encompass arbitrary judgments on different values of human life.
There are several recent studies, Greenstone and Nigam (2020), for
example, that use differences in health expenditures and labor market
behavior to establish VSL as a function of age. Using our notations,
the optimal policy choice problem is
 I  

min Ni · fi (X, Bi ) · Vi + C(X) , (1)
X
i=1

where C(X) is the cost of policy X, and the marginal costs are
positive and increase with the magnitude of the policy intervention.
X, in our example, is a scalar, but it can also be a vector. The optimal
choice of X is reached when the expected value of marginal gain in
lives saved is equal to the marginal cost of the policies,


I
 
− Ni · fi (X, Bi ) · Vi = C (X), (2)
i=1

where primes represent derivatives.

The COVID-19 pandemic led to many studies that evaluated
policy intervention. For example, Kaplan et al. (2022) showed that
an economic shutdown that reduced the U.S. GDP by about 25%
was worthwhile given the number of lives that it was supposed to
save. Goldstein and Lee (2020) presented a similar study about the
economic cost of the pandemic under different assumptions. Since on
average age of death for individuals who died from COVID-19 was
about 11.7 years lower than the pre-pandemic life expectancy, an
alternative approach to assess policies is to minimize the sum of the
aggregate expected life years lost and the cost of policy intervention.
Instead of using the VSL, one can use the value of quality-adjusted
310 Agricultural Economics and Policy

life year, which was $150,000 in the U.S. in 2015 (Young-Xu et al.,
2017).
Zilberman and Jin (2016) analyze policies to reduce the risk
of death during a famine in a region. The risk generation func-
tion depends on the aggregate availability of food in the region,
accessibility (dependent on distribution mechanisms), and vulner-
ability. The policy interventions considered were (1) increasing the
amount of food shipped to the region, (2) distribution methods (by
parachuting it from the air or providing meals through government
or soup kitchens), and (3) medical intervention. In some situations
where gangs may control parachuted food, it may be worthwhile
to reduce the total supply of food and distribute it individually.
There may be conditions that require both food supply as well
as medical care. The World Food Programme developed an agile
system that adjusted to different conditions in providing emergency
food assistance (Cozzolino et al., 2012). Wu et al. (2021) suggested
that delaying the introduction of golden rice to the Philippines and
Bangladesh resulted in a significant loss of life, particularly among
children. Wesseler and Zilberman (2014) use the notion of VSL to
assess the cost of delaying the introduction of golden rice to India.
They show that, even with relatively low VSL, this delay may lead to
the loss of tens of thousands of lives and cannot be easily justified.
Wu and Khlangwiset (2010) estimate the health cost of aflatoxin
in grains and food in Africa in terms of life years lost and identify
the cost of reducing aflatoxin. They find that the health benefits of
reducing aflatoxin far exceed the associated costs.
Thus far, we have established a conceptual framework to assess
interventions that affect health risk. Below we discuss large bodies
of literature documenting the extent of food security and food safety
problems, and some policies that have been introduced to address
these issues.

2. Food Security
According to United States Agency for International Development,
food security means “having, at all times, both physical and economic
access to sufficient food to meet dietary needs for a productive
 Health Risks, Food Security, and Food Safety 311

and healthy life”.1 Lack of food security results in malnourishment

conditions that may include stunting2 , wasting (acute malnutrition
during a short period of time that may lead to chronic stunting),
and being underweight or overweight (United Nations Children’s
Fund (UNICEF) et al., 2021). Close to 150 million children under
the age of 5 suffered from stunting in 2020, mostly in developing
countries (Nomura et al., 2023). But these conditions are not limited
to children. Body Mass Index (BMI), which is weight divided by
height squared (km/m2 ), is an indicator of health and nutritional
situation. A person is obese if BMI is greater than 30, overweight if
BMI is between 25 and 30, in the normal weight category if BMI is
in the range of 18.5 and 25, and underweight if BMI is below 18.5,
which is frequently associated with stunting. The classification is the
same for men and women.
Bansal and Jin (2023) review the literature on the relationship
between obesity and health at the global level. They find that average
life expectancy increases with average BMI within a range and then
declines, indicating that both undernutrition and obesity reduce
average life expectancy. Since the 1960s, life expectancy has increased
by about two months per year, and women on average live four years
longer than men. Developed nations have higher obesity levels, which
tends to reduce average life expectancy, but that is compensated
by higher medical expenditure, indicating a substitution between
medical costs and nutritional status (Bansal and Zilberman, 2020).
According to the Food and Agriculture Organization (FAO)
of the United Nations, food insecurity refers to “lack[ing] regular
access to enough safe and nutritious food for normal growth and
development and an active and healthy life,” and severe food

1
Available at: https://www.usaid.gov/agriculture-and-food-security (accessed
March 9, 2023).
2
According to the World Health Organization (WHO), “stunting is the impaired
growth and development experience from poor nutrition, repeated infection,
and inadequate psychosocial stimulation. Children are defined as stunted if
their height-for-age is more than two standard deviations below the WHO
Child Growth Standards median.” Available at https://www.who.int/news/item/
19-11-2015-stunting-in-a-nutshell (accessed March 9, 2023).
312 Agricultural Economics and Policy

Figure 1. Food insecurity across the world.

Source: The graph is reproduced from Smith and Meade (2019), which is in the
public domain.

insecurity refers to when one has “run out of food and gone a day
or more without eating.”3 Globally, 26% of the population suffers
from food insecurity, and 11% from severe food insecurity. Sub-
Saharan Africa suffers from the worst rates of food insecurity, with
55% of its population food insecure. About 30% of the population of
Latin America and South Asia suffer from food insecurity. We even
observed more than 10% of food insecurity in Europe as well (see
Figure 1).
Barrett’s (2002) survey of the food security literature sug-
gests that early lines of work emphasized the role of food avail-
ability in enhancing food security, the importance of increased

3
Available at: https://www.fao.org/hunger/en/ (accessed January 27, 2024).
 Health Risks, Food Security, and Food Safety 313

food productivity (especially in developing countries), as well as

the important role of inventory, transportation, and supply chain
(Barrett, 2002). Because of the randomness of food supply, the
availability of reliable storage with low levels of food waste is essen-
tial. Lichtenberg and Zilberman (2002) presented a simple model of
gains from storage to overcome seasonality and variability. Storage
may require the use of pest control devices as well as temperature
control. It increases social welfare and reduces environmental costs
by reducing transportation costs. Inventory development requires
public intervention because of the welfare gain to consumers that may
not be fully captured by the industry. Williams and Wright (1991)
analyze some of the complexity of inventory policy. Development of
transportation systems enables countries to import food that they
cannot produce from locations with comparative advantage, and it
is crucial for intervention in cases of drought or crop failures (Bakker
et al., 2018).
With broader and more complex links between food origins
and destinations, we are witnessing a transformation in global
food supply chains. The food systems are affected by urbanization,
changes in diets, intensification of production systems, expansion of
intermediation, digitization, the transformation from commodities to
differentiated products, and extensive regulation. The transformation
may increase overall food availability but may lead to increased
vulnerability to shocks, including climate change and pandemics.
Thus, to assure supply availability, the resiliency of supply chain can
be expanded by diversification of sources of input and production
location, increased redundancy, and continuous innovation (Barrett
et al., 2022; Reardon and Timmer, 2014).
Another element that affects food security, according to Barrett
(2002), is access to food. Access to food can be enhanced by physical
means, investment in infrastructure, improved transportation, and
storage. However, food accessibility is also affected by institutions
and policies. Behrman (1997) emphasized the issue of intra-family
allocation of resources based on the household production model,
suggesting that different household members may have different
preferences and different commands over resources within the family,
314 Agricultural Economics and Policy

with women and children tending to have inferior access to food.

Heiman et al. (2001) introduced a framework for analyzing food con-
sumption within the family. By using data from Israel, they showed
that consumption differences are affected by cooking capacity, time
availability, and income, among other factors. Their analysis suggests
that food consumption outside the home is growing in importance
with urbanization and food support programs, such as lunch at
school. Heiman et al. (2019) emphasize the importance of cultural
consideration, especially religion: using data from Israel, they found
that the level of observance of religion’s rules affects households’
diet and food supply chains. Religion, thus, has an ambiguous effect,
either leading to cooperation that increases food availability or
constraints that reduce access to food.
A third element affecting food security is interaction with other
factors, including health, employment, education, and housing. Sen
(1982) emphasized the important impact of political economy factors
on food security. Populations under colonialism are more vulnerable
to severe famines because of a lack of political power. That was the
case with the big famines of Bengal and Ireland. Similarly, famines
are more likely under dictatorships, and thus introducing democracy,
where the masses are given political power, would increase food
security, as evidenced by India after liberalization. The pursuit of
the Green Revolution and the increased food security after it was
influenced by global awareness of food security risk may reflect the
political powers of the masses in the post-World War II era. The
literature reviewed by Barrett (2002) emphasizes the important link
between food insecurity and poverty, which is linked to health. Food
insecurity may harm health, and bad health reduces the capacity to
produce income and leads to poverty and food insecurity, leading to
a vicious cycle called the “poverty trap” (Barrett et al., 2016).
Education is another important factor affecting food security. It
affects individual income on the one hand, as well as their ability
to make sound nutritional and health choices on the other hand. In
the case of the COVID-19 pandemic, Alvi and Gupta (2020) found
that lack of access to school as part of social distancing strategies
affected access to school lunch programs and reduced food security.
 Health Risks, Food Security, and Food Safety 315

Laborde et al. (2020) documented the negative impact of the

pandemic on food security in developing countries, finding that loss
of income because of unemployment, relocation, and lack of assets
prevented the poor from buying food. While governments were able
to develop various mechanisms to overcome or adapt to supply chain
disruption during the pandemic, supply chains throughout the world
were negatively affected by disruptions, especially in developing
countries, where small enterprises lost access to customers and labor.
On the other hand, supply chain modification through digitization
and developing new channels (food delivery) allowed for overcoming
some of the limitations of the pandemic. Schools’ closures and other
disruptions prevented access to food delivery mechanisms for the
poor, and loss of employment reduced cash transfers that limited
food consumption in developing countries.
Because food security is subject to random shocks, families
and governments develop risk-coping mechanisms to address them.
The World Food Programme and other aid organizations developed
mechanisms to deliver food aid and health services during acute
food crises. On a smaller level, networks through the community,
religious organizations, and families can ease the burden on those
dealing with food shortages. Food assistance programs, mostly
by governments, are a major mechanism to address food security
challenges. In this sense, it is useful to distinguish between domestic
and international food assistance programs. Domestic food assistance
programs include food aid programs such as food stamps in the
United States, the Special Supplemental Nutrition Program for
Women, Infants, and Children (WIC), food for work programs, food
subsidies and price stabilization, micronutrient fortification programs
(adding vitamins and other crucial micronutrients), and nutrition
education programs. International food assistance programs include
food aid, where a donor provides surplus food, usually for free, and
food-related international finance that enables developing countries
to purchase food. There is a large body of literature assessing these
policies over multiple dimensions. Examining these policies and their
design provides opportunities for future research. Food and nutrition
policies should be evaluated on their efficiency (in terms of direct and
316 Agricultural Economics and Policy

indirect costs), incentive compatibility, equity, additionality (expand-

ing food security), targeting (to whom the program is targeted and
who benefits from it), as well as their flexibility and adaptability to
changes over time (Barrett, 2002).

3. Food Insecurity in the United States

Despite its wealth, food insecurity is a major problem in the U.S.
Food insecurity ranges from 5.4% in New Hampshire to 15.3%
in Mississippi between 2019 and 2021. Figure 2 shows that food

Figure 2. Food Insecurity in the U.S. over 2019–2021.

Source: Reproduced from Chart Gallery of ERS. Available at: https://www.
ers.usda.gov/data-products/chart-gallery/gallery/chart-detail/?chartId=104693
(accessed March 9, 2023). Permission obtained.
 Health Risks, Food Security, and Food Safety 317

insecurity is especially prevalent in the southern states, but even

in the Corn Belt and California, food insecurity is substantial
(U.S. Department of Agriculture (USDA), 2022). Figure 3 shows
that obesity rates are especially high in states with food insecurity
problems, and Figure 4 shows that again, lower life expectancy is
connected to these regions. Bansal and Jin (2023) show evidence
that high rates of obesity are associated with diseases such as
diabetes, heart conditions, and other debilitating diseases, which
end up requiring significantly increased health expenditures. Thus,
addressing the issues of food and health has become a major policy
concern in the U.S. and in the rest of the world.
Gundersen et al. (2011) survey the literature on the economics
of food insecurity in the United States. Based on the results of the
current population survey (CPS), they report that between 2001 and
2009, food insecurity among children in the U.S. increased between
18% and 24% and severe food insecurity between 9% and 12%. The
literature has identified multiple factors that are associated with food
insecurity; for example, African American or Hispanic households,

Figure 3. Obesity rate in the U.S.

Source: Map reproduced from Landgeist. Available at: https://landgeist.com/
2021/05/07/prevalence-of-obesity-in-the-us/. Permission was obtained from
Landgeist.
318 Agricultural Economics and Policy

Figure 4. Life Expectancy at Birth for U.S. Census Tracts (2010–2015).

Source: National Center for Health Statistics, Centers for Disease Control and
Prevention. Map was designed by Tejada-Vera et al. (2020) and is in the public
domain.

single-parent households, and less-educated households are more

likely to be food insecure. Furthermore, households with children
are more food insecure than those without children. The likelihood
of being food insecure is consistent with the permanent income
hypothesis: average income is a better indicator than current income
for food insecurity. Moreover, limited assets and unemployment are
also major contributors to food insecurity.
The literature finds that food insecurity has severe health effects.
Food insecurity during pregnancy is associated with birth defects,
and food insecurity overall is associated with anemia, higher level of
anxiety, asthma, behavioral problems, and chronic disease. However,
there is a continued need to develop a more systematic understanding
of the circular links between various health effects and food insecurity
 Health Risks, Food Security, and Food Safety 319

(Gundersen et al., 2011). Econometric tools could be useful, but the

research may require multidisciplinary approaches and may develop
predictive tools.
The U.S. has a wide array of food assistance programs. There
has been a significant body of work on the historical food stamp
program and now, the Supplemental Nutrition Assistance Program
(SNAP). The food assistance programs in the United States were
part of the Farm Bill and have reflected a political economic alliance
between agricultural and urban regions (Pinstrup-Andersen, 1993).
The design of a food assistance program is challenging, needing to
set a limit on what goods can be purchased. There have been reports
of the sale of food stamps for a discount, and thus the participants
in the program may substitute food for income. Thus, some may
argue that food support programs are inferior to income transfer
programs. However, as Jensen and Wilde (2010) argue, there are
restrictions as to what can be supported by the program (e.g., it
cannot be used for cigarettes or lottery tickets). Furthermore, there
is evidence that the programs improve nutritional choices and food
security in the U.S. The SNAP program has been modified over the
years to incorporate budgetary and political considerations, taking
advantage of new electronic and data tools to make it more digital
and effective. The entitlement requires income and asset tests, and
the monthly benefits can reach $668 in 2010 dollars. Unfortunately,
significant amounts of people who are entitled do not participate,
maybe because of stigma, transaction costs, as well as low benefit-
cost ratio for some families. Because of low participation in the
program and data problems, it is not obvious to what extent SNAP
programs contribute to addressing food insecurity in the U.S., and
this is a subject for further research (Gundersen et al., 2011).
The National School Lunch Program (NSLP) is another federal
assistance program in the U.S. Half the participants receive subsi-
dized lunches, and most pay nothing. In 2010, 31 million children
participated in the program. The program aims to improve diets
and health. School lunch programs are prevalent throughout the
world, and some studies suggest that they provide significant benefits
(Barrett, 2002). However, Gundersen et al. (2011) suggest that the
320 Agricultural Economics and Policy

methodological and data availability constraints make it difficult to

quantify their impact.
Wide arrays of programs aim to improve nutritional intake,
including economic incentives, subsidies, and income support, and
educational efforts. For example, one mechanism to combat obesity
is taxing sugary products and subsidizing healthy products, like
vegetables (Cash et al., 2005). There is significant literature on the
impact of soda taxes, showing that they would change consumption
patterns, especially when imposed in some locations and not others
(Taylor et al., 2019). There is momentum to expand the soda tax
coverage globally, adjusting it to differences among countries (Roache
and Gostin, 2017). Soda taxes may be much more beneficial in
countries with high obesity rates, such as the U.S. and Mexico, but
may even be harmful in countries with undernutrition. So, the design
of food incentives should take into account heterogeneity and can
improve with better data and monitoring tools. Furthermore, Just
(2006) suggests that it is important to consider behavioral economic
tools like nudging, as well as framing to modify food choice behavior.

4. Food Safety
Food safety concerns food that is toxic or otherwise endangers human
health. It is an important component of food security. Every year,
approximately 600 million people globally fall ill after consuming
food contaminated by bacteria, viruses, parasites, or chemical sub-
stances, and among them, 420,000 die (Lee and Yoon, 2021). Belluz
(2015) shows that food safety challenges are mostly attributed to
vegetables and eggs, but also happen with meat and dairy.
The framework presented by Equations (1) and (2) has been
applied to address issues of food safety, for example in the control
of pesticides and water quality, as can be seen from the chapter
on pesticides in this book, as well as from Lichtenberg (2002). The
risk generation function is useful for modeling food safety problems.
The risks can be viewed as the result of multiple processes, in
particular, the emergence of a toxin, exposure, and vulnerability.
Production processes control the emergence of toxicity, such as the
 Health Risks, Food Security, and Food Safety 321

use of pest control. Exposure can be reduced by preventive activities:

boiling water, washing fruit, and other means of food-appropriate
processing. Individuals who are allergic to various types of food can
reduce their vulnerability simply by avoiding them; when an allergen
attack occurs then medical treatment is needed.
Food safety is affected by multiple policies, including regulation
of production processing, investment in health and safety education,
and other procedures. Food safety may be either a complement
or substitute for food security. Pest control activities to reduce
spoilage increase the quantity of food supplied and its quality. Yet
some food safety procedures (e.g., culling of animals suspected of
carrying diseases like tuberculosis (Olmstead and Rhode, 2015)) may
reduce food supply. Optimal food policy processes can be derived by
minimizing the cost of health risk and of risk prevention activities.
Antle’s (2001) survey of the economic literature on food safety
emphasizes that food safety are outcome of all food systems, starting
in production, going through processing, marketing, and returning.
It is affected by pathogens in the soil, chemical residue in production,
spoilage and treatment, and pathogens in processing. Reducing food
safety hazards requires continuous monitoring throughout the food
supply chain. There is a wide array of food safety regulations in
the United States, starting with the Meat Inspection Act of 1906.
Multiple agencies regulate food safety, including the Food and Drug
Administration, U.S. Environmental Protection Agency, and the
USDA, and there may be inefficiencies in regulation that need to
be evaluated often. One aspect of food safety regulation, pesticide
regulation, aims to minimize or eliminate harm. Since eliminating
harm is not feasible, one big challenge is the “reasonable certainty of
no harm.” Lichtenberg and Zilberman (1988) suggested that, with a
specific degree of statistical reliability in the risk assessment model,
regulation can set a target level of risk, expanding the risk scope
of the risk generation function. Another set of regulations stems
from the Hazard Analysis Critical Control Points (HACCP) (Food
and Safety Inspection Service, 1996), which involves monitoring
and controlling food safety hazards throughout the food system.
The strategies to monitor and reduce hazards need to be adapted
322 Agricultural Economics and Policy

over time based on new knowledge (including the occurrence of

accidents) and new technologies. The regulatory efforts also include
requirements to provide information and training to consumers,
and labeling design can be modified as knowledge and technology
improve (e.g., Wilson and Miao, in press). Since agricultural food
products are traded globally, there is another set of food safety
regulations affecting trade. On the one hand, for example, these
policies must be strict to control invasive species, but on the other
hand, these policies may be abused and serve as trade barriers.
The International Agreement and the World Trade Organization
are establishing some of these regulations, and economists need to
reevaluate them over time.
The extent to which the government should regulate food safety
is a subject of continuous research, especially given the uncertainty
about food safety outcomes and their costs. Social cost-benefit
analysis can contribute to effective food safety regulations, and to
the sharing of responsibility between the public and private sectors
in introducing and monitoring safety standards. Antle (2001) argues
that companies may engage in food safety activities to protect
their reputations and brands, and that consumers demand high
safety standards. Products may be evaluated differently based on
their perceived risk, which is a product characteristic. Food safety
characteristics and information provision may affect market equilib-
rium, product quality, and market structure. Because of asymmetric
information between producers and consumers, brands and pricing
serve as signals that allow consumers to discriminate among products
based on safety and other characteristics. There is large evidence of
heterogeneity among consumers in terms of their demand for and
understanding of food safety, which may lead to inefficiencies and a
continuous need to redesign food safety features and regulations.
Henson and Caswell (1999) provide a detailed analysis of the
categories of actions and division of responsibilities between the
public and private sectors in food safety regulation. To develop
sufficient regulation, it is important to have a sound framework
for risk assessment that identifies hazards and the processes of
generation and exposure. Risk assessment provides the grounds for
 Health Risks, Food Security, and Food Safety 323

risk management that evaluate hazards and treatment costs, and that
develop monitoring strategies and interventions. Finally, there is an
important role in risk communication since hazard control requires
action from producers and consumers. The public sector affects food
safety through direct regulation and product liability, and the private
sector may affect food safety through self-regulation and certification.
In most cases, public food safety regulations may be presented
as standard, and once they are violated, the liability rules apply.
Countries often have reasonable standards, but the implementation
varies significantly. Consumers and the public, through non-profit
organizations, frequently use the legal system to ensure that food
safety standards are implemented.
Antle (2001) provides evidence that existing regulatory regimes
may be suboptimal. There may be significant cases of excessive regu-
lations that may raise the price of food and harm consumers. Studies
by Just et al. (2006) also provide examples of inefficient regulations.
For example, harmonizing pesticide regulation among countries and
requiring that countries use more expensive and “safer” chemicals
may be inefficient and costly for poorer countries. Finally, there
is significant evidence that regulation of agricultural biotechnology
has resulted in welfare loss and reduced the adoption of modern
technologies, agricultural activity, and environmental quality.

5. Conclusion
Food security and safety are major global problems, and understand-
ing them requires multidisciplinary approaches that combine the
knowledge of the physical and economic processes that lead to food
insecurity, as well as the medical challenges associated with treating
it. Moreover, Clark et al. (2019) show how nutritional patterns and
food availability affect both health and environmental outcomes and
suggest the need for a holistic approach to address it.
Risk-generation function provides a mechanism to assess the
processes behind food safety and security; in both cases, we can
distinguish between the processes that generate risk, exposure, and
vulnerability. These factors can be affected by regulations and
324 Agricultural Economics and Policy

innovation. Economists can play an important role in integrating

different perspectives to assess the increasing availability of food
and mechanisms linking the food supply chain to human and
environmental health. But, economic analysis requires the integration
of knowledge from other disciplines and combines both assessments
of performance of past regulatory frameworks with the prediction of
the performance of future regulatory frameworks.
The high rate of agricultural and medical innovation is challenged
to increase food availability and improve human health efficiently
and equitably. The economic research tools developed to address
innovation, technology adoption, and impact assessment can apply to
food availability and security issues. The same analytical approach
that applies to understanding precision farming can be applied to
precision medicine and to the notion of personal diet that is likely to
develop first in the developed world, and then diffuses globally. The
World Health Organization suggested the introduction of a “One
Health” framework to integrate human and environmental health in
an integrated way. Economic analysis can assist in framing and tools
to provide solutions. The linkage between farming, food, and health
has become a major issue of research.

References
Alvi, M. and M. Gupta. 2020. Learning in Times of Lockdown: How Covid-19
is Affecting Education and Food Security in India. Food Security 12(4):
793–796.
Antle, J.M. 2001. Economic Analysis of Food Safety. Handbook of Agricultural
Economics 1: 1083–1136.
Bakker, C., B.F. Zaitchik, S. Siddiqui, B.F. Hobbs, E. Broaddus, R.A. Neff,
J. Haskett, and C.L. Parker. 2018. Shocks, Seasonality, and Disaggregation:
Modelling Food Security Through the Integration of Agricultural, Trans-
portation, and Economic Systems. Agricultural Systems 164: 165–184.
Bansal S. and Y. Jin. 2023. Heterogenous Effects of Obesity on Life Expectancy:
A Global Perspective. Annual Review of Resource Economics 15: 433–454.
Bansal, S. and D. Zilberman. 2020. Macrorelationship between Average Life
Expectancy and Prevalence of Obesity: Theory and Evidence from Global
Data. Agricultural Economics 51(3): 403–427.
Barrett, C.B. 2002. Food Security and Food Assistance Programs. Handbook of
Agricultural Economics 2: 2103–2190.
 Health Risks, Food Security, and Food Safety 325

Barrett, C.B., T. Garg, and L. McBride. 2016. Well-Being Dynamics and Poverty
Traps. Annual Review of Resource Economics 8: 303–327.
Barrett, C.B., T. Reardon, J. Swinnen, and D. Zilberman. 2022. Agri-Food
Value Chain Revolutions in Low- and Middle-Income Countries. Journal of
Economic Literature 60(4): 1316–1377.
Behrman, J.R. 1997. Intrahousehold Distribution and the Family. Handbook of
Population and Family Economics 1: 125–187.
Belluz, J. 2015. Fruits and vegetables poison more Americans than beef and
chicken. News article published on Vox.com. Available at: https://www.
vox.com/2015/3/6/8158289/food-poisoning (accessed March 9, 2023).
Cash, S.B., D.L. Sunding, and D. Zilberman. 2005. Fat Taxes and Thin Subsidies:
Prices, Diet, and Health Outcomes. Acta Agriculturae Scandinavica Section C
2(3–4): 167–174.
Clark, M.A., M. Springmann, J. Hill, and D. Tilman. 2019. Multiple Health and
Environmental Impacts of Foods. Proceedings of the National Academy of
Sciences 116(46): 23357–23362.
Cozzolino, A., S. Rossi, and A. Conforti. 2012. Agile and Lean Principles In
The Humanitarian Supply Chain: The case of the United Nations World
Food Programme. Journal of Humanitarian Logistics and Supply Chain
Management 2(1): 16–33.
Food Safety and Inspection Service. 1996. The final rule on pathogen reduction
and HACCP. U.S. Department of Agriculture, Federal Register, 61, No. 144
(July 25) 38805–38855.
Goldstein, J.R. and R.D. Lee. 2020. Demographic Perspectives on the Mortality of
COVID-19 and Other Epidemics. The Proceedings of the National Academy
of Sciences 117(36): 22035–22041.
Greenstone, M. and V. Nigam. 2020. Does Social Distancing Matter? University of
Chicago, Becker Friedman Institute for Economics Working Paper 2020–26.
Gundersen, C., B. Kreider, and J. Pepper. 2011. The Economics of Food Insecurity
in the United States. Applied Economic Perspectives and Policy 33(3):
281–303.
Heiman, A., D.R. Just, B. McWilliams, and D. Zilberman. 2001. Incorporating
Family Interactions and Socioeconomic Variables into Family Production
Functions: The Case of Demand for Meats. Agribusiness: An International
Journal 17(4): 455–468.
Heiman, A., B. Gordon, and D. Zilberman. 2019. Food Beliefs and Food Supply
Chains: The Impact of Religion and Religiosity in Israel. Food Policy 83:
363–369.
Henson, S. and J. Caswell. 1999. Food Safety Regulation: An Overview of
Contemporary Issues. Food Policy 24(6): 589–603.
Jensen, H.H. and P.E. Wilde. 2010. More Than Just Food: The Diverse Effects
of Food Assistance Programs. Choices 25(3): 1–5.
Just, D. 2006. Behavioral Economics, Food Assistance, and Obesity. Agricultural
and Resource Economics Review 35(2): 209–220.
Just, R.E., J.M. Alston, D. Zilberman. 2006. Regulating Agricultural Biotechnol-
ogy: Economics and Policy. Springer Science & Business Media.
326 Agricultural Economics and Policy

Kaplan, S., J. Lefler, and D. Zilberman. 2022. The Political Economy of

COVID-19. Applied Economic Perspectives and Policy 44(1): 477–488.
Laborde, D., W. Martin, J. Swinnen, and R. Vos. 2020. COVID-19 Risks to Global
Food Security. Science 369(6503): 500–502.
Lee, H. and Y. Yoon. 2021. Etiological Agents Implicated in Foodborne Illness
World Wide. Food Science of Animal Resources 41(1): 1–7.
Lichtenberg, E. 2002. Agriculture and the Environment. Handbook of Agricultural
Economics 2: 1249–1313.
Lichtenberg, E. and D. Zilberman. 1988. Efficient Regulation of Environmental
Health Risks. The Quarterly Journal of Economics 103(1): 167–178.
Lichtenberg, E. and D. Zilberman. 2002. Storage Technology and the Environ-
ment. Journal of Agricultural and Resource Economics 27(1): 146–164.
Nomura, K., Bhandari, A.K.C., Matsumoto-Takahashi, E.L.A., and
Takahashi, O. 2023. Risk Factors Associated with Stunting among Children
Under Five in Timor-Leste. Annals of Global Health 89(1): 63.
Olmstead, A.L. and P.W. Rhode. 2015. Arresting Contagion: Science, Policy, and
Conflicts Over Animal Disease Control. Harvard University Press.
Pinstrup-Andersen, P. 1993. The Political Economy of Food and Nutrition
Policies. John Hopkins University Press.
Reardon, T. and C.P. Timmer. 2014. Five Inter-Linked Transformations in
the Asian Agrifood Economy: Food Security Implications. Global Food
Security 3(2): 108–117.
Roache, S.A. and L.O. Gostin. 2017. The Untapped Power of Soda Taxes:
Incentivizing Consumers, Generating Revenue, and Altering Corporate
Behavior. International Journal of Health Policy and Management 6(9): 489.
Robson, M.G., W.A. Toscano, Q. Meng, and D. A. Kaden (Eds.). 2022. Risk
Assessment for Environmental Health. CRC Press.
Sen, A. 1982. Poverty and Famines: An Essay on Entitlement and Deprivation.
Oxford University Press.
Smith, M.D. and B. Meade. 2019. Who Are the World’s Food Insecure?
Identifying the Risk Factors of Food Insecurity Around the World. Amber
Waves, Economic Research Service, U.S. Department of Agriculture.
Taylor, R.L.C., S. Kaplan, S.B. Villas-Boas, and K. Jung. 2019. Soda Wars:
The Effect of a Soda Tax Election on University Beverage Sales. Economic
Inquiry 57(3): 1480–1496.
Tejada-Vera, B., B. Bastian, E. Arias, L.A. Escobedo, and B. Salant.
2020. Life Expectancy Estimates by U.S. Census Tract, 2010–2015.
National Center for Health Statistics. Available at: https://www.cdc.gov/
nchs/data-visualization/life-expectancy/ (accessed January 27, 2024).
United Nations Children’s Fund (UNICEF), World Health Organization, The
World Bank. 2021. Levels and Trends in Child Malnutrition: Key Findings
of the 2021 Edition of the Joint Child Malnutrition Estimates. Geneva: World
Health Organization.
U.S. Department of Agriculture. 2022. Food Insecurity Rates Differ across U.S.
States. Available at: https://www.ers.usda.gov/data-products/chart-gallery/
gallery/chart-detail/?chartId=104693 (accessed March 9, 2023).
 Health Risks, Food Security, and Food Safety 327

Viscusi, W.K. and J.E. Aldy. 2003. The Value of a Statistical Life: A Critical
Review of Market Estimates Throughout the World. Journal of Risk and
Uncertainty 27: 5–76.
Wesseler, J. and D. Zilberman. 2014. The Economic Power of the Golden Rice
Opposition. Environment and Development Economics 19(6): 724–742.
Williams, J.C. and B.D. Wright. 1991. Storage and Commodity Markets.
Cambridge University Press.
Wilson, R. and E.A. Crouch. 1987. Risk Assessment and Comparisons:
An Introduction. Science 236(4799): 267–270.
Wilson, N.L.W., and R. Miao. in press. Food Waste, Date Labels, and Risk
Preferences: An Experimental Exploration. Applied Economic Perspectives
and Policy 1–29. https://doi.org/10.1002/aepp.13507.
Wu, F., J. Wesseler, D. Zilberman, R.M. Russell, C. Chen, and A.C. Dubock.
2021. Allow Golden Rice to Save Lives. Proceedings of the National Academy
of Sciences 118(51): e2120901118.
Wu, F. and P. Khlangwiset. 2010. Health Economic Impacts and Cost-
Effectiveness of Aflatoxin-Reduction Strategies in Africa: Case Studies in
Biocontrol and Post-Harvest Interventions. Food Additives and Contami-
nants 27(4): 496–509.
Young-Xu, Y., R. van Aalst, E. Russo, J.K.H. Lee, and A. Chit. 2017. The Annual
Burden of Seasonal Influenza in the US Veterans Affairs Population. PLoS
One 12(1): e0169344.
Zilberman, D. and Y. Jin. 2016. A Probabilistic Approach to Food Security.
In Food Security in a Food Abundant World. Emerald Group Publishing
Limited.
This page intentionally left blank
 Chapter 11

Economics of Pesticides

Pesticides are chemicals used in controlling agricultural pests and

include three major classes: insecticides, fungicides, and herbicides.
The use of pesticides in agriculture presents some interesting aspects
to be considered. First, they need to be chemically updated over
time as pests build resistance. Second, they cause adverse human and
animal health effects. The adverse human health effects of different
types of pesticides depend on the similarity between human or animal
biology and the biology of the target pest. Insecticides, for example,
are generally worse for human health than fungicides.
From 1965 to 1980, growth in the relative price of labor increased
the use of herbicide as a factor of production. This occurred because
herbicide use was a substitute for labor. The 1970 creation of the
Environmental Protection Agency (EPA) and an increase in energy
prices led to a reduction in insecticide use through the 1970s. During
the 1980s, lower agricultural commodity prices and reduced crop
acreage led to a reduction in herbicide use. Overall, pesticide use
in the past four decades in the United States has been quite stable.
See Fernandez-Cornejo et al. (2014) for a comprehensive analysis
of pesticide use in U.S. agriculture. In this chapter, we discuss
how to incorporate pesticide use in production analysis and related
policies.

329
330 Agricultural Economics and Policy

1. Pesticides in a Damage Control Framework

Since pesticides are damage control agents, crop production can be
modeled as

Y = g(Z)[1 − D(n)],

where Y is actual output, g(Z) is potential output (i.e., the maximum

output that can be produced in the absence of damage due to the
presence of a pest), Z stands for all inputs not related to pest control,
and D(n) ∈ [0, 1] is the damage function, expressed as percent of
output lost to pest damage and assumed to be function of the pest
population, n.
The effect of pesticide use is that of controlling the pest
population, according to

n = h(n0 , X, A),

where n0 is the initial level of pest population before pesticide

application, X is the level of applied pesticide, and A is some
alternative pest control method, such as Integrated Pest Management
(IPM). We assume that hX ≡ ∂h/∂X < 0 and hA ≡ ∂h/∂A < 0.
Obviously, there are costs associated with pesticide application.
If the total damage from pests is less than the social cost associated
with a single application of a pesticide to a field (including marginal
external cost), then the welfare-maximizing level of pesticide use is
zero. Note that this implies toleration of some pests in the field and
the associated pest damage, such as less visibly appealing fruits and
vegetables.
When the level of pest damage rises above the social cost of
one pesticide application, then it is welfare-maximizing to apply the
pesticide. Profit maximization in the private market will determine
the economic threshold, n̄0 , as the pest population level at which
it becomes profit-maximizing to apply the pesticide. The economic
threshold is determined by setting total pest damage equal to the
total cost of a single pesticide application and solving for n̄0 :

P g(Z)D(n̄0 ) = w,
 Economics of Pesticides 331

where P is output price and w is the total cost of a single pesticide

application.
Given function forms for g(·) and D(·), one could solve the above
equation for n̄0 . In the models that follow, we assume that the pest
population is above the economic threshold.

2. Model of Pesticide Use with Known Pest

Population and Pest Control Alternatives
The optimal level of pesticide use is determined by solving
max {L = P · g(Z) [1 − D(h(n0 , X, A))] − V A − wX}, (1)
X,A

where V is the unit cost of alternative control methods, and other

symbols are as defined in the above section.
The first-order conditions are
dL
= −P · g(Z)Dn hX –w = 0, (2)
dX
dL
= −P · g(Z)Dn hA –V = 0. (3)
dA
Thus, one would maximize profits by applying pesticides until the
value of marginal product (marginal benefit) of pesticide application
equals the marginal cost of pesticide application. The model predicts
that the use of pesticides will increase as a result of the following:
— an increase in initial pest population (n0 ),
— an increase in the output price (P ),
— an increase in potential output (g(Z)),
— an increase in the price of alternative controls (V ), or
— a decrease in the price of pesticides (w).
Analogous results hold for the alternative pest control method.

2.1. A model with a secondary pest

Sometimes, a secondary pest may exist along with a primary pest,

defined as those targeted by pesticide use due to their more severe
effect on crops. Here we consider the presence of a secondary pest.
332 Agricultural Economics and Policy

For simplicity, we do not consider any direct alternative controls

besides the use of pesticides, but we assume a specific biological rela-
tionship between the two pest populations: both pests cause damage,
both population levels are known, and pest 1 is a predator of pest 2.
The damage function, D(·), will be function of both pests’
population levels: D(n1 , n2 ). Pest 1’s population is directly affected
by the pesticide so that n1 = h(n0 , X), while pest 2’s population
level is a decreasing function of pest 1’s population level: n2 =
Ψ(n1 ), with Ψn1 < 0. In other words, using pesticide to control
pest 1 may lead to an increase in the population of pest 2, since
pest 1 is a predator of pest 2. Thus, the problem of determining the
optimal level of pesticide to use becomes
max {P · g(Z) [1 − D(h(n0 , X), Ψ(n1 ))] − wX} (4)
X

with the first-order condition:

−P · g(Z)[Dn1 hX + Dn2 Ψn1 hX ] − w = 0.
Both the direct impact on n1 and indirect impact on n2 on crop
damage have to be considered on determining the optimal level of
X. Lack of recognition of the biological predator–prey relationships
may lead to economically inefficient over-application of pesticides,
since the beneficial effect of the predator pest on reducing pest 2 is
ignored.

2.2. Pesticide resistance

Through the biological process of natural selection, pests exposed to

pesticides gradually develop genetic resistance to pesticides. Higher
levels of pesticide application may accelerate buildup of resistance
due to genetic selection of resistant genes. Short run pesticide
control problems in a given season will be inefficient if long term
resistance effects are not considered. Therefore, the calculation of
optimal dosage of pesticide should take into account (i) resistance
buildup (pesticide effectiveness is an “exhaustible resource” and
should be modeled as such) and (ii) use of alternative chemicals
or alternative pest control methods (such as the use of alternative
cropping methods, crop rotation, natural diseases, and predator–prey
 Economics of Pesticides 333

relationships) should be considered in order to reduce resistance

buildup.

3. Unknown Pest Population and Pest Population

Monitoring
When pest populations are unknown, as is usually the case, one
can distinguish between preventive and reactive pesticide application.
Contrary to what is considered common wisdom in human medicine,
for agricultural pest control, preventing may be worse (less efficient)
than reacting.
With preventive application, pesticides are applied without
an attempt to determine potential pest populations. Instead, based
on experience or historical data, the farmer makes educated guesses
about the probabilities of various pest population levels occurring.
The farmer then chooses a level of pesticide use to maximize expected
profit. For example, suppose there are two possible initial pest
population levels, low (n1 ) with probability p and high (n2 ) with
probability 1− p. Assuming preventive application, then the problem
of deciding how much pesticide to apply is

max E(π) = p{P g(Z)[1 − D(h(n1 , X))] − wX}

X

+ (1 − p){P g(Z)[1 − D(h(n2 , X))] − wX}.

The first-order condition is

p{−P g(Z)Dh hX (n1 , X) − w}

+ (1 − p){−P g(Z)Dh hX (n2 , X) − w} = 0.

Given specific g, D, and h functions, one could solve this first-

order condition for X. Plugging the optimal value, X, back into
the objective function would then give the level of expected profit
associated with preventive pesticide application. Note that because
the pest population is uncertain, X will be the same regardless
of which pest population level, n1 or n2 , actually occurs. This is
inefficient because we would like to use less pesticide if n1 occurs
and more if n2 occurs.
334 Agricultural Economics and Policy

To decide between preventive and reactive application methods,

we need to compare the level of expected profits under preventive
pesticide application with the level of expected profits under the
following model of reactive application.
With reactive application, a fixed monitoring cost, m, is paid
to determine the pest population level, and then the optimal X is
chosen for the specific pest level. This enables more precise pesticide
use. The problem is then

max E(π) = p{P g(Z)[1 − D(h(n1 , X1 ))] − wX1 }

X1 ,X2

+ (1 − p){P g(Z)[1 − D(h(n2 , X2 ))] − wX2 } − m.

The first-order conditions are

∂E(π)
= p{−P g(Z)Dh hX1 (n1 , X1 ) − w} = 0,
∂X1
∂E(π)
= (1 − p){−P g(z)Dh hX2 (n2 , X2 )− w} = 0.
∂X2

Given specific g, D and h functions, one could solve the first-

order condition’s for the optimal X1 and X2 . Plugging X1 and X2
back into the objective function gives the level of expected profits
under reactive pesticide application. Note that the resulting equation
for expected profits will contain monitoring costs, m. With reactive
application, there is a tradeoff between the monitoring costs and
the savings in pesticide costs made possible by monitoring. Two
points need to be underlined. First, if the difference between X1
and X2 is large, and m is relatively small, then reactive application
will give a higher level of expected profits than would preventive
application. Monitoring and reactive application are key components
of modern IPM programs. Second, even if the difference between
X1 and X2 is large, farmers may still prefer preventive application
whenever the price of pesticides is very low or monitoring cost is
very high to ensure higher expected profits. In order to combine
the profit-maximizing decision with social-welfare maximization, an
 Economics of Pesticides 335

appropriate tax would be imposed on pesticides so that effective

prices would reflect marginal external cost.
Pests do not recognize property rights. Pest control activities
often lead to externality problems. For instance, a recent study
published in Science shows that pest control approaches adopted
by organic crop producers may increase overall pesticide uses on
neighboring farms (Larsen et al., 2024). Pest control districts are
introduced to overcome these problems (e.g., mosquito control
districts). The activities of such districts encompass joint effort in
monitoring activities, coordinate crop management and rotation, and
coordinate pesticide spraying.

4. Health-risk and Environmental Effects

of Pesticide Use
Health risk is the probability that an individual selected randomly
from a population contracts adverse health effects (mortality or
morbidity) from a substance. The health risk-generating process
contains three stages: contamination, exposure, and dose–response.
Contamination is the presence of toxic pesticide or its derivatives
on the agricultural product. It is direct result of pesticide appli-
cation: the chemicals are spread through the air and water and
become absorbed by the product. Exposure is the contact of toxic
substances with human organisms, resulting from eating, breathing,
or touching contaminated product by consumers or agricultural
workers. Response translates exposure to probability of contracting
certain diseases. We usually distinguish between acute and chronic
risks. Acute risks are the immediate risks of poisoning. Chronic risks
are risk that may depend on accumulated exposure and which may
take time to manifest themselves (e.g., the higher incidence of certain
type of cancer in populations that are exposed to a certain pesticide
for a long time).
The processes that determine contamination, exposure, and the
response relationship are often characterized by heterogeneity, uncer-
tainty, and random phenomena (e.g., weather). Thus, contamination,
336 Agricultural Economics and Policy

exposure, and the response relationship need to be analyzed with

models that accounts for the inherent uncertainty. Risk assessment
models estimate health risks associated with pesticide application by
making use of estimated probabilities.
Let r be the represent individual health risk. It can be expressed
as
r = f1 (X, B1 )f2 (B2 )f3 (B3 ),

where X is the level of pollution on site (i.e., the level of pesticide

use), B1 is the damage control activity at the site (e.g., protective
clothing or re-entry rules), B2 is the averting behavior by potentially
exposed individuals (e.g., washing fruits and vegetables), and B3
reflects the dosage of pollution (e.g., the type of pesticide residual
consumed) or the level of medical treatment employed. The health
risk of an average individual is modeled as the product of three
functions. The first one is f1 (X, B1 ), the contamination function that
relates contamination of an environmental medium to activities of an
economic agent (i.e., relates pesticide residues on apples to pesticides
applied by growers). The second is f2 (B2 ), the human exposure
coefficient, which depends on an individual’s actions to control
exposure (e.g., relates ingested pesticide residues to the level of
rinsing and degree of food processing an individual engages in). The
last one is f3 (B3 ), the dose–response function which relates health
risk to the level of exposure of a given substance (i.e., relates the
proclivity of contracting cancer to the ingestion of particular levels
of a certain pesticide), based on available medical treatment methods,
B3 . Dose–response functions are usually estimated in epidemiological
and toxicological studies of human biology.
The product f1 (X, B1 )f2 (B2 ) is the overall exposure level of an
individual to a toxic material (e.g., the amount of pesticide present on
an apple times the percentage not removed by rinsing the apple). The
degree of overall exposure can be affected by improved technology
and by a greater dissemination of information.
Estimating these functions involves much uncertainty. Scientific
knowledge of dose–response relationships of pesticides is generally
incomplete, especially for pesticides ingested in small doses over
long periods of time. Contamination function depends partly on
 Economics of Pesticides 337

assimilation of pollution by natural systems, which can differ region-

ally (e.g., wind distributed residues). Exposure coefficient depends
on education of population (e.g., consumers’ awareness of pesticide
residue averting techniques, such as washing).
Uncertainty could be included in the economic model by using
a safety-rule approach. The policy goal of pesticide use regulation
should be to maximize welfare subject to the constraint that the
probability of health risk remains below a certain threshold level, R,
and that the safety level, α, is above a certain threshold. Here α could
be a measure of social risk aversion or the degree of confidence we
have in our target risk. For any target level of risk and any degree of
safety level, the model can be solved for the optimal levels of pesticide
use, damage control activities, and averting behavior by consumers.
General implications of this way of approaching the modeling
include the following. First, the optimal solution involves some
combination of pollution control and exposure avoidance. Second, the
cost of reaching the target risk level increases with the safety level α.
And finally, the shadow price of meeting the risk target depends on
the degree of significance (α) we have that the target is being met.
The higher α, or the greater the uncertainty we have in our estimate
of risk, the higher the shadow value of meeting the constraint.

Example 1 (a model without uncertainty):

Suppose there is no uncertainty regarding the health effects of
pesticide use. That is, toxicologists know with certainty a point
estimate of the dose–response function, f3 (·). Moreover, let
Y = f (X, A) be the farm production function,
X = the level of pesticides used on a field,
A = the level of alternate pest control activities,
P = the value of farm output (e.g., the price of a basket of
produce),
Y = the level of farm output,
W = the price of pesticide,
V = the price of alternative controls (V > W ),
R = f1 (X, B1 )f2 (B2 )f3 (B3 ) be the level of health risk in society,
where the notation is defined as before,
338 Agricultural Economics and Policy

C(R) = the cost to society of health risk R, and

C(B1 , B2 , B3 ) = the cost incurred by conducting B1 , B2 , and B3 .
For simplicity, we assume away other production costs.
Then, the objective of the society is to
max P f (X, A) − C(R) − C(B1 , B2 , B3 ) − WX − VA (5)
X,A,R,B1 ,B2 ,B3

subject to
R = f1 (X, B1 )f2 (B2 )f3 (B3 ).
The optimization problem can be written in Lagrangian form as
max L = P f (X, A) − C(R) − C(B1 , B2 , B3 ) − WX − VA
X,A,R,B1 ,B2 ,B3

+ λ[R − f3 (B3 )f2 (B2 )f1 (B1 , X)].

The first-order conditions are
dL
= P fA − V = 0, (6)
dA
indicating that the marginal revenue product (MRP) of the
alternative control equals the marginal cost of the alternative control,
and
dL
= −C  (R) + λ = 0, (7)
dR
indicating that the marginal social cost of health risk equals the
shadow value of risk (i.e., the marginal cost of risk in terms of social
damages is equal to the shadow price of reducing societal risk).
Furthermore, we have
 
dL df1
= P fX − W − λ f3 f2 = 0, (8)
dX dX
 
dL df1
= −CB1 − λ f3 f2 = 0, (9)
dB1 dB1
 
dL df2
= −CB2 − λ f3 f1 = 0, (10)
dB2 dB2
 
dL df3
= −CB3 − λ f2 f1 = 0. (11)
dB3 dB3
 Economics of Pesticides 339

We can rewrite Equations (8)–(11) using Equation (7) as

 
 df1
P fX = W + C (R) f3 f2 ,
dX
which implies that the marginal revenue product of pesticide use
to the farm is equal to the marginal private cost of pesticides plus
the product of marginal cost of risk and marginal contribution of
pesticides to the risk;
 
 df1
CB1 = −C (R) f3 f2 ,
dB1
which implies that the marginal cost of damage control equals the
product of avoided marginal cost of risk and marginal improvement
in risk from engaging in damage control activities;
 
df2
CB2 = −C  (R) f3 f1 ,
dB2
which implies that the marginal cost of averting behavior equals the
product of avoided marginal cost of risk and marginal improvement
in risk from engaging in averting behavior; and
 
 df3
CB3 = −C (R) f2 f1 ,
dB3
which implies that the marginal cost of medical treatment equals the
product of avoided marginal cost of risk and marginal reduction in
risk from engaging in medical treatment.
The optimal solution involves equating all six first-order condi-
tions. Equations (7)–(11) can be expressed as
P fX − W −CB1 −CB2 −CB3
λ = C  (R) =  = = = ,
df1 df1 df2 df3
f3 f2 dX f3 f2 dB1 f3 f1 dB2 f2 f1 dB3

which states that the optimal solution involves equating the shadow
price of risk with a series of ratios. The denominator of each expres-
sion transforms marginal benefits or marginal costs of health-related
activities into changes in health risk. When parameters are known,
the model can be solved for the optimal levels. The model has some
general implications. First, if there is no tax on pesticide use and no
340 Agricultural Economics and Policy

subsidy on farm-level damage control, then the farm will not recog-
nize the effect of pesticide use on societal health, and will operate as
if λ = 0. As a consequence, an inefficiently high level of pesticides
will be used, and an inefficiently low level of damage control will be
applied. Second, the optimal solution may involve a large level of
pesticide use, little damage control, little medical treatment, and a
high degree of averting behavior. Rinsing and washing produce might
be the least expensive method of reducing health risk in society.

Example 2 (a model with uncertainty):

Let r be the probability of an individual contracting a disease and
r = c·e·d·x, where c stands for contamination probability, e exposure
probability, d dose-response probability, and x amount of pesticide
applied. Let

0.1 with probability .5
c=
0.2 with probability .5,

0.1 with probability .5
e=
0.3 with probability .5,
 −4
10 with probability .5
d=
10−5 with probability .5.
Suppose x = 1, then we have
⎧
⎪
⎪ 1 × 10−6 with probability 1/8
⎪
⎪
⎪
⎪ 2 × 10−6 with probability 1/8
⎪
⎪
⎪
⎪ 3 × 10−6 with probability 1/8
⎪
⎨
6 × 10−6 with probability 1/8
r=
⎪
⎪ 1 × 10−7 with probability 1/8
⎪
⎪ 2 × 10−7 with
⎪
⎪ probability 1/8
⎪
⎪
⎪
⎪ 3 × 10−7 with probability 1/8
⎪
⎩
6 × 10−7 with probability 1/8.
Note that r = 1 × 10−6 means “one person per million people”
contracts the disease. Here the expected risk is
13.2
× 10−6 = 1.65 × 10−6 ,
8
 Economics of Pesticides 341

or 1.65 people per one million people, on average, contract the

disease.
Based on the distribution of r, the objective of the society is
to maximize the expected net benefits from using pesticides. The
objective function can be readily written by expanding optimization
problem (5). In many cases, however, the highest value (worst case
estimator) of each probability is used to solve optimization problem
(5) when the risk generation processes are broken down to many
sub-processes. This creates a “creeping safety” problem, in that the
multiplication of many “worst case” estimates may lead to wildly
unrealistic risk estimates. The variability and uncertainty associated
with risk estimates can be reduced by expenditures on research and
through information-sharing.

5. Pesticide Policy
Federal Insecticide, Fungicide and Rodenticide Act (FIFRA) was
passed into law by Congress in 1972 and is enforced by the EPA,
which has the power to prohibit the sale, distribution, or use of
pesticides such as insecticides, fungicides, and rodenticides under
the Act. If a threatened or endangered species could be adversely
affected, the EPA can also issue an emergency suspension of certain
pesticides. FIFRA requires that farmers, utility companies, and
other users of pesticides register when they purchase pesticides,
for which these individuals also have to pass a certification exam.
FIFRA also contains provisions that require that all pesticides
used in the United States be approved and licensed by the EPA.
In 1972, the EPA banned the use of DDT (dichloro-diphenyl-
trichloroethane), the pesticide featured in Silent Spring (Carson,
1962) and required extensive review of all pesticides. In 1996, the
Food Quality Protection Act was passed to tighten standards for
pesticides used to grow food, with special protections to ensure that
foods are safe for children to eat. In 2015, EPA revised Agricultural
Worker Protection Standards to increase protections from pesticide
exposure for agricultural workers and their families in the US.
Modern pesticide policies tend to be triggered solely by health
considerations: when a chemical is found to be carcinogenic or
342 Agricultural Economics and Policy

damaging to the environment, it is banned or “cancelled.” The

impacts of pesticide cancellation depend on the available alternatives.
Cancellation often causes losses in crop yields due to higher pest
damages and increases costs, since alternative methods of control, if
there is any, are generally more expensive. To estimate overall short–
term impacts, the impacts on yield per acre and cost per acre are
evaluated using one of the following methods:
(1) Delphi method: The Delphi method was named after the
famous “Oracle at Delphi” in ancient Greece. It uses “guesstimates
by experts,” which are often straightforward to obtain but are
arbitrary and sometimes baseless.
(2) Experimental studies: These studies are based on data from
agronomical experiments, but experimental plots often do not reflect
real farming situations.
(3) Econometric studies: Statistical methods are based (ideally)
on data gathered from real farming operations. However, these
studies are often not feasible because of data limitations and the
difficulty of isolating the specific effects of pesticides.
(4) Cost budgeting method: Let yij denote output per acre of
crop i at region j with pesticide, Pij price of crop i in region j,
Aij acreage of crop i in region j, Δyij yield reduction per acre
because of cancellation, and Δcij cost increase per acre because
of cancellation. Under a partial crop budget, impacts on farmers’
surplus are estimated as
I J
(Pij Δyij + Δcij )Aij .
i=1 j=1

A pesticide cancellation causes losses in revenue from lowered yields

and increased costs per acre, which is multiplied the total acreage in
all regions and across all types of crop affected by the ban. This cost
budgeting approach has at least two limitations. First, it ignores
the effect of a change in output on output price. This tends to
overestimate producer loss and underestimate consumer loss. Second,
it ignores feedback effects from related markets. In general, this
method does not consider the interaction of supply and demand,
 Economics of Pesticides 343

and does not attempt to find the new market equilibrium after the
application of a pesticide ban.
(5) General equilibrium method: This method is based on analyz-
ing the impact of a pesticide ban on equilibrium prices and output,
taking into account the interaction of supply and demand and any
feedback effects from related markets. In addition, this method offers
a better assessment of equity effects by computing welfare changes
for various groups. As a result of a pesticide ban, marginal cost
per acre increases, output declines, and output price increases. The
magnitude of the change in output price depends on the elasticity
of demand and any feedback effects from related markets, such as
markets for substitute goods. General equilibrium analysis recognizes
heterogeneity in welfare effects: the welfare of non-pesticide-using
farmers increases due to the increase in output price, but the welfare
of pesticide-using farmers decreases if demand is elastic (but may
increase if demand is inelastic). Consumer welfare is harmed by price
increases, but is enhanced by reducing health risk due to pesticide
regulations. A recent study by Ye et al. (2021) provides an analysis on
the impact of restricting glyphosate use on social welfare, considering
the health, environmental, and economic impact of this herbicide.
Chapter 5 in this book provides an analysis of the welfare impact of
pesticide regulation accounting for the existence of other agricultural
policies, such as price support policies.
Pesticide effects include several related issues such as food safety,
worker safety, ground water contamination, and other environmental
damage. Pesticide bans aim to address all these issues. However,
a pesticide ban can be an inefficient policy because pesticide uses
and impacts vary significantly across regions. Economic mechanisms
(taxes and partial bans) that discriminate across different types
of uses may eliminate most of the pesticide damage but retain
most pesticide benefits. Although a pesticide ban may provide the
incentive to develop new, less dangerous pest control methods, a
pesticide tax may serve the same purpose and also allow a more
gradual and efficient transition to the new technology. However,
if a pesticide tax is used, policymakers should keep in mind that
pesticide use patterns could shift significantly across geographic
344 Agricultural Economics and Policy

regions. Other policy tools can affect different stages of the risk
generation process. For instance, regulations on pollution controls
and protective clothing can affect contamination and exposure;
regulations on residual tolerance standard can address food and
water safety concerns; re-entry regulation can address worker and
public safety concerns; and finally, water disposal regulation can
reduce ground water contamination.

6. Conclusion
In this chapter, we have discussed the optimal pesticide use under
various scenarios (e.g., known and unknown pest populations as well
as the presence of a secondary pest). We have also discussed the
health risk and environmental effects of pesticide use, followed by a
brief discussion of pesticide policy in the United States. Pesticides
are important inputs in agricultural production, and the tradeoff
between their positive impact on production and negative impact on
the environment will be a critical issue for societies in years to come.
Tackling this issue has attracted much attention from researchers,
policymakers, non-governmental organizations, and general public.
Since pesticide use is closely related to agricultural production and
is determined by farmers’ choices, research in agricultural economics
is well positioned to provide behavioral insights and policy recom-
mendations. With the increased availability of data for pesticide use
as well as for health and environmental measurement, a burgeoning
field of study in economics has tried to quantify the adverse impact
of pesticide uses on human health and the environment by using
national or regional scale data (e.g., Li et al., 2020; Dias et al., 2023),
complementing experimental or laboratory findings from natural
sciences.

References
Carson, R. 1962. Silent Spring. Houghton Mifflin Harcourt.
Dias, M., R. Rocha, and R.R. Soares. 2023. Down the River: Glyphosate Use in
Agriculture and Birth Outcomes of Surrounding Populations. The Review of
Economic Studies 90(6): 2943–2981.
 Economics of Pesticides 345

Fernandez-Cornejo, J., R. Nehring, C. Osteen, S. Wechsler, A. Martin, and

A. Vialou, 2014. Pesticide Use in U.S. Agriculture: 21 Selected Crops, 1960–
2008. U.S. Department of Agriculture, Economic Research Service, Economic
Information Bulletin Number 124.
Larsen, A.E., F. Noack, and L.C. Powers, 2024. Spillover Effects of Organic
Agriculture on Pesticide Use on Nearby Fields. Science 383(6689).
Li, Y., R. Miao, M. Khanna. 2020. Neonicotinoids and Decline in Bird Biodiversity
in the United States. Nature Sustainability 3: 1027–1035.
Ye, Z., F. Wu, and D.A. Hennessy. 2021. Environmental and Economic Concerns
Surrounding Restrictions on Glyphosate Use in Corn. Proceedings of the
National Academy of Sciences 118(18): e2017470118.
This page intentionally left blank
 Chapter 12

Economics of Space: Interaction

of Urban and Rural Sectors

One of society’s biggest challenges is allocating land between the

urban sector, agriculture, forestry, and wilderness. Within agricul-
ture, land allocation is also critical. Land management has been
one of the main interests of economists for millennia, and we will
cover some of the major findings in this section. We will start with
one of the oldest economic models: the von Thünen land allocation
model. Then, we will discuss market failures associated with land
use, implications of land valuation, and targeting for environmental
services. We will also discuss issues arising between energy prices,
land, and housing, as well as the linkages between land use and
climate change policies.

1. Models for Allocation of Land

The von Thünen model (Samuelson, 1983) represents a simple system
for allocating resources over space, setting the central business
district (CBD) of a region as an origin and reflecting everything else
located in relation to it. This CBD can be considered a port, and
all the producers are exporting their products to the port. A simple
model allocates land over a line, where the zero point is a CBD. Let
us assume that the distance of each particular point from the CBD
is denoted by x and farmers can produce multiple activities. Let i

347
348 Agricultural Economics and Policy

be an activity indicator, and i takes value from 1 to I. In our simple

model, the net benefit per acre of production of activity i at location
x is denoted by:

Bi (x, N ) = bi (x, N ) − ti · x,

where the net benefit Bi (x, N ) is the difference between the gross
benefit of production per acre, bi (x, N ), and transportation costs
per acre for activity i. For every activity, the gross benefits per acre
are functions of location and population (N ). We assume that the
net benefit per acre of activities declines with location, while for the
same location, the gross benefit per acre is lower for higher i (namely,
bi (x, N ) > bi+1 (x, N )), and benefit per acre increases with the size
of the population. We further assume that the transportation costs
per acre tend to be lower for activities with higher i’s (ti > ti+1 ).
Assuming that land is allocated under competition, under our
assumption, the area under the CBD, from x = 0 to x∗1 , will be
allocated to the highest value activity (activity 1). At the switching
point, x∗1 , b1 (x, N ) − t1 · x = b2 (x, N ) − t2 · x. Similarly, the switching
point, x∗i , where bi (x, N ) − ti · x = bi+1 (x, N ) − ti+1 · x, separates the
region where activity i is from the farther region where activity i + 1
is. The rent per acre for land at location x is equal to the net benefit
per acre of the optimal activity for each location.
Figure 1 illustrates the outcome of the model: the envelope (the
bold line) traces the rent as a function of location. The figure suggests
five activities, and the wilderness area (the area between x∗5 and x∗N ),
or the end of the region. The interpretation has varied: in much of
the literature interpreting the von Thünen model, the area near the
business circle is a commercial area, the next area is the residential
area, and the agricultural area starts from high-value agricultural
activities and moves to low-value agricultural activities. In some
regions, livestock production is near the urban sector because of its
relatively high transportation costs compared to field crops. In other
regions, perishable high-value fruits and vegetables are closer to the
city, followed by livestock, feedstock, and wildland. Clark (1973) also
presents an exposition of the von Thünen model and demonstrates
its applicability with historical data.
 Economics of Space: Interaction of Urban and Rural Sectors 349

Figure 1. Net benefit and rent as a function of distance from CBD.

Because we assume that the benefit at each location is a

function of the population, an increasing population tends to increase
the benefit of all activities, and thus, is likely to increase the
land rent and, furthermore, reduce the wildland area. Indeed,
human settlement historically resulted in reductions in wildlands
and biodiversity. While technological change has diminished the
relevance of the von Thünen model since the 19th century, some
of its features remain applicable. Over time, the model has been
expanded and modified to address changing reality. Hochman et al.
(1977) assume that some benefits from land use are associated with
production and introduce production functions that rely on input
(e.g., fertilizer) and pollution. Their model allows us to consider
pollution from production activities as well as agricultural activities,
and their analysis considers two types of externalities. First, pollution
accumulates toward the urban center and causes damage there or
on contiguous natural areas (normally bodies of water). In this case,
social optimum requires introducing regulations, like taxes, that may
350 Agricultural Economics and Policy

reduce production intensity close to the urban area, and thus may
change the rent structure, production intensity, or land use over
space to reduce pollution generation. This approach, for example,
is appropriate for designing policies that address the accumulation
of pollution from farming activities along the Mississippi, or activities
that result in a dead zone in the Gulf of Mexico (Rabotyagov et al.,
2014). The second type of externalities is those spread around the
production area, such as odor or air pollution. These externalities
can be controlled by penalties or zoning. For example, there was an
attempt to reduce California’s rice straw burning in order to reduce
air pollution in Sacramento (Carey et al., 2000).
Land use has major implications for greenhouse gas emissions.
There is a large body of literature on the agricultural emissions that
contribute to climate change and attempts to reduce emissions
through land use changes, technological innovation, and better
understanding of the interaction between the economy and ecological
systems (Levin and Xepapadeas, 2021). Several outstanding studies
by Karen Seto and collaborators established that increasing popula-
tion and rising living standards are likely to contribute to greenhouse
gas emissions and biodiversity loss (Seto et al., 2011, 2012). They
suggest that urbanization, especially in the developing world, is likely
to double the urban landscape if there are no interventions. This will
result in a significant loss of high-quality agricultural land that will
further enhance deforestation and loss of biodiversity, contributing
massively to climate change and mostly hurting the disadvantaged.
Thus, policies increasing input-use efficiency on urban land, among
other policies, could reduce the carbon footprint from the urban
sector and offer benefits to agricultural production. Further under-
standing of the link between urban and agricultural development
and policies to control the externalities of both are major research
priorities and may require multidisciplinary collaboration.
Because climate change and other stock pollution problems
(e.g. groundwater contamination) are dynamic phenomena, modeling
them requires optimal control over space and time. Xabadia et al.
(2008) develop a model where economic agents are heterogeneous
and generating stock pollution over differing areas. The optimal
 Economics of Space: Interaction of Urban and Rural Sectors 351

intervention design is to maximize social welfare by maximizing

the net present value of producer, consumer, environmental, and
government surpluses. Through their approach, one can suggest
a policy that sets the optimal pollution level to maximize social
welfare. An optimal policy is spatially and temporally differentiated
(varies by location over time). Real life has some limitations: lack
of information and implementation capacity may lead to static and
spatially uniform policy; or dynamic but spatially uniform policy; or
static but spatially differentiated policy. Using a simulation based
on data from California, they show that a simple, static, spatially
uniform tax to control pollution results in about half of the gain of the
optimal spatially and temporally differentiated policy, while policies
that vary only by space or only by time may lose a quarter of the gain
compared to the optimal policy. This analysis suggests a significant
gain potential based on policies that can differentiate behavior over
space and time. Certainly, designing such policies requires analytical
tools to be developed.
The use of new remote sensing technologies and improved
computational capacity led to the development of powerful tools that
will eventually improve policies to incorporate variability over space
and time. Spatial econometrics (Irwin and Geoghegan, 2001) modifies
techniques of time series analysis to estimate technical and behavioral
parameters over space and provides a foundation for quantitative
policy making. Usually, there is a significant preference for open space
and aesthetic values (Irwin and Bockstael, 2007), offering room for
land use analysis to extend the von Thünen model to consider both
distance from CBD and land value. One possible pattern of land
use is consistent with maximizing utility from income and aesthetic
value and assuming well-functioning land markets. For instance, the
rich may live in areas with the best views next to the city and
the poor may live near the city but in less appealing locations. Wu
and Plantinga (2003) suggest that open space policies and aesthetic
considerations are crucial for land use management and are likely
to be an immigration magnet. Their analysis suggests that zoning
can be a major mechanism affecting land use, taxation, and institu-
tional arrangements (e.g., land distribution between municipalities).
352 Agricultural Economics and Policy

According to Glaeser et al. (2005), housing shortages do not reflect

physical restraints but are zoning-biased in favor of landlords and
reduce housing supply. Glaeser and Khan (2004) argue that urban
sprawl is mostly a result of technology (car-based living), and its
main deficiencies relate to inequality in access and transportation
use. Their analysis did not, however, consider the environmental
externalities of sprawl (Seto et al., 2011).
The analysis of land use in modern society, especially in the
urban sector, has also discovered the importance of agglomeration
economics, where productivity increases with density and the decline
of the importance of transaction costs, leading to dense downtowns
and urban sprawl (Glaeser and Gottlieb, 2009). Wu (2006) suggests
that the geographic features of various locations, in addition to
economic considerations, may explain changing density and non-
contiguity along the urban–rural fringe and leapfrog development,
segregation, and jurisdictional fragmentation. Newburn and Berck
(2011) present evidence that exurban development, residential areas
outside suburban areas, is comparatively more threatening to farm-
land and the natural environment than are the urban and suburban
areas. They argue that while suburban development is constrained by
water, especially sewer availability, the development of septic systems
in the exurban areas is a major contributor to fragmentation and loss
of intact agricultural and forested areas.
The literature on the evolution and economics of land use pattern
development has benefited from improved empirical capabilities. New
data sources allow using instrumental variables, matching techniques,
differences-in-differences, regression discontinuity design, random-
ized controlled trials to understand individual land use choices in
response to market conditions, policy incentives, and policy and mar-
ket outcomes. Plantinga (2021) reviews recent progress in empirical
studies of land use and suggests that new capabilities could provide
new research opportunities, leading to improved understanding and,
eventually, better policy choices.
Policy choices frequently reflect improved data, understanding of
human behavior, and decision-making rules. The shift of agricultural
policy to address environmental challenges led to the emergence of
 Economics of Space: Interaction of Urban and Rural Sectors 353

research on allocating resources to environmental amenities, which

will be discussed in the following section.

2. The Economics of Payment for Ecosystem

Services (PES)

For a long time, agricultural policies’ main priorities were providing

good and affordable food (consumer surplus), maintaining the well-
being of farmers (producer surplus), and reducing government
costs. This led to traditional agricultural policies that included
land set-aside programs to control supply. Increased concern about
environmental externalities and awareness of the importance of a
sustainable environment for human well-being in the 1970s led to the
introduction of programs with explicit environmental goals. Thus,
the land set-aside program of the U.S. was converted in the 1980s
to become a conservation reserve program, where landowners were
paid to convert their land use from agricultural activities to less
environmentally intensive activities that benefit the environment.
This was a pioneering effort in establishing Payment for Ecosystem
Services (PES). Over time, PES has become global tools for resource
management in agriculture, forestry, and other resource forms (Bulte
et al., 2008; Engel et al., 2008).
Currently, the United States has multiple agricultural PES.1
A prominent example is the Conservation Reserve Program, a land
set-aside program that provides environmental amenities like water
protection, soil erosion reduction, wildlife habitat preservations,
forest and wetland preservation, and enhanced resilience against
natural disasters. Another example is the Environmental Quality
Incentives Program (EQIP), a working land program that pays farm-
ers to improve their production practices. PES programs may change
over time to include support for activities that reduce greenhouse
gas emissions and other objectives. The European Union’s (EU)

1
See Farm Service Agency’s (FSA) conservation program webpage that is
available at: https://www.fsa.usda.gov/programs-and-services/conservation-pro
grams/index (accessed January 19, 2023).
354 Agricultural Economics and Policy

Common Agricultural Policy (CAP) includes programs that aim to

protect against environmental and biodiversity loss and contribute
to climate change mitigation. These programs include payments for
set-aside activities and working land programs. The requirements of
the programs may also change over time to reflect changing food and
agricultural contexts. For instance, the environmental payment and
requirement may decline during tighter food supply situations.
There is growing economic literature on the economics of PES.
One body of literature considers the issue of targeting budgets for
environmental services payments. Babcock et al. (1997) developed a
conceptual framework to associate allocating a budget to purchase
agricultural land for environmental services. The model assumes
that land within a region is heterogeneous (regarding productivity
and environmental amenities). Each parcel of land has a fixed
economic rent per acre (denoted by c, 0 ≤ c ≤ c). Assuming
the environmental agency will pay the farmer the rent, then c
reflects costs per acre for the environmental agency. Each piece
of land also has a fixed environmental quality coefficient per acre
(denoted by e, 0 ≤ e ≤ e). The density distribution of land is
 e=e  c=c
f (c, e), and the total land (A) is e=0 c=0 f (e, c)dcde = A. In
this case, land can be allocated by three criteria. First is cost
targeting: when given a certain amount of budget B, the government
buys all the land whose cost per acre is smaller than c1 such that
 e=e  c=c1
e=0 c=0 c · f (e, c)dcde = B (the expenditure meets the available
budget). The second criterion is environmental benefit maximization:
the government buys all the land above a critical land quality e2
 e=e  c=c
so that e=e2 c=0 c · f (e, c)dcde = B. A third criterion is benefit-
cost ratio maximization: the government buys all the land, where
the benefit-cost ratio is greater than be , and this critical value is
e=e  c=e/b
determined by solving e=0 c=0 e c · f (e, c)dcde = B. In Figure 2,
the areas that correspond to these three criteria are the area below
the horizontal dashed line, the area to the right of the vertical dashed
line, and the area below the sloping dashed line, respectively.
The cost targeting maximizes the acreage but is preferred by
farmers because they are left with their best land. This cost targeting
is also optimal from an environmental perspective if there is a
 Economics of Space: Interaction of Urban and Rural Sectors 355

Figure 2. Targeting of environmental amenities.

full negative correlation between rent and environmental quality.

In this case, buying the cheapest land will maximize environmental
amenities. The environmental benefit maximization may result in
the lowest purchase of land, but it would not maximize aggregate
environmental benefit if some high-environmental-quality land has
very high rent. Therefore, it may be worthwhile to buy cheaper lands,
which together have higher aggregate environmental quality. The
benefit-cost ratio targeting maximizes the environmental benefit for
a given budget and should be considered by environmental agencies.
The analysis presented here is simple and can be expanded. Wu
et al. (2001) consider the case where the industry faces negatively
sloped demand, and differences in rent represent differences in land
productivity. In this case, PES programs have expected negative
consequences, such as slippage: if output demand is inelastic, the
PES program increases agricultural prices leading to the farming
of previously idle lands, potentially leading to a reduction of
environmental benefits. Thus, PES may need to pay for keeping lands
with high environmental services out of production. In this case,
purchasing funds that are the sole buyer of environmental amenities
356 Agricultural Economics and Policy

may pay monopolistic pricing and short-change the landowners.

PES programs also affect multiple agents in addition to government
agencies and landowners. In particular, they affect consumers and
farmworkers. Zilberman et al. (2008) develop a model that considers
heterogeneity in both environmental and economic factors to assess
the impact of PES programs. They argue that these programs
tend to benefit landowners, but land diversion programs (from
agricultural activities) may harm consumers by increasing prices
and farmworkers by reducing employment opportunities. So, there
may be a conflict between environmental protection and equity,
especially in development. However, working land programs where
farmers are paid for improved environmental amenities (more precise
use of chemicals, water conservation, integrated pest management
(IPA)) may benefit consumers by increasing supply and improving
environmental quality. For workers, it may provide more employment
opportunities and thus are desirable from an equity perspective.
Quantifying the benefits and costs of PES programs is a significant
and growing research topic.
The environmental benefit may have an agglomeration effect,
namely that the benefit from an environmental amenity increases
nonlinearly with the amount of land. For example, designing reserves
for wildlife over space may require establishing corridors to allow
wildlife movement. Furthermore, the same piece of land may provide
multiple environmental amenities such as wetland protection, air
quality protection, or wildlife shelter. Wu and Boggess (1999)
developed an extended model to design PES that considers multiple
attributes and agglomeration effects.
Over the years, land purchasing funds moved from targeting
cheap land to using the benefit-cost criteria. The funds are used to
achieve several objectives (e.g., soil erosion reduction, water quality
protection, and wildlife habitat protection). The different features of
parcels of land are quantified, and each feature is assigned a benefit
weight. The benefit per acre is the sum of the benefits of the different
features of the field. Each landowner should state a willingness
to accept for providing environmental services. The environmental
 Economics of Space: Interaction of Urban and Rural Sectors 357

agencies select the plots with the highest environmental benefit per
unit of land cost.
Political economy considerations of PES are important for
designing PES programs. Different regions may benefit differently
from emphasizing one program feature over another. For example,
some regions may benefit from wetland protection, while others may
benefit from improved air quality. So, weight and target of different
features are subject to political negotiation. Furthermore, much
of the environmental amenities are provided by non-governmental
agents that benefit from government policies. For example, non-
governmental organizations (NGOs) like the Nature Conservancy
or World Wildlife Federation purchase environmental land or other
resources with significant amenities, but their purchases may benefit
from preferential tax treatment. Thus, understanding how tax poli-
cies affect land diversion to the agricultural sector is important for
further research.

3. Land Pricing
The previous analysis suggests that land resources provide multiple
benefits to agricultural, rural, urban, municipal, and environmental
sectors. The von Thünen model and the analysis of PES suggest
mechanisms for assessing land rents, namely, payments per unit of
land per unit of time. However, the transition from land rent to land
prices is quite tricky. According to Poterba et al. (1991), the return
on housing investments is equal to a user cost, μ, which is
μ = (1 − θ)(t + r) + δ + m + α − g,
where θ is the marginal tax rate, r is the interest rate, t is the real
estate tax rate, δ is depreciation, m is maintenance, α is the insurance
risk premium, and g is expected capital gain. The ratio of net rent
per acre to the price per acre is equal to the user cost. The net rent
per acre in the von Thünen model is the periodical benefit per unit
of land at a given location minus the user cost. For given agricultural
activities and locations, the price of the land is equal to the net rent
divided by the user cost, which suggests that land values increase
358 Agricultural Economics and Policy

when the interest rate, property tax, depreciation, maintenance costs,

and insurable risk are declining, and when the expected capital
gain is increasing. Just and Miranowski (1993) expanded the model
to incorporate inflation and risk aversion, showing that expected
capital gain, the opportunity cost of capital, and inflation (which
tend to reduce prices) are important explanatory variables for price
variability and profitability in the 1980s and 1990s, which suggests
the importance of monetary policy and macro situations to the farm
sector.
Agricultural land prices are affected directly by factors that affect
agriculture. For example, increasing the price of food and introducing
government programs that assure farm income, ceteris paribus, tend
to increase the price of land. One reason farmers fight strongly
for maintaining government support programs is that it benefits
them directly in each period and enhances their wealth. The price-
rent ratio formula above suggests that changes in taxes on land,
macroeconomic policies that affect interest rates, and expectations
for economic growth that affect land pricing affect the well-being of
the farm sector. As such, farmers are keen on macro-policies. The
link between macro policies and the agricultural sector is an area of
importance (Rausser et al., 1986) although not emphasized in this
chapter.
The land market and the agricultural sector are closely linked to
the energy sector. Wu et al. (2019) argued that over the 2007–2008
oil shock, the high-energy prices devastated the real estate market
and triggered the financial crisis. The oil shock almost doubled the
cost of transportation, thus reduced the willingness to pay for new
houses farther away from cities. Land rent and house values declined
faster with distance, and the expenditure of people who lived farther
away from town increased. Thus, the net value of highly mortgaged
individuals with houses farther from town became negative, and
they declared bankruptcy and moved to towns to reduce their
transportation costs. Many of the home bankruptcies during the
financial crisis occurred in California. For example, home prices
in commuting areas 40+ miles away from San Francisco, including
Tracy, Stockton, and Antioch, declined more than 56 %. The increase
in gas prices between 2005 and 2008 increased the transportation cost
 Economics of Space: Interaction of Urban and Rural Sectors 359

of individuals who paid $4,000 for commuting in 2005 to $7,600 in

2008 annually, and this difference may push people to insolvency and
deter others from wanting to move to these locations. Similar losses
occurred in southern California, resulting in more than $200 billion
in bankruptcies.
As the von Thünen-type analysis suggests, the value of agri-
cultural land is affected by the value of urban land, and shocks
(e.g., financial crises) that reduce the value of residential land affect
agricultural land as well, which suggests that the value of agricultural
land is affected by energy prices through their impact on residential
land. But energy prices also have a direct impact on agriculture. In
some areas, an increase in energy price would also increase the price of
transportation and fertilizers, thus reducing the price of agricultural
land. In other areas — for example, the Midwest — higher energy
prices incurred a search for substitutes (e.g., biofuels) that increased
agricultural crop prices and land prices.

4. Conclusion
This chapter emphasized the importance of urban–rural relation-
ships. The footprint of agriculture and its economy is affected
by the economics and conditions of the adjacent urban center.
Policies that affect the urban sector, in turn, impact rural regions
and agriculture. We found that both urban and rural areas are
affected by fuel and commodity prices as well as various government
policies. In particular, taxation and interest rates, directly and
indirectly, influence land rent and land prices. Zoning plays an
important role in establishing agricultural regions and the borderline
between urban and rural areas. Government policies in agriculture,
specifically payment for ecosystem services, are affecting land use and
income distribution and become an important part of agricultural
policy packages. Analyses of the impact and design of these policies
must recognize heterogeneity across locations and agents and use
new analytical tools that take advantage of spatial information
and advanced computational methods. One point that we must
remember is that continued urban expansion, if not well-managed,
will deprive us of much agricultural land and will worsen climate
360 Agricultural Economics and Policy

change. In conclusion, in order to obtain a full understanding of the

agricultural sector and its environmental implications, one has to
understand urbanization and urban–rural interactions, and there is
much more to discover and research regarding the subject matter.

References
Babcock, B.A., P.G. Lakshminarayan, J. Wu, and D. Zilberman. 1997. Targeting
Tools for the Purchase of Environmental Amenities. Land Economics 73(3):
325–339.
Bulte, E.H., L. Lipper, R. Stringer, and D. Zilberman. 2008. PES and Poverty
Reduction: Concepts, Issues, and Empirical Perspectives. Environment and
Development Economics 13(3): 245–254.
Carey, M., D.A. Sumner, and R.E. Howitt. 2000. The Value of Tradable Credits
for Rice Straw Burning. AIC Issues Brief 12.
Clark, C. 1973. The Value of Agricultural Land. Elsevier.
Engel, S., S. Pagiola, and S. Wunder. 2008. Designing Payments for Environmen-
tal Services in Theory and Practice: An Overview of the Issues. Ecological
Economics 65(4): 663–674.
Glaeser, E.L. and M.E. Kahn. 2004. Sprawl and Urban Growth. In Handbook of
Regional and Urban Economics, Vol. 4, pp. 2481–2527. Elsevier, Amsterdam.
Glaeser, E.L., J. Gyourko, and R.E. Saks. 2005. Why Have Housing Prices Gone
Up? American Economic Review 95(2): 329–333.
Glaeser, E.L. and J.D. Gottlieb. 2009. The Wealth of Cities: Agglomeration
Economies and Spatial Equilibrium in the United States. Journal of Eco-
nomic Literature 47(4): 983–1028.
Hochman, E., D. Pines, and D. Zilberman. 1977. The Effects of Pollution Taxation
on the Pattern of Resource Allocation: The Downstream Diffusion Case. The
Quarterly Journal of Economics 91(4): 625–638.
Irwin, E.G. and J. Geoghegan. 2001. Theory, Data, Methods: Developing Spatially
Explicit Economic Models of Land Use Change. Agriculture, Ecosystems &
Environment 85(1–3): 7–24.
Irwin, E.G. and N.E. Bockstael. 2007. The Evolution of Urban Sprawl: Evidence
of Spatial Heterogeneity and Increasing Land Fragmentation. Proceedings of
the National Academy of Sciences 104(52): 20672–20677.
Just, R.E. and J.A. Miranowski. 1993. Understanding Farmland Price Changes.
American Journal of Agricultural Economics 75(1): 156–168.
Levin, S. and A. Xepapadeas. 2021. On the Coevolution of Economic and
Ecological Systems. Annual Review of Resource Economics 13(1): 355–377.
Newburn, D. and P. Berck. 2011. Exurban Development. Journal of Environmen-
tal Economics and Management 62(3): 323–336.
Plantinga, A.J. 2021. Recent Advances in Empirical Land-Use Modeling. Annual
Review of Resource Economics 13: 1–15.
 Economics of Space: Interaction of Urban and Rural Sectors 361

Poterba, J.M., D.N. Weil, and R. Shiller. 1991. House Price Dynamics: The Role
of Tax Policy and Demography. Brookings Papers on Economic Activity 2:
143–203.
Rabotyagov, S.S., C.L. Kling, P.W. Gassman, N.N. Rabalais, and R.E. Turner.
2014. The Economics of Dead Zones: Causes, Impacts, Policy Challenges,
and a Model of the Gulf of Mexico Hypoxic Zone. Review of Environmental
Economics and Policy 8(1): 58–79.
Rausser, G.C., J.A. Chalfant, H.A. Love, and K.G. Stamoulis. 1986. Macroe-
conomic Linkages, Taxes, and Subsidies in the US Agricultural Sector.
American Journal of Agricultural Economics 68(2): 399–412.
Samuelson, P.A. 1983. Thünen at Two Hundred. Journal of Economic Literature
21(4): 1468–1488.
Seto, K.C., M. Fragkias, B. Güneralp, and M.K. Reilly. 2011. A Meta-Analysis
of Global Urban Land Expansion. PloS One 6(8): e23777.
Seto, K.C., B. Güneralp, and L.R. Hutyra. 2012. Global Forecasts of Urban
Expansion to 2030 and Direct Impacts on Biodiversity and Carbon Pools.
Proceedings of the National Academy of Sciences 109(40): 16083–16088.
Xabadia, A., R.U. Goetz, and D. Zilberman. 2008. The Gains from Differentiated
Policies to Control Stock Pollution When Producers are Heterogeneous.
American Journal of Agricultural Economics 90(4): 1059–1073.
Wu, J.J. 2006. Environmental Amenities, Urban Sprawl, and Community Char-
acteristics. Journal of Environmental Economics and Management 52(2):
527–547.
Wu, J.J. and W.G. Boggess. 1999. The Optimal Allocation of Conservation Funds.
Journal of Environmental Economics and Management 38(3): 302–321.
Wu, J.J. and A.J. Plantinga. 2003. The Influence of Public Open Space on Urban
Spatial Structure. Journal of Environmental Economics and Management
46(2): 288–309.
Wu, J.J., S. Sexton, and D. Zilberman. 2019. Energy Price Shocks, Household
Location Patterns and Housing Crises: Theory and Implications. Energy
Economics 80: 691–706.
Wu, J.J., D. Zilberman, and B.A. Babcock. 2001. Environmental and Distribu-
tional Impacts of Conservation Targeting Strategies. Journal of Environmen-
tal Economics and Management 41(3): 333–350.
Zilberman, D., L. Lipper, and N. McCarthy. 2008. When Could Payments for
Environmental Services Benefit the Poor? Environment and Development
Economics 13(3): 255–278.
This page intentionally left blank
 Chapter 13

Climate Change and Agriculture

Climate change is humanity’s most challenging issue, and it is

changing agriculture and agricultural policy worldwide. There is a
large body of research addressing climate change, and every few
years, a report is published by the Intergovernmental Panel on
Climate Change (IPCC) to assess scientific findings related to climate
change.1 In this chapter, we present some of the modeling and policy
issues, emphasizing the impact of climate change on agriculture
and the possible role of agriculture in adaptation and mitigation.
Adaptations are changes in behavior that will allow us to withstand
the impact of climate change, and even prosper. Effective adaptation
builds resilience. Mitigation involves activities that will allow to slow
or even reverse the buildup of greenhouse gases (GHGs). Agriculture
and food systems must adapt to climate change, and this can be
a major contribution to its mitigation. Adaptation and mitigation
are substitute activities. Mitigation may slow the pace of climate
change and reduce the high cost of adaptation. It will be useful
for policymakers to be able to quantify the cost-optimal mitigation
strategies given the level of mitigation and optimize adaptation and
mitigation strategies.

1
For instance, the most recent report, the Sixth Assessment Report, can be found
at: https://www.ipcc.ch/assessment-report/ar6/ (accessed October 21, 2023).

363
364 Agricultural Economics and Policy

After a brief review of the impact of climate change on agri-

culture, this chapter will address how policies, both direct and
indirect, can affect how agriculture will adapt and how agriculture
can contribute to mitigating climate change. The first section of the
chapter will assess the impact of climate change on agriculture, and
then we will present several perspectives on how agriculture will
adapt to climate change. We will consider impacts and adaptation
in the context of individual markets, as well as supply chains for
mitigation. The final section will discuss changes in policy and
perception that are important to address climate change.

1. Impact of Climate Change on Agriculture

The IPCC reports list the effects of climate change, including shift
of climate from the tropics to the poles, rising sea levels, greater
likelihood of extreme weather events, melting snowpack, and more
(Rosenzweig and Tubiello, 2007). These impacts will damage ecosys-
tems, increase human vulnerability, and increase political instability.
Most models predict that warmer climates will shift from the
tropics to the poles, and temperatures will rise. More concretely,
we could see the environmental conditions in Mexico appear in
California, while that in California could shift to Oregon. Most of
Texas and Oklahoma are set to become deserts, but parts of Canada
could become agriculturally viable. Similarly, there could be areas
in Russia and northern Europe that are also viable for agriculture,
but more southern areas will become uninhabitable. Expansion of
the Sahel could expand political instability in the region.
Historically, those working in agriculture have been able to adapt
to certain conditions. Farmers will change the input and switch
crops. Water systems will be redesigned and reconstructed. Areas
close to the equator will be deserted, and some near the poles will
be farmed. Therefore, it is important to distinguish between the
aggregate and regional effects. Lu, Buchsbaum, and Zilberman (2021)
develop a conceptual model to assess the impact of climate change
on agriculture. The model assumes heterogeneity across locations,
where l is the locational variable, varying from 0 (e.g., equator)
 Climate Change and Agriculture 365

to L (e.g., the North Pole). They assume that only two crops are
available, and i is the crop indicator, which is assuming the value
1 for a warmer-temperature crop (e.g., corn) and 2 for a lower-
temperature crop (e.g., wheat). The production function per unit
of land at time t for crop i at location l is yit (l) = f (h(st , l), t),
where yit is output, h(st , l) is the temperature, which depends on
the time t carbon stock st and location l. It is assumed that output
increases with technological change, which, itself, increases over time.
Output first increases and then decreases in temperature (thinking
of extreme heat that causes crop damages). Temperature increases
with the stock of carbon st and declines (for a given st ) as l increases
(i.e., locations get closer to the poles). At each location, a producer
must allocate land between the two crops. Let δit denote the land
share of crop i such that δ1t (l) + δ2t (l) ≤ 1.
We assume that producers aim to maximize profits; profits per
unit of land at location l are given by:

πt (l) = δ1t (t)[p1t y1t (l) − c1 (y1t (l)) − T1 (t)]

+ δ2t (t)[p2t y1t (l) − c2 (y2t (l)) − T2 (t)],

where p1t and p2t are prices of goods 1 and 2 at period t, c1 and
c2 are the cost of production per acre as functions of output. We
assume that marginal cost is positive and increasing in output and
that there are transition costs from increasing the share of crop
i, at period t. We denote this transition cost per unit of land as
Ti (t). So, land allocation at period t and location l is determined
by choosing the land share of the technologies and their output
per acre to maximize profits per acre, πt (l). This is a micro-land
allocation problem. By aggregating over all land to obtain supply
to each of the crops, as equated with demand, one can solve a
macro-problem that determines prices and aggregate quantity of the
two crops.
The solution to the micro and macro problems is discussed by
Lu et al. (2021). Their analysis suggests that climate change will
invoke a transition away from the equator and toward the poles. Some
lands near the equator will be deserted, and some lands near the poles
will be reclaimed. Because of the transition costs, there may be a
366 Agricultural Economics and Policy

delay in the transition to a new land that may affect supply. However,
their analysis suggests that the total agricultural productivity on a
global level may not necessarily decrease with climate change after
some adjustments, as production in the areas in Canada and Russia
may replace lost productivity in Oklahoma and China. Furthermore,
if technology continues to increase resiliency, there might not be a loss
in production. The conceptual model thus suggests that total food
production may not decrease under a changing climate. However,
it may have very severe negative distributional effects; there may
be new opportunities in Canada and Russia, but domestic markets
in Mexico and several African countries will be damaged. So, the
challenge of climate change is how to deal with these severe spatially
heterogeneous effects.
There are several significant bodies of literature on the impact
of climate change on agriculture. The Ricardian approach, pioneered
by Mendelsohn et al. (1994) assesses the impact of climate change on
yield and cost over space (translated as land rent per unit of land),
suggesting the aggregated effect is not as big as the distributional
effect. However, the approach was criticized for ignoring irrigation
and other effects and a new econometric literature emerged, esti-
mating the impact of climate change on crop yields (Schlenker et al.,
2005; Schlenker and Roberts, 2009; Deschênes and Greenstone, 2007;
Fisher et al., 2012; Burke and Emerick, 2016; Miao et al., 2016;
Malikov et al., 2020).
Carter et al. (2018) connect the past performance of crop
breeding with restricted capacity to indicate that technological
innovation may overcome much of the yield losses caused by rising
temperatures. Ortiz-Borbea et al. (2021) find that climate change
might have reduced agricultural productivity gains in agriculture by
21%, but the losses are more substantial in developing countries.
Since the utilization of agricultural biotechnology and other advanced
technology in the developing world has been limited, there seems to
be a significant possibility of maintaining and improving yield despite
the effects of climate change in many regions, and this, combined with
new farming opportunities in cold areas, suggest that aggregate food
supply is much less of a problem associated with climate change than
the significant losses in different locations.
 Climate Change and Agriculture 367

While the effects of changing environmental conditions directly

on agriculture are spatially determinant, the expansion of areas hab-
itable for pests paints a darker picture. Skendžić et al. (2021) suggest
that climate change would increase the range of multiple insects,
the transmission of plant diseases, and the risk of overwintering the
survival of insects and their invasiveness. Adaptation will require fast
development and modification of pest control strategies.
Assessment of the impact of climate change on agriculture is
a multistage process. One approach is to assess impacts on yield
distributions at various locations, which depends on climatic scenar-
ios, technology improvement, and policies. Yield changes will affect
the profitability of different crops, potentially resulting in changes
in land use, supplies of crops, and input demands. These changes, in
turn, will affect prices and welfare distribution among groups. Several
alternative modeling strategies to assess these impacts are reviewed
by McCarl and Hertel (2018).
Yield effects were estimated econometrically and through simula-
tions. Blanc and Schlenker (2017) review econometric studies using
large panel data. These studies have identified nonlinear response
functions that may indicate significant yield loss as temperatures
increases. Lobell et al. (2011) suggest that global maize and wheat
yields may decline significantly with climate change, eliminating
the gains from technological improvement. However, the role of
temperature in relation to soil humidity on yield is subject to
ongoing research: a recent study by Lobell et al. (2020) docu-
ments that U.S. corn yield has become increasingly sensitive to
drought, citing the increased plant density as one of many possible
reasons.
Several papers present more ways agricultural economics can
function with other fields of science to study the impact of climate
change on agriculture. For instance, agricultural economists can work
with scientists and obtain their assessment of yields, technologies,
and pest dynamics, incorporate these impacts on supply to generate
supply functions, and then intersect it with demand relationships to
assess the overall effect. Adams et al. (1990) presented an early model
integrating relationships from atmospheric science, plant science, and
agricultural economics, along with the impact of climate change on
368 Agricultural Economics and Policy

temperature and carbon, suggesting that land-use patterns in the

United States will shift with a growing reliance on irrigation, and
the overall effects depend on the specification of the model. Fei et al.
(2017) develop a simulation relying on multiple models of the impact
of climate change on yields under different conditions, as well as
a model of the grain market, indicating the relocation of wheat
production could increase the relative share of soybean compared
to corn, and reduce the acreage of cotton. Beach et al. (2015)
combine economics, agronomy, and forestry simulations to assess the
impact of climate change in the U.S., comparing scenarios where
climate change is mitigated to those where it is unabated. The results
suggested that there could be gains from enacting mitigation policies
(e.g., carbon sequestration), particularly for agriculture, forestry, and
ranchlands.
There is also plenty of literature concerning sea-level rise, an
important consideration with vulnerabilities to coastal agriculture.
For example, much of the population in Bangladesh and Pakistan is
vulnerable to floods, and this vulnerability will drastically increase
from climate change. Sea water intrusion will affect the land supply
and groundwater supply. Global rice production could decline by
3%, increasing prices substantially (Chen et al., 2012). Melting snow
pacts mean that water storage is even more crucial: irrigation water
supply stands to decrease by 40% in areas such as Central Asia,
the Western U.S., and the Andes (Qin et al., 2020). Mitigation to
confront these effects includes building dams to contain and store
runoff water.

2. Adaptation to Climate Change

Climate change will increase the likelihood of extreme events, includ-
ing floods, storms, and typhoons. Thus, climate change adaptation
will have to consider not only responses to average scenarios but also
how to improve resiliency to extreme events. Mechanisms to protect
against these events are crucial, such as dikes, grain silos, and even
emergency response plans. The development of these mechanisms
and capacities is a multidisciplinary effort where economists can gain
 Climate Change and Agriculture 369

input from other disciplines, and it is one of the challenges of future

research.
Predictive capacity for weather and climate is another step
to resiliency in the face of climate change, as one of the main
challenges of engaging in both adaptation and mitigation strategies
is uncertainty regarding the impacts of climate change at the macro
and micro levels. Lemoine and Rudik (2017) develop a recursive
integrated assessment model to develop decision trees and make
decisions that involve major uncertainties at the macro level. At the
micro level, being able to assess when and where snowmelt will occur
is important to discover alternative water supplies. In this sense,
increased resilience is a result of policy geared toward protecting and
evacuating victims when the situation arises.
Fankhauser (2017) reviews some of the economic modeling
approaches that are taken to assess adaptation strategies. An
integrated assessment model that represents both the economic and
biophysical conditions associated with a changing climate can be
a starting point for assessing adaptation strategies, and they can
provide a background on the conditions that require adaptation, as
well as an assessment of changes in economic and climatic conditions
over time (Kling et al., 2017).
Econometric analysis is important to assess individual behavioral
responses to different types of shocks and opportunities. For example,
one could figure out how likely certain people are to adopt some
technology within different locations, as well as the factors that
shift the capacity to adopt. Burke and Emerick (2016) employ
a long-differences approach to study the adaptation within U.S.
agriculture; using corn as an example, they found that the long-run
responsiveness of corn yield to overheating temperature changes is
not statistically different from the short-run responsiveness. They,
therefore, conclude that there is no clear adaptation in the corn
yield to climate change. Yu et al. (2021) expand the long differences
approach by allowing the responsiveness to differ over two periods far
apart (e.g., 50 years) and find that there is a significant adaptation of
U.S. corn and soybean yields to overheat temperature over the past
six decades. Zhang et al. (2023) further expand the long differences
370 Agricultural Economics and Policy

approach by allowing the responsiveness of crop yield to climate to

vary across both time and space. They find that for corn, about half
of the studied counties experience adaptation while the other half
experience no adaptation or even maladaptation to climate change.
On a more aggregated level, it is important to be able to use
an economy-wide simulation model that may include general equi-
librium models and programming models. Khanna and Zilberman
(2012) demonstrate how different models that may address different
degrees of detail can be used together to assess the impact of policies;
in their case, land use changes that are associated with policies that
aim to reduce greenhouse gas emissions. When it comes to specific
situations, various decision-making tools have been proposed to
address choices in specific situations, considering both economic and
technological uncertainties. There are several microeconomic tools
to assess individual choices, including cost-benefit analysis, cost-
effectiveness analysis, as well as others to assess different dimensions
of the impact of adaptation policies (Fankhauser, 2017). These
calculations also distinguish between choices in the private and public
sectors that point toward future collaboration, and the constraints
on lower-income individuals with limited resources and those who are
either climate deniers or have negative attitudes toward adaptation
(Fankhauser, 2017).
There are further considerations that have to be taken into
account for large-scale individual behavior: large-scale geoengineer-
ing that modify the vulnerability of continents requires international
agreements and thus may result in conflict (Schneider, 1996). In this
sense, geoengineering should be an extreme measure, a last resort
that may be utilized only when mitigation has not been effective
(Brown, 2010).
Zilberman et al. (2012) identify several strategies for adaptation
to climate change in agriculture. First is investment and support
for research and development that result in technological and
institutional innovations. Second, once appropriate solutions for tech-
nologies are identified, mechanisms to enhance the adoption of new
technologies and crops should be introduced. Frequently, adaptation
to climate change may require shifting of crops across regions, as
 Climate Change and Agriculture 371

well as the development of new crop varieties that can overcome

some of the challenges of climate change (e.g., drought-tolerant
varieties). A third strategy that may occur when economic variability
is limited is migration, a drastic strategy. The development of an
international environment that will enable migration is a major
challenge and the global community needs to develop mechanisms
to enable migration and relocation of climate refugees. On this line,
institutions have to focus on the changes stemming from shifting
economic and trade activities, which would allow capitalizing on new
conditions in response to climate change (Dellink et al., 2017).
Regarding the first strategy mentioned above (i.e., innovations),
a recent study by Moscona and Sastry (2023a) shows that agri-
cultural innovation reduces the negative impact of climate change
on U.S. agriculture by only 20%, indicating that innovation should
not be relied upon as a sole approach of adaptation. Moreover,
technological innovations mainly occur in developed countries, and
those adaptations may not transfer well to developing countries.
In another paper, Moscona and Sastry (2023b) show that this
technology mismatch reduces global agricultural productivity by 58%
while increasing inequality of productivity by 15%. It is crucial,
therefore, to focus on R&D in developing countries to avoid such
disparities.
The build-up of large inventories of commodities and the capacity
to utilize them is another useful mechanism to stabilize uncertain
supply situations, especially in agriculture (Wright, 2012). Insurance
mechanisms for farmers and intermediaries can allow investment,
stabilize income, and prevent some of the instability associated with
climate change, but the design of a solid insurance program is a major
challenge, and insurance may even hinder innovation (Surminski
et al., 2016; Miao, 2020). Additionally, consumers and producers who
rely on supply chains will be even more vulnerable when adaptation
is stunted (Reardon and Zilberman, 2018). For example, frequent
droughts or floods may harm affected regions, as well as regions
that export products to producers in the affected regions. Therefore,
suppliers may develop diversified sourcing and enhanced inventory.
However, the capacity to adapt will vary among different economic
372 Agricultural Economics and Policy

agents, and some groups may become more vulnerable to shocks. In

these cases, transportation and infrastructure will have to change
to include safeguards in the event of a shock. Finally, adaptation
strategies should recognize the likelihood that major disaster events
may occur, and the development of emergency aid is an important
strategy for climate change adaptation.

3. Mitigation
Agriculture is a major contributor to climate change, especially
through emissions of carbon and non-carbon dioxide GHGs, such as
methane and nitrous oxide. Agriculture can contribute to mitigation
by reducing the emission of these GHGs, providing cleaner alter-
natives to fossil fuels, and by sequestering carbon through various
agricultural and biological practices.
Paustian et al. (2001) suggest three avenues for agricultural
mitigation of GHGs. First, sequestering carbon by moving to low
and non-tillage practices as well as growing more environmentally
friendly cover crops. They furthermore suggest that grazing practices
can be enhanced by using improved species, sowing legumes, fertil-
izing, and irrigating. Other activities for carbon sequestration are
restoring degraded soils and ecosystems, reforesting and afforesting,
retiring marginal land through programs such as the conservation
reserve program, and controlling desertification. Second, farmers
could reduce nitrous oxide emissions through more precise nitrogen
application, planting genetically improved varieties, and the use
of cover crops, and better management of animal and food waste
to reduce leakages, involving the storage of wastes anaerobically.
Third, methane emission could be reduced through fermentation
from livestock manure, reduction of waste of livestock, and residue
burning. This may require technology like introducing new varieties,
changing feeding ratios, changing waste management, and improving
storage practices. Production of biofuels is another important avenue
for reducing GHG emissions as they replace more carbon-intensive
fossil fuels.
 Climate Change and Agriculture 373

Rosenzweig and Tubiello (2007) suggest that carbon sequestra-

tion can be an effective mitigation strategy that prevents carbon
emissions from agriculture and restores soil carbon that was emitted
in the past. Regarding sequestration, the strategy addresses the one-
third of carbon dioxide emissions in the atmosphere that come from
land use (Lal, 2004), but it may only capture up to half of this total.
Therefore, mitigation has to be considered in relation to adaptation
(Rozenweig and Tubiello, 2007). However, farmers themselves may
stand to gain through mitigation strategies, as they often involve
more efficient yields.
Introducing all these new technological changes requires sig-
nificant policy changes and incentives to modify behavior, such
as changes in land use practices, adoption of new technologies,
and improved management of waste products from animal and
agricultural production. Policies must be incorporated at multiple
levels—the national and international levels—since GHGs are global
externalities that will not be managed correctly unless some mutual
agreement, monitoring, and accounting exist. One key element
to successful policy is full GHG accounting from all agricultural
activities, both in terms of land use, crop production, and animal
waste. This calls for the use of the tools of life cycle analysis
that can and should be adjusted to incorporate economic principles
(Rajagopal et al., 2017). Effective emission accounting needs to
be able to assign responsibility for emissions to the source: there
has to be a way to track carbon fees and subsidies and evaluate
compliance with regulation. Because many agricultural emissions
are non-point sources, mechanisms have to point to observable
behavior. For example, one may estimate greenhouse gas emissions
based on known acreage, crop production, and technologies employed
on-farm.
One of the main challenges of policies to reduce greenhouse gas
emissions is dealing with additivity issues, especially permanence.
A farmer may move to non-tillage, and their land may sequester car-
bon for multiple years, but changes in prices or other conditions may
lead a farmer to de-plow the field and emit much of the sequestrated
carbon. Incentives can be important to reducing the risk of deplowing
374 Agricultural Economics and Policy

and retaining the appeal of sequestration activities for landowners

(Thamo and Pannell, 2016). Feng et al. (2002) develop a dynamic
model to investigate the optimal path of emission sequestration
and carbon stocks. They suggest efficient carbon sequestration with
three mechanisms: a pay-as-you-go system, contracting, and carbon
annuities. Van’t Veld and Plantinga (2005) suggest that the timing
of sequestration may be affected by the price of carbon, and when
one expects carbon price to decrease over time, conservation will
be delayed. Another challenge is to make sure individuals get paid
for GHG mitigation activities that would not have been done rather
than activities that would have been performed anyway (Konidari
and Mavrakis, 2007).
McCarl and Schneider (2001) develop a mathematical program-
ming model that relies on multidisciplinary data to assess the costs
and impacts of greenhouse gas mitigation in agriculture and forestry
in the U.S. They show that optimal policies depend on the price of
carbon. Their analysis incorporates the agricultural crop, livestock,
and forestry sectors. They find that local strategies to reduce carbon
mitigation include carbon sequestration, afforestation, and livestock-
related non-carbon emissions. High-price carbon mitigation may lead
to changes in forestry practices and biofuels. Van Meijl et al. (2018)
project that, by assessing integration, partial equilibrium, biophysical
crop models, and computer-generated equilibrium models, there
would be a negative but small impact on agriculture by 2050.
However, mitigation strategies that aim to contain temperature
increases by 2◦ C have a significant impact on global agricultural
production. Their results vary based on modeling, basic economic
assumptions, and policy.
Climate change mitigation may play a major role in agriculture
in the future, depending on policies. Different mitigation strategies
may be introduced at different locations, and mitigation strategies
will change land use patterns as well as patterns of trade. Developing
the capacity to assess the impact of mitigation is a major challenge
to economists and will require further development of predictive
multidisciplinary models.
 Climate Change and Agriculture 375

4. Climate Smart Agriculture

The current centuries have witnessed both the unraveling of the
Kyoto Protocol and the limitations of the Paris Climate Accords.
There has been a realization, then, that environmental adaptation
does not take precedence over all societal factors. Climate-smart agri-
culture (CSA), which harmonizes the development of food systems,
especially in developing countries (Lipper et al., 2014), with other
objectives by having incentive-compatible mechanisms (Groves and
Ledyard, 1987), will really be smart in achieving its objectives.
The notion of CSA led to a research agenda that aimed to develop
technological solutions and policy tools that result in changes in
agricultural systems that will both enhance productivity and well-
being, as well as address climate change. The notion of climate-
smart agriculture and its associated policies have evolved with
climate change policies. As far back as the Kyoto Protocol, CSA
was offering a framework for projects in developing countries that
could earn the support of the people and governments there. Over
time, climate-smart agriculture activities emphasize both adaptation
and mitigation activities that provide new income opportunities
and growth. It entails researching, utilizing, and adopting new
technologies that provide new opportunities to address climate
change and contribute to sustainable growth. Thus, CSA is con-
sistent with the utilization of modern biotechnologies and other
advanced technologies in developing countries around the world
(Lipper et al., 2017).
A key element of CSA is the development of innovation and
associated supply. However, there exist several obstacles to imple-
mentation, and policies have evolved to overcome these constraints.
A lack of technical knowledge on the impact of climate change and
the properties of different solutions requires investment in R&D
and mechanisms for technology transfer across locations. Access to
technologies may be hampered by intellectual property, and thus
it is useful to develop mechanisms that would allow special access
to technologies, especially those that serve developing countries
376 Agricultural Economics and Policy

(Graff and Zilberman, 2001). Another constraint is regulation:

precaution is often warranted, but regulation has to balance benefits
with the risks in order to avoid missing the adoption of beneficial
technologies. Adoption of some of the new climate-smart solutions
needs to enhance community participation, and therefore to develop
technologies that fit the needs of the community. Solutions require
finance, and because of the positive externalities associated with
CSA, it is important to have targeted farms, and the global-social
benefit will be considered so that the cost of finance will be reduced.
As noted previously in the chapter, effective solutions need to account
for barriers to adoption, so it is useful to integrate incentives such
as including extension activities, demonstrations, and discounted
pricing that will accelerate adoption and lead to learning by doing
and learning by using (Lipper et al., 2017).
Adaptation to climate change requires a new environmental
thinking and framework. The traditional thinking of environmental
groups is to preserve, protect, and conserve. However, in a changing
world, one must transition to methods of controlling climate change
and sustaining the improvements to human wellbeing (Gates, 2022).
Preservation is challenging in an evolving world, and we need to find
a new balance of change with environmental protection. One must
recognize that different priorities in the developing world result in
diverging policies. While the developing world is interested in food
security and increasing well-being, the developed world places more
emphasis on preservation. CSA represents a strategy that enforces
mitigation while maintaining preservation ideals, and it embraces
new technologies that may enable sustainable development (Belton
et al., 2020).

5. Conclusion
Climate change will dramatically affect agricultural systems, so
developing policies and technologies to address climate change should
be a major emphasis of environmental policy research. This research
should assess the impacts of climate change on agriculture, identify
alternative mechanisms for adaptation and mitigation, and recognize
 Climate Change and Agriculture 377

that the solutions to climate change need to be achieved using

climate-smart policies that address other societal issues.
Climate change consideration emphasizes the development and
adoption of innovations that allow adaptation and mitigation. They
include drought and heat tolerance, pest control strategies, tech-
nologies that enhance input-use efficiency of water and other inputs,
mechanisms to quickly identify and respond to changes in weather,
mechanisms that lengthen shelf life, and resilience of livestock to
respond to weather conditions. In addition to technological innova-
tion, there is a need to develop institutions to facilitate peaceful
migration and relocation in response to climate change, mechanisms
to relocate rights to natural resources that will become productive
as the weather is changing, institutions that will allow low-cost
flood protection and storage facilities (Lipper et al., 2017). Many
of these new technologies require the use of modern biotechnologies
and innovations. In some cases, these technologies will lead to drastic
changes in land use and may affect ecosystems, sometimes negatively,
at least in the short run.
Several promising technological directions have been identified to
address climate change, including carbon capture and sequestration,
battery storage systems, improved air conditioning, solar microgrids,
nuclear energy, and new porous materials. These technologies have
mostly relied on breakthroughs in physics, but the 21st century is
likely to enjoy new capabilities developed by the life sciences and
biology. These solutions include modern biotechnologies, like GMOs
and CRISPR, that develop new traits to accelerate photosynthesis,
fix nitrogen, and produce alternative meats. Algae can play a major
role in sequestering carbon, insects can provide new sources of protein
and minimize waste and thus reduce emissions, soil and plants
can sequester carbon, compost can serve as a carbon sink, biofuels
can provide fuel for airlines, trucks, and boats, especially for the
reduction of emissions associated with the production of nitrogen,
and lumber can be used for housing. Finally, wind and solar energy
represent further solutions, but they are likely to compete with
agriculture for land. Pascaris (2021) suggests that introducing solar
energy in rural locations encounters objection because of its impact
378 Agricultural Economics and Policy

on land use, but developing technology and policies that enhance

cooperation between energy and agriculture may increase income and
are part of smart government policies (Miao and Khanna, 2020).
Developing new technological solutions requires significant insti-
tutional and policy change, challenging agricultural and environmen-
tal policy research. This research should have multiple priorities.
First, assess the benefit of new technologies combining economic
and scientific knowledge. Second, design solutions that allow efficient
resource allocation, inducing economic agents to pay the price
of carbon emissions. Third, understand public acceptance of new
technologies and develop new policies that enhance the adoption
of socially desirable technologies. Fourth, understand the political
economy of climate change and assist in developing international
and domestic mechanisms to address the issues of climate change
in general and especially in climate change and natural resources.
Fifth, better understand the operation of the supply chain in natural
resources and develop mechanisms that lead to a more efficient
market and supply chain (Rausser and Zilberman, 2022).

References
Adams, R.M., C. Rosenzweig, R.M. Peart, J.T. Ritchie, B.A. McCarl, J.D. Glyer,
R.B. Curry, J.W. Jones, K.J. Boote, and L.H. Allen. 1990. Global Climate
Change and US Agriculture. Nature 345: 219–224.
Beach, R.H., Y. Cai, A. Thomson, X. Zhang, R. Jones, B.A. McCarl, A. Crimmins
et al. 2015. Climate Change Impacts on US Agriculture and Forestry:
Benefits of Global Climate Stabilization. Environmental Research Letters
10(9): 095004.
Belton, B., T. Reardon, and D. Zilberman. 2020. Sustainable Commoditization
of Seafood. Nature Sustainability 3(9): 677–684.
Blanc, E. and W. Schlenker. 2017. The Use of Panel Models in Assessments of
Climate Impacts on Agriculture. Review of Environmental Economics and
Policy 11(2): 258–279.
Brown, M.A. 2010. Policy Update: The Multiple Policy Dimensions of Carbon
Management: Mitigation, Adaptation and Geoengineering. Carbon Manage-
ment 1(1): 27–33.
Burke, M. and K. Emerick. 2016. Adaptation to Climate Change: Evidence
from US Agriculture. American Economic Journal: Economic Policy 8(3):
106–140.
 Climate Change and Agriculture 379

Carter, C., X. Cui, D. Ghanem, and P. Mérel. 2018. Identifying the Economic
Impacts of Climate Change on Agriculture. Annual Review of Resource
Economics 10: 361–380.
Chen, C.C., B.A. McCarl, and C.C. Chang. 2012. Climate Change, Sea Level Rise
and Rice: Global Market Implications. Climatic Change 110(3–4): 543–560.
Dellink, R., H. Hwang, E. Lanzi, and J. Chateau. 2017. International Trade
Consequences of Climate Change. OECD Trade and Environment Working
Papers, OECD Publishing.
Deschênes, O. and M. Greenstone. 2007. The Economic Impacts of Climate
Change: Evidence from Agricultural Output and Random Fluctuations in
Weather. The American Economic Review 97(1): 354–385.
Fankhauser, S. 2017. Adaptation to Climate Change. Annual Review of Resource
Economics 9: 209–230.
Fei, C.J., B.A. McCarl, and A.W. Thayer. 2017. Estimating the Impacts of
Climate Change and Potential Adaptation Strategies on Cereal Grains in
the United States. Frontiers in Ecology and Evolution 5: 62.
Feng, H., J. Zhao, and C.L. Kling. 2002. The Time Path and Implementation of
Carbon Sequestration. American Journal of Agricultural Economics 84(1):
134–149.
Fisher, A.C., W.M. Hanemann, M.J. Roberts, and W. Schlenker. 2012. The
Economic Impacts of Climate Change: Evidence from Agricultural Output
and Random Fluctuations in Weather: Comment. American Economic
Review 102(7): 3749–3760.
Gates, B. 2022. How to Avoid a Climate Disaster: The Solutions We Have and
the Breakthroughs We Need. Penguin Random House LLC, New York.
Graff, G., and D. Zilberman. 2001. Towards an Intellectual Property Clearing-
house for Agricultural Biotechnology. IP Strategy Today 3: 1–13.
Groves, T. and J. Ledyard. 1987. Incentive Compatibility since 1972. Information,
Incentives, and Economic Mechanisms: Essays in Honor of Leonid Hurwicz
84(2): 48–111.
Khanna, M. and D. Zilberman. 2012. Modeling the Land-Use and Greenhouse-
Gas Implications of Biofuels. Climate Change Economics 3(3): 1250016.
Kling, C.L., R.W. Arritt, G. Calhoun, and D.A. Keiser. 2017. Integrated
Assessment Models of the Food, Energy, and Water Nexus: A Review and
an Outline of Research Needs. Annual Review of Resource Economics 9:
143–163.
Konidari, P. and D. Mavrakis. 2007. A Multi-Criteria Evaluation Method
for Climate Change Mitigation Policy Instruments. Energy Policy 35(12):
6235–6257.
Lal, R. 2004. Soil Carbon Sequestration Impacts on Global Climate Change and
Food Security. Science 304(5677): 1623–1627.
Lemoine, D. and I. Rudik. 2017. Steering the Climate System: Using Inertia to
Lower the Cost of Policy. American Economic Review 107(10): 2947–2957.
Lipper, L., P. Thornton, B.M. Campbell, T. Baedeker, A. Braimoh, M. Bwalya,
P. Caron et al. 2014. Climate-Smart Agriculture for Food Security. Nature
Climate Change 4(12): 1068–1072.
380 Agricultural Economics and Policy

Lipper, L., N. McCarthy, D. Zilberman, S. Asfaw, and G. Branca. 2017. Climate

Smart Agriculture: Building Resilience to Climate Change. Springer Nature.
Lobell, D.B., J.M. Deines, and S.D. Tommaso. 2020. Changes in the Drought
Sensitivity of US Maize Yields. Nature Food 1, 729–735.
Lobell, D.B., W. Schlenker, and J. Costa-Roberts. 2011. Climate Trends and
Global Crop Production since 1980. Science 333(6042): 616–620.
Lu, X., J. Buchsbaum, and D. Zilberman. 2021. Economics of the Impact of
Climate Change on Agriculture, F. Castillo, M. Wehner and D.A. Stone
(Eds.), Extreme Events and Climate Change: A Multidisciplinary Approach.
Elsevier, Amsterdam.
Malikov, Emir, R. Miao, and J. Zhang. 2020. Distributional and Temporal
Heterogeneity in the Climate Change Effects on U.S. Agriculture. Journal
of Environmental Economics and Management 104: 1–10.
McCarl, B.A. and T.W. Hertel. 2018. Climate Change as an Agricultural
Economics Research Topic. Applied Economic Perspectives and Policy 40(1):
60–78.
McCarl, B.A. and U.A. Schneider. 2001. Greenhouse Gas Mitigation in US
Agriculture and Forestry. Science 294(5551): 2481–2482.
Mendelsohn, R., W.D. Nordhaus, and D. Shaw. 1994. The Impact of Global
Warming on Agriculture: A Ricardian Analysis. The American Economic
Review 89: 753–771.
Miao, R. 2020. Climate, Insurance, and Innovation: The Case of Drought and
Innovations in Drought-Tolerant Traits in U.S. Agriculture. European Review
of Agricultural Economics 47(5): 1826–1860.
Miao, R. and M. Khanna. 2020. Harnessing Advances in Agricultural Technologies
to Optimize Resource Utilization in the Food-Energy-Water Nexus. Annual
Review of Resource Economics 12: 65–85.
Miao, R., M. Khanna, and H. Huang. 2016. Responsiveness of Crop Yield and
Acreage to Price and Climate. American Journal of Agricultural Economics
98(1): 191–211.
Moscona, J. and Sastry, K. 2023a. Does Directed Innovation Mitigate Climate
Damage? Evidence from US Agriculture. Quarterly Journal of Economics
138(2): 637–701.
Moscona, J. and Sastry, K. 2023b. Inappropriate Technology: Evidence from
Global Agriculture. Working Paper. Harvard University.
Ortiz-Bobea, A., T.R. Ault, C.M. Carrillo, R.G. Chambers, and D.B. Lobell. 2021.
Anthropogenic Climate Change Has Slowed Global Agricultural Productivity
Growth. Nature Climate Change 11(4): 306–312.
Pascaris, A.S., C. Schelly, L. Burnham, and J.M. Pearce. 2021. Integrating Solar
Energy with Agriculture: Industry Perspectives on the Market, Community,
and Socio-Political Dimensions of Agrivoltaics. Energy Research & Social
Science 75: 102023.
Paustian, K., B. Babcock, J.L. Hatfield, R. Lal, B.A. McCarl, S. McLaughlin, C.
Rice et al. 2001. Agricultural Mitigation of Greenhouse Gases: Science and
Policy Options. In 2001 Conference Proceedings, First National Conference
 Climate Change and Agriculture 381

on Carbon Sequestration. Washington, DC: Conference on Carbon Seques-

tration, 2001.
Qin, Y., J.T. Abatzoglou, S. Siebert, L.S. Huning, A. AghaKouchak, J.S. Mankin,
C. Hong, D. Tong, S.J. Davis, and N.D. Mueller. 2020. Agricultural Risks
from Changing Snowmelt. Nature Climate Change 10(5): 459–465.
Rajagopal, D., C. Vanderghem, and H.L. MacLean. 2017. Life Cycle Assessment
for Economists. Annual Review of Resource Economics 9: 361–381.
Rausser, G. and D. Zilberman. 2022. Resource Economics and Modern Science
to the Rescue. Annual Review of Resource Economics 14: v–xvi.
Reardon, T. and D. Zilberman. 2018. Climate Smart Food Supply Chains In
Developing Countries in an Era of Rapid Dual Change in Agrifood Systems
and the Climate, L. Lipper, N. McCarthy, D. Zilberman, S. Asfaw, G. Branca
(Eds.), Climate Smart Agriculture: Building Resilience to Climate Change,
pp. 335–351. Springer Nature.
Rosenzweig, C. and F.N. Tubiello. 2007. Adaptation and Mitigation Strategies in
Agriculture: An Analysis of Potential Synergies. Mitigation and Adaptation
Strategies for Global Change 12: 855–873.
Schlenker, W., W.M. Hanemann, and A.C. Fisher. 2005. Will U.S. Agriculture
Really Benefit from Global Warming? Accounting for Irrigation in the
Hedonic Approach. The American Economic Review 95(1): 395–406.
Schlenker, W., and M.J. Roberts. 2009. Nonlinear Temperature Effects Indicate
Severe Damages to U.S. Crop Yields Under Climate Change. Proceedings of
the National Academy of Sciences 106(37): 15594–98.
Schneider, S.H. 1996. Geoengineering: Could — Or Should — We Do It? Climatic
Change 33(3): 291–302.
Skendžić, S., M. Zovko, I.P. Živković, V. Lešić, and D. Lemić. 2021. The Impact
of Climate Change on Agricultural Insect Pests. Insects 12(5): 440.
Surminski, S., L.M. Bouwer, and J. Linnerooth-Bayer. 2016. How Insurance Can
Support Climate Resilience. Nature Climate Change 6(4): 333–334.
Thamo, T. and D.J. Pannell. 2016. Challenges in Developing Effective Policy
for Soil Carbon Sequestration: Perspectives on Additionality, Leakage, and
Permanence. Climate Policy 16(8): 973–992.
van Meijl, H., P. Havlik, H. Lotze-Campen, E. Stehfest, P. Witzke, I.P.
Domı́nguez, B.L. Bodirsky et al. 2018. Comparing Impacts of Climate Change
and Mitigation on Global Agriculture by 2050. Environmental Research
Letters 13(6): 064021.
Van’t Veld, K. and A. Plantinga. 2005. Carbon Sequestration or Abatement? The
Effect of Rising Carbon Prices on the Optimal Portfolio of Greenhouse-Gas
Mitigation Strategies. Journal of Environmental Economics and Management
50(1): 59–81.
Wright, B.D. 2012. International Grain Reserves and Other Instruments to
Address Volatility in Grain Markets. The World Bank Research Observer
27(2): 222–260.
Yu, C., Miao, R. and Khanna, M. 2021. Maladaptation of U.S. Corn and Soybeans
to a Changing Climate. Scientific Reports 11: 12351.
382 Agricultural Economics and Policy

Zhang, J., E. Malikov, R. Miao, and P. Ghosh. Geography of Climate Change

Adaptation in U.S. Agriculture. Working paper (2023), the Department of
Agricultural Economics and Rural Sociology, Auburn University.
Zilberman, D., J. Zhao, and A. Heiman. 2012. Adoption Versus Adaptation, with
Emphasis on Climate Change. Annual Review of Resource Economics 4:
27–53.
 Epilogue: Sustainable Development

Sustainable development has become a major guiding principle for

policymaking around the world. The notion of sustainable develop-
ment, in a way, is an oxymoron given that “sustainable” implies
preserving and staying in place, and “development” means moving
forward. In practice, sustainable development is a compromise
between the desire of the developed world to preserve and protect
the environment and the developing world’s wishes for progress. The
Brundtland Commission defines sustainable development as “devel-
opment that meets the needs of the present without compromising the
ability of future generations to meet their own needs” (Brundtland,
1987). This definition of sustainable development is amenable to
economic interpretation as optimization, subject to constraint. This
definition comes from an anthropocentric perspective, as the benefits
are viewed from human perspectives and may not be acceptable to
some environmentalists.
There is a growing economic perspective on sustainable devel-
opment. First, we will review some macro-economic approaches
to sustainable development and then the micro approaches. In
both cases, we will emphasize the implications of the notion to
policymaking.

383
384 Agricultural Economics and Policy

1. The Macro-Economic Approach to Sustainable

Development
The surveys by Pearce et al. (1994) and Zilberman (2014) provide an
overview of the growing macroeconomics of sustainable development,
where the analysis of sustainable development is built on economic
growth theory (Jones, 2016). One approach to sustainable develop-
ment is to maximize the net present value of average utility under the
constraint that average utility doesn’t decline over time. A question
is: What is the source of average utility? Is it solely consumption,
or does it come in a multi-dimensional form? Many policymakers
view sustainable development as a three-legged stool that includes
economic, social, and environmental sustainability (Sachs, 2015),
which may imply that utility comes from consumption, environ-
mental amenities, and social well-being. Therefore, environmental
policy has a role in public policy, given that environmental amenities
provide utility. Furthermore, given heterogeneity among people and
the concern about social sustainability, the pursuit of sustainable
development may aim to protect the utility of the poor (say, the
bottom 10%) over time and ensure that it does not decline. To do so,
policymakers can either employ the max-min approach to maximize
the utility of the lowest-well-being group over time or reduce the
variability of the income of the poor.
Economic research on sustainable development also recognizes
that the environment and society are subject to random shocks, such
as natural disasters, pandemics, financial crises, political dramas, and
wars. Therefore, resilience is a growing emphasis as a key element
of sustainable development (Perrings, 2006). Namely, sustainable
development policies must be developed to allow economic systems
to withstand and recover from shocks, which implies that analy-
sis of sustainable development must incorporate randomness and
uncertainty, and solutions may need to be multiple-dimensional.
Sustainable development policies may also require diversification
of economic activities and increased capacity to move resources in
response to shocks.
Consumption and environmental amenities are outcomes of pro-
duction activities, and economic growth models have a specification
 Epilogue: Sustainable Development 385

of production systems. Some theoretical growth models, originating

from Solow’s pioneering work, present output as the outcome of
variable inputs, especially labor and physical capital (Solow, 1994).
However, as we model sustainable development, we have to recognize
that there are even more variants of capital, including human,
natural, and social. The stock of each type of capital changes over
time. Physical capital could increase with investment and decline
with depreciation. Human capital, which is the knowledge and skills
of humans, could increase with education and investment in research
and development but decline due to aging and depreciation. Natural
capital, a combination of the regulatory forces of nature that provide
environmental services (Bulte et al., 2008), could decrease through
mining and exploitation but, in turn, increase through restoration
and care. Finally, social capital represents a difficult concept to
define and measure, although it is generally viewed as a stock of
social cohesion which includes trust, collective action, and reduced
transaction costs. Several lines of research emphasize the impor-
tance of social arrangement and organization for economic growth,
underlying the major contributions of social capital. Nobel Laureate
Elinor Ostrom (2009) emphasizes the role of social arrangement in
sustaining effective water systems in developing countries. Acemoglu
et al. (2005) stress the crucial role of institutions in economic growth.
Development of indicators of institutions and arrangements and other
factors that contribute to social capital, and the factors that affect
economics, is a major challenge.
Much economic analysis considers population growth is given.
But, as Becker et al. (1999) argued, population choices are economic
activities. Furthermore, countries like China and India have imple-
mented policies that aim to affect population growth. Population
change has also been crucial to understanding the environmental
side of sustainable development.
The research on the economics of sustainable development
distinguishes between the notion of weak sustainability and strong
sustainability. The notion of weak sustainability recognizes signifi-
cant substitution between different types of capital goods and the
capacity of economic activities to restore and improve environmental
conditions. The notion of strong sustainability limits the substitution
386 Agricultural Economics and Policy

between different types of capital and other inputs and emphasizes

concern about irreversible outcomes. The assumptions of weak sus-
tainability lead to increased investments in research and development
and taking environmental risks; for example, increased mining of
resources with the belief that alternatives will be available over
time. Strong sustainability assumes that thresholds of environmental
protections are very costly to exceed and, thereby, suggests more
constraints in action. Leading economists developed a foundation for
macroeconomics based on the two ideas, which we will demonstrate
later in this epilogue. Most of this analysis assumes weak sustain-
ability. One paramount scientific challenge is identifying humanity’s
constraints and where and when the weak or strong sustainability
assumptions hold.

2. Quantitative Foundation for Measuring

Sustainable Development
There is significant quantitative modeling of the macroeconomics of
sustainable development based on the economic growth literature,
which is a large and sophisticated literature. In this section, we
provide an overview and some references that can lead the readers
to study the topic in depth.

2.1. Technological change as a function

of economic activity

Macro-analysis of sustainable development is constructed signifi-

cantly upon the works of Robert Solow. Solow (1957) and other
scholars of economic growth stated as a limitation of their models
that changes in input like capital and labor could not explain all the
changes in output. Solow’s (1957) econometric analysis utilized time
series of national data on the output and input of various nations to
demonstrate that economic growth, measured by output, is caused by
traditionally measured inputs plus a “residual.” This residual, called
Total Factor Productivity (TFP), measures the economy’s long-term
technological change.
 Epilogue: Sustainable Development 387

The unexplained residual has been attributed to major factors

associated with technological change, such as innovation, learning,
and technological improvement. Early studies considered this tech-
nological change to be determined by forces outside the economic
system, but new endogenous growth theories argue that technological
change is an outcome of economic activities and is determined
within the economic system (Romer, 1994). Technological change is
frequently originated in research activities that result in knowledge,
which has properties of a public good in that a large population
can utilize it simultaneously. This property of research may lead
to increasing economics of scale to production when inputs for
innovation and knowledge are considered (Jones, 2022).
The challenge of sustainable development is substituting non-
renewable resources and natural capital with human, physical, and
social capital. Solow espouses the economic perspective toward
sustainability by saying that almost all the natural environment
and resources are substitutable (except for, perhaps, rare minerals
and unique locations like Yosemite and the Grand Canyon). This
perspective is consistent with “Weak Sustainability” to be distin-
guished from “Strong Sustainability”, which limits the substitution
of one form of capital or resource with another (Daly, 1990).1 ,2

2.2. Alternative approaches to model weak

sustainability

There are two different approaches to modeling weak sustainability,

which assumes that resources are interchangeable. One approach,
which we will refer to as the Solow–Rawls (SR) approach, is more
normative in that it aims to find a unique resource allocation over

1
These two concepts of sustainability were first presented by Professor Daly
at the Conference on Human Demography and Natural Resources, Stanford
University, Hoover Institution and Morrison Institute, February 1–3, 1989, p. 15.
See: www.popline.org/node/360084#sthash.gwfL2zbx.dpuf.
2
Under Strong Sustainability, economic growth should be optimized subject to
specific constraints setting a threshold that limits the extraction or depletion of
specific resources. These concepts are explored in the following subsection.
388 Agricultural Economics and Policy

time that will maximize the net expected discounted utility of

consumption per capita, subject to three constraints: technological
capacity, limits to various capital, and non-decreasing consumption.
This approach calculates for each period the amount of capital and
physical labor utilized, the output produced, resources extracted, and
pollution generated, subject to the equation of motions of various
types of capital and a constraint that the utility of consumption per
capita does not decline over time. An illustration of this approach
considers the case in which individuals gain utility from consumption
and environmental quality, insofar as the aggregate utility does not
decline over time.
A variation of this approach, referred to herein as the Arrow–
Dasgupta (AD) approach, assumes greater flexibility of the utility
function such that utility does not decline below some threshold
over time (Arrow et al., 2004). The AD approach has a strong
element of accounting and measurement of actual national and global
performance. It aims to measure whether actual aggregate resource
allocation is consistent with the definition of sustainability by the
Brundtland Commission. This approach developed several criteria
before one can assess whether the evolution of the economies of
nations follows a path leading to sustainable development. A major
advantage of this approach is that under certain assumptions,3
resource allocation can be achieved by a competitive economy
with price incentives (taxes and subsidies) for non-market goods or
externalities (Arrow et al., 2004).
The AD approach starts from the assumption that sustainable
development occurs when intertemporal social welfare is not declining
over time, where intertemporal social welfare is defined as the sum
of the net present values of the aggregate utility of consumption
at a given moment in time, given the equations of motion of the
various types of capital.4 Arrow et al. (2004) suggest that this

3
The aggregate production function demonstrates diminishing returns of scale
and non-decreasing marginal costs.
4
This sustainability criterion holds true if the sum of the net discounted present
values of the utility of consumption from now to infinity is smaller or equal to
the sum of the infinite discounted present values of utility of consumption from
next year to infinity.
 Epilogue: Sustainable Development 389

sustainability criterion holds if the economy’s productive base, the

sum of all the capital stocks used for the production and provision
of services, does not decline over time. Thus, this sustainability
criterion—of non-declining intertemporal social welfare—is akin to
the non-declining productive base.
Further, this criterion implies that the value of the sum of
all types of capital will not decline over time. This interpretation
assumes that one can monetize the value of all types of capital and
sum them. In a well-functioning economy, the value of each type of
capital at each moment represents the sum of the discounted value of
future benefits that this capital good will provide. The value of the
sum of all types of capital can also be referred to as genuine wealth,
and the change in the value of genuine wealth can be referred to
as a genuine investment. This first sustainability criterion, which we
denote AD1.1, states that the rate of change of genuine investment
is not declining over time.
However, Arrow et al. (2004) realized that this sustainability
criterion must be adjusted for population growth and technological
change. Therefore, they suggested a modified criterion for sustainable
development that compares the rate of change of genuine capital and
population growth. This sustainability criterion, denoted as AD2.1,
states that the genuine investment per capita will not decline over
time. However, this criterion does not consider technological change
(i.e., the Solow residue). Thus, the third criterion for sustainable
development, which we denote as AD3, is that the genuine investment
per capita adjusted for technological change will not decline over time.
In particular,

genuine Investment population growth

−
genuine capital P opulation size
T otal F actor P roductivity
− ≥ 0,
output elasticity of genuine capital

where the output elasticity of genuine capital (EGP) measures

the contribution of genuine capital to growth. This technological
change contributes to societal welfare at a growth rate equivalent to
TFP/EGP in genuine capital. The inclusion of this fraction generates
390 Agricultural Economics and Policy

a measure of the overall change in productive capacity per capita,

including the contribution of technology.
The AD sustainability indicators are a practical tool applied by
Arrow et al. (2004) to various countries and regions. In general, the
AD approach is especially valuable for assessing the state of resource
allocation at present (based on AD3), but the present condition
does not imply that it will hold in the future. The AD measures of
sustainability provide an overall assessment of excess consumption
(of our overall capital), interpreted as a suboptimal genuine invest-
ment. Arrow et al. (2004) highlighted several important findings:

(1) During the last three decades of the 20th century, genuine
investment (AD1.1) was positive in most regions but negative
in sub-Saharan Africa and the Middle East (where resource
extraction was most concentrated).
(2) On the other hand, the growth rate of genuine capital per capita
(AD2.1) was negative for most of the developing world (except
China) due to the high rate of population growth in exceeding
the rate of growth of genuine capital.
(3) The AD3 criterion may result in different outcomes predicted by
the constrained expected net present value maximization, which
accounts for environmental amenities and resource dynamics. Yet
policies that aim to reduce pollution or set the optimal price for
renewable or non-renewable resources, based on constrained net
present value optimization, are likely to enhance sustainability
measured by AD3.
(4) AD3 measurements show that sustainability declined in Sub-
Saharan Africa and the Middle East, but did not change much in
developing countries like Bangladesh, India, and Pakistan, and
grew substantially in China (due to high investment rates in
manufacturing and human capital). These findings indicate that
AD3 tends to be lower than the growth rate of the Gross National
Product (GNP). Some countries with a significant growth rate
of GNP have a low negative measure of sustainability, like AD3.

The AD measures of sustainability are a work in progress. For

instance, the monetizing of forest depletion and greenhouse gas
 Epilogue: Sustainable Development 391

(GHG) emissions, as well as other elements contained in AD3, are

lively debated and subject to much uncertainty. Therefore, presenting
a confidence interval where the sustainability estimates may lie
in with a high degree of statistical significance may be valuable.
Furthermore, as we learn more about the costs of climate change and
the value of biodiversity, the measure of sustainability may change.
Finally, technology is evolving quickly and will remain a key source
of uncertainty, captured by the Solow residue.
The AD and SR measures of sustainable development are com-
plementary. The AD approach uses ex-post measurements reflecting
choices made in the past and indicates the urgency of interven-
tion in the future. The SR measures are ex-ante and suggest
the best plan for future performance. However, these measures
address issues of inter-temporal equity by providing conceptual
frameworks and asking: do humans consume too much (globally
and regionally)? Is the overall environmental situation improving —
and which regions need to improve most? Answering these big
questions is important, and thinking, in general, is crucial to
developing an effective set of strategies to address environmental
challenges. But policy implementation depends on the specifics.
For example, the importance of focusing on a single species in
developing environmental policies points to the emphasis on strong
sustainability.
The macro perspective on sustainable development suggests the
importance of enhancing the stock of different types of capital
over time through physical investment, education, environmental
protection, restoration, and action to improve trust and social
cohesion. This range of activities, combined with efficient use of
resources to increase net social benefit, is a policy foundation to
pursue sustainable development. Developing rigorous theoretical and
empirical economic research on sustainable development is a major
challenge for the future. The studies mentioned above are the
beginning of quantifying sustainable development, especially assum-
ing weak sustainability. It is challenging to expand the theoretical
foundation to analyze cases of strong sustainability and determine
when each set of assumptions is appropriate. These lines of research
392 Agricultural Economics and Policy

are likely to change national accounting to incorporate environmental

and social considerations.

3. Micro-Level Analysis of Sustainable

Development
To develop management practices and policies based on sustainable
development in agriculture, it is important to assess the impact
of this concept at the micro-level (Zilberman, 2014). Sustainable
development has implications both for analytical approaches and
policy strategies. The implications of sustainable development in
relation to research methodologies include:

(1) Emphasis on multidisciplinary work: As with environmen-

tal and resource economics, a basic understanding of biophysical
relationships is crucial for economic modeling. To manage water
well, one must understand the basic principles of hydrology
and soil-water relationships. Pest control analysis requires an
understanding of entomology and toxicology. This does not imply
that the technologists need to be specialists in the field. Still, they
need to understand key relationships and collaborators who can
provide specific details. As Khanna (2022) suggested, applied
economists not only need to have depth in economics but also
some breadths in other fields to combine multidisciplinary work.
(2) Production, pollution, and risk function: Economics
emphasizes the role of the production function that relates
outputs to inputs. But economic activities have side effects that
are important to consider in developing sustainable solutions.
Therefore, pollution functions that relate different types of
pollution to economic input and output levels are essential for
the complete analysis of systems that consider externalities.
Similarly, many economic activities, such as the use of pesticides,
may result in risk. So, risk-generating functions that relate
mortality probabilities to economic activities and recognize the
procedures of contamination, exposure, and dose-response are
essential to complete the analysis of economic activities.
 Epilogue: Sustainable Development 393

(3) Heterogeneity and randomness: The simple economic anal-

ysis assumes certainty and identical firms. In reality, agents are
different, and production activities are subject to random shocks.
Therefore, it is important to explicitly recognize these condi-
tions and develop policies that allow adaptations to different
conditions and shocks. The notion of precision is increasing the
productivity of inputs while reducing pollution; and resilience to
shocks is crucial to develop sustainable solutions. Our analysis
of the risk of adoption emphasizes the response to randomness
and heterogeneity. The development of improved monitoring
solutions (e.g., remote sensing and global positioning system),
communication tools (e.g., the internet), better analytic tools
(e.g., artificial intelligence), and better mobility tools (e.g.,
drones) allow the development of increased precision and fast
adaptation to shocks.
(4) Input-use efficiency: The distinction between applied and
effective inputs as well as the externalities caused by input
residues are key features of agricultural systems. The develop-
ment of methods that increase input-use efficiency, and thus
conserve resources and reduce waste is necessary. But, having
more precise technologies does not mean that they will be
adopted, so policy intervention and supply chain design should be
considered.
(5) Supply chain and life cycle analysis: Agriculture is part of
the supply chain of food, fuels, and other products. Therefore,
complete analysis in search of sustainable development requires
analysis of the complete supply chain. Furthermore, innovation
and product supply chains are associated with economic activ-
ities (Zilberman et al., 2022). Because of the multiple links
between economic activities and markets throughout the value
chain, assessing the environmental impact of policies may require
life cycle analysis from birth to death. Life cycle analysis methods
have been developed by engineers and may need adjustment to
be used in economic analysis. But this adjustment is essential
and has some foundations to guide it (Rajagopal, 2017).
394 Agricultural Economics and Policy

4. Conclusion: Avenues for Achieving Sustainable

Development Solutions
We conclude by discussing several avenues for achieving sus-
tainable development solutions. First, the transition from non-
renewable resources to renewable resources. In principle, non-
renewable resources are depletable, so their use is not sustainable.
But even when they are abundant, their use may cause negative
externalities, as with fossil fuels. Thus, transitioning to a renewable
economy is one of the guiding principles of sustainable development.
We can distinguish between two types of renewable resources. The
first is physical renewables, such as energy from the sun, wind, and
water. The second is living organisms, such as plants, animals, and
others. Early in human history, humans mostly harvested living
organisms. But, with the emergence of agriculture, they moved
to animal husbandry systems. Humans invested in breeding and
raising living organisms because they acquired higher yields and more
controllable and cheaper harvesting. We transitioned from hunter-
gatherer to agriculture; we now have a transition toward aquaculture,
and with research, we may breed other species.
Second, adoption of conservation strategies to improve input-
use efficiency and reduce residue. Such strategies are essential for
adapting and mitigating climate change (Zilberman et al., 2012).
Policy design and implementation are necessary to create conserva-
tion technologies, adapt them for different applications (crops and
environment), and enhance adaptations.
Third, recycling. Reusing materials reduces the financial and
environmental toll of production. Additionally, recycling plays a
valuable role in the global economy, as second-hand markets for
products, like clothes and cars, provide an efficient means for goods
to make their way to developing countries. Recycling is especially
critical when it comes to non-renewable resources like metals. One
of the most notable causes of crime worldwide is reusing equipment
(car parts) and precious materials. So, there is an important role in
analyzing the economics of recycling, its implications, and the role
of policies to improve it.
 Epilogue: Sustainable Development 395

Embedded within the idea of recycling is the circular economy.

Every good and product goes through processes of life and death, and
its residues are recycled. Production systems have been, historically,
linear, considering the inputs and outputs that are needed and useful
while ignoring the residues. In contrast, a circular economy recognizes
these residues and treats them. Policies to control environmental side
effects may transform the linear economy into a circular economy and
lead to increased resource conservation and reuse.
Lastly, the bioeconomy. The bioeconomy is an economic sector
that utilizes natural resources and takes advantage of new biological
capabilities. It has a traditional component, which involves agricul-
ture, fishery, forestry, agrotourism, bioenergy, and green chemistry,
and was essential for the development of human civilization to build
the capacity to preserve food and enhance health. The modern com-
ponent of the bioeconomy involves science-based discoveries among
medical treatment, sequestration of carbon, and the development of
new modes of production of chemicals and energy. These, according
to Zilberman et al. (2013), have important implications for continued
advancements in longevity for humanity and the environment.
However, there are a few major challenges for the development
of bioeconomy. First of all, regulations have to improve procedu-
ral safety. Currently, regulations are inconsistent globally and do
not provide incentives to integrate biotechnology solutions across
nations. Regulatory inconsistency and uncertainty are the major
barriers to the expansion of the bioeconomy and deter investments
and product development. Some countries are even considering
banning modern biotechnology. Moreover, there must be investment
in research for biotechnological innovations. The scientific knowledge
base for the bioeconomy is in its infancy, and further expansion of
research and education in the life sciences, microbial and cell biology,
and related organismal technology will identify new mechanisms
and technological opportunities. However, technology development
is likely to encounter many barriers, especially when it comes to
technologies that can address environmental challenges and pro-
vide solutions to problems that afflict the poor and developing
nations. Therefore, introducing policies that will further support the
396 Agricultural Economics and Policy

commercialization of biotechnologies that can address the challenges

of climate change and food security may require government support.
Policies that will provide incentives to address climate change, e.g.,
support for carbon sequestration or reduction of greenhouse gas
emissions, may benefit the bioeconomy. Finally, the bioeconomy may
provide capabilities that may endanger humanity, e.g., biological
weapons. International institutions must develop mechanisms to
avoid such development and use while allowing biotechnological
research to advance.

References
Acemoglu, D., S. Johnson, and J.A. Robinson. 2005. Institutions as a Fundamen-
tal Cause of Long-Run Growth. Handbook of Economic Growth 1: 385–472.
Arrow, K., P. Dasgupta, L. Goulder, G. Daily, P. Ehrlich, G. Heal, S. Levin
et al. 2004. Are We Consuming Too Much? Journal of Economic Perspectives
18(3): 147–172.
Becker, G., E.L. Glaeser, and K.M. Murphy. 1999. Population and Economic
Growth. American Economic Review 89(2): 145–149.
Brundtland, G.H. 1987. World Commission on Environment and Development.
Our Common Future. Oxford University Press, Oxford.
Bulte, E.H., L. Lipper, R. Stringer, and D. Zilberman. 2008. Payments for
Ecosystem Services and Poverty Reduction: Concepts, Issues, and Empirical
Perspectives. Environment and Development Economics 13(3): 245–254.
Daly, H.E. 1990. Sustainable Development: From Concept and Theory to
Operational Principles. Population and Development Review 16: 25–43.
Jones, C.I. 2016. The Facts of Economic Growth. Handbook of Macroeconomics
2: 3–69.
Jones, C.I. 2022. The Past and Future of Economic Growth: A Semi-Endogenous
Perspective. Annual Review of Economics 14: 125–152.
Khanna, M. 2022. Breakthroughs at the Disciplinary Nexus: Rewards and Chal-
lenges for Applied Economists. American Journal of Agricultural Economics
104(2): 475–492.
Ostrom, E. 2009. A General Framework for Analyzing Sustainability of Social-
Ecological Systems. Science 325(5939): 419–422.
Pearce, D.W., G.D. Atkinson, and W.R. Dubourg. 1994. The Economics of
Sustainable Development. Annual Review of Energy and the Environment
19(1): 457–474.
Perrings, C. (2006). Resilience and Sustainable Development. Environment and
Development Economics 11(4): 417–427.
Rajagopal, D., C. Vanderghem, and H.L. MacLean. 2017. Life Cycle Assessment
for Economists. Annual Review of Resource Economics 9(1): 1–21.
 Epilogue: Sustainable Development 397

Romer, P.M. 1994. The Origins of Endogenous Growth. The Journal of Economic
Perspectives 8(1): 3–22.
Sachs, J.D. 2015. The Age of Sustainable Development. Columbia University
Press.
Solow, R.M. 1957. Technical Change and the Aggregate Production Function.
The Review of Economics and Statistics (1957): 312–320.
Solow, R.M. 1994. Perspectives on Growth Theory. Journal of Economic Perspec-
tives 8(1): 45–54.
Zilberman, D. 2014. Fellows Address: The Economics of Sustainable Development.
American Journal of Agricultural Economics 96(2): 385–396.
Zilberman, D., J. Zhao, and A. Heiman. 2012. Adoption versus Adaptation, with
Emphasis on Climate Change. Annual Review of Resource Economics 4(1):
27–53.
Zilberman, D., E. Kim, S. Kirschner, S. Kaplan, and J. Reeves. 2013. Technology
and the Future Bioeconomy. Agricultural Economics 44(s1): 95–102.
Zilberman, D., T. Reardon, J. Silver, L. Lu, and A. Heiman. 2022. From the
Laboratory to the Consumer: Innovation, Supply Chain, and Adoption with
Applications to Natural Resources. Proceedings of the National Academy of
Sciences 119(23): e2115880119.
This page intentionally left blank
 Appendix A

Elements of Optimization Theory

In this appendix we briefly introduce some basic techniques for

static and dynamic optimization that are often used in agricul-
tural economics. These techniques include Kuhn–Tucker conditions
for constrained optimization, linear programming with constraints,
the bordered Hessian and its applications in comparative statics,
decision-making over time, and finally, optimal control. For complete
treatment of these topics, we refer readers to Chiang (1992), Dixit
and Pindyck (1994), as well as Simon and Blume (1994).

1. Kuhn–Tucker Conditions
With reference to Figure 1, consider the problem
max f (x)
x

s.t. x ≥ 0,
where f (·) is a concave function.
If function f (x) is f A as depicted in Figure 1, the optimum occurs
at x1 , where (df A /dx)|x∗ =x1 = 0. In the case of f B in Figure 1, the
global maximum is not achievable so that the constrained optimum

399
400 Agricultural Economics and Policy

Figure 1. A constrained maximization problem.

occurs at x = 0, where (df B /dx)|x∗ =0 < 0. Therefore, in a general

case, the first-order conditions can be characterized by

fx (x∗ ) ≤ 0, x∗ fx (x∗ ) = 0,

where fx (x) is defined as df (x)/dx.

Consider now

max f (x)
x

s.t. g(x) ≤ b, and x ≥ 0,

where x = (x1 , . . . , xn ) , g(x) = (g1 (x), . . . , gm (x)) , and b =

(b1 , . . . , bm ) .
Form the Lagrangian

L = f (x) + λ  [b − g(x)],

where λ = (λ1 , . . . , λm ).
 Appendix A: Elements of Optimization Theory 401

Then the Kuhn–Tucker conditions can be written as

Lxi = fxi (x∗ ) − (λ
λ∗ ) gxi (x∗ ) ≤ 0, ∀i = 1, . . . , n, (1)
x∗i · Lxi = x∗i [fxi (x∗ ) − (λ
λ∗ ) gxi (x∗ )] = 0, ∀i = 1, . . . , n, (2)
Lλj = bj − gj (x∗ ) ≥ 0, ∀j = 1, . . . , m, (3)
λ∗j Lλj = λ∗j [bj − gj (x∗ )] = 0, ∀j = 1, . . . , m, (4)
x∗ ≥ 0, (5)
λ ∗ ≥ 0. (6)
These conditions are sufficient if f (·) is quasiconcave and g(·) is
quasiconvex.

2. Linear Programming and Constrained

Optimization
2.1. Primal vs. Dual

Consider the following optimization problem:

max c x
x

s.t. Ax ≤ b, and x ≥ 0,
where x is an n×1 vector of choice variables (e.g., production levels);
c is an n × 1 vector of net returns of the choice x; A is the m × n
matrix of technical coefficients, where the element {aij } indicates the
requirement of resource i needed to support one unit of choice j; and
b is the m × 1 vector of resource availability. In this problem, we
have n variables and m constraints, with n > m.
It is an optimization problem subject to inequality constraints.
An optimal solution can be characterized by the necessary Kuhn–
Tucker conditions. We first write the Lagrangian as
L = c x + λ  [b − Ax],
where λ is an m × 1 vector of shadow prices.
402 Agricultural Economics and Policy

The optimality conditions are given by

Lx = c − Aλ ≤ 0,
Lλ = b − Ax ≤ 0,
and the following complementary slackness conditions:

m
xi Lxi = xi [ci − (λj aji )] = 0, ∀i = 1, . . . , n,
j=1


n
λj Lλj = λj [bj − (aji xi )] = 0, ∀j = 1, . . . , m.
i=1

This problem is called the primal.

Now consider the following:
min λ  b
λ

s.t. Aλ ≥ c, and λ ≥ 0,

where the parameters are defined as above. This optimization
problem is called the dual.
The fundamental result in linear programming is that the primal
and the dual have the same first-order necessary conditions. Readers
can easily prove this conclusion by using Kuhn–Tucker conditions
(hint: for the Lagrangian of the dual problem, set the Lagrangian
multipliers as x). These lead to some of the following results:

1. At the optimal solution, c x∗ = λ ∗  b or, in a scalar notation,


n 
m
ci x∗i = λ∗j bj ;
i=1 j=1

2. if ni=1 aji x∗i < bj then λ∗j = 0;
 ∗ ∗
3. if mj=1 aji λj > ci , then xi = 0.

2.2. The bordered Hessian and its applications

Consider the following optimization problems:

 Appendix A: Elements of Optimization Theory 403

maxx f (x) minx f (x)

s.t. g(x) = b s.t. g(x) = b

where x is an n × 1 vector of decision variables, b is an m × 1 vector

of resource availability, f : Rn → R, and g : Rn → Rm .
Using Lagrange multiplier techniques, these problems become

maxx f (x) + λ  [b − g(x)] minx f (x) + λ  [b − g(x)],

where λ is a m × 1 vector of Lagrange multipliers. The first-order

conditions for both problems are

∇ f (x) − ∇ g(x) λ = 0
 x   x   
n×1 n×m m×1 n×1

b − g(x) = 
0 .
  
m×1 m×1

The bordered Hessian of these problems is

⎡ m×m m×n
⎤
   
⎢ 0 −(∇x g(x)) ⎥
⎢ ⎥
⎢ ⎥
H=⎢ ⎥.
⎢ ⎥
⎣ −(∇x g(x)) ∇2xx f (x)n×1 − ∇x [∇x g(x)λ
λ] ⎦
     
n×m n×n

The second-order conditions of the maximization and minimiza-

tion problems are

maximum minimum
(−1)r Hr > 0 ∀r = m + 1, . . . , n, (−1)m Hr > 0 ∀r = m + 1, . . . , n,

where Hr is the (m + r)th leading principal minor of the bordered

Hessian, H. Note that Hr consists of the first m+r rows and columns
of |H|, and Hn = |H|.
404 Agricultural Economics and Policy

2.3. Example of application

Consider the problem

max f (x)
x

s.t. g(x) = b,

with x ∈ R3 , b ∈ R2 , f : R3 → R, and g : R3 → R2 . That is, it is

a problem with three variables and two constraints. The Lagrangian
for this problem can be formed as

L = f (x) + λ1 [b1 − g1 (x)] + λ2 [b2 − g2 (x)],

with the first-order conditions:

∇x f (x) − ∇x g(x)λ
λ = 0,
b − g(x) = 0,

or, in scalar notation,

fx1 − gx11 λ1 − gx21 λ2 = 0,

fx2 − gx12 λ1 − gx22 λ2 = 0,
fx3 − gx13 λ1 − gx23 λ2 = 0,
b1 − g1 (x) = 0,
b2 − g2 (x) = 0.

To simplify notation, we let g11 denote gx11 , f12 denote fx1 x2 , and
so on. The bordered Hessian is
⎡ ⎤
0 0 (−g11 ) (−g21 ) (−g31 )
⎢ ⎥
⎢ 0 0 (−g12 ) (−g22 ) (−g32 ) ⎥
⎢ ⎥
⎢ − 1 λ − 1 λ − 1 λ ⎥
⎢(−g 1 ) (−g2 ) (f 11 g11 1 (f 12 g12 1 (f 13 g13 1⎥
⎢ 1 1
−g 2 λ ) −g 2 λ ) −g 2 λ ) ⎥
H=⎢ ⎢ 2 2 2 ⎥.
1 λ ⎥
11 12 13
⎢ (f 21 − g 1 λ
21 1 (f 22 − g 1 λ
22 1 (f 23 − g23 1⎥
⎢(−g21 ) (−g22 ) ⎥
⎢ −g21 λ2 )
2 −g22 λ2 )
2 −g23 λ2 ) ⎥
2
⎢ 1 λ ⎥
⎣ 1 2 (f31 − g31 1 λ
1 (f32 − g32 1 λ
1 (f33 − g33 1⎦
(−g3 ) (−g3 )
−g31 λ2 )
2 −g32 λ2 )
2 −g33 λ2 )
2
 Appendix A: Elements of Optimization Theory 405

2.4. Comparative statics

By totally differentiating the first-order conditions, the following

comparative statics results can be derived:
⎡ ⎤ ⎡ ⎤
dλ1 −1 0
⎢dλ ⎥ ⎢ 0 −1⎥  
⎢ 2⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥ db1
H ⎢dx1 ⎥ = ⎢ 0 0 ⎥
⎢ ⎥ ⎢
⎥ db .
⎢ ⎥ ⎢ ⎥ 2
⎣dx2 ⎦ ⎣ 0 0 ⎦
dx3 0 0
We have
⎡ ⎤ ⎡ ⎤
dλ1 −db1
⎢ ⎥ ⎢ ⎥
⎢dλ2 ⎥ ⎢−db2 ⎥
⎢ ⎥ ⎢ ⎥
⎢dx1 ⎥ = H −1 ⎢ 0 ⎥ .
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥
⎣dx2 ⎦ ⎣ 0 ⎦
dx3 0
Note that
⎡ ⎤
C11 C21 C31 C41 C51
⎢ ⎥
⎢C12 C22 C32 C42 C52 ⎥
1 ⎢⎢ ⎥
H −1 = C C C C C ⎥,
|H| ⎢ ⎥
13 23 33 43 53
⎢ ⎥
⎣C14 C24 C34 C44 C54 ⎦
C15 C25 C35 C45 C55
where Cij is the (i, j)th cofactor of matrix H. Specifically, Cij =
(−1)i+j |H−i,−j |, where H−i,−j is a submatrix of H obtained by
deleting the ith row and jth column of matrix H. Therefore, we
further have
⎡ ⎤ ⎡ ⎤
dλ1 −C11 db1 − C21 db2
⎢ ⎥ ⎢ ⎥
⎢dλ2 ⎥ ⎢−C12 db1 − C22 db2 ⎥
⎢ ⎥ 1 ⎢ ⎥
⎢dx1 ⎥ = ⎢ ⎥
⎢ ⎥ |H| ⎢−C13 db1 − C23 db2 ⎥ .
⎢ ⎥ ⎢ ⎥
⎣dx2 ⎦ ⎣−C14 db1 − C24 db2 ⎦
dx3 −C15 db1 − C25 db2
406 Agricultural Economics and Policy

Solving, for example, for dx3 /db1 , one obtains

dx3 C15
=− = −(−1)1+5 |H−1,−5 |/|H|.
db1 |H|
From the second-order conditions we know that the determinant of
H is negative. Therefore, the sign of dx3 /db1 is the same as that of
|H−1,−5 |.

3. Decision-Making Over Time

A meaningful analysis of the agricultural and resource system cannot
be done without understanding their evolution over time. Some of the
most important decisions that economists are asked to evaluate are
investment decisions, which entail assessments of dynamic flows of
incomes and expenditures. In this section, we discuss two concepts
that are critical to such assessments: discounting and interest rate.
There is a basic theoretical explanation for determination of the
interest rate (sometimes we use the term discount rate). It is an
equilibrium price for a one-period delay in use of one unit of value
(e.g., a dollar). Figure 2 illustrates the determination of the discount
rate.
The curve AB is a production possibility frontier, and it denotes
all the possible tradeoffs between consumption in period t and period
t+1. Point B corresponds to a situation with maximum consumption
in period t and no consumption in period t + 1. At A, all the
consumption is delayed to period t + 1, and the points connecting A
to B correspond to positive consumption in both periods. The figure
is drawn under the assumption that delay in consumption will lead
to expanded resource availability. At point D, BE units of resource
are not consumed in period t and lead to the availability of OG units
of resources in period t + 1. The assumed concavity of the tradeoff
curve is the reason for this outcome. Each of the curves I0 , I1 , and
I2 is a locus of consumption patterns that result in the same level
of utility; therefore, consumers are indifferent to movements between
points on the same curve. Higher indifference curves correspond to
higher utility levels. The optimal resource allocation is at D where
 Appendix A: Elements of Optimization Theory 407

ct+1 M

D
G

I2

I1
N

I0

O ct
E B

Figure 2. Consumption decision over time.

the highest feasible indifference curve is tangent to the production

possibility frontier. The absolute value of the slope of the tangency
line MN is 1 + r, where r is defined as the discount rate.
Under certain assumptions (competition, full information, and no
externalities), the market gives rise to the socially optimal discount
rate. These conditions are not likely to hold in many cases, and
market discount rates are likely to be different and, in most cases,
higher than the social interest rate. The importance of monetary
policy considerations in the determination of interest rates is one
reason for the difference between market rates and socially optimal
rates. Market rates may be higher for the following reasons:

(a) Risk considerations: Society has a much better capacity than

any individual to carry risks, especially ones that are not correlated.
Since society has many members that carry and share uncorrelated
risks, the risk cost to the society approaches zero, and societal choices
should be conducted using an interest rate reflecting risk neutrality.
408 Agricultural Economics and Policy

Many market rates reflect risk aversion. This argument was first
introduced by Arrow and Lind (1970).
(b) Externality considerations: When individuals decide about
future investments, they consider only the benefits to themselves and
their direct families in the future. However, their sons or daughters
may get married, and other unrelated people may benefit from the
activities generated by this investment. However, these benefits are
not taken into account by the private parties. Therefore, they tend
to underinvest from a social perspective and that can be interpreted
as having a private interest rate that is higher than the social one
(note that higher interest rates mean that benefits in the present
are weighted more heavily relative to benefits in the future). This
argument was advanced by Marglin (1963). Moreover, the interest
rates paid by producers and consumers are adjusted to incorporate
a lot of other elements besides society’s equilibrium value of time
preference, such as inflation and transaction costs associated with
facilitation of loans.

3.1. The decomposition of interest rates

Let I denote interest rates paid by individuals to organizations, which

can be decomposed into several elements: real market discount rate
(R), rate of inflation (RI), transaction cost (T C), and risk factor
(RF ). That is, I = R + RI + T C + RF .
Examples:

• Banks pay to Federal Reserve: R + RI.

• Low-risk customers of banks pay prime interest rate: I = R +
RI + T Cm + RFm , where T Cm and RFm are factors that reflect
minimum traction and risk cost levels.
• Loans backed by assets generally pay lower interest rates than
loans that are not backed by assets, reflecting risk factor.
• Credit ranking and other devices are used by lenders to assess
riskiness of loan and to determine the risk factor.
• Lenders assess riskiness of investment and new projects before
financing them.
 Appendix A: Elements of Optimization Theory 409

• If the real interest rate is 3%, the inflation rate is 4%, and the risk
and transaction cost of banks is 1%, then it could be that
– Federal Reserve loan to banks at 7%,
– the consumer-paid interest rate is 8%.
• If home mortgage loans are 9%, then it could be that
– lender receives 7%,
– risk-transaction cost is 2%.
• If the nominal interest rate is 12%, then it could be that
– inflation rate is 14%,
– real interest rate is –2%.

When interest rates are higher, individuals are less likely to invest
money in projects and instead will deposit their money in banks and
buy government bonds.

3.2. Methodology to assess investments

Here we only consider deterministic scenarios and refer readers

to Dixit and Pindyck (1994) for a comprehensive treatment of
investments under stochastic scenarios. Two steps are involved to
assess investments in risk-free scenarios: (1) Assess economic cost and
benefit over time and (2) use discount rate to compute net present
value or compute internal rates of return.
For a potential investment project, let Bt be benefits and Ct
be costs in period t. The net present value (NPV) of this project
N
t=0 (1/(1 + r)) (Bt − Ct ) in a discrete time framework and is
is t
 T −rt
0 e (Bt − Ct )dt in a continuous time framework. Investment on a
project will make sense if its NPV is positive.
An alternative way to determine whether to invest is to calculate
N
the internal rate of return (IRR) by using equation t=0 (1/(1 +
IRR))t (Bt − Ct ) = 0. If IRR is greater (respectively, less) than the
market interest rate, then the investment should (respectively, should
not) be made.
Example. Suppose a project will need $100 million to be
established. Once established, it will last for two years, generating
benefits at $66 million in year one and $60.5 million in year two.
410 Agricultural Economics and Policy

Assuming that the interest rate is 10%, then the NPV of this project
is −100+66/(1+0.1)1 +60.5/(1+0.1)2 = 10. The IRR of this project
is solved from equation −100+66/(1+IRR)1 +60.5/(1+IRR)2 = 0:
IRR ≈ 17.5%. If the benefits in each year were to be $70 million, then
the NPV of this project would be −100+70/(1+0.1)1 +70/(1+0.1)2 ≈
21.5, and the IRR would be about 25%.
Paradox of Internal Rate of Return. Consider a flow of net
benefits −a, b, −c across three periods. The IRR solves
b c
−a + − = 0.
1 + IRR (1 + IRR)2
The quadratic equation may have two positive solutions. To see this,
suppose a = 10, b = 30, and c = 20. In this case,

−10(1 + IRR)2 + 30(1 + IRR) − 20 = 0

√
30 ± 100 30 ± 10
⇒ 1 + IRR = =
20 20
⎧
⎨1
⇒ (1 + IRR) =
⎩2.

What is the true IRR? How to analyze investments? To avoid such a

paradox, one could use IRR only in cases when there is one switch of
the signs of net benefits across periods (e.g., investment occurs first
and returns later). Or, one could just use the NPV approach in most
cases.

4. Dynamic Systems
Dynamic systems are widely applied in agricultural economics such
as water management models and soil carbon sequestration models.
Optimal control is often used to derive optimal policies for a dynamic
system. The analytic methodology for deriving the optimal solution
to deterministic dynamic problems was developed by Pontryagin
et al. (1962) and has been applied to a wide range of economic
problems. In this section, we first obtain optimality conditions for
 Appendix A: Elements of Optimization Theory 411

discrete optimal control problems and then analyze the necessary

optimality conditions of a continuous control problem.

4.1. Discrete time optimal control

A dynamic system typically has a policy variable (labeled xt ) and

a stock variable (labeled st ), where the policy variable is chosen in
each period by the decision maker and the stock variable is influenced
by the choice. The relationship between st and xt is reflected by
an equation of motion, gt (st , xt , θ t ), where θ t stands for a set of
exogenous model parameters. The goal of the decision maker is to
optimize his or her objective function value, ft (xt , st , θ t ), under initial
conditions and any given constraints. A discrete time deterministic
control model can be written as

T
max β t f (xt , st , θ t ) + β T +1 V (sT +1 )
xt
t=0

s.t. st+1 − st = g(st , xt , θ t ), ∀t = 0, . . . , T, and

s0 = s¯0 ,
where β = 1/(1 + r) is a discounting coefficient and r is interest rate.
For example, f (xt , st , θ t ) can be the profits from a fishing
operation, st may be the stock of fish, xt may be the fishing effort,
and θ t may be a vector of prices. The equation of motion may denote
the change in the fishing population and may combine the effect of
fishing (reduces stock) and growth. The function, V (sT +1 ), is the
terminal value of the stock of fish at the end of the planning horizon.
The Lagrangian approach may be useful for solving this problem
(we omit θ t for convenience):
⎛ ⎞

T ⎜ ⎟
⎜ t ⎟
L= ⎜β f (xt , st ) −λt [st+1 − st − g(st , xt )]⎟ + β T +1 V (sT +1 ) ,
⎜    ⎟   
t=0 ⎝ ⎠
period t scrap value
net benefit

subject to the initial condition s0 = s¯0 .

412 Agricultural Economics and Policy

The first-order conditions are

Lxt = β t fxt + λt gxt = 0, t = 0, . . . , T, (7)

Lst = β t fst + λt (1 + gst ) − λt−1 = 0, t = 1, . . . , T, (8)
Lλt = st+1 − st − g(st , xt ) = 0, t = 0, . . . , T, (9)
LsT +1 = β T +1 VsT +1 − λT = 0, (10)

given s0 .
The Lagrange coefficient, λt , denotes the marginal value (i.e.,
shadow price) of extra units of stock at the end of period t discounted
to period 0.
Condition (7) suggests that the decision variable, xt , is set so
that its discounted marginal net benefit at period t is equal to its
marginal costs in terms of stock growth. We expect gxt < 0.
Conditions (8) and (10) can be rewritten as

−(λt − λt−1 ) = β t fst + λt gst , t = 1, . . . , T, (11)

λT = β T +1 VsT +1 . (12)

Equations (11) and (12) provide a set of difference equations

establishing the dynamics of the shadow price of the stock. The
shadow price of time T is equal to the discounted marginal con-
tribution of the residual stock at time T + 1. All shadow prices are
discounted to time 0, so we do not have a time difference problem.
The net shadow price of stock increases as time approaches period
zero because the right-hand side of Equation (11) is positive. The
increase (λt−1 − λt ) reflects the marginal contribution to production
(fst ) and growth (gst ) of the stock at this extra period. Note the
earlier the stock is introduced, the more it can contribute because it
lasts longer.
The difference equations (11) and (12) reflect the dynamics of the
dual variables (shadow price of stock variables), while the difference
equation (9) with the initial s0 sets the dynamics of the real stock
variables. Therefore, we have to solve for both the dynamics of
(shadow) price and quantities in solving optimal control problems.
 Appendix A: Elements of Optimization Theory 413

To better understand the problem, f (st , xt ) may be viewed as

profits from fishing, i.e., f (st , xt ) = pt k(st , xt ) − wt xt , where pt is the
fish price, k(st , xt ) is the quantity of catch, st is the fish stock, xt is
the effort, and wt is the cost per unit of effort. The stock equation
may be st+1 − st = h(st ) − k(st , xt ), where h(st ) denotes the stock
growth. The shadow price, λt , is the discounted value of fish in the
water.

4.2. Continuous time optimal control

It is much more convenient and elegant to use a continuous optimal

control model for analytic purposes. We derive the optimality con-
ditions rather heuristically using the approach of Intriligtor (1971).
The continuous version of the model presented earlier is
 T
max e−rt f (xt , st ) dt + e−rT V (sT )
xt 0

s.t. ṡt = g(st , xt ).

One can write the Lagrangian function

 T
L= {e−rt f (xt , st ) − λt [ṡt − g(st , xt )]}dt + e−rT V (sT ), (13)
0

where λt is the dynamic shadow price of the equation of motion. It

reflects the discounted marginal value of stock added at time t. This
Lagrangian function can be rewritten in an alternative way. To see
it, note
 T  T  T
L= e−rt f (xt , st )dt − λt ṡt dt + λt g(st , xt ) dt + e−rT V (sT ).
0 0 0

Since
 T 
T  T
λt ṡt dt = (λt st ) − λ̇t st dt
0 0 0
414 Agricultural Economics and Policy

by using integration by parts, the alternative formulation of the

Lagrangian is
 T
L= [e−rt f (xt , st ) + λ̇t st + λt g(st , xt )]dt + V (sT )e−rT
0
− λT sT + λ0 s0 , (14)
where s0 = s¯0 .
The optimality conditions are obtained by differentiating L, using
Equation (13) or (14) as appropriate:
∂L
= e−rt fxt + λt gxt = 0, (using equation (13)),
∂xt
∂L
= e−rt fst + λ̇t + λt gst = 0, (using equation (14)),
∂st
∂L
= ṡt − g(st , xt ) = 0, (using equation (13)),
∂λt
∂L
= λT − e−rT V  (sT ) = 0, (using equation (14)),
∂sT
∂L
= s0 , (using equation (14)).
∂λ0
Pontryagin et al. (1962) have derived a simple approach to obtain this
optimality condition. They define the Hamiltonian function where
H(xt , st , λt ) = e−rt f (xt , st ) + λt g(st , xt )
and proved that at each t the optimality conditions are
∂H
= e−rt fxt + λt gxt = 0, (15)
∂xt
∂H
ṡt = = g(st , λt ), (16)
∂λt
∂H
−λ̇t = = e−rt fst + λt gst , (17)
∂st
given s0 and λT = e−rt V  (sT ).
These sets of conditions allow one to solve a dynamic optimal
control problem as a succession of static choice problems corrected by
 Appendix A: Elements of Optimization Theory 415

the dynamics of stock variables and their shadow prices. The dynamic
of stock is given by Equation (16) and s0 = s¯0 , and the dynamic of
shadow prices is presented by Equation (17) and λT = e−rt V  (sT ).

4.3. Optimal control techniques and applications

In the previous section, we presented the optimality conditions using

shadow prices discounted to period 0. Many times we are interested in
temporary prices and values, rather than discounted ones. To obtain
such prices, we derive first temporary optimality conditions. Consider
the problem discusses earlier:
 T
max e−rt f (xt , st ) dt + e−rT V (sT )
xt 0

s.t. ṡt = g(st , xt ).

The (discounted) Hamiltonian of this problem is

HD = e−rt f (xt , st ) + λD
t g(st , xt ),

where λDt is the shadow price of the equation of motion (or the
shadow price of stock in time t) in values discounted to time 0.
The optimality conditions are

∂H D
= e−rt fxt + λD
t gxt = 0, (18)
∂xt
∂H D
ṡt = = g(st , xt ), (19)
∂λD
t

∂H D
−λ̇D
t = = e−rt fst + λD
t gst , (20)
∂st
−rT 
λD
T =e V (sT ), (21)

given s0 .
Let H = ert H D denote the temporal (or current value) Hamilto-
nian and λt = ert λD
t denote the temporal shadow price of the stock.
416 Agricultural Economics and Policy

The alternative presentation of the optimality condition is

∂H
= fxt + λt gxt = 0, (22)
∂xt
∂H
ṡt = = g(st , xt ), (23)
∂λt
∂H
−λ̇t + rλt = = fst + λt gst , (24)
∂st
λT = V  (sT ), (25)

given s0 .
To see that conditions (18)–(21) and (22)–(25) are consistent,
note that
∂H D ∂H
−λ̇t = −rλt − ert λ̇D
t = −rλt + e
rt
= −rλt + .
∂st ∂st

4.3.1. The simple model of economic growth

The first and perhaps the most important early applications of

optimal control in economics pertain to the study of economics
growth. These are macro models, but the analysis can apply to micro
problems. The economic growth model presented here considers the
case when one good is both consumed and invested. Suppose the
production function is

yt = f (kt ),

where yt is the output and kt is the capital stock. Let ct be

consumption in time t and investment is yt − ct = f (kt ) − ct .
Utility derived from consumption is measured by u(ct ), where
u(·) is a utility function with u > 0 > u . Thus, the objective is
 ∞
max e−rt u(ct )dt,
ct 0

and the equation of motion is

k̇t = f (kt ) − ct − γkt .

 Appendix A: Elements of Optimization Theory 417

Capital increases at the investment rate but may decline because

of depreciation where γ is the depreciation coefficient. The initial
capital stock is k0 . The temporary Hamiltonian is

H = u(ct ) + λt [f (kt ) − ct − γkt ],

and the optimality conditions are given by

∂H
= uc − λt = 0, (26)
∂ct
∂H
−λ̇t + rλt = = λt fk − λt γ, (27)
∂kt
∂H
k̇t = = f (kt ) − ct − γkt . (28)
∂λt
First, optimality condition (26) equates the shadow price of the
stock to marginal utility of consumption; this is reasonable since
a unit of the good is either traded or invested and marginal benefit
from both activities has to be equal at the optimal solution. The rate
of change in the nominal shadow price is affected by three elements
(from Equation (27), we can obtain λ̇t /λt = r − fk + γ). They are as
follows: (1) discounting, which has the effect of increasing nominal
prices over time, (2) depreciation, which has a similar effect as that of
discounting, and (3) marginal productivity of capital (i.e., fk ), which
operated by reducing nominal prices over time. The higher fk is, the
more capital will be available in the future and the less valuable it is.

4.3.2. The dynamics of consumption

Total differentiation of Equation (26) yields ucc ċt − λ̇t = 0. From

(27), we obtain λ̇t = λt (r − fk + γ). Thus, ucc ċt − λ̇t = 0 becomes

ucc ċt + λt [fk − r − γ] = 0.

Since uc = λt , the expression leads to a condition defining the

rate of changes in consumption over time:
ċt uc
=− [fk − r − γ].
ct uccct
418 Agricultural Economics and Policy

uc
Let us define η(ct ) = ucc ct . It can be interpreted as the elasticity
of demand for the good. To see this point, suppose we have a
utility-maximizing consumer who derives additive utility from c and
expenditure. The optimal consumption choice of this individual is

max{u(c) + I − pc},
c

where I is the income and p is the commodity price. Optimality

condition is uc − p = 0. Total differentiation of this condition yields
dc 1
ucc dc − dp = 0 ⇒ = < 0.
dp ucc
Therefore,
dc p uc
= = η(ct ) < 0.
dp ct ucc ct
Using this definition, note that
c˙t
= −η(ct )[fk − r − γ]. (29)
ct
If fk − r − γ > 0, then the productivity of capital is substantial
and the shadow prices decline over time. In this case, Equation (29)
suggests that consumption increases and the increase in consumption
is inversely related to the demand elasticity. The intuition is as
follows. The price of today’s consumption is today’s investment. If the
demand elasticity is smaller (i.e., more negative), then the decision
maker will reduce today’s consumption and increase investment,
which results in larger increase in tomorrow’s consumption.

4.3.3. Steady state

At steady state, both state and co-state (shadow price of stocks)

variables do not change over time. Economists are interested in
knowing if steady-state situations exist, what will be the dynamic
path which leads to them, and whether they are stable. They are
stable when the system returns to steady state in spite of some
random shocks that move it away. Economists are enamored with
 Appendix A: Elements of Optimization Theory 419

Figure 3. A phase diagram.

steady states because they are the dynamic equivalent of long-

run equilibria, which represent the outcome for which the system
converges.
In the growth theory model, steady state occurs when

k̇ = f (k) − c − γk = 0, (30)
ċ = λ̇ = fk (k) − r − γ = 0. (31)

Figure 3 presents a phase diagram. The loci of all the k and c

combinations that lead to k̇ = 0 and ċ = 0 are depicted as well as
their intersection(s) which are the steady states of the system.
The locus of all the points with ċ = 0 is at k = k∗ . The curve
c = f (k) − γk is the locus of all the points with k̇ = 0. It intersects
c = 0 at k = 0 and k = km . The steady state of the system occurs in
our case at Z with k = k∗ and c = c∗ .
To study the dynamic properties of the system, note that the
diagram indicates that consumption deceases when k > k∗ and
consumption increases when k < k ∗ . Similarly, k declines above the
curve with k̇ = 0 (when f (k) − γk − c < 0) and increases below it.
420 Agricultural Economics and Policy

The curve AZB denotes optimal consumption-capital path. If

initial capital is k01 , initial consumption should be c10 and movement
along BZ will lead to equilibrium. If initial capital is k02 , initial
consumption should be c20 and, for a period before steady state,
consumption will be greater than production.
The analysis here denotes the importance of time preference. If
r = 0, optimal solution will be at kx > k∗ . But with r > 0, even if
initial stock is kx , there will be periods of excess consumption along
the line AZ until the steady-state capital at k = k ∗ is attained.

References
Arrow, K.J. and R.C. Lind. 1970. Uncertainty and the Evaluation of Public
Investment Decisions. The American Economic Review 60(3): 364–378.
Chiang, A.C. 1992. Elements of Dynamic Optimization. Waveland Press, Inc.
Long Grove, Illinois.
Dixit, A.K. and R.S. Pindyck. 1994. Investment under Uncertainty. Princeton
University Press. Princeton, NJ.
Intriligator, M.D. 1971. Mathematical Optimization and Economic Theory.
Prentice-Hall, Inc., Englewood Cliffs, NJ.
Marglin, S.A. 1963. The Social Rate of Discount and The Optimal Rate of
Investment. The Quarterly Journal of Economics 77(1): 95–111.
Pontryagin, L.S., V.G. Boltyanskii, R.V. Gamkrelidze, E.F. Mischenko. 1962. The
Mathematical Theory of Optimal Processes (trans: Tririgoff KN; Neustadt
LW (ed)). Wiley, New York, NY.
Simon, C.P. and Lawrence Blume. 1994. Mathematics for Economists. W. W.
Norton & Company, Inc. New York, NY.
 Appendix B

Problem Sets

In this appendix, we present six problem sets that can be used during
teaching agricultural economics and policy. The first two problem
sets focus on production and the third on risk analysis. The last
three problem sets are comprehensive and can be used as exercises
or exams. The solutions to the first three problem sets are provided
in Appendix C.

Problem Set 1
(1) Assume that a product is used initially without any regulation.
The price, which is equal to the marginal cost of producing the
product, is equal to 2 and the quantity is equal to 8. We know
that the demand is linear and, at the initial outcome, the price
elasticity is
∂Q P 1
ηD = − = ,
∂P Q 4
where Q is output and P is price. Suppose that the marginal
social cost of externality is MSC = Q.
A. What is the social optimum and the resulting price and
quantity of the output?

421
422 Agricultural Economics and Policy

B. What is the consumer surplus, producer surplus, government

revenue, and environmental cost before the regulation and (1)
under pollution tax, (2) under pollution reduction subsidy,
and (3) under tradable permits?
C. Suppose that the product is sold by a monopoly. What is the
optimal pollution policy in this case?
(2) The micro-level production function is Von-Liebig
y = ax for x ≤ x̄,
where y is output per acre, x is water, and a is the land fertility
parameter. The land density function as a function of fertility is

g(a) = M − na, g(a) ≥ 0.

A. Given output price, P , and water price, W , what is aggregate

supply? What is aggregate water demand?
B. Attempt to derive an aggregate production function and
interpret its coefficient.
(3) Let the ex-ante production function be

Y = AK α L1−α ,

where Y is output, K is capital, and L is labor. Let P denote

output price and W denote wage rate. Suppose output price is
increasing at rate γ, wage rate at rate δ, and r>δ >γ where r is
the discount rate. In a putty-clay framework, write a model to
answer the following:
A. What is the optimal capital/output and labor/output ratio
of a new plant?
B. What is the planned economic life of this plant?
C. How will the optimal life of capital change as γ and δ
increase?

Problem Set 2

(1) A crop is produced with a constant return-to-scale technology

using land and effective water (water attaining the plants).
 Appendix B: Problem Sets 423

The production function is

q = f (e),

where q is output per acre and e is effective water per acre.

Effective water is a production of applied water per acre, a, and
land quality (α) 0 ≤ α ≤ 1, i.e.,

e = aα.

Let P denote the output price and w water price.

A. How will per-acre output supply and actual and effective
water demands be affected by changes in P , w, and α?
B. Assume a competitive land market. How will the rental rate
behave as a function of P , w, and α?
C. Assume the production function is a C.E.S. Derive (1) per-
acre output supply and water demands and (2) the rental-rate
function.
(2) Suppose a household is managed like a business. It controls N̄
hours of labor. It runs a small enterprise that produces output
Y with Nh hours of labor, with a Just–Pope production function
Y =f (Nh ) +αNh ε. The production function f is concave and
α > 0. The price of output is p and ε is a random variable with
mean 0. Family members can also work outside the firm and earn
w dollars per hour.
A. Establish the appropriate assumptions and find the optimal
decision rules for time allocation under risk neutrality. What
is the likelihood of an internal solution (when family members
diversify their time between work in the family business and
the labor market) for a risk-neutral individual? How will a
change in the price of the variable input or the output affect
the time allocation and the input use?
B. Suppose that the family is risk averse and utility is a
function of wealth with initial wealth denoted by W0 What
are the optimal allocation rules for a risk-averse household?
Derive the first-order conditions and interpret them. How will
changes in the price of output and the wage rate affect time
allocation and expected output?
424 Agricultural Economics and Policy

C. Compare analytically the output of a risk-averse family with

a risk-neutral one. If the family has an initial wealth of
W0 , how a change in wealth affects its labor allocation and
expected output?
D. Solve the problem for the case where the household has a
negative exponential utility function and the random variable
is distributed normally (specify the assumptions). How will
an increase in the variance of the random variable affect the
time allocation?
E. Thus far the household was modeled as a firm. Explain how
your modeling will change if we consider a household that
gains utility from both income and leisure and total labor is
not given but is a decision variable.

Problem Set 3
(1) A risk-averse individual evaluates the following five portfolios:

I II III IV V
X P (X) X P (X) X P (X) X P (X) X P (X)
3 1/6 5 1/3 2 1/4 3 1/3 2 1/6
7 1/6 7 1/6 6 1/12 7 1/6 8 1/6
10 1/3 12 1/12 10 1/3 12 1/12 9 1/3
13 1/6 14 1/12 14 1/12 14 1/12 11 1/6
17 1/6 15 1/3 16 1/4 17 1/3 17 1/6

A. Order the distributions (whenever possible) according to

their riskiness using the mean-preserving spread concept.
B. Using the second-order stochastic dominance concept, order
these distributions. What are the implications of the
ordering?
C. Order the distribution using a mean-variance criterion.
D. Using Roy’s minimum probability rule — min [P (X ≤ 6)] —
to order these distributions.
E. Using Kataoka’s safety-fixed rule — Max d, s.t., P (X ≤ d) ≤
1/6 — order these distributions.
F. Using U (X) = ln (X + 1) to order these distributions.
G. Compare and evaluate the different orderings.
 Appendix B: Problem Sets 425

(2) A farmer operates L acres of land using the production function

defined on a per-acre basis:

q = f (l, m) + g(l, m)ε,

where ε is a random variable, fl > 0, fm > 0, gl > 0, and

gm < 0 — f and g are concave in m and l.
Let the total profit be

Π = L · (pq − wl − rm) − F,

where p is output price, r and w are input prices, and F is fixed

cost.
A. Assume the farmer maximizes the expected utility of income
with u > 0, u < 0, and E(ε) = 0.
i. How has behavior under uncertainty differed from behav-
ior under certainty when ε ≡ 0?
ii. Assuming decreasing absolute risk aversion, how does
reduction in fixed costs affect input use?
iii. Assuming constant relative risk aversion, how will
increase in farm size affect input use?
Prove all your results mathematically.
B. Assume ε ∼ N (0, σ 2 ), and U (Π) = 1 − e−δΠ .
What are the optimal decision rules of the farmers? How are
they affected by farm size and the fixed cost level?
C. Assume that ε is gamma distributed — E(ε) = 1. What are
the optimal decision rules?
D. Assume that ε ∼ N (0, σ 2 ) and the farmer pursues Kataoka’s
safety-fixed rule. What are his optimal decision rules?
E. Compare the behavior under the alternative models.

Problem Set 4
(1) Consider L units of land and many farms. There is hetero-
geneity because different locations are exposed differently to
pest damage, but otherwise land productivity is the same. The
industry is producing an output with a constant return to scale
426 Agricultural Economics and Policy

technology and the production function per acre is a damage

control function y = f (z) [1 − d(N, j)], where y is output per
acre, f (z) is potential output per acre, z is fertilizer use per
acre, and d(N, j) is pest damage that assumes value from 0 to
1. The pest damage function increases with the pest population
per acre N , which assumes values from 0 to N̄ and is affected
by pest control treatment j. It can assume three values, j = 0
when there is no treatment, j = 1 when there is a chemical
treatment, and j = 2 when GMO is introduced. The damage
function d(N, j) = h(N )g(j) can be decomposed to h(N ),
damage without interference, and g(j), fraction of damage after
treatment j. We assume that g(0) = 1 (when j = 0 there is no
damage reduction). We also assume that 1 ≥ g(1) ≥ g(2) ≥ 0,
namely there is more damage after treatment with chemical
pesticides than GMO (this is hypothetical). For example, when
g(1) = 0.5 and g(2) = 0, the chemical treatment eliminates half
the damage and GMO all of it. Let output price be denoted
by P , fertilizer price by V , and the prices of the pest control
treatment j, w(j) where w(0) = 0, 0 < w(1) ≤ w(2). The pest
exposure per acre is a random variable with a density function
 N̄
ψ(N ) where 0 ψ(N )dN = 1. So, if the marginal land operating
has a pest population 0 ≤ Na ≤ N̄, the total land in production
N
is L 0 a ψ(N )dN .

A. First consider the case where the two pest control treatments
are not available, so j = 0. Write the optimization problem
of a farm with initial pest level N . If farms maximize profits,
what will be the aggregate output of the industry (hint, first
define a critical level of N above which production is not
profitable and then aggregate). Show how output will change
in response to changes in P and V .
B. Suppose the chemical pesticide is becoming available, so
either j = 0 or j = 1. Analyze the adoption of the pesticide.
What subgroup of farms will adopt it? What will be the
impacts on land use and output? How will changes in output
price or pesticide price affect adoption?
 Appendix B: Problem Sets 427

C. Suppose the GMO is available. So j may assume three

values (0, 1, or 2). Write the farmer optimization problem.
Identify conditions under which the GMO will be adopted,
and in particular when it will be adopted by all farmers or
only by subgroups of farmers (with or without adoption of
chemical pesticides). How will the introduction of the GMO
affect supply and pesticide use? Derive supply for various
conditions. How will the adoption of pesticides affect fertilizer
use?
D. Consider the case where the agriculture industry is facing
inelastic demand. Discuss graphically how the introduction
of GMO will affect prices and quantity of output.
(2) An entrepreneur is considering producing a new biofuel. He
considers a processing plant to refine the biofuel from a feedstock
crop. He can either grow the feedstock or buy if from farmers.
The entrepreneur has limited initial credit that he has to allocate
among the refinery and feedstock production faculties that he
may invest in.
A. Develop a model to present the entrepreneur’s decision
problem.
i. Find the optimal decision rules: How much biofuel to
produce? How much feedstock to produce in-house and
how much to buy from others? Interpret the optimality
conditions.
ii. How will the credit constraint affect the outcome?
B. Suppose the refinery receives the subsidy for greenhouse gases
that the biofuel saves relative to gasoline. Expand the model
to include the subsidy and analyze its implications.
C. Suppose the entrepreneur is risk averse and has an exponen-
tial utility function. Assumes that the cost of processing of
one unit of feedstock has a random component with a Normal
distribution.
i. What will be the optimal allocation rule? How will the
risk aversion affect the overall production and feedstock
production choices?
428 Agricultural Economics and Policy

ii. Now suppose also that the cost of feedstock is a linear

combination of a deterministic and random component
with a Normal distribution. Derive the supply chain design
problem and characterize the solution.
(3) Buyers want to buy a product that provides the benefit of b if it
fits and 0 otherwise.
The probability of fit is q. The buyer has three options — (1)
to buy it with a money-back guarantee (MBG) at a price PG that
allows return and refund if the product does not fit, but with a
return fee of R dollars, or (2) to buy it without guarantee at
price PN , or (3) not to buy.

A. Under what condition will a buyer utilize the MBG option?

What is the value of the option as a function of the key
parameters?
B. There are M buyers and the probability of fit is distributed
uniformly from qL to 1. How many people will buy the
product with the MBG? How many people will buy the
product without the MBG?
C. Suppose the buyer has a disappointment cost of S if the
product does not fit. How will it affect the answers to B?
D. Suppose that a buyer can pay a fee of D dollars and take a
demonstration that with probability γ will tell whether the
product fits or not — while keeping the buyer uncertain with
probability 1−γ. Under what conditions will the buyer prefer
the demonstration to the MBG?

(4) A small project.

Write a brief (2 pages) on an agriculture or agribusiness region,
organization, or a company (e.g., California, Monsanto, John
Deere, Iowa, UC cooperative extension) providing

A. Basic facts — what is unique about the region or company?

B. What are the main challenges and opportunities?
C. What policies (actions) will you recommend?
 Appendix B: Problem Sets 429

Problem Set 5
(1) A farmer is growing wheat with a Just–Pope production function,
where the output and input are denoted by y and x, respectively.
The prices of the output and the input are p and w, respectively.
A. Establish the appropriate assumptions and find the optimal
decision rules for input choice under risk aversion and risk
neutrality.
B. Compare input use and expected output under both risk
neutrality and risk aversion, under various assumptions
about the technology (e.g., risk-increasing or risk-decreasing
technologies).
C. How will an increase in the initial wealth of the farmer affect
input use?
D. Solve the problem for the case where the farmer has a
negative exponential utility function and the random variable
is distributed normally (specify the assumptions).
E. For the case of normal distribution of the random element of
the yield function, what will be the decision rule if the firm
wants to maximize the level of profits that will be exceeded
95% of the time (the 5% of the profits distribution)?
(2) A. Compare the optimal stock of groundwater in a steady state
to the one arising from a decentralized, private property
solution.
B. Give some intuition for why they differ as they do.
C. Discuss the role of natural recharge.
D. Discuss how an increase in water use efficiency would change
the steady-state level of pumping or the groundwater stock.
(3) A. What is the difference between diffusion and adoption? What
are the measurements of these concepts?
B. What are the main theories explaining observed diffusion
patterns? What are the main assumptions and implications?
What are their strengths and weaknesses?
C. Derive a threshold model for the adoption of an irrigation
technology where output per acre y depends on effective
430 Agricultural Economics and Policy

input αx where α is a land quality index assuming value

from 0 to 1. Modern technology costs I dollars per acre but
doubles input use efficiently. Fixed cost per acre regardless
of water cost is F . Output price is p and water price is w.
Prices of inputs vary over time. There is a distribution of land
1
quality with density function g(α) where 0 g(α)dα = L and
L is total land.
i. Establish basic assumptions. Derive the microlevel opti-
mization rule for each period.
ii. Divide lands into three groups: adopters, non-adopters,
and idle. What will determine to which category each
land quality belongs? Derive aggregate supply.
iii. If the fixed cost of the technology will decline over time,
how will it affect diffusion? How will an increase in output
price affect diffusion?
iv. How does risk consideration affect the outcomes?
(4) A. What are the main stages in the evolution of U.S. agriculture,
and correspondingly in the evolution of U.S. agricultural
policy?
B. To what extent did agricultural policy serve to enhance
efficiency versus other social objectives?
C. How does agricultural policy affect the environment?
D. What institutions, policies, and arrangements have emerged
to address various types of risks in different contexts?
E. What are likely to be the economic impacts of these institu-
tions and policies (e.g., how are they likely to affect prices,
quantities, welfare of various agents, etc.)?
F. What are the likely drawbacks of some of these
arrangements?
Use basic models to illustrate some of your answers.

Problem Set 6
(1) A firm is producing a product using energy with a production
function y = f (x), where the output and input are denoted
 Appendix B: Problem Sets 431

by y and x, respectively. The production function has normal

properties. The prices of the output and the input are p and w,
respectively. The price of energy w is a random variable w = w̄ε,
where ε has a zero mean.
A. Establish the appropriate assumptions and find the optimal
decision rules for input choice under risk aversion and risk
neutrality.
B. Compare input use and expected output under both risk
neutrality and risk aversion.
C. Suppose there is an energy tax of t dollars per unit. Show
how will it affect the optimal energy use.
(2) On pesticide use.
A. Explain the damage control function approach to analyze
pesticide use. Use this specification to develop a model where
output depends on fertilizer use and pest population. The
pest population is affected by initial pest population and
pesticide treatment. Specify your assumptions.
B. Show how changes in the initial pest population and changes
in the prices of output, pesticides, and fertilizers will affect
pesticide and fertilizer uses, pest damage, and output in the
case of a profit-maximizing firm.
C. Suppose there is an upper bound limiting pesticide use. Show
how it will affect fertilizer use and output.
D. Suppose there is a pesticide tax. How will it affect optimal
choices?
E. What will the regulated farm prefer: A pesticide tax or an
upper bound of pesticide use? Explain.
F. Suppose a new genetically modified variety is available. It
reduces pest damage but requires extra cost. Expand the
model and identify the conditions under which it will be
adopted. How will it affect output, pesticide use, and fertilizer
use?
(3) Consumers derive utility from health and fun. They consume
three goods. Each good has fixed coefficients of fun and health
per unit consumed. Good A has the highest fun-to-health ratio,
432 Agricultural Economics and Policy

good C has the lowest fun-to-health ratio, and the fun-to-health

ratio of good B is in between.
A. Develop a model to analyze the choices of a consumer with a
given income. Specify your assumptions.
B. How will a change in the price of Good A affect its consump-
tion? Explain. How will it affect the consumption of fun and
health?
C. An investor is considering introducing a new product that will
dominate all existing products. What are the requirements it
must meet?
(4) On U.S. agricultural policies.
A. How have the agricultural policies of the U.S. changed over
time? Please explain these changes.
B. Discuss how the energy situation and climate change affect
agriculture.
C. Discuss how the energy situation and climate change are
likely affecting agricultural and related policies.
 Appendix C

Solutions to Selected Problem Sets

Solutions to Problem Set 1

(1) Solution

A. What is the social optimum and the resulting price and quantity
of the output?
Since demand is linear, P = 2 and D = 8, based on
∂Q P 1
ηD = − = ,
∂P Q 4
we have
∂Q 1 Q 1 8
= − × = − × = −1.
∂P 4 P 4 2
Thus, we know that the linear demand curve has a slope of −1.
Given the slope and a point on the demand curve (i.e., P = 2
and D = 8), we can readily check that the demand is Q = 10 − P
and the inverse demand is P = 10 − Q.
Since the marginal cost of producing the product is 2 and the
marginal social cost of externality is Q, the total social marginal
cost of producing the product is 2+ Q. So, at the social optimum,
we have 10 − Q = 2 + Q so Q∗ = 4 and P ∗ = 6.

433
434 Agricultural Economics and Policy

B. Before Regulation:
CS0 (consumer surplus) = (10 − 2) × 8/2 = 32
PS0 (producer surplus) = 0
EC0 (externality cost) = 8 × 8/2 = 32
SS0 (social surplus) = 0
Under a pollution tax:
Because the marginal cost of the product is 2 and because the
socially optimal price and quantity are 6 and 4, respectively, the
socially optimal tax is 4. Under this tax, the equilibrium quantity
demanded is 4. Therefore,
CS T = (10 − 6) × 4/2 = 8
PS T = 0
EC T (externality cost) = 4 × 4/2 = 8
GRT (Government revenue) = 4 × 4 = 16
SS T = CS T + PS T − EC T + GR T = 16
Under a pollution reduction subsidy:
Under such a subsidy, the producer takes actions to mitigate
the externality. Assuming the cost of the actions is equal to the
marginal externality cost. Then the optimal subsidy rate is 4.
In this case, the producer will only invest in pollution reduction
technology for the first 4 units of product because investing in
the pollution reduction technology for the 5th unit will reduce
the profit. Readers can readily check this.
CS S = (10 − 6) × 4/2 = 8
PS S = 4 × 4/2 + 4 × 4 = 24
EC S (externality cost) = 0
GR S = −16
SS S = 16
Under a tradable permit:
Subsidy = Tax = 0
CSR = (10 − 6) × 4/2 = 8
PSR = 4 × 4 = 16
 Appendix C: Solutions to Selected Problem Sets 435

ECR = 4 × 4/2 = 8
GRR = 0
SSR = 16
C. Outcome under a Monopoly:
MR = 10 − 2Q
At Monopoly optimum MR = MC → 10 − 2Q = 2. So,
QM = 4, P M = 6
CSM = (10 − 6) × 4/2 = 8
PSM = 4 × 4 = 16
ECM = 4 × 4/2 = 8
SSM = 16
Monopoly results in social optimum in this case.

(2) Solution

A. Farmers will produce only if profits are positive, i.e., if

π = pax − wx ≥ 0,

or
w
a≥ .
p
If profits are strictly positive, they will produce the maximum
amount possible per acre, i.e., y = ax̄. Aggregate supply is
therefore given by the aggregated maximum output of each
fertility grade of land that can be farmed profitably, or
 M      
n M3 M W 2 n W 3
Y (P, W ) = ax̄(M − na)da = x̄ − +
W 6n2 2 P 3 P
P

and aggregate demand for water by

 M      
n M2 W n W 2
X(P, W ) = x̄(M − na)da = x̄ −M + .
W 2n P 2 P
P
436 Agricultural Economics and Policy

B. Solving the water demand equation for W/P yields


W M 2X
= −
P n nx̄
and substituting this into the supply equation gives the aggregate
production function
⎡   ⎤
2 3
M 3 M M 2X n M 2X ⎦
Y (X) = x̄ ⎣ 2 − − + −
6n 2 n nx̄ 3 n nx̄
  
M 2 2X
= − X.
n 3 nx̄

(3) Solution

A. A producer building a new plant chooses capital, labor, and the

planned economic life of the plant (denoted T ) so as to maximize
the discounted sum of expected gross profits less the cost of
capital. The optimization problem is, therefore,
 T
max e−rt [Pt AK α L1−α − wt L]dt − K
K,L,T t=0

subject to the no-shutdown condition

Pt AK α L1−α − wt L ≥ 0.

Using that

Pt = P0 eγt
wt = w0 eδt ,

where P0 and w0 are initial price and wage rate. We can rewrite
the objective function as

(1 − e(γ−r)T ) (1 − e(δ−r)T )
max P0 AK α L1−α − w0 L − K.
K,L,T r−γ r−δ
 Appendix C: Solutions to Selected Problem Sets 437

The first-order conditions are (assuming an interior solution)

(1 − e(γ−r)T )
K: αP0 AK α−1 L1−α −1=0
r−γ
(1 − e(γ−r)T ) (1 − e(δ−r)T )
L: (1 − α)P0 AK α L−α − w0 =0
r−γ r−δ
T : P0 AK α L1−α e(γ−r)T − w0 Le(δ−r)T = 0,

whence

K (1 − e(γ−r)T )
k= = αP0
AK α L1−α r−γ
L P0 (1 − e(γ−r)t ) (r − δ)
l= = (1 − α)
AK α L1−α w0 (1 − e(δ−r)t ) (r − γ)

and
  α 
p0 K   k α 
ln A ln p0
A
w0 L w0 l
B. T = = .
δ−γ δ−γ
C. Plugging k and l into the above equation and using the implicit
function theorem, readers can obtain comparative statics of
dT/dδ and dT/dγ.

Solutions to Problem Set 2

(1) Solution

A. The producer’s optimization problem is

max[P f (aα) − wa]

a
subject to P f (aα) − wa ≥ 0

with first-order condition (assuming an interior solution)

P fe α − w = 0.
438 Agricultural Economics and Policy

Applying the implicit function theorem to this condition, we find

da fe α fe
=− 2
=−
dP P fee α P fee α
da 1
=
dw P fee α2
da P fe + P fee aα fe a
=− 2
=− 2
−
dα P fee α fee α α
Thus,
de da fe
=α =−
dP dP P fee
de da 1
=α =
dw dw P fee α
de da fe
=α =− −a
dα dα fee α
and
dq de f2
= fe =− e
dP dP P fee
dq de fe
= fe =
dw dw P fee α
dq de f2
= fe = − e − afe .
dα dα fee α

B. The rental rate will equal profits as a function of land quality:

r(α) = π(α) = P f (aα) − wa.

By the envelope theorem, we need only consider the direct effects

of changes in P , w, and α on this rate, so that
dr dπ
= = f (aα)
dP dP
dr dπ
= = −a
dw dw
dr dπ
= = pfe a,
dα dα
 Appendix C: Solutions to Selected Problem Sets 439

all evaluated at the optimal solution.

C. If the production function is CES, we have
Q = (c1 E ρ + c2 Lρ )1/ρ ,
where E denotes total effective water demand and L denotes land.
Written in intensive form (i.e., in per-acre basis), this becomes
q = (c1 eρ + c2 )1/ρ .
The producer’s optimization problem is therefore
max[P · (c1 αρ aρ + c2 )1/ρ − wa].
a

The first-order condition

P · (c1 αρ aρ + c2 )(1/ρ)−1 c1 αρ aρ−1 − w = 0,
implicitly defines the optimal actual water demand
a∗ = a∗ (c1 , c2 , P, w, α, ρ).
The effective demand, output supply, and rental rate are therefore
given by
e∗ = αa∗
q ∗ = (c1 αρ (a∗ )ρ + c2 )1/ρ
r ∗ = P · (c1 αρ (a∗ )ρ + c2 )1/ρ − wa.

(2) Solution

A. If there is an internal solution, the profit-maximization problem

under risk neutrality becomes very simple:
maxNh π = p (f (Nh )) − wNh .
The first-order condition is:
dπ
= p(f  (Nh )) − w = 0
dNh
N0 = N̄ − Nh .
If p(f  (0)) < w, Nh = 0. If p(f  (N̄ )) > w, N̄ = Nh . One can
readily check that an increase in output price will increase the
optimal time allocated to the small enterprise, whereas if the
440 Agricultural Economics and Policy

opportunity cost of the labor (w) increases then the optimal time
allocated to the enterprise will decrease.
B. Assume an internal solution. Here, we need to consider the entire
income of the household.
 
maxNh EU (W0 + p (f (Nh ) + αNh ε) + N̄ − Nh w).

The first-order condition is:

EU  (·) (p(f  (Nh ) + α ) − w) = 0

−E[U  (·) pα ]
pf  (Nh ) = w + .
EU  (·)
The value of the marginal product of labor is equal to the wage
plus a risk element that reflects the negative correlation between
marginal utility and . Based on the discussion in the chapter
about risk and uncertainty, one can show that E[U  (·) pα ] < 0.
By comparing the first-order condition here with that under risk
neutrality, one can see that the optimal input under risk aversion
will be smaller than that under risk neutrality.
C. A risk-averse family will produce less than a risk neutral one
based on our interpretation of the first-order conditions. The
impact of a change in initial wealth depend on the decision
maker’s risk preference. Under decreasing (respectively, constant
or increasing) absolute risk aversion in wealth, an increase in
the initial wealth will increase (respectively, have no impact
or decrease) expected output and increase labor use inside the
household (relative to outside the household).
D. Based on Chapter 4, if we have negative exponential utility
and a normal distribution, the objective function of the firm
is to maximize mean profit minus measure of absolute risk
aversion times variance divided by 2 (i.e., a mean-variance utility
function). Note that E (1 − e−rπ ) = 1 − E (e−rπ ) and that
E (e−rπ ) is a moment-generating function of random variable π
at r. The moment-generating function of a random variable x
is etx . If x is normally distributed with mean μ and variance
1 2 2
σ 2 , then the moment-generating function is etμ+ 2 σ t . If we
 Appendix C: Solutions to Selected Problem Sets 441

replace t with −r, maximizing exponential utility is equivalent

to maximizing μ − σ 2 r/2.
We denote the measure of absolute risk aversion as r. Thus the
objective function (in case of internal solution) is
1
maxNh pf (Nh ) − Nh w − rα2 Nh2 σ 2 p2 .
2
The first-order condition is:
pf  (Nh ) − w = rα2 Nh σ 2 p2 .
Obviously, risk reduces labor inside the household, and higher
variance does the same.
E. In this case, we have a household production function, but
now the household gains utility from leisure. In this case, we
add another variable, Nl , and the time allocation constraint is
N̄ = Nh +Nl +No . In this case, one can have several formulations.
If we have expected utility, we get utility from income and leisure,
and the optimization problem becomes:
 
maxNh ,Nl EU (W0 + p (f (Nh ) + αNh ε) + N̄ − Nh − Nl w, Nl ).
In this case, you have utility from wealth, which is initial wealth
plus income, and leisure.

Solutions to Problem Set 3

(1) Solution

A. According to the mean-preserving spread criterion, portfolios

with the same mean may be compared. In this case, I, II, and
IV have E(X) = 10. The means of III and V are 9.5 and 28/3,
respectively. The variances of portfolios I to V are 19.3, 19.8, 27.4,
35.8, and 19.6, respectively. Riskier portfolios have “more weight
in the tail.” The following figure shows the probability density
for portfolios I, II, and IV.
By observing the figure, we can draw the following conclusions:
II dominates IV (i.e., IV is a mean-preserving spread of II);
I dominates IV; and I and II cannot be compared because, for
442 Agricultural Economics and Policy

a small tail, I is riskier; but for a large tail, II is riskier. To

better see that I and II cannot be compared, note that conditional
on I and II having equal means, I is a mean-preserving spread
of II if and only if for any x, the area to the left of x under
the cumulative distribution function (CDF) curve of portfolio
I should be larger than or equal to that under CDF curve of
portfolio II. Mathematically, the condition can be expressed as,
 x  x
S (x) ≡
I
F (t)dt ≥ S (x) ≡
I II
F II (t)dt, ∀x,
−∞ −∞

where F I (·) and F II (·) are the CDFs of portfolios I and II,
respectively. One can easily check that S I (3) = S II (3) =
0, S I (5) = 1/3 > S II (5) = 0, S I (7) = S II (7) = 2/3, and
S I (10) = 10/6 < S II (10) = 13/6, showing that the condition
does not hold (see the following figure).
B. Second-order stochastic dominance: It is necessary to calculate
and plot S I (x) to S V (x) as a function of x described above. Any
 Appendix C: Solutions to Selected Problem Sets 443

portfolio for which the plot lies wholly to the right of another
dominates the other.
Conclusions: I dominates III, IV, and V; II dominates III and
IV but not V. I and II are not comparable in the sense of second-
order stochastic dominance.
C. Order the distributions using the mean-variance criterion.

I >II because it has the same mean and lower variance.

I >III because it has a higher mean and a lower variance.
I >IV because it has the same mean and lower variance.
I >V because it has a higher mean and a lower variance.
II >III because it has a higher mean and a lower variance.
II >IV because it has the same mean and lower variance.

Therefore, the ordering is I >II >III and IV and I >V. III and
IV, III and V, as well as IV and V are not comparable without
knowing further information about the penalty that the mean-
variance criterion puts on variance.
444 Agricultural Economics and Policy

D. Order the distributions using Roy’s minimum probability rule:

min [P (X ≤ 6)].

Portfolio I II III IV V
P (X ≤ 6) 1/6 1/3 1/3 1/3 1/6
So, I and V are preferred to II, III, and IV.
E. Order the distributions using Kataoka’s safety-fixed rule: Maxi-
mize d subject to P (X ≤ d) ≤ 1/6.

I II III IV V
max d 7-ε 5-ε 2-ε 3-ε 8 - ε.
where ε is a positive infinitesimal number. Thus, V >I >II >IV
>III.
F. Order the distributions using U (X) = ln (X+ 1).
E(U ) = ΣP (X) ln (x + 1)
I II III IV V
E(U ) 2.2984 2.3075 2.1702 2.2115 2.2127
Thus, II >I >V >IV >III.
G. The mean-preserving spread, second-order stochastic dominance,
mean-variance criterion, and Roy’s minimum probability rule
(min [P (X ≤ 1/6)]) give partial orderings only. Kataoka’s safety-
fixed rule and the utility function give complete orderings.
The mean-preserving spread and second-order stochastic dom-
inance give rankings (albeit partial) and the basis of any risk-
averse utility function. That is, the orderings implied by these
two methods will be true for any risk-averse individual. Note that
the partial orderings implied by these two methods are consistent
with the orderings implied by the utility function.
The orderings implied by the mean-variance criterion are con-
sistent with expected utility maximization if the utility function is
defined over only the mean and variance (or if the utility function
is exponential and the distributions are all normally distributed).
Note that the orderings implied by M −V are not consistent with
those implied by the log utility function.
The orderings implied by the safety rules are not consistent
with any sort of utility or expected utility maximization.
 Appendix C: Solutions to Selected Problem Sets 445

(2) Solution

A. The farmer maximizes his expected utility of income:

max EU [L[pf ( , m) + pg( , m) · ε − w − rm] − F ] .
,m

The first-order condition with respect to is:

∂EU  
= E U  · [L(pf + pg · ε − w)] = 0,
∂
where U  is shorthand for U  (L[pf ( , m)+pg( , m)·ε−w −rm]−
F ) (the same for the below-mentioned U  ). Based on this first-
order condition and on a property of the expected value of the
product of two random variables (i.e., E(X · Y ) = E(X) · E(Y ) +
cov(X, Y )), we have
E(U  ) (Lpf − Lw) + Lpg cov(U  , ε) = 0.
Since E(ε) = 0,
−pg cov(U  , e)
pf − w = .
E(U  )
With g > 0 and cov(U  , ε) < 0, this indicates that pf − w > 0.
Under certainty, ε = 0, which implies that cov (U ’, ε ) = 0.
Therefore, with complete certainty, pf −w = 0. With diminishing
returns to the factor 1 (i.e., f < 0), this implies that, with
uncertainty, less of the input will be used. The result is, of course,
that, with everything else the same, the output will be less. This
seems intuitive for increased use of the input l increases not only
the expected yield but also the variation in yield. A risk-averse
farmer would, therefore, limit the use of the input because, by so
doing, he reduces the risk.
Analogously, we can study the optimal condition for input m:
∂EU  
= E U  [L(pfm + pgm · ε − r)] = 0,
∂m
L · E(U  ) (pfm − r) + L pgm cov(U  , ε) = 0
−pgm cov(U  , ε)
pfm − r =
E(U  )
because gm < 0, we can conclude that pfm − r < 0.
446 Agricultural Economics and Policy

As above, certainty (or risk neutrality) implies that

cov(U  , ε) = 0 and, hence, pfm − r = 0. Comparing the case
of certainty with that of uncertainty, it is clear that more of
the input m is used when uncertainty is introduced, because
increasing m reduces the variance in yield.

Comparative statics

To examine the effect of a change in fixed costs and a change in

farm size on input use, we differentiate totally the two first-order
conditions with respect to , m, f, and L.
The total derivative of first-order condition E[U  Π ] = 0 can be
written as,

   
E U  Π + U  Π Π d + E U  Πm + U  Π Πm dm
     
h11 h12
  
+ E U  ΠF + U  Π ΠF dF + E U  ΠL + U  Π ΠL dL = 0.
     
a11 a12

For first-order condition E [U  Πm ] = 0, the result of total differenti-

ation is

   
E U  Πm + U  Πm Π d + E U  Πmm + U  Πm Πm dm
     
h21 h22
 
  
+E U ΠmF + U Πm ΠF dF + E U ΠmL + U  Πm ΠL dL = 0.
 
     
a21 a22

In sum, we have

       
h11 h12 d a11 a12 dF
+ = 0,
h21 h22 dm a21 a22 dL
 Appendix C: Solutions to Selected Problem Sets 447

where
Π = L(pf + pg · ε − w)
Πm = L(pfm + pgm ε − r)
Π = L(pf + pg · ε)
Πm = L(pf m + pgm · ε)
Πm = L(pfm + pgm · ε)
Πmm = L(pfmm + pgmm · ε)
ΠF = 0
ΠmF = 0
ΠL = (pf + pg · ε − w)
ΠmL = (pfm + pgm · ε − r)
ΠF = −1
ΠL = (pf + pgε − w − rm).

Solving for d and dm , we obtain:

       
d 1 h22 −h12 −a11 −a12 dF
= ,
dm |H| −h21 h11 −a21 −a22 dL
 
h11 h12
where H ≡ .
h21 h22
From this, we can see that:
d 1
= [−a11 h22 + h12 a21 ]
dF |H|
dm 1
= [a11 h21 − a21 h11 ] .
dF |H|
If we substitute for a11 and a21 , we obtain:
d 1    
= (E[U  Π ])h22 − E U  Πm ]) · h12 .
dF |H|
To sign this expression, we assume that the second-order condition
for the maximization problem hold, i.e., H is a negative definite
matrix. This implies that h11 and h22 are negative and that, with an
448 Agricultural Economics and Policy

even number of decision variables (namely, and m), |H| is positive.

To sign the term E [U  Π ] , we can proceed as what we have discussed
in the production chapter:
 >
E U  · L(pf + pg · ε − w) 0.
<

Define Π̃ as a point where L(pf +pg ε−w) = 0. If pf +pg ε > w, then
Π > Π̃ and, with decreasing absolute risk aversion, Ra (Π) < Ra (Π̃).
If pf + pgε < w, then Π < Π̃ and Ra (Π) > Ra (Π̃). Then, for all
values of pf · pgε − w,

Ra (Π) (pf + pg ε − w) < Ra (Π̃) (pf + pg ε − w)

U  (Π)
⇒− (pf + pg ε − w) < Ra (Π̃) (pf + pg ε − w)
U  (Π)
⇒−U  (Π)(pf + pg ε − w) < Ra (Π̃) U  (Π) (pf + pg ε − w)
 
⇒−E U  (Π)(pf + pg ε − w)] < Ra (Π̃) E[U  (Π) (pf + pg ε − w) .

Hence, E [U  Π ] > 0 because E [U  Π ] = 0 from the first-order con-

dition. By an analogous argument, we should be able (in principle) to
show that E [U  Πm ] < 0. If this is the case and if we can assume that
h12 < 0, then (d /dF ) < 0 and (dm/dF ) > 0. These results indicate
that under decreasing absolute risk aversion, reduction in fixed costs
will increase the use of “risk-increasing” input and decrease the use
of “risk-decreasing” input m. This is intuitive because when fixed
costs F decreases, the decision maker has more wealth and thus
becomes less sensitive to risk. Therefore, she is more willing to use
risk-increasing input and less willing to use risk-decreasing input.
To examine the effect of farm size on input use, we look at the
expressions

d 1
= (−a12 h22 + a22 h12 )
dL |H|
dm 1
= (a12 h21 − a22 h11 ).
dL |H|
 Appendix C: Solutions to Selected Problem Sets 449

Although it has not been shown mathematically, the intuitive answer

to this is that, with constant relative risk aversion, (d /dL) =
(dm/dL) = 0. That is, when the farm is large, the farmer is just
as unwilling to risk, say, 5% of his farm as he is when it is small.
As a result, the level of input usage per acre should remain exactly
the same as it was when the farm was small. Intuitively, because the
decision maker is constant-relative-risk averse, a proportional change
in her wealth does not affect her decisions.

B. The farmer’s maximum expected utility is:

 
E(U ) = 1 − E e−δΠ
 
= 1 − E e−δ[L(pf +pgε−w−rm−F )]
 
= 1 − e−δ[L(pf −w−rm−F )] E e−δLpgε .

 
Because ε ∼ N (0, σ), for any number t, we have E etε =
2 2   2 2 2 2 2
et σ /2 . If we define t = −δLpg, then E e−δLpgε = eδ L p g σ /2 .
Therefore, to maximize E(U ), we maximize

2 L2 p2 g 2 σ 2 /2
E(U ) = 1 − e−δ[L(pf −w−rm−F )] eδ
2 L2 p2 g 2 σ 2 /2
= 1 − e−δ[L(pf −w−rm−F )]+ δ .

The first-order condition is:

 
∂E(U ) 1 2 2 2
= e−δ[·] 2
−δL(pf − w) + δ L p 2gg σ = 0
∂ 2
⇒ pf − w = δLp2 gg σ 2 .

With g and g > 0, this implies that pf − w > 0. This means
that less of input is used as compared with the case of complete
certainty (σ 2 = 0) or the case of risk neutrality (δ = 0), both of
450 Agricultural Economics and Policy

which imply that pf − w = 0.

∂E(U )
= 0 ⇒ pfm − r = δLp2 ggm σ 2 .
∂m
Because gm < 0, this implies that pfm − r < 0. In this case, more
of input m is used as compared to the case of complete certainty
or risk neutrality.

Comparative statics

Because fixed costs do not appear in the first-order conditions,

a change in fixed costs will not affect the level of input use (unless
the increase in fixed costs is so great that it becomes unprofitable for
the firm to produce and it is shut down).
To determine the effect of a change in farm size, we differentiate
totally the two first-order conditions and obtain:
    
pf − δLp2 σ 2 gg + g2 d + pfm − δLp2 σ 2 (ggm + g gm ) dm
     
h11 h12

−δp2 gg σ 2 dL = 0.
  
a1
   
pfm − δLp2 σ 2 (ggm + gm g ) d + pfmm − δLp2 σ 2 (ggmm + gm
2
) dm
     
h21 h22

−δp2 ggm σ 2 dL = 0.
  
a2

Therefore, we have,
       
h11 h12 d a1 0
+ dL =
h21 h22 dm a2 0

    
d 1 h22 −h12 −a1
⇒ = dL
dm |H| −h21 h11 −a2
⎧
⎪ d 1
⎨ = |H| (−h22 a1 + h12 a2 )
⇒ dL
⎪
⎩ dm = 1 (h a − h a ) .
|H| 21 1 11 2
dL
 Appendix C: Solutions to Selected Problem Sets 451

To sign these two expressions, we assume that the second-order

condition for the maximization problem holds, i.e., H is a negative
definite matrix. This implies that h11 and h22 are negative and that,
with an even number of decision variables (namely, and m), |H| is
positive. By inspection, it is clear that a1 <0 and a2 >0. If h12 =
h21 <0, then d /dL < 0 and dm/dL > 0. Intuitively, these results
are reasonable. The exponential utility function implies increasing
relative risk aversion (Rr = δ · Π). Thus, as farm size increases, it
seems likely that the farmer will use less of the risk-increasing input,
, and more of the risk-decreasing input, m. However, as may be seen
from an examination of the expression h12 , the sign of this term will
depend on how f and g change with m (and how fm and gm change
with ). Note, however, that h12 < 0 is a sufficient condition; if its
sign should change, this sign on d /dL and dm/dL could remain as
above.

C. Assume now that ε ∼ gamma with E(ε) = 1. We saw that

 
E(U ) = 1 − e−δ[L(pf −w−rm−F )] E e−δLpgε .

For the gamma distribution, E(etε ) = (1 − βt)−α , where α, β

are parameters of the distribution, and E(ε) = αβ = 1; V (ε) =
αβ 2 ; α, β > 0; and ε >0. If we define t = −δ Lpg, we have
E(U ) = 1 − e−δ[·] (1 + βδLpg)−α .
The problem for the farmer is to maximize
E(U ) = 1 − e−δ[·] (1 + βδLpg)−α
∂E(U )
⇒ = −e−δ[·] (1 + βδLpg)−α (−δ [L(pf − w)])
∂
− e−δ[·] (−α) (1 + βδLpg)−α−1 βδLpg = 0
−βpg α
⇒ pf − w = .
1 + βδLpg
Now let us assume that there is no risk. Replacing ε with 1
in the production function, profit maximization yield first-order
−βpg α −pg
condition pf − w = −pg < 1+βδLpg = 1+βδLpg . Note that the
last equality holds because E(ε) = αβ = 1. Therefore, we can see
452 Agricultural Economics and Policy

that with risk, less will be used than under the case of certainty.
Similarly, we have
∂E(U ) βpgm α pgm
= 0 ⇒ pfm − r = − =− < −pgm .
∂m 1 + βδLpg 1 + βδLpg
The last inequality holds because gm < 0. This implies that more
of input m will be used than under complete certainty.
D. Assume that ε ∼ N (0, σ 2 ). The farmer acts under Kataoka’s
safety-fixed rule:

Maximize d s.t. p(Π ≤ d) ≤ α.

If ε ∼ N (0, σε2 ), then Π ∼ N ([L · (pf − w − rm − F )] , (Lpgσ)2 ),

or Π ∼ N (uπ , σπ2 ). Thus,
 
Π − uπ d − uπ
p ≤ =α
σπ σπ
or
d − uπ
= Φ−1 (α),
σπ
where Φ(·) is the cumulative distribution function of standard
normal distribution. Suppose α = 0.05, then Φ−1 (α) = −1.64.

d = uπ + Φ−1 (α)σπ
d = L(pf − w − rm − F ) − 1.64 Lpgσ
∂d
= L(pf − w) − 1.64 Lpg σ = 0
∂
pf − w = 1.64 pg σ > 0.

Again, note that, since pf − w > 0, less of input will be used
than certainty (σ = 0). Similarly,
∂d
= 0 ⇒ pfm − r = 1.64 pgm σ < 0.
∂m
Because pfm − r < 0, more of input m will be used than under
uncertainty.
 Appendix C: Solutions to Selected Problem Sets 453

E. The behavior implied by all of the models was very similar. As

compared with the certainty case, less of input λ was used and
more of input m was used. The precise levels of usage of these
inputs differed from model to model. In general, however, the
level of input use depended on the risk-averseness of the farmer
(e.g., δ and α) and the degree of riskiness of production (e.g., σ
and φ).
This page intentionally left blank
 Index

A climate-complementing technology,
221
absolute risk aversion, 82–83
climate-substituting technology, 221
adoption, 227
Cobb–Douglas function, 19, 21
adoption behavior, 295
Cobb–Douglas production function,
advertisements, 297
31
advertisements, brands, 298 common pool problem, 217
advertisements, timely, 298 common ratio effect, 128
aggregate production function, 37 compensated demand, 58
almost ideal demand system (AIDS), conservation reserve program, 13, 353
55, 60, 62 constant elasticity of substitution
ambiguity, 137 demand, 60
ambiguity averse, 138 consumer surplus, 150, 175
ambiguity seeking, 138 consumption technology, 70
Arrow–Dasgupta (AD) approach, 388 creative destruction, 267
crop insurance, 155
B cumulative prospect theory, 134, 136
bioeconomy, 395
D
biofuel tax credit, 163
Bracero program, 14 deadweight loss, 167
decreasing absolute risk aversion, 87
C deficiency payment, 12, 14, 151, 161
demand elasticity, with respect to
Cambridge controversy, 37–38 own price; with respect to income,
certainty effect, 122, 126 57
certainty equivalence, 157 demand theory, 260
climate change, adaptation, 363, 368 diffusion, 201, 227
climate change, geoengineering, 370 directed innovation hypothesis, 204
climate change, mitigation, 363, 372 directed innovation hypothesis,
climate smart agriculture, 375 market size effect, 208

455
456 Agricultural Economics and Policy

directed innovation hypothesis, price health risk-generating process,

effect, 208 dose-response, 335
direction of innovation, 208 health risk-generating process,
disaster assistance programs, 13 exposure, 335
disembodied hypothesis, 198 hedonic pricing, 263
disembodied innovations, 267 Hicksian demand, 58
Hotelling’s lemma, 23
E household production function, 260
economies of scale, 19 hyperbolic discounting, 247
education–industry complex, 270
efficiency substitution, 73 I
elasticity of input use efficiency, 28 implicit Marshallian demand, 65
elasticity of marginal product (EMP), increasing relative risk aversion, 87
28, 30 increasing returns to scale, 19
elasticity of substitution, 21 independence axiom, 126
embodied innovations, 267 independence of irrelevant
embodied-disembodied argument, 198 alternatives, 183–184
environmental protection agency indirect utility function, 56
(EPA), 13 induced innovation hypothesis, 201,
environmental quality incentives 203
program (EQIP), 353 inferior good, 57
Exact Affine Stone Index (EASI) innovation, 201, 266–267
model, 55, 60, 67 innovation process, 267, 269
Exact Stone Index (ESI) demand, 66 innovation, institutional, 268
expenditure function, 57 innovation, managerial, 268
innovation, process, 267
F innovation, product, 267
farm problem, 10 innovation, resource use, 268
food assistance programs, domestic, input elasticity, 21
315 integrated pest management (IPM),
food assistance programs, 330
international, 315 interest group, 179
food insecurity, 311, 316 inventory control policies, 12
food insecurity, severe, 311 isoquants, 19
food safety, 308, 320
food security, 310 J
Jeffersonian Vision, 5
G Just and Pope production model, 113
Giffen good, 57
government revenue, 150 L
L-augmenting, 204
H L-biased, 204
health risk, 308 Lancaster’s model, 71
health risk-generating process, learning-by-doing, 289
contamination, 335 Leontief production function, 19, 21
 Index 457

linear invariance axiom, 183 P

loading factors, 156 Pareto optimality, 183
loss aversion, 132 payment for ecosystem services
luxury good, 57 (PES), 353
payment for ecosystem services
M (PES), slippage, 355
α-MaxMin model, 138, 141 payment for ecosystem services
Machina–Marschak triangle, 123 programs, 13
marginal rate of technical peasants, 5
substitution (MRTS), 21 pesticide application, economic
marketing, 294 threshold, 330
marketing, awareness, 296 pesticide application, preventive,
marketing, evaluating alternatives, 333
296 pesticide application, reactive,
marketing, information search, 296 333
marketing, product feedback, 296 pesticide ban, cost budgeting method,
marketing, purchasing decision, 342
296 pesticide ban, Delphi method, 342
marketing, reminders, 298 pesticides, 329
marketing, rental and leasing, 301 political economic resource
marketing, salespeople, 302 transactions (PERTs), 172
marketing, warranties, 301 political economy, 171, 189
Marshallian demand, 56, 58 poverty trap, 314
MaxMax model, 138, 141, 144 Preemption Act of 1841, 4
MaxMin model, 137, 139, 141, 144 price support, 150, 161–162
mean–variance approach, 100 price support policies, 11
mean-preserving spread, 82, primary pest, 331
108–109 private substitution, 73
median voter model, see also median probability weighting, 132
voter theory producer surplus, 150, 175
median voter theorem, 171 production function, macro, 38
median voter theory, 173–174 production function, micro, 38
moment-generating functions, 88 production regulation, 161
money-back guarantee (MBG), prospect theory, 82, 131, 246
299 putty-clay model, 37

N Q
national school lunch program quadratic production functions, 19
(NSLP), 319
normal good, 57 R
recycling, 394
O reference point, 132
one health framework, 324 reflection effect, 122
ordinary good, 57 relative risk aversion, 82, 86
output elasticity, 28 relentless innovation, 252
458 Agricultural Economics and Policy

rent, economic, 174 supply chains, product, 277

rent-seeking, 174–175 supply chains, risk, 287
risk, 81 supply chains, symbiotic, 277
risk assessment models, 336 supply chains, three-segment, 273,
risk generation function, 307 276
risk, additive, 95 supply chains, two-segment,
risk, linear, 95 271
risk, multiplicative, 95 supply controls, 153–154
Roy’s identity, 58 sustainable development, 383
Roy’s safety rule, see also safety rules, sustainable development, micro-level
Roy’s minimum probability rule analysis, 392
sustainable development, strong
S sustainability, 385
S-shaped diffusion curves, 228 sustainable development, the
Safety First approach, see also safety macro-economic approach,
rules, Roy’s minimum probability 384
rule sustainable development, weak
safety rules, 115, 337 sustainability, 385
safety rules, Kataoka’s safety-fixed symmetry axiom, 183
rule, 118
safety rules, Roy’s minimum T
probability rule, 116–117 technological bias, 199
safety rules, Telser’s safety-first rule, technological change, 198
117 technological change, Harrod’s
Salter model, 39, 41 measure, 199
scale elasticity, 21 technological change, Hicks’ measure,
Schumpeterian view, 284 199
set-aside program, 13, 154 technological change, Solow’s
set-aside requirements, 12 measure, 199
sharing of spoils, 177 technology adoption, dynamic
Shephard’s lemma, 58 models, 236
Slutsky equation, 59–60 technology adoption, imitation
Solow–Rawls (SR) approach, 387 model, 229
stochastic dominance, 82, 110 technology adoption, imitators,
stochastic dominance, first-order, 250
110 technology adoption, innovators,
stochastic dominance, second-order, 250
111 technology adoption, real option
supplemental nutrition assistance models, 243
program (SNAP), 319 technology adoption, static expected
supply chains, 259, 270 utility model, 234
supply chains, credit, 286 technology adoption, threshold
supply chains, innovation, 277 model, 231
supply chains, innovation-induced, The Homestead Act, 4
275 total social welfare, 150
 Index 459

U W
uncertainty, 81, 143 welfare analysis, 149
uncompensated demand, 56 welfare analysis, basic, 150
utility functions, Cobb-Douglas, 88 welfare analysis, overlapped policies,
utility functions, exponential, 88 158
utility functions, logarithmic, 88 westward movement, 4
utility functions, quadratic, 88 willingness to pay (WTP), 150
willingness to sell (WTS), 150
V
value of marginal product (VMP), 30 Z
value of statistical life (VSL), 309 Z-augmenting, 204
von Thünen model, the, 347 Z-biased, 204