Introduction to economic analysis

Alberto Bisin
Dept. of Economics
NYU

January 12, 2011

Contents

Preface ix

1 Introduction 1
1.1 Theory, Models, and Empirical Analyis . . . . . . . . . . . . . 4
1.2 Debates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Empirical analysis . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 Some successful results . . . . . . . . . . . . . . . . . . . . . . 13
1.6 Fun readings . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.7 Useful references . . . . . . . . . . . . . . . . . . . . . . . . . 15

2 Rational choice 17
2.1 Preferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Rationality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.3 Utility representation . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.5 Social choice . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.6 Classic references . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.7 Useful references . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3 Consumer choice 35
3.1 The Budget constraint . . . . . . . . . . . . . . . . . . . . . . 36
3.2 Indi¤erence curves . . . . . . . . . . . . . . . . . . . . . . . . 38
3.2.1 Monotonicity . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.2 Convexity . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.3 Types of preferences . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.1 Perfect substitutes . . . . . . . . . . . . . . . . . . . . 45

v
vi CONTENTS

3.3.2 Perfect complements . . . . . . . . . . . . . . . . . . . 45

3.4 Consumer problem . . . . . . . . . . . . . . . . . . . . . . . . 46
3.5 Demand Functions . . . . . . . . . . . . . . . . . . . . . . . . 49
3.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

4 Equilibrium and e¢ ciency 61

4.1 The Economy . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.1.1 Notation and De…nitions . . . . . . . . . . . . . . . . . 62
4.2 Pareto e¢ ciency . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2.1 The Social Planner problem . . . . . . . . . . . . . . . 64
4.3 Competitive equilibrium . . . . . . . . . . . . . . . . . . . . . 66
4.4 Welfare Economics . . . . . . . . . . . . . . . . . . . . . . . . 67
4.4.1 The Edgeworth box . . . . . . . . . . . . . . . . . . . . 68
4.5 Externalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

5 Equilibrium in the labor market 79

5.1 Competitive labor markets . . . . . . . . . . . . . . . . . . . . 79
5.2 Search frictions . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.2.1 Pareto ine¢ ciency of search equilibrium . . . . . . . . 82
5.3 Empirical implications of search equilibrium . . . . . . . . . . 82
5.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

6 Time and uncertainty 85

6.1 Choice over time; a.k.a. Saving . . . . . . . . . . . . . . . . . 86
6.1.1 Intertemporal preferences for consumption . . . . . . . 86
6.1.2 The intertemporal budget constraint . . . . . . . . . . 87
6.1.3 Choice over time, for given r . . . . . . . . . . . . . . . 88
6.1.4 Determinants of savings . . . . . . . . . . . . . . . . . 90
6.1.5 The interest rate r . . . . . . . . . . . . . . . . . . . . 90
6.1.6 In‡ation and Monetary Neutrality . . . . . . . . . . . . 91
6.2 Choice under Uncertainty; a.k.a. Portfolio choice . . . . . . . 93
6.2.1 Probability . . . . . . . . . . . . . . . . . . . . . . . . 93
6.2.2 Expected Utility . . . . . . . . . . . . . . . . . . . . . 94
6.2.3 Risk Aversion . . . . . . . . . . . . . . . . . . . . . . . 95
6.2.4 State contingent consumption . . . . . . . . . . . . . . 95
6.3 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
CONTENTS vii

7 Growth 121
7.0.1 Production . . . . . . . . . . . . . . . . . . . . . . . . 122
7.0.2 Growth and Savings . . . . . . . . . . . . . . . . . . . 123
7.0.3 Other Determinants of Growth . . . . . . . . . . . . . 125

8 Finance 129
8.1 Asset Pricing and the CAPM . . . . . . . . . . . . . . . . . . 129
8.1.1 Portfolio Choice . . . . . . . . . . . . . . . . . . . . . . 130
8.1.2 Time and Risk Correction . . . . . . . . . . . . . . . . 130
8.1.3 Beta Representation . . . . . . . . . . . . . . . . . . . 132
8.1.4 The Term Structure of Interest Rates . . . . . . . . . . 134
8.1.5 The Market Rate of Return and the betas . . . . . . . 135
8.1.6 Equivalent Risk Neutral Representation . . . . . . . . . 136
8.2 E¢ cient Market Hypothesis . . . . . . . . . . . . . . . . . . . 137
8.2.1 Idiosyncratic Risk Does Not A¤ect Prices . . . . . . . . 137
8.2.2 Prices Adjust Immediately to All Available Information 137
8.2.3 Risk Adjusted Prices Are Not Predictable . . . . . . . 138
8.2.4 Arbitrage Opportunities Are Not Present in Financial
Markets . . . . . . . . . . . . . . . . . . . . . . . . . . 138
8.3 Modigliani-Miller Theorem . . . . . . . . . . . . . . . . . . . . 140
8.3.1 Modigliani-Miller More Formally . . . . . . . . . . . . 140
8.3.2 Other Implicit Assumptions . . . . . . . . . . . . . . . 142
8.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

9 Incentives 157
9.1 Principal-Agent Problems . . . . . . . . . . . . . . . . . . . . 158
9.1.1 The symmetric information benchmarks . . . . . . . . 159
9.1.2 Moral hazard . . . . . . . . . . . . . . . . . . . . . . . 161
9.1.3 Adverse selection . . . . . . . . . . . . . . . . . . . . . 165
9.1.4 Social security . . . . . . . . . . . . . . . . . . . . . . . 169
9.1.5 Other examples [to be added] . . . . . . . . . . . . . . 172
9.2 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172

10 Fiscal policy 175

10.1 Ricardian Equivalence - a.k.a. Arithmetics . . . . . . . . . . . 175
10.1.1 Borrowing-Constraints . . . . . . . . . . . . . . . . . . 177
10.1.2 Distortionary Taxes . . . . . . . . . . . . . . . . . . . . 178
10.1.3 Krugman, ....., and the army of …scal policy hawks . . 181
viii CONTENTS

10.2 Optimal taxation . . . . . . . . . . . . . . . . . . . . . . . . . 186

10.2.1 Distortionary Taxation . . . . . . . . . . . . . . . . . . 186
10.2.2 Public Goods Provision . . . . . . . . . . . . . . . . . . 187
10.2.3 Time Inconsistency . . . . . . . . . . . . . . . . . . . . 191
10.3 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192

11 Empirical economics 193

11.1 Correlation and regression analysis . . . . . . . . . . . . . . . 195
11.2 Controlled experiments . . . . . . . . . . . . . . . . . . . . . . 198
11.3 Natural experiments . . . . . . . . . . . . . . . . . . . . . . . 199
11.4 Structural econometrics . . . . . . . . . . . . . . . . . . . . . 202
11.4.1 Problems with regression analysis . . . . . . . . . . . . 202
11.4.2 Back to models . . . . . . . . . . . . . . . . . . . . . . 205
11.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

12 Game theory 207

12.1 Strategic Games . . . . . . . . . . . . . . . . . . . . . . . . . . 208
12.1.1 Nash equilibrium . . . . . . . . . . . . . . . . . . . . . 209
12.1.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . 209
12.1.3 Mixed Strategies . . . . . . . . . . . . . . . . . . . . . 213
12.2 Dynamic Games . . . . . . . . . . . . . . . . . . . . . . . . . . 214
12.3 Bargaining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
12.4 Reputation and repeated games . . . . . . . . . . . . . . . . . 215
12.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216

13 Mathematical Appendix 217

13.1 Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
13.2 Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
13.3 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
13.4 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
13.4.1 Some Properties of Functions . . . . . . . . . . . . . . 220
13.4.2 Optimization . . . . . . . . . . . . . . . . . . . . . . . 224
13.4.3 Solving the Cobb-Douglas Consumer’s Problem . . . . 229
13.5 Linear algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
13.5.1 Reduction to echelon . . . . . . . . . . . . . . . . . . . 233
13.5.2 Independent columns . . . . . . . . . . . . . . . . . . . 234
13.5.3 Inverse matrix . . . . . . . . . . . . . . . . . . . . . . . 235
13.5.4 Determinant . . . . . . . . . . . . . . . . . . . . . . . . 235
CONTENTS ix

13.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235

13.6.1 Sets and numbers . . . . . . . . . . . . . . . . . . . . . 235
13.6.2 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . 236
13.6.3 Maximization . . . . . . . . . . . . . . . . . . . . . . . 240
13.6.4 Linear algebra . . . . . . . . . . . . . . . . . . . . . . . 245
 Preface

Introduction to Economic Analysis is the introductory class I taught for

several years at NYU. It is the …rst class of the theory concentration major in
Economics. Though this is not a plain vanilla Economics 101, it is nonetheless
intended for students who wish to begin their formal study of economics.
Its goal is to develop concepts and analytical tools that economists use to
understand general social and economic phenomena. This book originates as
lecture notes for this class.
As such the book relies on a relatively high level of abstraction and
focuses on the concepts and the techniques of economic analysis rather than
on the understanding of speci…c economic phenomena or institutions. It is
particularly well suited for students/readers who are interested in pursuing
careers or higher degrees in economics or in quantitative …elds such as …nance.
But readers who favor an analytic approach to their intellectual endeavors
might …nd the book enjoyable.
Some math background is needed to read this book. I o¤er no apolo-
gies. Economists very often express their ideas using mathematical concepts
simply because these allow them to express themselves more precisely than
with spoken language. Virtually, all the math used in the book is (or should
be) taught in high school. We shall explain thoroughly any mathematical
concept we use and we shall refer to the Math Appendix for details when
necessary. How much the reader will get out of the book will depend more
on how much fun he/she has doing math than on how much math he/she
knows.
My gratitude goes to James Ramsey, Director of Undergraduate Studies
at NYU, for convincing me to teach in the undergraduate program and for
having …rst envisioned and then supported the whole idea of a course like
Introduction to Economic Analysis. My gratitude goes also to the many
students who have su¤ered through various versions of my notes over the
years. I am proud to report that two of them are now successful academic
economists, Ahu Gemici at NYU and Laura Pilissoph at the University of
Chicago. The book also owes a lot to an exceptional list of T.A.’s who have
helped me and who wrote (and solved) most of the problems: Shachar Kariv,
Mark Dean, Victor Archavski, Ignacio Esponda, Luke Geldermans. They are
also all by now successful academic economists.
The reader with some knowledge of the methodological distinctions in
the economics profession will certainly note a certain Chicago freshwater

ix
x Preface

style prevailing in the book. My graduate school years at the University of

Chicago in the early nineties have fundamentally marked the way I do (and
teach) economics. I enthusiastically take this occasion to thank Gary Becker,
Jose’Scheinkman, Mike Woodford, Bob Lucas, John Cochrane, ... for their
help and their inspiration, as well as my closest peers there, Danilo Guaitoli,
Afonso Mello Franco, Giorgio Topa, Fulvio Ortu.
Jess Benhabib, Andy Schotter, Chuck Wilson in particular, have shaped
my

The book is dedicated to my son Vittorio. He was just learning to read

when I started designing Introduction to Economic Analysis. Now he could
read this book.
Preface xi
Preface 1
Chapter 1

Introduction

[...] the di¤erence between economics and sociology is very simple. Eco-
nomics is all about how people make choices. Sociology is all about why they
do not have any choices to make. (James S. Duesenberry).

Adam Smith David Ricardo

Economics is a broad-ranging discipline in scope. Consequently, various

de…nitions of economics have been attempted, with mixed success. Perhaps
the most in‡uential de…nition is due to Lionel Robbins (in his 1932 essay on
The Nature and Signi…cance of Economic Science):

economics is "the science which studies human behavior as a relation-

ship between ends and scarce means which have alternative uses."

1
2 CHAPTER 1 INTRODUCTION

The point of this de…nition is to stress the issue of scarcity, essentially

to point out that economists would be useless in Nirvana, where everybody
is happy with an unlimited supply of anything they might desire. Without
scarcity there are no choice problems; and no choice - no economics!
More pragmatically, it is nowadays commonplace to implicitly de…ne eco-
nomics functionally, as follows:

economics is the study of a speci…c set of phenomena we call ‘eco-

nomic;’ demand and supply of commodities, market equilibrium and
prices, e¤ects of monetary and …scal policies, [......].

But this (as well as any) functional de…nition is very narrow. Economists
actively study phenomena which are traditionally attributed to sociology, an-
thropology, political science, law, and even biology. Examples include: crime,
family, fertility, primitive societies (like hunter-gatherers), voting, compara-
tive analysis of political and legal institutions, social norms, social networks,
genetic and cultural evolution of preferences, and so many others we do not
have space to list.1
If a functional de…nition of economics is narrow, as the boundaries of the
subjects studied in the discipline keep expanding, then we better rely on a
methodological de…nition. An appropriate one could go something like this:

economics is the study of aggregate phenomena as equilibrium out-

comes of individual choices.

Individual choice and equilibrium analysis are in fact at the core of any
economic analysis of whatever issue. These terms are somewhat jargony and
hence it’s better to be clear about how we interpret them at the outset.
Individual choice. Individuals - not groups - make choices. When choos-
ing individuals evaluate rationally costs and bene…ts: they will choose the
best bundle of goods for their money; they will sell assets they know are over-
valued and will buy assets they know are under-valued; they will engage less
1
Applications to economics outside of typically economic phenomena has a long tradi-
tion in the discipline. The economist which is mostly responsible for this, who has studied
strange subjects when it was not cool to do so but who ultimately has received a Nobel
prize for his e¤orts along these lines, is Gary Becker, from the Univeristy of Chicago. His
view has been popularized recently by Steven Leavitt, also from the Univeristy of Chicago,
in a recent series of books/blog entries/NYTimes columns/movies/... under the title of
Freakonomics.
 3

in criminal activities if the probability of detection is higher, if punishment

is more severe, if more alternatives to crime are o¤ered ex-ante, if crime is
not "cool" in their social group; they will migrate when expeciting improve-
ments in their economic and social lifestyle. Rationality can be substantially
relaxed - and it is nowadays in economics. But it still represents the stan-
dard assumption in most of the discipline. Rational choice is referred to (and
criticized) as the method of economics in sociology and political science.
Equilibrium. Individuals in an economy or society interact through mar-
kets and through di¤erent institutions (families, …rms, schools, peers). Choices
of di¤erent individuals are connected in the economy, and an economic analy-
sis of a speci…c phenomenon considers all the relevant connections, the direct
and indirect e¤ects of a change in the determinants of such phenomenon, for
example. For instance, a preference shift of young people in favor of beef over
chicken will have an e¤ect in their demand, which will in turn have an e¤ect
on the relative price of beef and chicken, which will have an e¤ect on the de-
mand of all people, and of the producers of beef and chicken. An equilibrium
is the level of demand of young and old people, the price, the supply of beef
and chicken after all these e¤ects have been taken into account. Similarly, a
technological innovation which changes the demand for skilled workers will
have an e¤ect on their wage, and on the wage of un-skilled workers too; in
turn this will have an e¤ect on their supply (e.g., because more people will
acquire the demanded skills following an increase in the wage rate for skilled
workers).
Often the notion of economic equilibrium is criticized by other social
scientists (and even by some economists who should know better) on the
grounds that modern economies should be better represented as in a sort of
"permanent disequilibrium." This critique is based on a fundamental mis-
conception of what an economic equilibrium is, a misconception induced by
the meaning that the word "equilibrium" has in common parlance (but not
in economics!). An economic equilibrium is not in any sense a state in which
economic variables are constant or slowly moving or anything like that. It is
in fact possible to show that the notion of economic equilibrium, even when
combined with individual rationality, is perfectly consistent with any sort of
complex dynamics of e.g., aggregate capital or GNP. Even chaotic dynamics,
that is dynamics which do not converge to a constant not a cyclical state and
which depend greatly from initial conditions, are possible.2
2
The word "chaotic" also has a precise meaning in the mathematics of dynamical
4 CHAPTER 1 INTRODUCTION

Let’s try to see the economic method (individual choice+equilibrium) at

work on an issue which many would characterize as sociological, just to beat
the dead parrot some more. Consider the issue of the e¤ects of a minimum
drinking age law. Such a law might reduce alcohol consumption in minors.
You impose a constraint on a group (minors) - making it harder for them to
drink - and they’ll drink less. But this argument is based on group choice -
the group drinks less. If we think of the group as composed by individuals
- each one choosing rationally - we need to ask ourselves why it is rational
for each one of them to drink. A possibility is that alcohol consumption
is a manifestation of preferences for risky and rebellious behavior. In this
case, making drinking illegal would make it even more a risky rebellious
choice and hence would give minors a new incentive to drink. In this case a
minimum drinking age law might even exacerbate under-age drinking. But
suppose that the law punishes drinking with such severity that in fact it
reduces the incentives of minors to drink. In this case, in equilibrium, they
will satisfy their demand for risk and rebellion with some other substitute
behavior. Let’s say bungee-jumping from tall buildings. Then, to evaluate
the e¤ect of a minimum drinking age law it is not enough to see the e¤ects
on drinking, but rather the equilibrium e¤ects on bungee-jumping - which
might be worse than drinking - might be considered.... Now is the time
for you to be impressed about how deep the economic method (individual
choice+equilibrium) can get with ease and speed.

1.1 Theory, Models, and Empirical Analyis

An economist is the only professional who sees something working in practice
and then seriously wonders if it works in theory (Ronald Reagan)
Nothing is less real than realism. Details are confusing. It is only by
selection, by elimination, by emphasis that we get to the real meaning of
things (Georgia O’Kee¤e)3

Models are theoretical exercises of abstraction: ignoring many details in

order to focus on the most important elements of the problem.
systems that is not well represented by its meaning in common parlance.
3
Thanks to Andrea Moro for suggesting, and in turn to Narayana Kocherlakota for
discovering, the quotation. .
1.1 THEORY, MODELS, AND EMPIRICAL ANALYIS 5

Figure 1.1: Math in economics

There is no such thing as the right degree of abstraction for all analytic
purposes. The proper degree of abstraction depends on the objective of
the analysis. A model that is a gross oversimpli…cation for one purpose
may be needlessly complicated for another. A map might be an appropriate
metaphor for a model: we rarely need 1:1 maps; and sometimes we need a
map of the whole American continent, sometimes one of the upper east side
of NYC.
Models are not necessarily mathematical models. The following example
(taken from Krugman, 1995) illustrates this point:

Dave Fultz at the University of Chicago in the late 40’s showed

that a dishpan …lled with water, on a slowly rotating turntable, with
an electric heating device bent around the outside of the pan provides
a good representation of the basic pattern of weather. The dishpan
was build to model the temperature di¤erential between the poles and
the equator and the force generated by the earth’s spin (abstracting
from most of the the intricacies and complexities of the earth geogra-
phy) and was successfully shown to exhibit phenomena which could be
6 CHAPTER 1 INTRODUCTION

interpreted as tropical trade winds, cyclonic storms of the temperate

regions, and the jet stream.
But often models are in fact mathematical models: the most sophisti-
cated weather forecasts nowadays require the estimation of the parameters
of a large number (very large, hundreds) of equations. Most economic mod-
els, in particular, are mathematical models. This is in part due to the fact
that math is a very e¢ cient language for abstract arguments (especially, be-
cause it facilitates the manipulation of complex logical arguments and the
identi…cation of logical and conceptual mistakes in abstract arguments). But
also, mathematics, especially when coupled with fast computers, allows the
constructions of models which can be used as laboratories, that is, mathemat-
ical imitation-economies that generate simulated data which can be compared
with actual data from real economic systems (Lucas, 1980 develops this point
with clarity).
Models can therefore be tested with the methods of statistics and econo-
metrics. While models are not easily "falsi…ed" because their implications
are mostly not deterministic but rather stochastic (statistical, if you wish;
more on this later), econometric analysis supports some models against oth-
ers and in the end guides economic theorizing. For instance in the theory
of economic decision and in game theory, ample evidence of deviations from
rationality has signi…cantly a¤ected the models economists use. Also, the
econometric evidence regarding real e¤ects of monetary policy (e.g., of liq-
uidity injections) has severly a¤ected macroeconomic theory as well as actual
monetary policy.

1.2 Debates
Why does public discussion of economic policy so often show abysmal igno-
rance of the participants? Why do I so often want to cry at what public
…gures, the press, and television commentators say about economic a¤airs
(Robert M. Solow).
Politicians and reporters are fond of pointing out that economists can
be found on both sides of many issues of public policy. If economics is a
science, why do economists quarrel so much? After all, physicists do not
debate whether the earth revolves around the sun or vice versa.
The question re‡ects a misunderstanding of the nature of science! Dis-
putes are normal at the frontier of any science. Clearly, nowadays physicists
1.3 PREDICTIONS 7

do not argue whether the earth revolves around the sun but they did (quite
vociferously). They do now argue however how useful is superstring theory,
though their disagreement on this matter goes mostly unnoticed to the pub-
lic because only few of us understand what they are talking about. When
physicists debate about issues that politicians and the general public care
about and think they understand, the outcome is the same, as is it obvious
e.g., on the issues regarding the causes (and even the existence) of global
warming.
Disputes about economics are generally aired to the public and thus all
sorts of people are eager to join in the ensuing debates, relying on their
common sense. Unfortunately, common sense is not always a reliable guide in
economics since many economic relationships are counterintuitive. Hopefully,
by the end of our course we will have a better sense of when common sense
works and when it fails.4

1.3 Predictions
Many economic commentators have observed in the course of the last few
years that few economists have anticipated the …nancial market crisis in the
fall of 2008. A cursory look at the time series in the This is fact true, though
a handful of economists did in fact publish warnings about an impending
crisis. These economists include Nouriel - a.k.a. Doctor Doom - Roubini
(NYU Stern School of Business and RGE Monitor), Robert Shiller (Yale
University), Raghu Rajan (University of Chicago Booth School of Business).
A cursory look at Figure 1.2, which report the time-serie of the Dow Jones
Industrial Index (one of the most common U.S. stock market indexes), may
make it seem as if at least in the last quarter of 2008 the impending crisis
had been clearly manifesting itself. In fact crying wolf in the last quarter
of 2008, predicting a crisis would have been easy enough (with a guarantee
of get it right with about 50% probability). Many have done this in fact.
But this is not what we mean by predictions in economics. What we mean
is what for instance R. Rajan at the Federal Reserve Bank of Kansas City’s
Annual Jackson Hole Symposium in 2005, while head of the IMF. He did
not just predict a crisis, but he provided a coherent analysis of the reasons
for a possible impending crisis. This is the crux of the matter.
4
Besides, economists tend to agree on many issues, not lastly e.g., the economic gains
of free trade.
8 CHAPTER 1 INTRODUCTION

Figure 1.2: Dow Jones Industrial index

Let me try and better explain with a metaphor what I mean when I talk
of generating predictions out a coherent analysis.

Inter Milan will soon be playing A.C. Milan at San Siro. I

predict Inter Milan will win. How did I come up with this prediction?
Three possibilities:
1. I am an A.C. Milan fan and superstitiously (like any Italian) I
believe that saying that Inter Milan will win brings bad fortune to it.
2. A tarot reader/an oracle told me.
3. By selling Kaka’A.C. Milan has no valid attacking mid…elder.
A.C. Milan’s game will have to start in the defensive side of their
mid…eld, with A. Pirlo. Inter Milan, on the other hand, can play a
solid 4-3-1-2, with W. Sneijder behind the attackers. This will allow
Inter Milan to press A.C. Milan in its half, stopping A.C. Milan’s
souce of play in A. Pirlo. Inter Milan’s two new great forwards, D.
Milito and S. Eto’o will do the rest.
Ok, you do not need to know everything about Italian soccer to
understand what I am talking about. And you guessed it, generating
predictions out a coherent analysis means doing as in predictions of
type 3.
1.3 PREDICTIONS 9

Generating predictions out of a coherent analysis, not just predicting, is

the objective of economics as a discipline. You might already guess that by
a coherent analysis economists mean the outcome of a model. It must also
be added that economic models are typically stochastic models, that is they
predict stochastic processes over time. In other words, economic models do
predict the time path a quantity of interest, say the Dow Jones index, but
predict the probability associated to each possible time path of the index.5
If an event, say a …nancial crisis, has a very small probability of occurring -
it does not mean it can’t occur. If it does occur, it does not mean that the
model predicting a very small probability associated to the event is wrong.
Several investment banks had in fact predicted very accurately what would
have happened if real estate prices would have stopped growing (or, God
forbid, started to decline); they just associated a very low probability to the
event that real estate prices would start declining.
Stochastic models are a way to explicitly admit one’s inability to predict
with certainty. In this sense, stochastic prediction are, in some sense, an
admission of partial failure.
Consider the example of a fair coin throw. It is typically modelled as a stochas-
tic process in which the event Heads occurs with probability 1/2 and the same for
Tails. But I could construct a model of the dynamics of a coin, depending e.g., on
the initial position of the coin (Heads or Tail), the weight of the coin, the force the
thumb operates on the coin, ... With such a model, and with the observation of
my determining variables, I could probably predict the outcome of the coin throw
with more precision, or maybe even predicting it with relative certainty. You see
then the sense in which modeling the coin throw as a stochastic process is an ad-
mission of failure: I cannot construct a model of the dynamics of a coin (or I rely
on approximations, which is something else we do not want to get into here).
Finally, a quick and dirty comparison between predictions in economics
and in other disciplines can be useful as an illustration. A couple of examples
will be enough.
Classical physics is very successful in developing deterministic predictions
(and in testing them). But remember however that not all physics is about
forecasting the trajectory of a bullet, which can be done with a certain de-
terministic precision thanks to Newton. Quantum theory broke the certainty

5
Even in the context of soccer predictions, when I say I predict Inter Milan will win I
mean something like I predict that Inter Milan has higher chances of winning.
10 CHAPTER 1 INTRODUCTION

spell for physicists - to the point of leading A. Einstein to famously complain

that God does not play dice. And then physicists are better scientists (no
doubts about this). Or may-be they also have it easier (Max Plank appar-
ently thought so: Professor [Max] Planck, of Berlin, the famous originator of
the Quantum Theory, once remarked to me that in early life he had thought of
studying economics, but had found it too di¢ cult!," John Maynard Keynes, 1933.)
Also, in other hard sciences, meteorologists produce stochastic predic-
tions about the weather, vulcanologists about eruptions, geologists about
earthquakes,...
Anthropology as a discipline is methodologically centered on collecting
data and stressing di¤erences across human cultures. For the most, anthro-
pologists refuse to order data through models. Anthropological accounts
tend to be exceptionally interesting, but, without the use of models, predic-
tions are impossible (nor are they an aim of anthropologists). Ever heard
anybody saying, In that unexplored forest there has to be two, possibly three,
hunter and gatherer tribes whose system of religious beliefs is animistic and
one which is evolving towards monotheism? Nonetheless, we owe to anthro-
pologists a much deeper understanding of human nature; this is what they
are after, with great success.
Sociology and most of political science6 have similar, while much less
extreme, methodological characteristics as anthropology. A large part of po-
litical science, for instance, is comparative, essentially a detailed institutional
analysis of di¤erent political systems.
In medicine, a doctor can (and will) tell you that smoking increases your
chances to develop lung cancer, but cannot tell you if and when you will
develop one, nor that you won’t if only you stop smoking. Similarly, an
economist can (and will) tell you that allowing large leverage ratios in …-
nancial markets (investing with borrowed money) increases the chances of a
…nancial crisis, but cannot tell you if and when one will occur, nor that it
won’t if only we strike a stringent regulation on …nancial markets.
Having said all this, there is another reason why only few economists
have anticipated the …nancial market crisis in the fall of 2008: our theory of
asset prices predicts that asset prices are not predictable, in the sense that
6
A sizeable and growing part of political science has essentially adopted the method of
economics. Rational choice political scientists, so they are called, are not distinguishable
from economists, though they sit in di¤erent academic departments. This is not the case
for sociology. Rational choice sociologists are few still leave in reservations next to Native
Americans.
1.4 EMPIRICAL ANALYSIS 11

we can only say that riskier assets have higher returns on average (this is not
a joke, nor an empty theory - you will understand what I mean later on in
the course when we’ll do …nance). Furthermore, economists do not have an
established theory of asset pricing bubbles and hence tend to underestimate
the probability that a sustained growth in an asset price is a bubble. Many
economists are back to the blackboard studying bubbles these days. As it
should be.

1.4 Empirical analysis

Many observers (social scientists, journalists, people you cannot refrain from
talking when they should not,...) complain that economics as a discipline
reduces to useless theoretical/mathematical exercises. This is simply false.
It is most probably the distorted view of a tribe whose system of beliefs
include tenet that any math is too much math!
Even a super…cial look at a good economics journal (Econometrica, Amer-
ican Economics Review, Journal of Political Economy, Quarterly Journal of
Economics,...) will give you the (correct) impression that economics is an
applied discipline, addressing well-de…ned real world problems. For instance,
these are the titles of the last issues, June and August 2010, of the Journal
of Political Economy and Quarterly Journal of Economics, respectively:

Does Professor Quality Matter? Evidence from Random Assignment of Students

to Professors

Innovation, Firm Dynamics, and International Trade

Housing Externalities

Identifying Agglomeration Spillovers: Evidence from Winners and Losers of Large

Plant Openings

Competition and the Structure of Vertical Relationships in Capital Markets

Estimating Welfare in Insurance Markets Using Variation in Prices

Measuring Beliefs and Rewards: A Neuroeconomic Approach

Monetary Non-Neutrality in a Multisector Menu Cost Model

12 CHAPTER 1 INTRODUCTION

Regulation and Distrust

Improved Access to Foreign Markets Raises Plant-Level Productivity. . . for Some

Plants

Sex and Science: How Professor Gender Perpetuates the Gender Gap

Can Exchange Rates Forecast Commodity Prices?

Counterparty Risk in Financial Contracts: Should the Insured Worry about the
Insurer?

The Geographic Determinants of Housing Supply

School Choice with Consent

Muslim Family Law, Prenuptial Agreements, and the Emergence of Dowry in

Bangladesh

No papers on Hilbert spaces nor on Martingales or Matrix algebra per

se (though all these mathematical concepts are indeed used by economists
when deemed useful). If you opened the journals, however, you would …nd
a lot of equations - often statistical analyses of data (which economists call
econometrics).
Statistical analyses in economics are commonplace. They are also typi-
cally di¢ cult. This is because economists generally do not have access to con-
trolled randomized experiments, the typical instrument of empirical analysis
of the hard sciences and of medicine and farmacology. Let me illustrate.

A test of the e¤ects of a new drug e.g., for hypertension involves

selecting a group of subjects with the same medical characteristics
- a fraction of which are chosen randomly to receive the drug (the
others typically receives a placebo). Since the subjects have ex-ante
the same characteristics, any di¤erential outcome in the two groups
after the experiment is reasonably interpreted to be an e¤ect of the
drug. An economist willing to test the e¤ects of monetary policy
e.g., on employment, on the other hand, cannot use anything like a
controlled randomized experiment: all individuals in the economy are
a¤ected by the policy, no two groups with the same characteristics can
be selected.
1.5 SOME SUCCESSFUL RESULTS 13

As a consequence, economists have had to develop sophisticated methods

to try and identify relationships of cause-e¤ect in the data.7 We shall talk
about these methods in a subsequent chapter.

1.5 Some successful results

At the cost of appearing cocky, let me …nish the chapter by listing some
examples of important non-obvious results of economics which illustrate its
intellectual success as a discipline:

- The characterization of gains from trade in general and of international

trades in particular. This is the Invisible hand result (or First Welfare
Theorem).

- Comparative advantage (and not absolute advantage) determines trade

‡ows between countries: a country which is less productive than any
other one at producing all tradable goods will still …nd international
trade markets where to sell some of its products.

- Risk adjusted returns in …nancial markets are unpredictable. This is the

No-arbitrage theorem.

- Several "neutrality" or "equivalence" results (careful! these results hold

under restrictive assumptions - somewhat like "in the vacuum" results
in physics - and hence never in real economies, but are fundamental
benchmarks that, when not recognized, induce important logical mis-
takes):

i) Doubling the amount of money in an economy has no real e¤ects;

equivalently, dividing all prices by a third (multiplying the value
of a Dollar by three) has no real e¤ects. This is called Monetary
neutrality.
7
Recently, economists have been able to design randomized experiments regarding the
e¤ects of small scale policies in development, like, for instance, the e¤ect of a new school
in a isolated village. In these contexts they can in fact select villages with similar charac-
teristics and then choose randomly some of them to set up a school in. The Abdul Latif
Jameel Poverty Action Lab at MIT is at the forefront of these methods.
14 CHAPTER 1 INTRODUCTION

ii) Given government expenditures in the present and future, how they
are …nanced, e.g., by taxing now, or by taxing later, or by printing
money (in‡ation) has no real e¤ects. This is called Ricardian
equivalence.
iii) Given a …rm’s production plan in the present and in the future,
how is the …rm …nanced, e.g., by equity or debt, has no e¤ect on
its value. This is the Modigliani-Miller theorem.

- The Malthusian theory of fertility, that is, economies will not grow be-
cause fertility growth will eat up all income growth, is inconsistent
with individual rationality.
- Evolutionary theory in biology can be accurately represented by the meth-
ods of game theory. The same for Foraging theory.
- Altruistic (or cooperative) behavior is consistent with individual rational-
ity in a well-de…ned series of conditions. This is the Folk theorem.
- Value and equilibrium prices coincide (once "value" is properly de…ned).
This is the Theory of value.
- The study of the e¤ects of economic policy is logically and empirically
‡awed when not embedded in equilibrium analysis. This is called the
Lucas Critique.
- Many empirical relationships and many stable correlations have been un-
covered by means of statistical and econometric techniques. Example
include, the e¤ects of taxes on labor supply, the determinants of busi-
ness cycles, the determinants of asset prices, and many many many
others.
- ..........

1.6 Fun readings

P. Krugman, Development, Geography, and Economic Theory, 1995, MIT
Press; Ch. 3.
G.S. Becker, The Economic Way of Looking at Life 1992, Nobel Lecture,
posted on my web page.
1.7 USEFUL REFERENCES 15

R.E. Lucas, ‘Methods and Problems in Business Cycle Theory,’Journal of

Money, Credit, and Banking, 12, 1980, Part 2.

On Freakonomics, read Ariel Rubinstein’s critique:

A. Rubinstein, ‘Freak-Freakonomics,’The Economists’ Voice, 3(9), Article

7, 2006.

On the …nancial crisis of 2008 and the debate on economics it has gener-
ated you might want to read

P. Krugman on the NYTimes Magazine, September 6th, 2009;

who has a very di¤erent take on the position of economists with respect
to the crisis than the one o¤ered here, but he reaches similar conclusions
regarding the economics of bubbles. See also

P. Krugman, How did economists get it so wrong?, New York Times Maga-
zine, Sept. 2nd, 2009.

J.H. Cochrane, How did Paul Krugman get it so wrong?, University of

Chicago, mimeo.

R.G. Rajan, ‘Has …nancial development made the world riskier?,’presented

at the Kansas City Fed Meeting at Jackson Hole, 2005.

A. Rolnick, ‘Interview with Thomas Sargent,’The Region, September 2010.

1.7 Useful references

C. Simon and L. Blume, Mathematics for Economists, Norton, New York,
1994.

A. Schotter, Microeconomics: A Modern Approach, Addison Wesley Long-

man, New York, 2000.

H. Varian, Intermediate Microeconomics: A Modern Approach, Norton,

New York, 1999. R.P. McAfee and T. Lewis, Introduction to Economic
Analysis, Flatworld Knowledge, 2008.
16 CHAPTER 1 INTRODUCTION

M. Osborne, An Introduction to Game Theory, Oxford University Press,

Oxford, 2004.

K. Binmore, Fun and Games: A Text on Game Theory, D. C. Heath and

Company, Lexington, MA, 1992.

H. Gintis, Game Theory Evolving, Princeton University Press, Princeton,

NJ, 2009.

R.E. Wright and V. Quadrini, Money and Banking, Flatworld Knowledge,

2008.
Chapter 2

Rational choice

Jeremy Bentham Vilfredo Pareto Gerard Debreu Ken Arrow

In this chapter we shall study how (economists postulate) individuals

make decisions. An important by-product of the theory is a precise de…nition
of rationality.
The central …gure of decision theory in economics is the individual decision-
maker (DM). The typical example of a DM is the consumer. We shall assume
that:

the DM has well de…ned preferences over a choice set;

the DM is rational.

We will show that these assumptions, especially rationality, once it is

de…ned properly, imply that the DM’s preferences can be represented by a

17
18 CHAPTER 2 RATIONAL CHOICE

utility function and that the DM chooses her preferred element in the choice
set as if she were maximizing her utility function in the choice set.

2.1 Preferences
The choice set is represented by X, an arbitrary set of objects; e.g., a
list or a set of consumption bundles. An example should clarify, X =
fDucati motorcycle, trip to Vegas, iPad+$500 on iTunes, year tuition at NYU,:::g :
More abstractly, we could have X = fx; y; z; k; p; :::g :
Preferences are represented by a preference relation, denoted by % (note
that it is not ), which allows us to compare any pair of alternatives x; y in
the set of objects X, from the point of view of the DM.
De…nition 2.1.1 We read x % y as "x is at least as good as y" or "x is
weakly preferred to y;" for the DM.
In words, we also say "x is weakly better than y," for the DM (the DM
weakly prefers x to y). The term "weakly" is math jargon: x is weakly better
than y means that either the DM likes x better than y or else he is indi¤erent
between x and y. From % we can derive two other important relations on X:
De…nition 2.1.2 The strict preference relation, , satis…es

x y i¤ x % y but not y % x
where i¤ means if and only if.
Read x y as "x is strictly preferred to y," for the DM (the DM strictly
prefers x to y).
De…nition 2.1.3 The indi¤erence relation, , satis…es

x y i¤ x % y and y % x
Read x y as "x is indi¤erent to y," for the DM (this is not proper English
but economists often get away with it; "the DM is indi¤erent between x and
y" is better said):
Imagine that we present a consumer with any pairs of alternatives, x and
y, and ask how she compares them: Is either x or y weakly better than the
other in your eyes? For each pair x and y, we can imagine four possible
responses to our question:
2.2 RATIONALITY 19

Figure 2.1: Milton Friedman and George Stigler, both Nobel Prize winners
in economics, strolling down University Ave. on the University of Chicago
campus.

i) x is better than y, but y is not better than x,

ii) y is better than x, but x is not better than y,

iii) neither is better,

iv) I am unwilling to commit to a judgment.

Make sure you understand that the answer that x is better than y and y
is better than x is logically possible only if when the DM says "better" she
means "weakly better." We need to be precise with our DM when we ask her
questions!
Finally, note that a preference ordering is not formally di¤erent from
other more common binary orderings like e.g., the height ordering, which we
use when we say that Uma Thurman is (weakly - and strictly) taller than
Reese Witherspoon; or that George Stigler was (weakly - and strictly) taller
than Milton Friedman.

2.2 Rationality
The hypothesis of rationality is embodied in two assumptions about the weak
preference relation %: completeness and transitivity.

De…nition 2.2.1 (Completeness) For all x; y 2 X, either x % y or y % x

or both.
20 CHAPTER 2 RATIONAL CHOICE

Completeness simply requires that any two elements of the choice set
can be compared by the DM. As such it is hardly objectionable as a …rst
order assumption. But can you think of situations in which it fails ? Try
introspection.

De…nition 2.2.2 (Transitivity) For all x; y; z 2 X, if x % y and y % z

then x % z.

Transitivity implies that a DM never faces a sequence of pairwise choices

in which her preferences cycle, that is, such that x % y, y % z and z x
(we also write cycles more compactly as x % y % z x): Is transitivity of
preferences a necessary property that preferences would have to have? Try
introspection again. Is it a plausible assumption?
To try and convince you that transitivity is actually a plausible assump-
tion about preferences, notice the following:
Lack of transitivity implies the possibility of money pumps. Here’s an
example of a money pump.

Consider an economy with 3 goods, x; y; z , and 2 agents, 1; 2..

Agent 1 is endowed with one unit of z , one unit of y , and some money.
Agent 2 is endowed only with one unit of good x. Agent 1 has non
transitive preferences % such that: x % y % z and z x. We do
not specify any preferences for agent 2 [make sure that at the end of
the example you understand why we do not need to]. Consider the
following sequence of trades between agent 1 and 2: Agent 2 gives
agent 1 a unit of good x in exchange for a unit of good y . - Agent 2
gives agent 1 a unit of good y in exchange for a unit of good z . - Agent
2 gives agent 1 a unit of good z in exchange for a unit of good x plus a
(small but positive) sum of money. Note that the sequence of trades is
constructed so that it is feasible (possible) from the given endowments,
and so that agent 2 is willing to enter each of these trades. At the end
of the sequence of trades agent 2 has the unit of x he was endowed
with, plus a sum of money. Agent 1 has instead the units of y and
z he was endowed with, but he has lost a sum of money. This trade
cycle can be repeated inde…nitely until agent 1 lost all his money.

On the other hand, experimental evidence is not always consistent with

transitivity. An important phenomenon, called endowment e¤ect, has been
2.3 UTILITY REPRESENTATION 21

…rst documented in laboratory experiments by Kahneman, Knetsch, and

Thaler in the late 80’s and often replicated since. Here’s a version of the
experiment.
A student if shown two objects, a pen and a co¤ee mug. She is
asked to say which object she likes better and she answers "the pen."
She is then asked how much she is willing to spend to buy the pen,
and she answers "$1." In the course of the experiment the student is
given the mug as a present. At the end of the experiment she is asked
if she wants to give back the mug in exchange for $1, and she refuses.
This behavior turned out to be relatively typical among subjects in
the experiment (in the original experiment the subjects were Cornell
University college students). It constitutes a violation of transitivity:
$1 % pen; pen % mug and mug $1:
In summary, completeness and transitivity are assumptions about DM’s
preferences, not statement of logical necessity. Convince yourself that com-
pleteness and transitivity are a reasonably accurate assumptions about a
DMs’preferences; or that they are not. Economists tend to think that they
are, in most relevant decision environments: a sub-…eld of economics, Exper-
imental economics, is about testing these assumptions by means of behavior
of subjects in the lab. This …eld now interacts extensively with psychology
and even neuroscience (neuroscience labs typically contain functional-MRI
machines, essentially big magnets, which allow the experimenter to observe
brain activation of subjects while the solve choice problems).
We conclude re-iterating the main point of this section: rationality in
economics essentially requires assuming completeness and transitivity of in-
dividual DMs’preferences.
Nothing more and nothing less.

2.3 Utility representation

Rationality of DM (completeness and transitivity of preferences) allows us
to represent the preference relation, %, by a utility function.
De…nition 2.3.1 A function u : X ! R is a utility function representing
preference relation % if, for all x; y 2 X,
x % y , u(x) u(y):
22 CHAPTER 2 RATIONAL CHOICE

As you will painfully learn studying these notes, it is much easier to work
with the binary relationship on real numbers than with % on arbitrary
sets of objects. This is exactly what we are after by introducing the concept
of a utility function. The following theorem is central to the analysis.

Theorem 2.3.2 (Representation) A preference relation % can be repre-

sented by a utility function if and only if it is complete and transitive.

Proof. Assume the set X contains a …nite number N of elements,

(x1 ; : : : ; xN ). [the general proof is more complicated and is omitted]
[only if] By contradiction. Suppose % is not complete. Then there exist
x; y 2 X such that neither x % y nor y % x. But if % can be represented by
some function u : X ! R, then either u(x) u(y) or u(y) u(x) or both
(the order on real numbers is in fact complete). This is a contradiction.
Suppose % is not transitive. Then : : : continue by analogy with the argument
used above for completeness.
[if] By construction. Construct a utility function u : X ! R as follows.
Rank (from worst to best) in terms of preferences % all elements (x1 ; : : : ; xN ).
Suppose, without loss of generality (make sure you understand this is in
fact the case), that the rank is exactly x1 ; : : : ; xN , that is, xi % xi+1 , for
i = 1; 2; :::; N 1:
Here’s the construction.

1. Let u(x1 ) = 1;

2. Let u(x2 ) = u(x1 ) = 1 if and only if x2 x1 ; otherwise, if x2 x1 let

u(x2 ) = u(x1 ) + 1;

3. Proceed as in step 2 for all u(xi ); i = 3; 4; :::N 1.

Now, you should be able to prove that the utility function, as we con-
structed it, is in fact a representation of the preference ordering % :
Do it. To make sure you really understood the proof, answer the following
questions:

i) Where did we use completeness and transitivity in the "if" part of the
proof? I hid this step on purpose. Find it.
2.3 UTILITY REPRESENTATION 23

ii) Is it important that, in step 2, that, if x2 x1 ; we let u(x2 ) = u(x1 ) + 1

as opposed to u(x2 ) = u(x1 ) + 1:39672 or some other real number?

iii) Is it important that, if x2 x1 and x3 x2 ; the utility function we

construct satis…es the property that u(x2 ) u(x1 ) = u(x3 ) u(x2 )?

If you answer correctly these questions, you have understood that the
Representation theorem implies that the only property of a utility function
that is important is how it orders the bundles. The level of utility and the
size of the utility di¤erence between any two bundles doesn’t matter. In the
economics jargon, we say that
preferences are ordinal and not cardinal.
This is an issue that has bothered the classical (English, 18th and 19th
century) economists quite a bit. More precisely stated, ordinality of prefer-
ences is implicitly de…ned by the following result.

Proposition 2.3.3 If a utility function u(x) represents the preference rela-

tion %, any monotonic strictly increasing transformation of u(x), f (u(x)),
also represents the same preference relation.

Proof. Suppose u represents some particular preference relation, %.

Then, by de…nition,

u(x) u(y) if and only if x % y

But if f (u) is a monotonic strictly increasing transformation of u, then

u(x) u(y) if and only if f (u(x)) f (u(y))

Thus,
f (u(x)) f (u(y)) if and only if x % y
and the function f (u) represents the preference relation % in the same way
as the function u.
Of course ordinality of preferences also implies that it is impossible to
compare two di¤erent agents’utility functions (impossibility of interpersonal
comparison of utility, economists say). Statement like "I like the Rolling
Stones more than you do" are absolutely meaningless in the context of eco-
nomic decision theory. You understand that this is an important, if disturb-
ing, property of our theory.. It in turn implies, for instance, that I cannot
24 CHAPTER 2 RATIONAL CHOICE

say that society would do good by taking $x form a rich individual to give
them to a poor individual, no matter how small is x (and how rich is the
rich and how poor is the poor). (We shall discuss later on in the course how
economists do/don’t resolve this issue). Vilfredo Pareto (Italian economist
and sociologist, end of 19th century; the father of the modern theory of pref-
erences) explained this to his students by saying, "my tooth-ache is di¤erent
than yours."

2.4 Choice
A rational agent chooses then the element x of the choice set X to which is
associated the highest utility u(x). Formally, a rational agent choice problem
is written:

maxx2X u(x)
Typically, however, the choice set faced by a rational agents is subject to
some constraint. This is the case, for instance, for the classic example of a
choice problem, the consumer problem. Let m be a …xed amount of money
available to the consumer and let p = (p1 ; :::; pk ) be the vector of prices for
goods 1; :::; k. The set of a¤ordable objects, for the consumer, is given by

B = fx 2 X : p1 x1 + + pk xk mg
and is called the budget set (and p1 x1 + + pk xk m is called the budget
constraint). Note that we can also write p1 x1 + +pk xk m more compactly
in vector notation as px m. The consumer problem can then be written
as:

choose x to maximize u(x) subject to x 2 B;

or equivalently as:

choose x to maximize u(x) subject to x 2 X and px m:

More compactly, we write:

maxx2B u(x):
2.5 SOCIAL CHOICE 25

2.5 Social choice

We might debate if completeness and transitivity represent accurate assump-
tions about a DMs’preferences, but we can show that they are certainly not
accurate description of group preferences.
Consider an economy with 3 goods, a; b; c and 3 agents, 1; 2; 3. Suppose
that each agent has complete transitive preferences over the 3 goods, and that
no agents is ever indi¤erent between any of the possible binary comparisons.
We can then represent their preferences by rankings. Agent 1’s rank (from
best to worst) is a; b; c; agent 2’s is b; c; a; and agent 3’s is c; a; b. Suppose
that group preferences, represented by a preference ordering %G ; are formed
by majority voting from the preferences of the 3 agents. That is, the group
prefers x to y if and only if at least 2 of the agents do. Construct now the
preferences of the group, %G . Do it! It’s instructive: associate a statement of
the form x % y to any 2 elements x; y of the choice set X = fa; b; cg- where
x %G y if and only if for at least two agents x % y). If you have done this
correctly, you will see that

x %G y %G z and z G x:

That is, the group has non-transitive preferences.

The fact that group preferences can easily display non-transitivity is a
fundamental result. It is due to the marquis de Condorcet (an 18th century
French philosopher, mathematician, and political scientist). Instances of non-
transitivity are usually called Condorcet cycles. The existence of Condorcet
cycles means that we cannot treat groups as individuals, that we typically
cannot disregard the fact that groups are formed of di¤erent individuals when
we study group behavior: groups do not necessarily display rational (complete
and transitive) preference ordering.
In other words, social choice and individual choice are distinct problems,
which cannot in general be dealt with using the same conceptual tools. I
can’t resist noticing that this is a major blow to classical sociology, de…ned
as the study of groups per se, independently of the individuals constituting
the group.
But notice that I (as Condorcet) postulated that group preferences are
formed by majority rule. May-be here is the problem. May-be there exist
a di¤erent mechanism to aggregate individual preferences into group prefer-
ences which guarantees that the resulting preference ordering is always (that
26 CHAPTER 2 RATIONAL CHOICE

is, for any underlying individual preferences) complete and transitive. Well,
... not really.

Theorem 2.5.1 (Arrow impossibility) The only mechanism which ag-

gregates individual preferences into a complete and transitive preference or-
dering, for any underlying complete and transitive individual preferences, is
dictatorship, that is a mechanism which selects one individual of the group
and identi…es group preferences with the preferences of that individual.

The theorem takes its name from Ken Arrow.1 This result has opened
up a whole new sub-…eld of economics: social choice theory.

2.6 Classic references

The …rst version of the co¤ee mug experiment is in:

Daniel Kahneman, Jack Knetsch, and Richard Thaler, "Experimental tests

of the endowment e¤ect and the Coase theorem," Journal of Political
Economy, 98(6), 1990.

The …rst representation theorem is in:

Gerard Debreu, "Representation of a Preference Ordering by a Numerical

Function", in Thrall et al., editors, Decision Processes, 1954.

The impossibility theorem is contained in Ken Arrow’s dissertation:

Kenneth Arrow, Social choice and individual values, Cowles Foundation,

1
We shall see his name again in other parts of the course: he is certainly one of the
most in‡uential economists of this century.
2.7 USEFUL REFERENCES 27

Yale University Press, 1951.

2.7 Useful references

A beautiful very simple introduction to decision theory is in

David Kreps, A course in microeconomic theory, Chapter 2.1, Prentice Hall,

1990.

Two masterly treatments of the theory (somewhat di¢ cult) are:

David Kreps, Notes on the theory of choice, Westview Press, 1988.

Ariel Rubinstein, Lecture Notes in Microeconomic Theory: The Economic

Agent, Chapters 1-3, Princeton University Press, 2006.

A nice and simple introduction to behavioral anomalies can be found in:

Daniel Kahneman; Jack Knetsch; Richard Thaler, Anomalies: The Endow-

ment E¤ect, Loss Aversion, and Status Quo Bias, Journal of Economic
Perspectives, 5(1), 1991.

A relatively simple formal statement and proof of Arrow impossibility

theorem are in:

John Geanakoplos, Three brief proofs of Arrow’s impossibility theorem,

mimeo, Cowles Foundation, Yale University.
28 CHAPTER 2 RATIONAL CHOICE

2.8 Problems
After a brief review of decision theory, on preference relations, we list a
series of problems. Some of the problems have solutions, the others are left
as exercises for the reader.2

Problem 2.8.1 Assume the weak preference relation, , is complete and

transitive. Prove that this implies that the strict preference relation, , is
transitive. Is also complete?

Answer. Let x; y; z 2 X such that x y and y z. Firstly we have

x y and y z
) x z (by transitivity of )

Secondly we have
y x and z y
From here we would like to conclude that z x, but we not to shown the
transitivity of the negation of the weak preference relation. Suppose it was
the case that z x. We also have from above that x y. By transitivity of
it must be the case that z y. However y z implies z y, and z y
and z y cannot hold at the same time. Thus z x.Summing up we have

x z and z x
=) x z (by de…nition of )

and so we have the transitivity of . Try and answer the completeness

question.

Problem 2.8.2 Assume the weak preference relation, , is complete and

transitive. Then the indi¤erence preference relation, , de…ned as above is
transitive. Is also complete?

Answer. Let x; y; z 2 X such that x y and y z. Firstly we have

x y and y z (by de…nition of )

) x z (by transitivity of )
2
Recall that X denotes a choice set and a preference relation. Let and be the
associated strict preference and indi¤erence relations, respectively.
2.8 PROBLEMS 29

Secondly we have

z y and y x (by de…nition of )

) z x (by transitivity of )

Summing up we have

x z and z x
=) x z (by de…nition of )

and thus we have shown the transitivity of . Try and answer the complete-
ness question.
A few of the problems follow straightforwardly from the analysis of Chap-
ter 1. Some others require a signi…cant amount of ingenuity.

Problem 2.8.3 (i) Consider a group of individuals A; B and C and the

relation at least as tall as, as in A is at least as tall as B. Does this relation
satisfy the completeness and transitivity properties? (ii) Consider instead
a group of individuals A; B and C and the relation taller than, as in A is
taller than B. Does this relation satisfy the completeness and transitivity
properties? (iii) Consider the same group of individuals A; B and C and the
relation at least as smart as, as in A is at least as smart as B. Does this
relation satisfy the completeness and transitivity properties?

Answer. (i) The relation at least as tall as is complete and transitive.

To verify completeness, pick any two individuals A and B. Clearly, either
individual A is at least as tall as individual B or individual B is at least
as tall as individual A or both. For transitivity, pick three individuals A,
B and C and suppose that individual A is at least as tall as individual B
and individual B is at least as tall as individual C. Obviously, individual A
must be at least as tall as individual C. Thus, the relation at least as tall as
satis…es the transitivity property.
(ii) is left to the reader. (iii) is as (i) if smartness is de…ned, e.g., by the
I.Q. number. But we can certainly fudge the de…nition so as to generate an
incomplete or a non-transitive relationship. Try doing this.

Problem 2.8.4 Determine if completeness and transitivity are satis…ed for

the following preferences de…ned on x = (x1 ; x2 ) and y = (y1 ; y2 ): (i) x % y
if f (if and only if) x1 y1 and x2 y2 ; (ii) x % y if f x1 y1 and x2 y2 ;
30 CHAPTER 2 RATIONAL CHOICE

(iii) x % y if f x1 > y1 or x1 = y1 and x2 > y2 ; (iv) x % y if f x1 < y1 and

x2 y2 ; (v) x % y if f maxfx1 ; x2 g maxfy1 ; y2 g; (vi) x % y if f x1 y1
and x2 y2 :

Answer. (i) The preference relation x % y if f (if and only if) x1 y1

and x2 y2 is not complete. Consider the following counter example: x =
(0; 1) and y = (1; 0). Clearly, neither xi yi for all i nor yi xi for all i.
So, neither x % y nor y % x. Hence, the bundles x = (0; 1) and y = (1; 0)
can not be compared. The preference relation x % y if f (if and only if)
x1 y1 and x2 y2 is however transitive. Pick x = (x1 ; x2 ), y = (y1 ; y2 ) and
z = (z1 ; z2 ) and suppose that x % y and y % z towards showing that x % z.
By assumption, x % y then xi yi for all i and since y % z, yi zi for all i.
That is,
x1 y1 and x2 y2
and
y1 z1 and y2 z2
Hence,
x1 z1 and x2 z2
Therefore, x % y and y % z imply that x % z.
(ii) Preferences de…ned by x % y if f (if and only if) x1 y1 and x2 y2
are not complete. Consider the following counter example: x = (0; 1) and
y = (1; 0). Clearly, neither xi yi for all i nor yi xi for all i. So, neither
x % y nor y % x. Hence, the bundles x = (0; 1) and y = (1; 0) can not be
compared. On the other hand, preferences de…ned by x % y if f (if and only
if) x1 y1 and x2 y2 are transitive. Pick x = (x1 ; x2 ), y = (y1 ; y2 ) and
z = (z1 ; z2 ) and suppose that x % y and y % z. Need to show that this
implies x % z. By assumption, Since x % y then xi yi for all i and since
y % z then yi zi for all i. That is,

x1 y1 and x2 y2

and
y1 z1 and y2 z2
Hence,
x1 z1 and x2 z2
Therefore, x % y and y % z imply that x % z. Note that this is called Pareto
preference relation.
2.8 PROBLEMS 31

(iii) Preferences de…ned by x % y if f x1 > y1 or x1 = y1 and x2 > y2

are not complete. For a counter example pick two bundles x and y such that
x = y. For example, x = (1; 1) and y = (1; 1). Clearly, since x1 = y1 and
x2 = y2 neither x % y nor y % x. Hence, the two bundles x = (1; 1) and
y = (1; 1) can not be compared by this preference relation. On the other
hand, preferences de…ned by x % y if f x1 > y1 or x1 = y1 and x2 > y2 are
transitive. Pick x = (x1 ; x2 ), y = (y1 ; y2 ) and z = (z1 ; z2 ) and suppose that
x % y and y % z. Need to show this implies x % z. By assumption, Since
x % y then
either x1 > y1 or if x1 = y1 then x2 > y2
and since y % z then

either y1 > z1 or if y1 = z1 then y2 > z2

Hence, it must hold that

either x1 > z1 or if x1 = z1 then x2 > z2

which implies that x % z. Note this is called Lexicographic preference rela-

tion.
(iv) The preference relation x % y if f (if and only if) x1 < y1 and
x2 y2 is not complete. Consider the following counter example: x = (0; 1)
and y = (1; 0). Although x1 < y1 , x2 > y2 . Therefore x = (0; 1) and
y = (1; 0) can’t be compared. However, it is transitive. Pick x = (x1 ; x2 ),
y = (y1 ; y2 ) and z = (z1 ; z2 ) and suppose that x % y and y % z towards
showing that x % z. By assumption, x % y then x1 < y1 and x2 y2 and
since y % z, y1 < z1 and y2 z2 . That is,

x1 < y1 and x2 y2

and
y1 < z1 and y2 z2
Hence,
x1 < z1 and x2 z2
Therefore, x % y and y % z imply that x % z.
(v) The preference relation x % y if f maxfx1 ; x2 g maxfy1 ; y2 g is
complete. Pick any x = (x1 ; x2 ) and y = (y1 ; y2 ). Clearly, either

maxfx1 ; x2 g maxfy1 ; y2 g
32 CHAPTER 2 RATIONAL CHOICE

holds or,
maxfx1 ; x2 g maxfy1 ; y2 g

holds or both. Hence, either x % y or y % x or both. It is also transitive.

Pick any x = (x1 ; x2 ), y = (y1 ; y2 ) and z = (z1 ; z2 ) and suppose that x % y
and y % z. To show that transitivity we need that x % z. Since x % y

maxfx1 ; x2 g maxfy1 ; y2 g

and since y % z
maxfy1 ; y2 g maxfz1 ; z2 g

So, we conclude that

maxfx1 ; x2 g maxfz1 ; z2 g

which implies that x % z. Therefore, x % y and y % z imply that x % z.

(vi) is left to the reader.

Problem 2.8.5 For the following choice environments and preference re-
lations, is the relation complete and transitive: (i) X R2 , x y i¤
2
max(x1 ; x2 ) max(y1 ; y2 ); (ii) X R; x y i¤ x1 y1 ; (iii) X R,
2
x y if x is a rational number and y is not; (iv) X R , x y i¤ x1 > y2 ;
(v) X R, x y i¤ x y 2; (vi) X R3 , x y i¤ xi > yi for 2
2
of the
p 3 i; (vii) X R , x py i¤ x1 + x2 = y1 y2 ; (viii) X R2 , x y
2 2 2 2 2
i¤ (x1 3) + (x2 7) (y1 3) + (y2 7) ; (ix) X R , x y i¤
x1 + x2 y1 + y2 and x1 x2 y1 y2 ; (x) X R2 , x y i¤ x1 + x2 y1 + y2
or x1 x2 y1 y2 ; (xi) X R2++ , x y i¤ x1 + x2 y1 + y2 and x1 x2 y1 y2 :

Problem 2.8.6 For the following preferences over bundles containing both
apples and oranges, determine whether or not there is a utility representation,
and if there is, write down two utility functions that represent the preferences:
(i) the DM prefers the bundle with the most apples; (ii) the DM prefers the
bundle with the most items of fruit; (iii) the DM prefers the bundle that is
closest to having the same number of apples and oranges; (iv) the DM ‡ips
a coin: if it comes up heads, he prefers the bundle with the most apples; if it
comes up tails he prefers the bundle with the most oranges.
2.8 PROBLEMS 33

Problem 2.8.7 Consider the following examples of choice sets X and asso-
ciated preference relations : (i) X = f1; 2; 3g and is de…ned by

1 1; 1 2; 1 3
2 3
3 1;

(ii) X = fall the people in the worldg and is de…ned by

f or any x; y 2 X
x y if x "shares at least one given name with" y;

(iii) X = R and is de…ned by

f or any x; y 2 X
x y if x y;

(iv) X = R and is de…ned by

f or any x; y 2 X
x y if jx yj > 1;

(v) X = R and is de…ned by

f or any x; y 2 X
x y if jx yj is an integer multiple of 2:

Show the following: (i) is NOT complete and NOT transitive; (ii) is NOT
complete and NOT transitive; (iii) is complete and transitive; (iv) is NOT
complete and NOT transitive; (v) is NOT complete but is transitive. Fur-
thermore, [Note that this part of the problem goes beyond the di¢ culty level
of this course and is only for those who are interested] consider the following
potential properties of the above preference relations: Weakly complete:
on X is weakly complete if for all x; y 2 X, either x = y, x y or y x;
Negatively transitive: on X is negatively transitive if for all x; y; z 2 X
such that x y and y z, it is the case that x z; Re‡exive: on X is
re‡exive if for all x 2 X, x x. Irre‡exive: on X is irre‡exive if for all
x 2 X, x x. Symmetric: on X is symmetric if for all x; y 2 X such
that x y, it is the case that y x. Asymmetric: on X is asymmetric if
34 CHAPTER 2 RATIONAL CHOICE

for all x; y 2 X such that x y, it is the case that y x. Antisymmetric:

on X is antisymmetric if for all x; y 2 X such that x y and y x, it is
the case that x = y; Acyclic: on X is acyclic if for all x1 ; :::; xn 2 X, for
n 2 N , such that x1 x2 ::: xn 1 xn , then it is the case that x1 6= xn .
For each of the examples (i)-(v), show whether or not the preference relation
has any of the properties listed above. That is, if the preference relation has
the property, provide a proof; if not, provide a counter-example.

Problem 2.8.8 The DM cannot detect small di¤erences. He consumes two

goods: x and y and he strictly prefers bundle (x0 ; y 0 ) to bundle (x; y), (x0 ; y 0 )
(x; y), if and only if
x0 y 0 xy > 1
(if the di¤erence is less than one in absolute value he is indi¤erent). (i) Show
that the strict part of his preferences is transitive, i.e, that if (x1 ; y1 ) (x2 ; y2 )
and (x2 ; y2 ) (x3 ; y3 ) then (x1 ; y1 ) (x3 ; y3 ) : (ii) Show that the indi¤erence
part of the preferences is not transitive, i.e., that if (x1 ; y1 ) (x2 ; y2 ) and
(x2 ; y2 ) (x3 ; y3 ) then it is not necessary that (x1 ; y1 ) (x3 ; y3 ).

Answer. (i) Consider three bundles (x1 ; y1 ); (x2 ; y2 ) and (x3 ; y3 ):If
(x1 ; y1 ) (x2 ; y2 ); then it follows that x1 y1 x2 y2 > 1: If (x2 ; y2 ) (x3 ; y3 );
then x2 y2 x3 y3 > 1:Summing up these two inequalities we obtain that
x1 y1 x3 y3 > 2; which implies that x1 y1 x3 y3 > 1: Notice now that this
last inequality holds only if (x1 ; y1 ) (x3 ; y3 ): (ii) Consider the following
counterexample: (x1 ; y1 ); (x2 y2 ) and (x3 ; y3 ) : x1 y1 = 1; x2 y2 = 1:8 and
x3 y3 = 2:2: Observe that (x1 ; y1 ) (x2 ; y2 ) since x2 y2 x1 y1 = 0:8 < 1:
At the same time, (x2 ; y2 ) (x3 ; y3 ) since x3 y3 x2 y2 = 0:4 < 1: However,
(x3 ; y3 ) (x1 ; y1 ) since x3 y3 x1 y1 = 1:2 > 1.

Problem 2.8.9 Consider the preference ordering over the composition of

Congress induced by the following procedure: I ‡ip a coin; if heads I strictly
prefer a Congress with a Democratic majority; if tails I strictly prefer a
Congress with a Republican majority. (i) Is this preference relationship com-
plete? (ii) Is it transitive ?
Chapter 3

Consumer choice

Alfred Marshall John Hicks

Consider an agent choosing her consumption of goods 1 and 2 for a given

budget and recall his choice problem in the form introduced in the previous
chapter

choose x to maximize u(x) subject to x 2 X and px m:

This is the workhorse of microeconomic theory. We shall study it in detail.

35
36 CHAPTER 3 CONSUMER CHOICE

3.1 The Budget constraint

There are only two goods, X = R2+ .1 A consumer who consumes x1 units of
good 1 and x2 units of good 2 is said to consume the consumption bundle
(x1 ; x2 ) 2 X. Any bundle can be represented by a point on a two-dimensional
graph.
The prices of these goods, p1 and p2 , are given and known to the consumer
who is a price taker. The consumer’s income, that is, the amount of money
the consumer has to spend, is m. Then the consumer’s budget constraint can
be written as p1 x1 + p2 x2 m.
The budget set consists of all bundles that are a¤ordable at a given prices
and income. More precisely,

B = fx 2 X : p1 x1 + p2 x2 mg
Thus, the budget line is the set of bundles that cost exactly m:
p1 x1 + p2 x2 = m;
and can be writtten also as:
p1 m
x2 = x1 + :
p2 p2
Note that this is a linear function with a vertical intercept m=p2 , a horizontal
intercept m=p1 and a slope of p1 =p2 ; as graphed, in Figure 3.1.
The slope of the budget line measures the rate at which the market is
willing to exchange good 1 for good 2. Thus, economists say that the slope
measures the opportunity cost of consuming good 1: the consumer has an
alternative opportunity to consuming 1 unit of good 1, consisting in selling
the unit of good 1, for an amount equal to p1 , and then using the proceeds
to buy good 2 - that is pp21 units of good 2.
Clearly, when prices and income changes, the set of goods that a consumer
can a¤ord changes as well. Increasing income shifts the budget line outward,
while increasing the price of good 1 (resp. good 2) makes the budget line
steeper (resp. ‡atter). Make sure you understand this. What do you think
happens to the budget line when both prices are changed at the same time?
For example, both prices become t times as large. Give a graphical and an
analytical answer.
1
The reader should note that extending the analysis to n goods, X = Rn+ , is straight-
forward, though the graphical representation of the problem is obviously lost.
3.1 THE BUDGET CONSTRAINT 37

Budget constraint: p x +p x =m , w ith p =1, p =2, m =8

8
1 1 2 2 1 2

5
(slope -p /p )
1 2
2

4
x

(x ,x )
1 2
3

0
0 0.5 1 1.5 2 2.5 3 3.5 4
x
1

Figure 3.1: Budget constraint

38 CHAPTER 3 CONSUMER CHOICE

3.2 Indi¤erence curves

Recall that the theory of consumer choice can be formulated by preferences
that satisfy completeness and transitivity. A useful graphical description of
preferences exploits a construction known as indi¤erence curves.
Indi¤erence curves are level curves, that is, they represent the set of
(x1 ; x2 ) for which the utility is constant at some level, say U: This is not
unlike topographic maps, as in the …gures below.

A peak Earthquake Lake, Montana - Idaho

Formally, the indi¤erence curve at level U is represented by the equation

u (x1 ; x2 ) = U:
Can indi¤erence curves representing distinct levels of utility cross ?
Note that the equation u (x1 ; x2 ) = U is non-linear (since u (x1 ; x2 ) is non-
linear, in general) in (x1 ; x2 ) : As a consequence, a solution in the form x2 =
f (x1 ; u) is often impossible to write down (even when it exists). Nonetheless
we can characterize the slope of the indi¤erence curve, known as the marginal
rate of substitution, by relying on the Implicit Function theorem of calculus.2
The marginal rate of substitution, M RS(x1 ; x2 );is precisely de…ned as the
2
The Implicit Function Theorem is one of the fundamental theorems of calculus, and
one of the most used in economics. A (imprecise) statement of the theorem is as follows.
3.2 INDIFFERENCE CURVES 39

Family of indifference curves: u(x ,x )=ln(x )+ln(x )=U, for various levels of U
1 2 1 2
8

6 (x ,x )
1 2

5
2
x

s lope=-MRS (x ,x )= Uti l i ty l eve l U '>U

1 2
4 - ∂ u/∂ x /∂ u/∂ x
1 2

3
Uti l i ty l eve l U

1
1 2 3 4 5 6 7 8
x
1

Figure 3.2: Indi¤erence curves and the marginal rate of substitution

40 CHAPTER 3 CONSUMER CHOICE

in…nitesimal change in good 2 that keeps the utility unchanged after a unitary
in…nitesimal increase in good 1.
Formally, consider a change (dx2 ; dx1 ) that keeps the level of utility con-
stant:

@u (x1 ; x2 ) @u (x1 ; x2 )
du = dx1 + dx2 = 0:
@x1 @x2
Rearranging, we obtain
@u(x1 ;x2 )
dx2 @x1
= @u(x 1 ;x2 )
:
dx1
@x2

But dx
dx1
2
is exactly the rate at which the consumer is willing to substitute an
in…nitesimal amount of good 2 for good 1; that is, the change of x2 for x1
which keeps the agent’s utility constant. That is,
@u(x1 ;x2 )
@x1
M RS(x1 ; x2 ) = @u(x1 ;x2 )
@x2

Consider the Cobb-Douglas utility function u(x1 ; x2 ) = x1 x21 ; for

0 < < 1; as an example. In this case the marginal rate of substitu-
tion can be easily solved for:
@u(x1 ;x2 )
@x1
M RS(x1 ; x2 ) = @u(x1 ;x2 )
=
@x2
x1 1 x21 x2
= == :
x 1 x2 1 x1
In fact, in this case we can even solve for the indi¤erence curve:

1
1
(x1 )
x1 x12 = U () x2 = U 1 :
Let u : <2 ! < be continuous and smooth (at least twice di¤ erentiable). Then there
exists a continuous di¤ erentiable function f : < ! < such that, locally,

u (x1 ; f (x1 )) = U

and
@:u(x1 ;x2 )
df (x1 ) @:x1
= @u(x1 ;x2 )
dx1
@:x2
3.2 INDIFFERENCE CURVES 41

At this stage we shall procede by describing some general assumptions

- monotonicity and convexity - which we will typically make about prefer-
ences. We shall also illustrate the implications of these assumptions for the
associated indi¤erence curves.

3.2.1 Monotonicity
We typically assume that more is better, that is, we consider goods, not bads.
More precisely, consider (x1 ; x2 ) as a bundle of goods and let (x01 ; x02 ) be any
other bundle with at least as much of both goods and more of one. That
is, x01 x1 and x02 x2 with at least one strict inequality, or, in short,
0 0
(x1 ; x2 ) > (x1 ; x2 ).
(Strict) monotonicity of preferences requires that

if (x01 ; x02 ) > (x1 ; x2 ) then (x01 ; x02 ) (x1 ; x2 ) :

Thus, (strict) monotonic preferences imply that more of (resp. less) of both
goods is a better (resp. worse) bundle. Monotonicity of preferences implies
that the utility function which represents preferences is monotonic increasing
in both its arguments:

@u (x1 ; x2 ) @u (x1 ; x2 )
; > 0; for any (x1 ; x2 ) 2 X:
@x1 @x2

It also implies that indi¤erence curves have a negative slope:

M RS (x1 ; x2 ) < 0:

Obviously, since the bundle (x1 ; x2 ) was chosen arbitrarily, we can draw an
indi¤erence curve through any bundle.
Starting at an arbitrary bundle (x1 ; x2 ) and moving up and to the right,
by monotonicity we must be at a preferred position. Thus, at a higher
indi¤erence curve the consumer is strictly better.

3.2.2 Convexity
A set X is convex if x + (1 ) y 2 X whenever x; y 2 X and 2 [0; 1] :
In words, A set X is convex if whenever it contains two elements x; y, it
42 CHAPTER 3 CONSUMER CHOICE

Fam ily of indifference curves: u(x ,x )=ln(x )+ln(x )=U, for various le vels of U
1 2 1 2
8

7 (x ',x ')
1 2

6 (x ,x )
1 2

5
2
x

Utility level U'>U

4

3
Utility level U

1
1 2 3 4 5 6 7 8
x
1

Figure 3.3: Indi¤erence curves and monotonicity

3.3 TYPES OF PREFERENCES 43

contains the entire segment connecting them.

A convex set A non-convex set

Convexity of preferences requires that the set of weakly preferred alloca-

tions to any given allocation is convex. Formally,

for any (x1 ; x2 ) 2 X; the set f(y1 ; y2 ) 2 X : (y1 ; y2 ) (x1 ; x2 )g is convex.

Taking two arbitrary bundles on the 0 same

1 indi¤erence0 curve,
1 (x1 ; x2 ) and
x1 x01
0 0
(x1 ; x2 ), by convexity the bundle @ A + (1 ) @ A is preferred to
0
x2 x2
each of the initial bundles. Convexity of preferences implies that the utility
function representing preferences is concave and that indi¤erence curves are
convex. Important examples where convexity does not hold include indivisi-
ble goods (e.g., shoes) or addictive goods. Can you see why ?
Often, monotonicity and convexity of preferences are taken as the de…ning
features for well-behaved indi¤erence curves. We shall always assume so in
this book.

3.3 Types of preferences

Some examples to get more used with the indi¤erence curve representation
of preferences.
44 CHAPTER 3 CONSUMER CHOICE

Family of indifference curves: u(x1,x 2)=ln(x 1)+ln(x 2)=U, for various levels of U
8

(x 1,x2 ) Utility level U'>U

7

5 α (x1,x2)+(1- α)(x 1 ',x 2 ')

2
x

4
(x 1',x 2')

Utility level U
2

1
1 2 3 4 5 6 7 8
x1

Figure 3.4: Indi¤erence curves and convexity

3.3 TYPES OF PREFERENCES 45

F a m ily o f in d if f e r e n c e c u r v e s f o r v a r io u s le v e ls o f U

5
2
x

4
U t ilit y le v e l U '> U

3 U t ilit y le v e l U

1
1 2 3 4 5 6 7 8
x
1

Figure 3.5: Indi¤erence curves for perfect substitutes

3.3.1 Perfect substitutes

Two goods are perfect substitutes if the consumer is willing to substitute one
good for the other at a constant rate. Thus, the indi¤erence curves have a
constant slope.
The utility function in the case x1 and x2 are perfect substitutes is of the
(linear) form
u(x1 ; x2 ) = x1 + x2 ;
and indi¤erence curves are also linear.

3.3.2 Perfect complements

Two goods are perfect complements if they are always consumed together
in …xed proportions. For example, right shoes and left shoes. The utility
function in the case x1 and x2 are perfect complements is called Leontief,
46 CHAPTER 3 CONSUMER CHOICE

F a m i l y o f i n d i ffe r e n c e c u r v e s fo r d i ffe r e n t l e v e l s o f U
8

5
2
x

U t ilit y le v e l U '> U
4

3
U t ilit y le v e l U

1
1 2 3 4 5 6 7 8
x
1

Figure 3.6: Indi¤erence curves for perfect complements

after Wassily Leontief. It is of the form

u(x1 ; x2 ) = min f x1 ; x2 g

and the associated indi¤erence curves are L-shaped.

3.4 Consumer problem

We are ready to solve now the consumer problem:

max u(x1 ; x2 ) s.t. p1 x1 + p2 = m;

x2X

under the assumption that the preferences represented by u(x1 ; x2 ) are well-
behaved: monotonic and convex. In fact we consider three di¤erent ways to
study the problem of consumer. We restrict the analysis to the example of
Cobb-Douglas preferences to simplify the algebra.
3.4 CONSUMER PROBLEM 47

The analysis in this section requires an understanding of the mathemat-

ical theory of maximization. The reader is strongly suggested to study the
Mathematical Appendix on this topic (or any calculus book). We shall take a
shortcut in the analysis, however, in the interest of simplicity: we shall only
derive the …rst order conditions (foc’s) for the maximum in the consumer
problem (or an equivalent appropriately manipulated version of it). This is
a shortcut because in principle, the foc’s do not guarantee a maximum -
formally, they are necessary but not su¢ cient conditions. Essentially three
problems may arise: (i) the maximum is at a corner (not an interior solu-
tion), (ii) the foc’s characterize a local - not a global - maximum, (iii) the
foc’s characterize a minimum - not a maximum. Possible corner solutions
need to be checked individually. The possibility of the foc’s not representing
a maximum is avoided when the maximization problem is convex, that is, a
set of second order conditions are satis…ed. We refer to the Math Appendix.
The reader should however check that in fact, in the consumer problem as
we formulated, the solution we obtain by the foc’s is a (global) maximum.

1. Geometric analysis of indi¤erence curves.

A consumption bundle (x1 ; x2 ) is an optimal choice for the consumer if

the set of bundles that the consumer prefers to (x1 ; x2 ) - the set of bundles
above the indi¤erence curve through (x1 ; x2 ) - has an empty intersection with
the bundles she can a¤ord - the bundles beneath her budget line.
It follows that at (x1 ; x2 ) the indi¤erence curve is tangent to the budget
line:

x2 p1 3
M RS(x1 ; x2 ) = = ;
x1 p2
and, naturally, at (x1 ; x2 ) the budget constraint must be satis…ed:

p1 x1 + p2 x2 = m
We have now two equations in two unknowns that can be solved for the
optimal bundle. Thus, substituting:
3
Note that, clearly, the above holds for an interior optimum and not for a boundary
optimum. Can you see why a Cobb-Douglas preferences induce an interior solution? How
would you deal with preferences implying perfect substitutes ?
48 CHAPTER 3 CONSUMER CHOICE

p 1 x1
p1 x1 + p2 =m
p2
and hence
m
x1 =
+ p1
This is the demand function for good 1. Similarly, the demand function for
good 2:
m
x2 =
+ p2

2. Algebraic analysis by substituting the budget constraint in to the objec-

tive function.

After the substitution, the consumer problem is reduced to:

m p1
max ln x1 + ln x1
x1 0 p2 p2

The foc for this problem is

p2 p1
= 0:
x1 m p1 x1 p2

By a (very) little algebra,

m
x1 =
+ p1
m
x2 = :
+ p2

3. General algebraic analysis of constrained maximization.

De…ne the Lagrangian:

L= ln x1 + ln x2 (p1 x1 + p2 x2 m)
Di¤erentiate it to obtain the foc’s for a maximum of the Lagrangian with
respect to (x1 ; x2 ) and a minimum with respect to :
3.5 DEMAND FUNCTIONS 49

@L
jx =x ; x =x ; = = p1 = 0
@x1 1 1 2 2 x1
@L
jx =x ; x =x ; = = p2 = 0
@x2 1 1 2 2 x2
@L
jx1 =x1 ; x2 =x2 ; = = m p1 x1 p2 x2 = 0
@
These are three equations with three unknowns. The best way to proceed
(which you should do!) is to …rst solve for . You will see that you get
back to the algebraic condition obtained in the …rst method, the geometric
analysis of indi¤erence curves.

3.5 Demand Functions

Demand functions give the optimal bundle (amounts of each of the goods)
as a function of the prices and income faced by the consumer

xl (p; m); l = 1; 2:
Note that when two goods are perfect substitutes, the consumer is willing to
substitute the goods on a one to one basis, and p2 > p1 (p2 < p1 ) then the
slope of the budget line is ‡atter (steeper) than of the indi¤erence curve.
Thus, the demand function for good 1 will be
8
>
> m=p1 if p2 > p1
>
<
x1 = [0; m=p1 ] if p2 = p1
>
>
>
: 0 if p < p
2 1

How is the demand function for good 1 if the goods are perfect complements?
Does the demand of a good l depend necessarily negatively on its own price?
A few de…nitions which exploit the demand function construction are
useful to remember.
A Normal good is a good l for which the quantity demanded increases
with income,

@xl (p; m)
> 0:
@m
50 CHAPTER 3 CONSUMER CHOICE

An Inferior good is a good for which the quantity demanded decreases

with income,

@xl (p; m)
< 0:
@m
An Engle curve is a graph of the demand for a good as a function income,
with all prices held constant, that is, a graph with quantity of good demanded
on the horizontal axis and income on the vertical axis.
An Income expansion path is a graph of the bundles of goods 1 and 2 that
are demanded at di¤erent levels of income, keeping prices held constant.

3.6 Problems
Problem 3.6.1 Let p1 ; p2 ; m be the set of prices and income. The utility
function is given by:
u (x1 ; x2 ) = x1 x21 :
(i) Determine the demanded bundle as a function of the prices, p1 and p2 ,
the income, m, and the parameter . (ii) What fraction of his income will
a consumer with this utility function spend on good 1? Does this fraction
depend on income m? Does it depend on the price ratio, pp21 ? (iii) Is good 2
a normal good? Explain.

Answer. (i) To derive the consumer’s demand for good 1, x1 , and for
good 2, x2 , given prices p1 and p2 and income m, we need to solve the
consumer maximization problem:
max x1 x12
(x1 ;x2 )2X
s.t. p1 x1 + p2 x2 m
Note that X = R2+ and hence the restrictions on (x1 ; x2 ) are equivalent to
the following three inequality constraints
x1 0
x2 0
p1 x1 + p2 x2 m
But we can simplify the problem. By the nature of the utility function, it
is never optimal to have either x1 = 0 or x2 = 0 (implicitly assuming …nite
3.6 PROBLEMS 51

prices and m > 0). Thus the …rst two inequality constraints will not bind
and can be dropped from the maximization problem.
Since the utility function is strictly increasing in both arguements and
"savings" do not derive any utility, the last inequality will bind for sure, i.e.
p1 x1 + p2 x2 = m (to prove this, suppose that you had an optimal choice that
did not use all of the consumer’s income, then …nd another bundle that gives
higher utility and is still in the budget set - this is proof by contradiction).
The maximisation problem is now:

max x1 x21
(x1 ;x2 )2R2+

s.t. p1 x1 + p2 x2 = m

Set up the Lagrangian

L = x1 x12 + (p1 x1 + p2 x2 m)

where is the Lagrange Multiplier.

The foc’s for an (interior) solution to this problem are:

@L
= 0
@x1
@L
= 0
@x2
@L
= 0
@
where the optimum choices x1 and x2 solve these three equations simul-
taneously (also note that there are three unkowns since the value of is
unknown).
This gives the following equations:

(x1 ) 1 (x2 )1 + p1 = 0
(1 )(x1 ) (x2 ) + p2 = 0
p1 x1 + p2 x2 = m

The suggested way of soving this system of simultaneous equations is to

re-arrange the …rst two equations so that is separated on the right hand
side. Then equating these two gives (dropping the " " notation for the time
52 CHAPTER 3 CONSUMER CHOICE

being but still understanding that these are the optimal values rather than
generic values of x1 and x2 ):

x1 1 (1
) x1
( ) = = ( )
p1 x2 p2 x2
x1 p2
=) =
x2 (1 ) p1
p2
x1 = x2
(1 ) p1
and
p1 x1 + p2 x2 = m

Putting these two equations (in two unkowns) together gives the solution:
m
x1 =
p1
m
x2 = (1 ) :
p2

(ii) The solution implies that

p1 x1 = m
p2 x2 = (1 )m;

so that the fraction of income spent on good 1 is and on good 2 is 1 :

(iii) The solution also implies that

@x2 (1 )
= > 0;
@m p2

and hence good 2 is a normal good.

Problem 3.6.2 Solve consumer problems of consumers with the following

utility functions. (i) u(x1 ; x2 ; x3 ) = x1 x2 x13 ; (ii) u(x1 ; x2 ) = 3 log x1 +
1
2 log x2 ; (iii) u(x1 ; x2 ) = min(x1 ; x2 ); (iv) u(x1 ; x2 ) = (x1 +x2 ) ; (v) u(x1 ; x2 ) =
3
x1 + x24 ; (vi) u(x1 ; x2 ) = x1 + x2 . In each case you will have to construct
the budget constraint, assuming that any good xl is available at price pl , and
there is m amount of money available
3.6 PROBLEMS 53

Answer. We just solve a few of these.

1
(iv) u(x1 ; x2 ) = (x1 + x2 ) . If the solution to the consumer problem is
interior4 , then the following equilibrium condition has to hold
@u
@x1 p1
@u
= (3.1)
@x2
p2
In this case this condition is given by
1
1 1 1 1
(x1 + x2 ) x1 x1 p1
1 = =
1
(x1 + x2 ) 1
x2 1 x2 p2
implying that
1
p1 1
x1 = x2
p2
Plugging this expression into the budget constraint, we obtain
p1 x1 + p2 x2 = m
1
p1 1
p1 x2 + p2 x2 = m
p2
m m
) x2 = 1 ; x1 = 1
p1 1 p1 1
p1 p2
+ p2 p1 + p2 p2
@u
@x1 1 p1
(vi) u(x1 ; x2 ) = x1 + x2 . Notice that if @u = 1
> p2
, the utility per
@x2
dollar spent in good 1 is higher that the utility per dollar spent in good 2,
irrespective of the values of x1 and x2 : Hence, in that case, the individual will
only consume good 1. Similarly, in case 1 < pp12 ; the individual only demands
good 2. When 1 = pp12 ; any allocation on the budget constraint is optimal.
Having this in mind, we can write the demand function as
8
>
> x1 = pm1 x2 = 0 if 1 > pp12
>
<
(x1 ; x2 ) = x1 = 0 x2 = pm2 if 1 < pp12
>
>
>
: f(x ; x ) 2 R2 : p x + p x = m if 1 = p1
1 2 + 1 1 2 2 p2
4
Notice that:
(i) when = 1 the indi¤erence curves are linear
(ii) as ! 0; this utility function comes to represent the same preferences as u(x) = x1 x2
(iii) as ! 1; indi¤erence curves become "right angles" as with u(x1 ; x2 ) =
minfx1 ; x2 g
54 CHAPTER 3 CONSUMER CHOICE

Problem 3.6.3 Let the utility function be:

p
u (x1 ; x2 ) = x1 + x2
(If you think there is something strange here because the utility is linear in x2 ,
you are right! But trust me, do not worry and proceed as usual). (i) Write
down the budget constraint: call income m and normalize the price p1 , that
is, p1 = 1. Determine the demanded bundle as a function of price p2 and
income, m. (Substitution works; Lagrange too; whatever is easier for you).
(ii) Draw the demand for good x2 and the demand for x1 as a function of
price p2 . (Now is the moment to think about the fact that the utility function
is linear in x2 . Does your solution generate always non-negative quantities
x1 and x2 ?).
Answer. (i) The budget constraint is

x1 + p2 x2 m

To be formal, the consumer chooses (x1 ; x2 ) 2 R2+ to

p
maxf x1 + x2 g s.t.
x1 0
x2 0
x1 + p2 x2 m

where we have written non-negativity constraints explicitly.

p
Since u(x1 ; x2 ) = x1 +x2 is an increasing function in x1 , x2 the constraint
x1 + p2 x2 m will always bind. Thus always have

x1 + p2 x2 = m

However, note that for this utility function we cannot say anything about
the non-negativity constraints.
The procedure is to analyze the problem in two stages. First assume the
solution is interior, i.e., solve the problem of choosing (x1 ; x2 ) 2 R2+ to
p
maxf x1 + x2 g s:t:
x1 + p2 x2 = m
3.6 PROBLEMS 55

ignoring the non-negativity constraints. Then look for corner solutions and
for conditions on the parameters of the problem (p2 ; m) that give rise to such
problems.
There are two methods for solving this problem: the substitution method
and the Lagrangian method.
Substitution Method. Assume x1 > 0, x2 > 0: Since we have the budget
constraint holding with equality and only two choice variables, rearrange the
budget constraint to solve for x2 :
m 1
x2 = x1
p2 p2
Putting this into the objective function gives, choose x1 2 R+ to
p m 1
max x1 + x1
p2 p2
The foc is:
1 1 1
p = 0
2 x1 p2
1 2
p =
x1 p2
p p2
x1 =
2
p2 2
x1 =
2
Putting this back into the expression for x2 :
m 1 p2 2
x2 =
p2 p2 2
m p2
x2 =
p2 4
Note that x2 0 i¤ pm2 p42 0, and hence i¤ 4m (p2 )2 :
Now let’s check for corner solutions. The solution obtained assuming
x1 > 0, x2 > 0 suggests that x2 = 0 could be a potential problem, since when
4m (p2 )2 we have x2 0: Postulating x2 = 0 and solving the problem
using the budget constraint, we obtain

x1 = m:
56 CHAPTER 3 CONSUMER CHOICE

This gives a corner solution of:

x1 = m
x2 = 0:
Lagrangian Method. Assume x1 > 0, x2 > 0. Set up the Lagrangian
p
L = x1 + x2 + (x1 + p2 x2 m) ;
and its foc’s:
@L 1 1
= p + =0
@x1 2 x1
1
=) p = 2
x1
p 1
=) x1 =
2( )
@L
= 1 + p2 = 0
@x2
1
=) =
p2
@L
= x1 + p2 x2 m = 0
@
=) x1 + p2 x2 = m
p
Plugging in the solution for into the equation for x1 gives
p p2
x1 =
2
p2 2
x1 =
2
Plugging this into the budget constraint gives
m p2
x2 = ;
p2 4
and therefore x2 0 i¤ 4m (p2 )2 :
Let’s check for corner solutions. Note that, re-arranging the foc’s of the
Lagrangian, we obtain
@u(x1 ;x2 )
@x1 p1
@u(x1 ;x2 )
= :
p2
@x2
3.6 PROBLEMS 57

The Lagrange method requires equating the slope of the indi¤erence curves
with the slope of the budget constraint. Corner solutions come about when
the parameters of the problem (p2 ; m) are such that the slope of the budget
constraint is never equal to that of an indi¤erence curve.
Note that lim @x 2
@x1
= +1. This means that we do not have a corner
x1 !0
solution for x1 = 0. However this is not the case for x2 = 0. In fact
r
@x2 1 1
= :
@x1 2 m
q
1 1
When p2 2 m
we have a corner solution with x2 = 0 and x1 = m: Note
q p
…nally that p2 12 m1 can be equivalently written p2 2 m; or, as before,
4m (p2 )2 :
(ii) We leave this to the reader. We note though that the demand for
either of the goods, as a function of p2 when m is kept constant, has a point
of non-di¤erentiability; but despite this, it is continuous.

Problem 3.6.4 Let the utility function be:

u (x1 ; x2 ; x3 ) = (x1 ) (x2 ) (x3 )1

Write down the budget constraint: call income m and normalize the price p1
of good x1 , that is, p1 = 1. Determine the demanded bundle as a function of
prices p2 ; p3 ; and income, m.
Answer. We solve the following problem:

max (x1 ) (x2 ) (x3 )1

x1 ;x2 ;x3

s:t: p1 x1 + p2 x2 + p3 x3 = m

Write the Lagrangian and its foc’s to obtain:

(x1 ) 1 (x2 ) (x3 )1 p1 = 0

(x1 ) (x2 ) 1 (x3 )1 p2 = 0
(1 ) (x1 ) (x2 ) (x3 ) p3 = 0;
58 CHAPTER 3 CONSUMER CHOICE

and hence
x1 p1
=
x2 p2
x1 p1
= :
(1 ) x3 p3

After some algebra, we obtain

(1 )
(p1 + p1 + p1 )x1 = m

Hence,
m
x1 =
p1
m
x2 =
p2
(1 )m
x3 = :
p3

Problem 3.6.5 Let the utility function be:

p
u (x1 ; x2 ) = x1 + (x2 ) (x3 )1 ; 0< < 1:
(i) Determine the demanded bundle as a function of the prices, p1 , p2 , p3
and income, m. (ii) Draw the demand for good x1 as a function of its price
p1 (taking p2 , p3 , and m as given). (iii) Let the utility function be instead:

1
u (x1 ; x2 ) = log x1 + 2 log x2 :
2
(It does not get any easier than this. But be careful and precise. Write
down the budget constraint: call income m and normalize the price p1 , that
is, p1 = 1. Determine the demanded bundle as a function of price p2 and
income, m. (substitution works; Lagrange too; whatever is easier for you).
Derive the demand for good x2 and the demand for x1 as a function of price
p2 . A 10% increase in income, at prices p1 = p2 = 1, has a bigger e¤ect on
x1 or on x2 ? Try and compute the e¤ect on x1 and on x2 :
3.6 PROBLEMS 59

Problem 3.6.6 An agent chooses consumption bundle according to the fol-

lowing utility function: U (x1 ; x2 ) = (x1 )2 +x1 x2 . (i) She has income m = 20,
and prices for goods x1 and x2 are p1 = 3 and p2 = 1. How much of good
x1 and x2 will she consume? (ii, HARD) What happens to the agent’s con-
sumption if …rm X introduces 50% discount on its product, i.e., p1 = 1:5?
Answer. (i) Proceed as in the previous problems to obtain:
2x1 + x2
x1 =
3
which implies that x1 = x2 : Substituting into the budget constraint, we
have x1 = x2 = 5:
(ii) Proceed as in (i) to obtain
2x1 + x2
x1 = ;
1:5
which implies that 0:5x1 = x2 : But since both x1 and x2 have to be non-
negative, the only possible solution to this condition is x1 = x2 = 0. However,
the budget constraint is not binding at this bundle and hence this bundle
cannot be maximizing u (x1 ; x2 ) : Observe for instance that the bundle x1 =
x2 = 1 is within the budget set and delivers a higher value of u: In fact, the
second order conditions on the concavity of u (x1 ; x2 ) are not satis…ed at the
bundle x1 = x2 = 0: There consumer problem then has no interior maximum,
but rather a corner solution. Suppose the consumer consumes only good x1 :
In this case, x1 = 13:33 and u(13:33; 0) = 13:332 : Suppose alternatively the
consumer consumes only good x2 : In this case, x2 = 20 and u(0; 20) = 0: We
can conclude then that the solution is x1 = 13:33; x2 = 0:
Problem 3.6.7 Consider an agent whose preferences over any couple (x1 ; x2 ) 2
R2+ of e.g., apples and oranges, is such that she prefers the bundle that is clos-
est to having the same number of apples and oranges. Write a utility function
u : R2 ! R which represent these preferences.
Answer. Let (x1 ; x2 ) denote a bundle with x1 apples and x2 oranges.
The agent prefers the bundle that is closest to having the same number of
apples and oranges,
(x1 ; x2 ) % (y1 ; y2 ) , (x1 x2 )2 (y1 y2 )2

Hence, u(x1 ; x2 ) = (x1 x2 )2 or u(x1 ; x2 ) = kjx1 x2 j; with k < 0;

represent preferences % :
Chapter 4

Equilibrium and e¢ ciency

Leon Walras Vilfredo Pareto Ken Arrow Gerard Debreu Lionel McKenzie

We investigate the fundamental economic problem of allocation and price

determination in a very simple economy. Our aim is to describe what out-
comes might arise by giving individuals the opportunity to voluntarily ex-
change goods.
Thus, we will follow two simple principles:
(i) Rationality - individuals choose the best patterns of consumption that
are a¤ordable for them, and
(ii) Equilibrium - prices adjust such that the amount that people demand of
some good is equal to the amount that is supplied.
We will determine equilibrium prices, equating demand and supply. We
will show that these prices solve the allocation problem e¢ ciently. Nonethe-
less the notion of equilibrium price is distinct from the notion of value, used

61
62 CHAPTER 4 EQUILIBRIUM AND EFFICIENCY

in common parlance; for instance in, "water is enormously valuable to each

of us." While this is not commonly made precise, we implicitly interpret the
value of a good as a measure of how di¢ cult it is to do without it. Read your
preferred rendition of King Midas’fable. The king understands the value of
simple things when his golden touch makes them completely unavailable to
him. There is no reason (it is a logical fallacy) to conclude that values and
not prices are what the noble man/woman (or the social scientist) should
care about. There is a long list of social scientists/phylosophers which have
fallen into a version of this fallacy.

4.1 The Economy

We shall consider a pure exchange economy: that is, an economy with no
production, in which consumers have …xed endowments of goods. In our
economy, markets are competitive: that is, consumers take prices as given,
independently of their trading choices. We say therefore that consumers are
price takers. Furthermore, we assume consumers are rational: that is, they
choose the consumption bundle which maximizes their utility (and, as you
know, they have a utility which represents their preferences to maximize if
and only if they are rational).
These assumpions are fundamental in our analysis. For notational sim-
plicity instead, and without loss of generality in fact, we assume that the
economy is a 2 2 economy: that is, it is composed of two (types of) con-
sumers who consume two (types of) goods.

4.1.1 Notation and De…nitions

Let f1; 2g denote the set of consumption goods. Let instead fA; Bg denote
the set of consumers. Consumer A’s and consumer B’s initial endowments
are denoted, respectively, wA = (w1A ; w2A ) and wB = (w1B ; w2B ). Similarly,
xA = (xA A
1 ; x2 ) and x
B
= (xB B
1 ; x2 ) denote consumer A’s and consumer B’s
consumption bundles respectively. Finally, Consumer A’s and consumer B’s
preferences are represented, respectively, by utility functions uA (x1 ; x2 ) and
uB (x1 ; x2 ) :

De…nition 4.1.1 An allocation is a pair of consumption bundles, xA and

4.2 PARETO EFFICIENCY 63

xB . An allocation (xA ; xB ) is a feasible allocation if:

xA B A B
1 + x1 = w1 + w1

and
xA B A B
2 + x2 = w2 + w2

That is, if the total amount consumed of each of the goods is equal to the
total amount available.

4.2 Pareto e¢ ciency

It is important to be precise about which de…nition of e¢ ciency we shall use.
Suppose we could aggregate each di¤erent (type of) agent’s preferences into
a preference ordering which i) satis…es completeness and transitivity, and
which ii) satis…es some general notion of fairness (e.g., it is in accordance
with preferences of the majority). Then we could construct a utility function
representing such preference ordering and require that it be maximized at an
e¢ cient allocation. If such a utility function would exist, we would refer to
it as a Social welfare function. But we have learnt in the previous chapter
(recall Ken Arrow’s Impossibility theorem) that such a function does not exist
in general.
Tough luck. We are then obliged to adopt a much weaker de…nition of
e¢ ciency. Here it is.

De…nition 4.2.1 A feasible allocation (xA ; xB ) is Pareto-e¢ cient1 if there

is no other feasible allocation (y A ; y B ) such that y A % xA and y B % xB ; with
at least one .

In words, an allocation is Pareto e¢ cient if it is feasible and there is no

other feasible allocation for which one consumer is at least as well o¤ and the
other consumer is strictly better o¤. This implies that at a Pareto e¢ cient
allocation (i) there is no way to make both consumers strictly better o¤,
(ii) all of the gains from trade have been exhausted, that is, there are no
mutually advantageous trades to be made.
Is a Pareto e¢ cient allocation fair ? De…ne fair and think about the
aswer.
1
In honor of Vilfredo Pareto (1848-1923)
64 CHAPTER 4 EQUILIBRIUM AND EFFICIENCY

4.2.1 The Social Planner problem

In this section will characterize the set of Pareto e¢ cient allocations as the set
of solutions to a maximization problem called the Social Planning Problem,
(SP).

max uA (xA A
1 ; x2 ) (4.1)
xA ;xB

subject to
xA B A B
1 + x1 = w1 + w1
xA B A B
2 + x2 = w2 + w2
uB (xB B
1 ; x2 ) u
Since both consumers’ utility is strictly increasing in both goods 1 and
2, it will never be optimal to allocate to consumer B a higher utility than
u. Thus the …nal constraint will be binding (i.e., it can be written as an
equality constraint).2
Notice that, for any given economy (de…ned by an endowment vector
and a utility function for each type of agent) the set of all Pareto e¢ cient
allocations de…ned by the Social planning problem will be a precisely ..... a
set, not a single allocation. This is what you get by having a weak de…nition
of e¢ ciency: a lot of things (allocations) satisfy it!
The set of e¢ cient allocations is obtained as the solution to the Social
Planning problem by varying u.
Convince yourself you understand this. It is as important a point as it
is subtle. Once you are convinced, note that the set of all Pareto e¢ cient
allocations can also be obtained as the solution to the following Modi…ed
Social planning problem:
max u(xA A
1 ; x2 ) + (1 ) u(xB B
1 ; x2 ) (4.2)
xA ;xB

subject to
xA B A B
1 + x1 = w1 + w1
xA B A B
2 + x2 = w2 + w2
2
To be precise, we should also impose non-negativity constraints on allocations:
x1A 0; x2A 0; x1B 0; x1B 0;
4.2 PARETO EFFICIENCY 65

Figure 4.1: Pareto frontier (PPF). Note that, di¤erently from the notation
in the text, in this …gure the two goods are denoted X and Y

Now the set of all Pareto e¢ cient allocations is obtained by varying the
relative weight of agent A in the planner’s objective, that is, by varying
between 0 and 1.
Can you prove that Social planning problem is equivalent to the Mod-
i…ed Social planning problem? Can you prove that they are equivalent to
the de…nition of Pareto e¢ ciency? Careful! It is not hard but requires some
ability/practice to transform words into precise logical statements. Can a
solution of the Social planning (resp. Modi…ed Social planning) problem not
satisfy the de…nition of Pareto e¢ ciency? Can a Pareto e¢ cient alloca-
tion not arise as a maximum of the Social Planning (resp. Modi…ed Social
planning) problem, that is, for no values of u (resp. )?
Note that the (solution of) the Social planning problem does not depend
on the whole endowment vector (w1A ; w1B ; w2A ; w2B ); but only on (w1A +w1B ; w2A +
w2B ): In other words, the (solution of) the Social planning problem does
not depend on the distribution of endowments across agents A and B, but
rather only on the aggregate (economy-wide) endowments of the goods. This
is a manifestation of the fact that, when the Social planner is choosing,
agents have no property rights - the planner chooses how to distribute the
endowments to agents for consumption as he/she pleases. This is not the
66 CHAPTER 4 EQUILIBRIUM AND EFFICIENCY

case in a Competitive equilibrium, which we’ll de…ne in turn.

4.3 Competitive equilibrium

De…nition 4.3.1 A competitive or Walrasian3 equilibrium in a 2 2 econ-
omy is a pair of prices (p1 ; p2 ) and allocations (xA ; xB ) such that:

(xA ; xB ) are demanded by agents A and B at prices (p1 ; p2 ); and

markets clear:

xA B A B
1 + x1 = w1 + w1
xA B A B
2 + x2 = w2 + w2

In words, in a competitive market / Walrasian equilibrium the total de-

mand for each good should be equal to the total supply. Put di¤erently, an
equilibrium is a set of prices such that each consumer is choosing her most
preferred (and a¤ordable) bundle, and both consumers’choices are compat-
ible in the sense that the total demand equals the total supply for each of
the goods. Notice the combination of the two principles which guide our
analysis: rationality and equilibrium.
Let’s start and analyze the notion of Competitive equilibrium. What hap-
pens to the budget set if both prices, p1 and p2 are proportionally increased
to p1 and p2 , for > 0? Note that your answer (well, the correct an-
swer) implies that one price can always be normalized when looking at a
competitive equilibrium, that is, p1 = 1 without loss of generality. Note that
this is equivalent to saying that only relative prices, like pp12 , are determined
at a competitive equilibrium. Furthermore, at a competitive equilibrium
(xA ; xB ; pp21 ) prices satisfy (check this!):

@uA (xA
1 ;x1 )
A @uB (xB
1 ;x1 )
B

p2 @xA
2 @xB
2
= = (4.3)
p1 A
(
@u xA A
1 ;x1 ) @uB (xB1 ;xB1 )
A
@x1 @xB
1

That is, at a competitive equilibrium relative prices are equal to both agents’
marginal rates of substitution evaluated at the equilibrium allocation! This
3
In honor of Leon Walras (1834-1910).
4.4 WELFARE ECONOMICS 67

implies, clearly, that the competitive equilibrium price of a good does not
contain any information about the value of this good. And there is no reason
why it should.

4.4 Welfare Economics

Is the competitive market mechanism Pareto e¢ cient ? More precisely, Are
Competitive equilibrium allocations Pareto e¢ cient? Or, in other words, Can
the competitive market mechanism really extract all the possible gains from
trade? The answer is yes.
Theorem 4.4.1 (First Theorem of Welfare Economics) Suppose pref-
erences are monotonic. Then, all competitive equilibria are Pareto e¢ cient.
Proof. We proceed by contradiction. Suppose there exist a feasible allo-
cation (y A ; y B ) that Pareto dominates (is weakly preferred by both agents
to, and is strictly preferred by at least one to) the competitive equilibrium
(xA ; xB ). Then, the allocation (y A ; y B ) must be not budget feasible for at
least one agent:
p1 (y1A w1A ) + p2 (y2A w2A ) 0
p1 (y1B w1B ) + p2 (y2B w2B ) 0
with at least one strict inequality sign. Summing up:

p1 (y1A w1A ) + p2 (y2A w2A ) + p1 (y1B w1B ) + p2 (y2B w2B ) > 0 (4.4)
By monotonicity of preferences, prices are positive (convince yourself of
this). Then equation (4.4) implies that either
(y1A + y1B w1A w1B ) > 0
or

(y2A + y2B w2A w2B ) > 0

which contradicts feasibility of (y A ; y B ).
The First Theorem of Welfare Economics says that all competitive equi-
libria are Pareto e¢ cient. This is the formal argument behind Adam Smith
invisible hand, in the ’Wealth of Nations,’(1776).
Is the converse true ? That is, Is any Pareto e¢ cient allocation a com-
petitive equilibrium for some endowments and prices ?
68 CHAPTER 4 EQUILIBRIUM AND EFFICIENCY

Figure 4.2: Note that, di¤erently from the notation in the text, in this …gure
an arbitrary agent is denoted h and endowments are denoted with the letter
e:

Theorem 4.4.2 (Second Theorem of Welfare Economics) Suppose pref-

erences are monotonic and have indi¤erence curves which are strictly convex.
Then any Pareto e¢ cient allocation is a competitive equilibrium for some
0 0
prices and endowments (wA ; wB ) such that

w10A + w10B = w1A + w1B

w20A + w20B = w2A + w2B

The Second Welfare theorem implies that all allocations in the Pareto
e¢ cient set (obtained by varying u in the Social planning problem or by
varying in the Modi…ed Social planning problem) are obtained also by
varying the distribution of aggregate endowments across (types of ) agents.

4.4.1 The Edgeworth box

The Edgeworth box4 is a convenient graphical way to analyze the aspects of
an economy with two consumers and two goods.

4
In honor of Francis Edgeworth (1845-1926).
4.4 WELFARE ECONOMICS 69

Figure 4.3: Edgeworth Box. Note that, di¤erently from the notation in the
text, in this …gure endowments are denoted with the letter e:

The Pareto e¢ cient set is constructed as the set of points (that is, the
set of allocations) in the Edgeworth box at which the indi¤erence curves of
the two consumers are tangent (in the interior of the box). If indi¤erence
curves are not tangent, it must be that there exist some advantageous trade
to explore and hence the corresponding allocation is not Pareto e¢ cient.

De…nition 4.4.3 The set of all Pareto e¢ cient allocations is called the con-
tract curve.

The proof that any competitive equilibrium is Pareto e¢ cient (First Wel-
fare Theorem) has a graphical representation in the Edgeworth box. The
proof goes something like this: A feasible allocation (in the Edgeworth box)
allocation is Pareto e¢ cient if the intersection of consumer A’s strictly pre-
ferred set and consumer B’s strictly preferred set is empty (that is, if the
indi¤erence curves of the two consumers are tangent in the interior of the
box). This is the case for a Competitive equilibrium, as in this case onsumer
A’s and consumer B 0 s strictly preferred sets cannot intersect since they lie
on di¤erent sides of the same budget set line.
70 CHAPTER 4 EQUILIBRIUM AND EFFICIENCY

4.5 Externalities
The First Welfare theorem is broken if the economy we consider is plagued
by externalities. In this case, the notion of Competitive equilibrium (the
competitive market mechanism) does not select Pareto e¢ cient allocations.
What are externalities? Here’s two fundamental examples.
Preference externality. Consider an economy in which the utility of
agent A depends also on the consumption choice of agent B; e.g., it’s:

uA (xA A B
1 ; x2 ; x1 ):

We say that the consumption of good 1 by agent B has an externality on

agent A (positive or negative depending on wether uA (:) is, respectively,
increasing or decreasing on x1B ).
Public good. Consider the case in which the utility of both agents
depend on a public good, which is bought adding the contributions of both
agents, that is:
uA (xA A B B B B A
1 ; x2 + x2 ); u (x1 ; x2 + x2 )

Examples of public good are bridges, parks, national defense, etc. They have
the property of being non-exclusive, that is, an agent’s consumption of it
does not preclude other agents’ consumption. You should immediately see
that they introduce a (positive) externality of A on B and viceversa.

4.6 Problems
Problem 4.6.1 Read Chapter 1 of Adam Smith, The Wealth of Nations,
…rst published in 1776; Edwin Cannan, ed., London: Methuen & Co., Ltd.,
5th edition, 1904. You will …nd the text at

http://www.econlib.org/library/Smith/smWN1.html#B.I,Ch.1,OftheDivisionofLabor

Answer (one page max; typed) the following question: What does labor spe-
cialization have to do with the Invisible Hand and market e¢ ciency?

Problem 4.6.2 Consider the following 2 2 economy (two consumers A and

B and two goods 1 and 2): wA = (w1A ; w2A ) = (1; 0); wB (= w1B ; w2B ) = (1; 2) ;
and the utility functions uA and uB for consumers A and B respectively are:

uA (xA A A A
1 ; x2 ) = x 1 x 2
4.6 PROBLEMS 71

uB (xA A B B
1 ; x2 ) = x 1 x 2

Find an algebraic formula for the the contract curve (Hint: write down and
solve the Social Planner problem)? Consider again the above 2 2 econ-
omy. Solve for the competitive equilibrium (prices and allocations). (Hint:
Solve for all the demand functions in the economy - how many are there?
Impose feasibility and solve for equilibrium prices - how many prices do you
have to solve for? Substitute prices in demand functions to …nd equilibrium
allocations.)

Answer. An allocation is a list of consumption bundles, one for each

consumer:
(xA A B B
1 ; x2 ; x1 ; x2 ):

A feasible allocation is an allocation that satis…es

xA B A B
1 + x1 = w1 + w1 = 2
xA B A B
2 + x2 = w2 + w2 = 2

i.e., satis…es the resource constraints of the economy. An allocation is Pareto

Optimal (PO) if it is feasible and @ another allocation that Pareto Dominates
it.The contract curve is the set of all Pareto Optimal allocations. Rewrite
the contract curve as the set of solutions to the following a maximisation
problem:
max xA A
1 x2
(x1 ;x2 ;x1 ;x2 )
A A B B

subject to

xA
1 0
xA
2 0
B
x1 0
B
x2 0
x1 + x1 = w1A + w1B = 2
A B

xA B A B
2 + x2 = w2 + w2 = 2
xB B
1 x2 u

Since both consumers’utility is strictly increasing in both goods 1 and 2, it

will never be optimal to give consumer B more untility than u. Thus the
…nal constraint will be binding (i.e. an equality constraint).
72 CHAPTER 4 EQUILIBRIUM AND EFFICIENCY

Assuming an interior solution (xA A B B

1 > 0; x2 > 0; x1 > 0; x2 > 0), construct
the Lagrangian

L = xA A B B
1 x2 + (x1 x2 u) + A
1 (x1 + xB
1 2) + A
2 (x2 + xB
2 2):

The f.o.c. of the associated maximization problem are:

@L
= xA
2 + 1 = 0 =) xA
2 = 1
@xA
1
@L
= xA
1 + 2 = 0 =) xA
1 = 2
@xA
2
@L
= 1 + xB B
2 = 0 =) x2 = 1
@xB
1
@L
= 2 + xB B
1 = 0 =) x1 = 2
@xB
2

Equating the equations with the same multiplier gives

xA
2 = xB
2
xA
1 = xB
1:

Dividing the …rst equation by the second gives

xA
2 xB2
A
=
x1 xB1
A B
x2 x
A
= 2B
x1 x1

This condition is the marginal rate of substitution condition

xA
2 A xB
2
A
= M RS12 (xA A B B B
1 ; x2 ) = M RS12 (x1 ; x2 ) =
x1 xB
1
4.6 PROBLEMS 73

Rewriting the two resource constraints we substitute out xB B

1 ; x2

xB
1 = 2 xA
1
B
x2 = 2 xA
2
=)
xA
2 2 xA
1
A
=
x1 2 xA
2
=)
xA
1 = xA2
=)
x1 = xB
B
2

Using the last equality constraint

xB B
1 x2 = u
2
=) xB
1 =u
p
=) xB
1 = u:

This gives us the following allocation

p
xA
1 = 2 u
p
xA
2 = 2 u
p
xB
1 = u
p
xB
2 = u

As long as u 2 (0; 4), the assumption of an interior solution is valid. If

u=0

xA A
1 = x2 = 2
xB B
1 = x2 = 0

and if u = 4

xA A
1 = x2 = 0
xB B
1 = x2 = 2

There does not exist a feasible allocation that can make u < 0 or u > 4.
74 CHAPTER 4 EQUILIBRIUM AND EFFICIENCY

At a competitive equilibrium in this economy, given prices (p1 ; p2 ), the

bundle (xi1 ; xi2 ) clears the markets:
xA B A B
1 + x1 = w1 + w1 = 2
xA B A B
2 + x2 = w2 + w2 = 2:

Since both consumers have strictly increasing utility in both goods, the
budget constraint will hold with equality. Assuming an interior solution,
xA A B B
1 > 0; x2 > 0; x1 > 0; x2 > 0 , the solution to the consumer optimisation
problem can be found constructing the Lagrangian
L = xi1 xi2 + (p1 xi1 + p2 xi2 p1 w1i p2 w2i ):
The f.o.c. of the associated maximization problem are
@L xi2
i
= xi2 + p1 = 0 =) =
@x1 p1
@L xi1
= xi1 + p2 = 0 =) =
@xi2 p2
which imply:
xi2 p1
i
=
x1 p2
and the budget constraint
p1 xi1 + p2 xi2 = p1 w1i + p2 w2i :
Normalising p2 = 1; we obtain
xi2 = p1 xi1
P1 xi1 + xi2 = p1 w1i + w2i
1
For i = A, substituting, we have xA1 = 2:
1 1
For i = B; substituting, we have xB1 = 2 + p1 . Using market clearing for
good 1
1 1 1
+ + = 2;
2 2 p1
which implies p1 = 1: To check that you have not made any mistakes, make
sure the market clearing condition for good 2 holds with p1 = 1:
p1 p1 1 1
+ 1+ = + 1 + = 2:
2 2 2 2
4.6 PROBLEMS 75

The Competitive Equilibrium price-allocation pair is then

1 A 1 B 3 B 3
p1 = 1; xA
1 = ; x2 = ; x 1 = ; x2 = :
2 2 2 2
Note that our assumption that the solution is interior is valid.
Also note that the First Welfare theorem holds. That is that the Com-
petitive Equilibrium is indeed Pareto e¢ cient.

Problem 4.6.3 Consider the following 2 2 economy (two consumers A

and B and two goods 1 and 2. The total endowments of the two goods are
w1A + w1B = 2 and w2A + w2B = 1. The utility functions uA and uB for
consumers A and B respectively are:

uA (xA A A A
1 ; x2 ) = x 1 x 2
uB (xA A B B
1 ; x2 ) = x 1 x 2 :

Suppose the weights of the planner for the two agents are A = 1=4 and
B
= 3=4. i) Find the Pareto e¢ cient allocation for this economy? ii)
Find the competitive equilibrium prices of the economy which has the Pareto
optimal allocation you have found as a competitive equilibrium? [this might
seem hard but it is not; think before jumping in the sea of computation]
iii) Find the distribution of the endowments across agents of the economy
which has the Pareto e¢ cient allocation you have found as a competitive
equilibrium? [this is hard and requires you indeed to jump]

Answer. i) The social planning problem is:

1 A A 3 B B
max x1 x2 + x1 x2
(xA1 ;xA2 ;xB1 ;xB2 ) 4 4
subject to

xA B A B
1 + x1 = w1 + w1 = 2
xA B A B
2 + x2 = w2 + w2 = 1:

Here we assumed non-negativity constraints on allocations are not binding,

so we have an interior solution. (Later on we will see this assumption is not
valid, so we will check the boundary solutions. However, for now we will solve
the questions as if the solution is interior). The problem can be simpli…ed in
76 CHAPTER 4 EQUILIBRIUM AND EFFICIENCY

the following way. Using xB 1 = 2 xA B

1 and x2 = 1 xA
2 , the social planner
solves:
1 A A 3
max x1 x2 + (2 xA 1 )(1 xA
2)
(x1 ;x2 )
A A 4 4
We now have an unconstrained optimization problem with two variables.
The associated f.o.c.’s are
1 A 3
x (1 xA
2) = 0 (4.5)
4 2 4
and
1 A 3
x (2 xA 1 ) = 0: (4.6)
4 1 4
1 1
Hence, xB A B A
1 = 2 x1 = 2 and x2 = 1 x2 = 4 . We can plug these number
back into the social planning welfare, to obtain 14 xA A 3 A A
1 x2 + 4 (2 x1 )(1 x2 ) =
12
32
. It can be easily checked that this solution is not a maximum. In fact, the
solution is at the boundary. It is not hard to see that it is optimal to give all
endowment to agent B: xB = (2; 1); and nothing to agent A, xA = (0; 0).
ii) At the competitive equilibrium, price ratio has to be equal to the
MRS of each agent. However, we have to be careful here since one agent has
no endowment, and his indi¤erence curve has a kink at (xA A
1 ; x2 ) = (0; 0).
Therefore any price line would be tangent at this point. The only condition
comes from agent B.
xB p1
M RS = 2B =
B
(4.7)
x1 p2
So price ratio is equal to
p1 1
= (4.8)
p2 2
iii) Since this is a boundary point, The only endowment across agents of
the economy which has this boundary Pareto e¢ cient allocation as a com-
petitive equilibrium is the Pareto optimal allocation itself. So, xB = (2; 1)
and xA = (0; 0).
Consider a variation of the above question:
Consider the same 2 2 economy, with utility functions :

uA (xA A A A
1 ; x2 ) = log x1 + log x2
uB (xA A B B
1 ; x2 ) = log x1 + log x2
4.6 PROBLEMS 77

Suppose the weights of the planner for the two agents are A = 1=4 and
B
= 3=4.
Assuming the non-negativity constraints on allocations are not binding,
so we have an interior solution (later on we will see this assumption is valid),
the proceed as above to obtain the following f.o.c.’s of the social planning
problem:
1 3
=0 (4.9)
4xA
1 4(2 xA
1)
and
1 3
= 0: (4.10)
4xA
2 4(1 xA
2)
3 3
Hence, xB1 = 2 xA B
1 = 2 and x2 = 1 xA
2 = 4 . . It can be checked that this
allocation is the maximum. At the competitive equilibrium of this economy,
p1
M RS A = M RS B = (4.11)
p2
At the Pareto optimal allocation, xA = ( 12 ; 14 ) and xB = ( 32 ; 34 ), so that
1 1 p1 1
M RS A = ; M RS B = ; and = : (4.12)
2 2 p2 2
Note that we implicitly already shown that the Pareto e¢ cient allocation
can be supported as a competitive equilibrium: we know one point on the
budget line, and we also know the slope of the budget line. So we can write
down the equation for the bugdet line as:
1 1 A 1
(xA
2 )= (x ) (4.13)
4 2 1 2
Any point on this line, which satis…es the feasibility constraints, will give
Pareto e¢ cient allocation as the competitive equilibrium allocation at prices,
p1
p2
= 12 . Note that agent B’s endowment is given by: xB 1 = 2 xA1 and
B A
x1 = 1 x2 .
Problem 4.6.4 Given a 2x2 economy with preferences and endowments

uA (xA A A
1 ; x2 ) = 4 log x1 + 5 log x2
A
wA = (3; 7)
uB (xB B B
1 ; x2 ) = 2 log x1 + log x2
B
wB = (2; 4);
compute the contract curve and a competitive equilibrium.
78 CHAPTER 4 EQUILIBRIUM AND EFFICIENCY

Problem 4.6.5 Below are utility functions and allocations for di¤erent economies.
In each case, give an equation for the contract curve and solve for the compet-
itive equilibrium. In each case show explicitly the three steps for solving for
a competitive equilibrium (…nd demand functions, substitute into feasability
and solve for prices, then substitute back into demand function to solve for
allocations:
i) uA (xA A A A (1
1 ; x2 ) = (x1 ) (x2 )
)
; wA = (3; 7);
uB (xB B B B (1
1 ; x2 ) = (x1 ) (x2 )
)
; wB = (2; 4);

ii) uA (xA A A A A A
1 ; x2 ; x3 ) = 4 log x1 + 5 log x2 + log x3 ; wA = (3; 7);
uB (xB B B B
1 ; x2 ) = 2 log x1 + log x2 ; wB = (2; 4);

iii) uA (xA A A A
1 ; x2 ) = 4 log x1 + 5 log x2 ; wA = (2; 1; 0);
uB (xB B B B B B
1 ; x2 ; x3 ) = 3 log x1 + log x2 + 2 log x3 ; wB = (1; 1; 1);

iv) uA (xA A A A
1 ; x2 ) = 4 log x1 + 5 log x2 ; wA = (1; 2);
uB (xB B B B
1 ; x2 ) = min(x1 ; 2x2 ); wB = (2; 3);

iv) uA (xA A A A
1 ; x2 ) = 3x1 + x2 ; wA = (1; 2);
uB (xB B B B
1 ; x2 ) = min(x1 ; 2x2 ); wB = (2; 3);
Problem 4.6.6 For the following economies, calculate the competitive equi-
librium prices and allocations
i) uA (xA A A A (1
1 ; x2 ) = (x1 ) (x2 )
)
; wA = (1; 2);
uB (xB B B B (1
1 ; x2 ) = (x1 ) (x2 )
)
; wB = (2; 1);
uC (xC C C C (1
1 ; x2 ) = (x1 ) (x2 )
)
; wC = (2; 2);

ii) uA (xA A A A
1 ; x2 ) = log x1 + 4 log x2 ; wA = (4; 4);
uB (xB B B B
1 ; x2 ) = 2 log x1 + 3 log x2 ; wB = (2; 1);
uC (xC C C C
1 ; x2 ) = 3 log x1 + 2 log x2 ; wC = (2; 2);
Problem 4.6.7 For the folloing economy, show that the First Welfare the-
orem doesn’t hold (in other words, calculate the competitive equilibrium, cal-
culate the set of Pareto e¢ cient allocations and show that the competitive
equilibrium is not Pareto e¢ cient,
uA (xA A A A (1
1 ; x2 ) = (x1 ) (x2 )
)
+ xB A
1 ; w = (3; 7);
uB (xB B B B (1
1 ; x2 ) = (x1 ) (x2 )
)
+ xA
1; wB = (2; 1):
Chapter 5

Equilibrium in the labor

market

William Beveridge George Stigler Peter Diamond Dale Mortensen Chris Pissaredes

5.1 Competitive labor markets

The consumer problem is:

max
s
u(c) + v(1 ls )
c;l
s
s.t. c wl
Note that the budget constraint can also be written as
c + w(1 ls ) w;

79
80 CHAPTER 5 EQUILIBRIUM IN THE LABOR MARKET

which stresses the interpretation of the wage rate w as the opportunity cost
of leisure.
The …rm problem is:

max ald wld :

ld

At equilibrium, market clearing requires:

ld = ls = l
c = al:

A characterization of the competitive equilibrium includes:

w = a
@u(c)
@c
w = @V (1 l)
:
@(1 l)

5.2 Search frictions

Let l be workers searching for a job and let v be vacancies o¤ered by …rms.
Matches are determined by an aggregate matching function M (l; v) which
satis…es

M (l; v) is increasing in both l and v

l l
M (l; v) = p( )l = q( )v:
v v
Let = vl denote the tightness of the labor market. The matching probability
that a searching worker …nds a job is

M (l; v)
p( ) =
l
while the probability that a …rm which posted a vacancy …lls it is

M (l; v)
q( ) = :
v
5.2 SEARCH FRICTIONS 81

It is easily shown that the postulated properties of M (l; v) imply that

p( ) is decreasing in ;
q( ) is increasing in :

A unit cost h is required to post a vacancy. A match between a worker and

a …rm produces a units of the consumption good. The relationship between
the …rm and the worker after matching is not competitive and hence the
wage is the outcome of bargaining:

w = a;

where is a measure of the bargaining power of the worker.

The consumer problem is:

max p( )u(c) + v(1 l)

c;l
s.t. c wl and w = a

The …rm problem is:

max q( ) (a w) v hv
v
s.t. w = a

At a search equilibrium,
l
= :
v
Note that both consumers and …rms take as given the tightness of the
market ; when choosing their labor supply and their vacancies. At equilib-
rium then, their choices are consistent and = vl : A characterization of the
search equilibrium includes the following conditions:

@u(c) @v(1 l)
ap( ) = 0; where c = al
@c @(1 l)
(1 )q( )a = h
l
= :
v
82 CHAPTER 5 EQUILIBRIUM IN THE LABOR MARKET

5.2.1 Pareto ine¢ ciency of search equilibrium

The search equilibrium is ine¢ cient. The Social planner problem is:
l
max p( )u(c) + v(1 l)
l;v v
s.t. c = wl;
w = a
l
q( ) (a w) = h:
v
The f.o.c.’s include
l @u(c) 1 @p( vl ) @v(1 l)
ap( ) + u(c) = 0; where c = al
v @c v @ vl @(1 l)
l
(1 )q( )a = h:
v
Compare with the equilibrium conditions, which can be written as:

l @u(c) @v(1 l)
ap( ) = 0; where c = al
v @c @(1 l)
l
(1 )q( )a = h:
v
l
1 @p( v )
Note that v @ l u(c) < 0 at a competitive equilibrium, as u(c) > 0 (recall
v
we normalized u(0) = 0) and p( vl ) is decreasing in = vl : As a consequence,
@p( vl )
at a competitive equilibrium ap( vl ) @u(c)
@c
+ v1 u(c) @v(1
@ vl
l)
@(1 l)
< 0 and an l
h i
such that ap( vl ) @u(c)
@c
+ @v(1 l)
@(1 l)
> 0 would increase e¢ ciency. By concavity
of p( vl )u(c) + v(1 l) in l; this requires a smaller l; that is, a decrease in
search.

5.3 Empirical implications of search equilib-

rium
At equilibrium l satis…es
l
(1 )q( )a = h;
v
5.4 PROBLEMS 83

Figure 5.1: Beveridge Curve

and unemployment is
l
u= 1 p( ) l:
v

An equilibrium relationship between u and v can then be obtained, as ex-

ogenous variables, e.g., productivity a; vary. It can be shown that it satis…es
@p
@v
< 0: This relationship is called Beveridge curve, and in the U.S. economy
it looks as in the Figure.

5.4 Problems
Problem 5.4.1 Prove directly that the equilibrium allocations of the com-
petitive labor market economy are Pareto e¢ cient. (Hint: write a social
planning problem and compare the f.o.c.’s of this problem with the equilib-
rium conditions.)
84 CHAPTER 5 EQUILIBRIUM IN THE LABOR MARKET

Problem 5.4.2 Consider the search friction economy with

u(c) = c;
v(1 l) = log(1 l);
M (l; v) = l v 1 :

Chacterize the search equilibrium as much as possible. Characterize also the

solution of the Social planner problem in this economy and compare it with
the equilibrium. Is unemployment larger at equilibrium or at the solution of
the social planner problem?
Chapter 6

Time and uncertainty

In this section, by using the techniques that we have already learned, we

examine two interesting choice problems:

i) saving over time, and

ii) allocating wealth to a portfolio of assets with risky returns.

In order to focus on saving and consumption over time and under uncer-
tainty, we shall restrict our attention to economies with a single consumption
good: an agent’s consumption is then, in fact, appropriately interpreted as
her total expenditure on consumption .
In this section, by using the techniques that we have already learned, we
examine two interesting choice problems:

i) saving over time, and

ii) insuring risky income.

In order to focus on saving and consumption over time and under uncer-
tainty, we shall restrict our attention to economies with a single consumption
good: an agent’s consumption is then, in fact, appropriately interpreted as
her total expenditure on consumption .

85
86 CHAPTER 6 TIME AND UNCERTAINTY

6.1 Choice over time; a.k.a. Saving

Milton Friedman Franco Modigliani

The economy evolves over 2 periods: t and t + 1, say the …rst and the
second part of an agent’s life. This is just for simpli…cation, but you can think
of an economy with any T periods to be interpreted as years, or quarters,
even T = 1: We denote the income that the consumer will have in each
period by (mt ,mt+1 ), and the amount of consumption in each period by
(ct ,ct+1 ). Income and consumption are measured in the same unit of account,
say dollars. (Do not think about in‡ation yet; we’ll discuss in‡ation later).
Assume that the consumer can borrow and save money at some …xed interest
rate r 0. Don’t be bothered by the fact that a single interest rate applies
to borrowing and saving (a counterfactual in the real world). This is a model,
remember, and the bid and ask spread on active and passive interest rates is
a detail of …nancial markets which is irrelevant for our purposes, that is, to
understand how savings are determined in a competitive economy.

6.1.1 Intertemporal preferences for consumption

The consumer preferences for consumption at time t and t+1 are represented
by the following utility function:

u(ct ) + u(ct+1 ); 0 < <1

Note that we have assumed that i) preferences for consumption are the same
in each period, and ii) future utilities are discounted at rate < 1. Further-
more, we assume that the function u : < ! < is di¤erentiable and strictly
6.1 CHOICE OVER TIME; A.K.A. SAVING 87

d2
concave (that is, dc2
u(c) < 0; for any c 0):1 We shall also assume:

lim u(c) = 1
c!0

that is, the marginal utility of consumption for an agent not consuming
anything is arbitrarily high (in…nity).2
Those of you who are trying to think about how to write the economy
when agents are in…nitely lived, rather than just two periods as we are as-
suming, should know that in this case, we typically write the utility function
as:
X
1
t
u(ct ); 0 < < 1
t=0

6.1.2 The intertemporal budget constraint

Let’s construct now the budget constraint of agents. In fact we have two
budget constraints, one for each period. Let the price of consumption in
periods t and t + 1 be the same (this amounts to assume there is no in‡ation;
we’ll deal with in‡ation later). We can then without l0oss of generality
assume that pt = pt+1 = 1 (this is just a normalization).We say that an
agent saves at t if mt > ct , that is, if he consumes less than she can a¤ord.
Otherwise, if mt < ct , she borrows. Let S denote the amount the consumer
takes to or from period 1. If S > 0 (resp. S < 0) we say the consumer saves
(resp. borrows). The budget constraint at time t is then

ct + S = mt ;

indicating that the consumer uses his income mt either to consume or to

save. The budget constraint in period t + 1 is:

ct+1 = mt+1 + (1 + r)S;

It indicates that the agent in the second period consumes her income plus
the proceeds of his saving (minus the repayment of the borrowing, if S < 0).
1
Note that strict concavity is a more stringent assumption than convexity of preferences
as required up to now. It’ll be clear later on why we need concavity.
2
This is for simplicity, and it can be easily relaxed.
88 CHAPTER 6 TIME AND UNCERTAINTY

We implicitely assumed that the agent cannot (is not allowed to) default on
her obligations.3
Fortunately, we can combine the two budget constraints into one, called
the intertemporal budget constraint, thereby transforming the choice problem
of the agent into one which looks like (and is soved as) the ones we have
studied in the previous chapter. Solve the system of budget constraints for
S, to have:
1 1
ct + ct+1 = mt + mt+1
1+r 1+r
Consider consumption in period t and consumption in period t + 1 as two
1
distinct goods. Then 1+r represents the relative price of consumption at time
t + 1; that is, the amount of period t consumption that the agent has to give
up to get an extra unit of consumption in period t + 1:Thus, the maximum
amount of period t consumption that the consumer can buy, if he borrows
1
as much money as she can possibly repay in period t + 1 is mt + 1+r mt+1 .
This is called permanent income; it is the whole present and future stream
of income of the agent evaluated as of time t.
The intertemporal budget constraint for an in…nitely lived agent will be:
X
1
1
t X
1
1
t
ct = mt
t=0
1+r t=0
1+r
and the right hand side is the permanent income.

6.1.3 Choice over time, for given r

We are ready now for the agent’s saving problem (we have preferences and
budget constraints, we do not need anything else):

max u(ct ) + u(ct+1 )

c1 ;c2

subject to
1 1
ct + ct+1 = mt + mt+1
1+r 1+r
3
Note also that we implicitly assume that the agent does not save nor borrow at time
t + 1: This is without loss of generality because at time t + 1: i) she will not want to save
as she receives no utility from consuming at time t + 2; and ii) she cannot borrow as she
cannot repay at time t + 2 (nor she can default).
6.1 CHOICE OVER TIME; A.K.A. SAVING 89

The …rst order conditions for the maximization problem are:

u0 (ct )
= 1+r (6.1)
u0 (ct+1 )
1 1
ct + ct+1 = mt + mt+1 (6.2)
1+r 1+r
Develop your intuition about what the condition means by drawing the
corresponding indi¤erence curve diagram.
We now solve the problem, that is, we provide some implications of the
solution. We distinguish three di¤erent cases.
1
1. Suppose = 1+r
. Then the …rst order condition (9.3) implies (convince
yourself)
ct+1 = ct ;
and the agent smooths consumption over time. Using the budget con-
straint:
1 1
ct+1 = ct = mt + mt+1 (6.3)
2 1+r
that is, consumption at time t (and t+1) depends on permanent income,
not just today’s income. The dependence of consumption on permanent
1
income rather than income, holds generally (not just when = 1+r ); it
is in fact the fundamental insight of the modern (after the 60’s) theory
of consumption, due to Milton Friedman and Franco Modigliani.
1
2. Suppose > 1+r : That is, the interest rate is relatively high (with
respect to discounting ). In this case, the …rst order condition (9.3)
implies (convince yourself)
ct+1 > ct+1 :
For a relatively high interest rate the agent will save more or borrow less
1
(with respect to the case in which = 1+r ), choosing a consumption
path which is increasing over time.
1
3. Suppose < 1+r : That is, the interest rate is relatively low (with
respect to discounting ). In this case, the …rst order condition (9.3)
implies (convince yourself)
ct+1 < ct+1 :
90 CHAPTER 6 TIME AND UNCERTAINTY

For a relatively low interest rate the agent will save less or borrow more
1
(with respect to the case in which = 1+r ), choosing a consumption
path which is decreasing over time.

6.1.4 Determinants of savings

Recall that we say that an agent saves if mt > ct and that she borrows if
mt < ct+1 . Suppose the interest rate in the economy 1 + r increases. What
happens to the savings of a saver ? And, to the borrowings of a borrower
? A …gure with the budget constraint and indi¤erence curves might help:
consider …rst the case in which mt = 0, so that the agent is certainly (under
our assumptions on preferences) a borrower; and consider than the case in
which mt+1 = 0 and the agent is a saver.

6.1.5 The interest rate r

The interest rate is just any other price. It is determined at a competitive
equilibrium by the market clearing (demand equal supply) condition. In this
case what is clearing is the demand and the supply of loans, that is, borrowing
must equal lending. Suppose there are many agents in our economy, indexed
by i 2 I. Let the income of an arbitrary agent i at t (resp. at t + 1) be
denoted by mit (resp. mit+1 ), and her consumption, respectively, cit and cit+1 .
The market clearing conditions in this economy are then written formally,
X X
cit mit = 0; cit+1 mit+1 = 0
i2I i2I

Note that the …rst condition captures borrowing equalPlending. What is

the role of the second condition then?
P In ifact, none: if i2I cit mit = 0,
i
then, it is necessarily the case that i2I ct+1 mt+1 = 0 (you can convince
yourself by summing all budget constraints over agents i 2 I).
A classical model of equilibrium interest rate is one where i) the economy
at t is populated by a group of young agents (with little income) and a group
of middle-aged agents (with relatively large positive income); and at t + 1
the young agents become middle-aged (with relatively large positive income)
and the middle-aged become young (with little income) - ok, if you think
Benjamin Button agents are too curious to be realistic, just assume that the
6.1 CHOICE OVER TIME; A.K.A. SAVING 91

middle-aged became old (with little income) and nothing is changed.4

6.1.6 In‡ation and Monetary Neutrality

It turns out that it is easy to modify our analysis to deal with the case
of in‡ation. To do this, let us suppose that the price of the consumption
good increases at time t + 1 (that is, the same amount of dollars buys less
consumption, there’s in‡ation). Keeping the normalization that pt = 1;
without loss of generality, we have pt+1 > pt = 1. Given the prices of
consumption in the two periods, the rate of in‡ation, , is given by
pt+1 pt
= = pt+1 1
pt
Suppose the income of the agent mt in the …rst period and mt+1 in the second,
are real, that is in units of goods (not dollars). Let i denote the interest rate
(we called it r before, but bear with me for a couple of lines). Then the agent
budget constraint is:
1 1
ct + pt+1 ct+1 = mt + pt+1 mt+1 :
1+i 1+i
Substituting our de…nition of ; the budget constraint can be re-written as:
1+ 1+
ct + ct+1 = mt + mt+1
1+i 1+i
1+i
Note that if you substitute 1 + r = 1+ in this budget set you get back
the budget set of the economy with no in‡ation. We are ready distinguish
between the nominal interest rate, i, and the real interest rate, r, which we
de…ne by
1+i 5
1+r = :
1+
The real interest rate is the interest rate in terms of goods; while the nominal
interest rate is in terms of dollars. If dollars lose value over time (because
4
These economies, called overlapping generation, were developed by Maurice Allais and
Robert Samuelson in the 50’s.
5
If you really want to see the exact expression for the real interest rate in terms of the
nominal interest rate and the in‡ation rate, here it is:
1+i i
r= 1= :
1+ 1+
92 CHAPTER 6 TIME AND UNCERTAINTY

of in‡ation), they are di¤erent. When there is no in‡ation, however, the

real and the nominal interest rates are equal (check this). In the previous
section there was no in‡ation, what we called r was both a real and a nominal
interest rate.
Does in‡ation have any e¤ect on the economy? Are equilibrium quanti-
ties changed because of in‡ation? Are prices? The answer to these questions
is straightforward once you noticed that in‡ation only a¤ects the units of
the budget constraint. The equilibrium condition determines the real inter-
est rate; the nominal interest rate is determined by adding in‡ation. If the
preferences and the incomes of the economy are not changed with the ad-
vent of in‡ation, the equilibrium real interest rate will not change because
of in‡ation per se, nor will equilibrium allocations. Make sure you really un-
derstand this! In‡ation will have no e¤ects in our economy unless something
is changed. For intance, the situation is di¤erent if the agent’s income is
nominal rather than real, that is, if mt and mt+1 are in dollar units. In this
case, the budget constraint becomes (convince yourself):

1+ 1
ct + ct+1 = mt + mt+1
1+i 1+i

and in‡ation acts as a tax, diminishing the purchasing power of the agent’s
income at time t + 1, and hence diminishing his/her permanent income. We
would expect that in the short run unexpected in‡ation has an e¤ect on wages,
and hence on agents’income, but not in the long run. See Bob Lucas’paper
in my webpage on monetary neutrality. Monetary neutrality is nothing else
that the equilibrium property we have just discussed, that is, in‡ation has
no e¤ects if incomes are real (in the long run). We call it monetary neutrality
because we usually think of the in‡ation rate as essentially the same as the
growth rate of the quantity of money. Can you see why is this the case?

For low in‡ation rates the real interest rate is approximately

r i :
6.2 CHOICE UNDER UNCERTAINTY; A.K.A. PORTFOLIO CHOICE93

6.2 Choice under Uncertainty; a.k.a. Portfo-

lio choice

John Von Neumann and

Oskar Morgenstern David Cass Joeseph Stiglitz

We …rst introduce random variables and expected values.

6.2.1 Probability
Let = f! 1 ; ! 2 g be the set of all possible events; for instance, ! 1 = it rains,
! 2 = it does not rain. Consider the following probability distribution (on
events): event ! 1 will occurr with probability p, 0 < p < 1, and event ! 2
with probability 1 p: Stop here and re‡ect on what does the wording "with
probability p" means.
We need now to introduce some simple concepts from statistics, concepts
which refer to random variables: a random variable y is a map from to R
which takes values
y1 = y(! 1 ); y2 = y(! 2 ):
The expected value of y is:

py1 + (1 p)y2 ;

denoted Ey.
94 CHAPTER 6 TIME AND UNCERTAINTY

The variance of y is:

p (y1 Ey)2 + (1 p) (y2 Ey)2 ;

p
denoted 2 (y); (y) = 2 (y) is the standard deviation of y. Both the

variance and the standard deviation are measures of the variability of y.

Note for instance that if y is constant, then y = Ey and 2 (y) = (y) = 0.
Consider two random variables y and z taking values

y1 = y(! 1 ); y2 = y(! 2 ); z1 = z(! 1 ); z2 = z(! 2 )

The covariance of y and z is:

p (y1 Ey) (z1 Ez) + (1 p) (y2 Ey) (z2 Ez) ;

denoted cov(y; z); cov(y;z)

(y) (z)
is the correlation coe¢ cient of y and z.
The covariance and the correlation coe¢ cient of y and z are a measure
of how the variables y and z move together, that is, in our example the
covariance and the correlation coe¢ cient are maximal when either y1 and z1
are both greater than their respective means Ey and Ez; or else y1 and z1
are both smaller than their respective means Ey and Ez.

6.2.2 Expected Utility

Consider an agent trying to evaluate the utility associated to consuming the
random variable y (de…ned above). Let the utility function of the agent be
u(c). His utility when consuming y is:

u(y1 ) with probability p; and u(y2 ) with probability 1 p

We assume an agent evaluates the utility of a random variable y by ex-

pected utility; that is,
pu (y1 ) + (1 p) u (y2 )
In other words, we assume that, when facing uncertainty, agents maximize
expected utility. A lot of experiments document failures of this assumption
in various circumstances. A lot of theoretical work addresses the failure,
postulating di¤erent (still optimizing) behavior on the part of agents. Most
economic theory still uses the assumption in …rst approximation.
6.2 CHOICE UNDER UNCERTAINTY; A.K.A. PORTFOLIO CHOICE95

6.2.3 Risk Aversion

Does an agent with utility function u(c) prefer to consume the random vari-
able y or its expected value Ey ?
Consider an agent with preferences represented by a continuous utility
function u : < ! <. Assume the agents is an expected utility maximizer.
Then the agent will prefer the expected value Ey to the (non-degenerate)
random variable y i¤ u(c) is strictly concave.
Proof. Draw a strictly concave utility function, and represent in the …gure
u(Ey) and pu(y1 ) + (1 p)u(y2 ). It will be apparent that
u(Ey) > pu(y1 ) + (1 p)u(y2 )
i¤ u is strictly concave.
If an agent has strictly concave utility function, we say he is risk averse.

6.2.4 State contingent consumption

How does the consumption problem of the agent looks like with uncertainty?
Suppose the agent’s income is a random variable y, that is, it takes value y1 =
y(! 1 ) with probability p and y2 = y(! 2 ) with probability 1 p: Consumption
in this case is also a random variable, denoted c: The agent distributes his
consumption across states (! 1 ; ! 2 ) by trading state contingent consumption
and income, that is, consumption and income to be delivered conditionally
on the realized state. Assume, without loss of generality, that y1 > y2 , so
that state ! 2 is the unfavorable state for the agent. Convince yourself that
allowing trading in state contingent consumption is equivalent to allowing
consumers trading insurance contracts: to insure the agent buys consumption
in the unfavorable state ! 2 and sells consumption in the favorable state ! 1 :
Let the price of consumption delivered in state ! 1 (resp. ! 2 ) to be q1 (resp.
q2 ): The agent’s choice problem is then:
max pu(c1 ) + (1 p)u(c2 )
fc1 ;c2 g
q1 c 1 + q2 c 2 = q1 y 1 + q2 y 2
The …rst order conditions of this problem are:
1 p u0 (c2 ) q2
=
p u0 (c1 ) q1
q1 c 1 + q2 c 2 = q1 y 1 + q2 y 2
96 CHAPTER 6 TIME AND UNCERTAINTY

We now solve the problem, that is, we provide some implications of the
solution. We distinguish three di¤erent cases.

1. Suppose 1 p p = qq21 . We say that, in this case, the implied insurance

price is fair. The …rst order condition (6.4) implies (convince yourself)

c1 = c2 ;

and the agent smooths consumption over states of uncertainty. In other

words, the agent fully insures himself, consuming the same amount
independently of the realized state of uncertainty. Using the budget
constraint:
q1 y 1 + q2 y 2
c1 = c2 = (6.4)
q 1 + q2

2. Suppose 1 p p > qq21 : We say that, in this case, the implied insurance price
is more than fair (the relative price of consumption in state insurance
! 2 is low and insurance is cheap). The …rst order condition (6.4) implies
(convince yourself)
c1 < c 2 ;
and the agent over-insures himself, consuming more in state ! 1 , when
his income is higher.

3. Suppose 1 p p < qq21 : We say that, in this case, the implied insurance price
is less than fair (the relative price of consumption in state insurance
! 2 is high and insurance is expensive). The …rst order condition (6.4)
implies (convince yourself)
c1 < c 2 ;
and the agent over-insures himself, consuming more in state ! 1 , when
his income is higher.

6.3 Problems
Problem 6.3.1 Consider a two period economy. Agents are all identical,
that is, there is one representative agent. The representative agent is alive
at time t and t + 1, and has preferences:

lnxt + lnxt+1 ; < 1:

6.3 PROBLEMS 97

This agent is endowed with 10 units of the consumption good at time t and
at time t + 1. There is no in‡ation in this economy, and hence you can
assume throughout that the price for the good at time t is 1. i)Write down
the consumer maximization problem of the representative agent (call the real
interest rate r in the budget constraint), …rst order conditions, and demand
functions. ii) Write down the market clearing conditions (that is, feasibility
conditions) for the whole economy. iii) Solve for the equilibrium interest rate
and for the representative agent equilibrium allocation. iv) Suppose the agent
cannot borrow and lend, that is, savings are zero and there is no interest rate
r: he/she has to consume his/her own endowment in each period. How would
you write the budget constraints? [hint: the plural is not a typo; there is a
budget constraint for each time period] Once again the price of consumption
at t and also at t + 1 now can be normalized to 1. Are his/her equilibrium
allocations changed? v) Suppose again that the representative agent cannot
borrow or lend, but he/she can now invest units of the consumption good
at time t (he/she still has his/her endowment as before). Call the amount
invested kt+1 ; the production function is

yt+1 = kt+1 ; >1

that is, the agent can give up kt+1 units of consumption at time t to get kt+1
extra units at time t + 1 (this is all per-capita; in terms of the notation used
in class, kt = 0). Assume > 1 (you will need this later). Write down the
budget constraints for the representative agent [be careful here! the plural is
still not a typo]. Write down the consumer maximization problem, using the
budget constraints you derived, to solve for the optimal choice of investment
kt+1 . Then solve for the equilibrium allocations. [Bonus Question; this
is hard] Suppose that borrowing and lending markets are now open, that is,
there is an interest rate to be determined. This of course together with the
investment and production technology as above. How does the budget con-
straint [yes, singular] look like? Are equilibrium allocations changed? What
is the equilibrium interest rate?

Answer. i) For the maximisation problem, we need the objective func-

tion, which is the utility function above, and a budget constraint. To get
the budget constraint we use the de…nition of savings and solve as we did in
98 CHAPTER 6 TIME AND UNCERTAINTY

class
s = mt xt = 10 xt
xt+1 = mt+1 + (1 + r)s = 10 + (1 + r)s
1 1 1
) xt + xt+1 = 10 + 10 = 10( + 1)
1+r 1+r 1+r
This gives the Lagrangian
1 1
L(x1 ; x2 ; ) = lnxt + lnxt+1 (xt + xt+1 10 10)
1+r 1+r
And …rst order necessary conditions:
@L 1
= =0
@xt xt
@L 1
= =0
@xt+1 xt+1 1+r
@L 1 1
= xt + xt+1 10 10 = 0
@ 1+r 1+r
Combining these gives us
xt+1
= (1 + r)
xt
) xt+1 = (1 + r)xt
Substituting into the constraint gives us
1
xt + xt = 10(1 + )
1+r
10 2 + r
) xt = ( )
1+ 1+r
10
) xt+1 = (2 + r)
1+
ii) As always, the market clearing (feasibility) conditions tell us that the
sum of the demand for a particular good across all agents has to equal the
total supply (or endowment). Here this is pretty easy, as there is only one
agent! So we have
xt = mt = 10
xt+1 = mt+1 = 10
6.3 PROBLEMS 99

Substituting in from above gives

10 2 + r
( ) = 10
1+ 1+r
10
(2 + r) = 10
1+

iii) Due to Walras’Law, we know we only need to use one of these condi-
tions to solve for the equilibrium interest rate. For ease, we’ll use the second
one. We can rearrange as follows:
10
(2 + r) = 10
1+
1+
2+r =
1+ 2
r =
1
r =

Substituting back into the demand functions gives

10 2 + r 1 1+ 1
xt= ( ) = 10
1+ 1+r 1+ 1+ 1
1 1+
= 10
1+ +1
= 10

and
10
xt+1 = (2 + r)
1+
10 1 +
=
1+
= 10

Note that we didn’t really have to substitute back in at this stage. We already
know from feasability that consumption has to equal endowment in each
period.
100 CHAPTER 6 TIME AND UNCERTAINTY

iv) If the agent cannot borrow or lend, then they have to balance the
budget in each period. In other words, savings have to be equal to zero.
Using the expressions we wrote down for savings before, we get

s = mt xt = 10 xt = 0
) xt = 10

and
xt+1 = mt+1 + (1 + r)s = 10 + (1 + r)s = 10
As the consumer cannot borrow and save, there is no choice for her to make:
she just has to eat her endowment each period. In fact, this is the same
as the equilibrium allocation from the previous problem though note that
they come about from very di¤erenct processes. Here, we have constrained
the agents to eat their own endowment each period. Previously it was the
feasibility constraint that led to the agent eating their own endowment.
v) We know that in period 1, the agent receives his endowment of 10
units. At this point, she can choose to consume these units or save them as
capital. This gives us our period 1 budget constraint:

xt + kt+1 = mt = 10

In the second period, the agent receives her endowment, plus the output from
investment in period 1. She will spend all this income on consumption. This
gives the second budget constraint:

xt+1 = yt+1 + mt+1 = kt+1 + 10

We solve for the equilibrium allocation. There are lots of di¤erent ways to
do this. The way we will follow here is as follows. First note that the only
thing that the agent really gets to choose is the amount she saves kt+1 . Once
we know this, we know ct from the …rst budget constraint, and ct+1 from the
second budget constraint. We are going to therefore proceed by using the
budget constraints above to get an expression for ct and ct+1 and substitute
this into the utility function

xt = 10 kt+1
xt+1 = kt+1 + 10
) U = ln(10 kt+1 ) + ln( kt+1 + 10)
6.3 PROBLEMS 101

This now an unconstrained maximisation problem in which the agents have

to choose kt+1 . Taking …rst order conditions gives:

@U 1 1
= + =0
@kt+1 10 kt+1 kt+1 + 10
1 1
) =
10 kt+1 kt+1 + 10
kt+1 + 10 = (10 kt+1 )
kt+1 + kt+1 = 10 10
(1 + )kt+1 = ( 1) 10
( 1) 10
kt+1 =
(1 + )

Note that this is only positive if 1. Otherwise (as we don’t allow the
person to invest negative amounts), the optimal choice will be kt+1 = 0.
Assuming that this condition is satis…ed, we can calculate xt and xt+1

( 1) 10 (1 + ) ( 1)
xt = 10 kt+1 = 10 = 10( )
(1 + ) (1 + )
( + 1)
= 10
(1 + )
( 1) 10 ( 1) 10
xt+1 = kt+1 + 10 = + 10 = + 10
(1 + ) (1 + )
(( 1) + (1 + )) (1 + )
= 10 = 10
(1 + ) 1+

[Bonus Question] First, let’s write down the budget constraint for each
period. In the …rst period, the number of bonds that the agent will buy is
equal to her income minus her spending on consumption minus her spending
on capital:
s = mt xt kt+1
In the second period, she will spend all the income she gets, which is equal
to her endowment, plus the return she gets from bonds, plus the return she
gets from her capital

xt+1 = mt+1 + (1 + r)s + kt+1

102 CHAPTER 6 TIME AND UNCERTAINTY

We can now combine these two budget constraints in the usual way:
1
s = (xt+1 mt+1 kt+1 )
(1 + r)
1
) mt xt kt+1 = (xt+1 mt+1 kt+1 )
(1 + r)
1 1 1
) xt + xt+1 = mt + mt+1 + ( 1)kt+1
(1 + r) (1 + r) 1+r
Now stare hard at the right hand side of the equation, remembering that the
left hand side is the amount that the person spends and the right hand side
1
is the amount that they have to spend. What will happen if 1+r 1 > 0?
Then the agent just gets richer and richer the more capital they buy. In this
case, they will just keep on borrowing money and spending it on capital, so
the demand for capital will be in…nite. This cannot be an equilibrium. Why
1
is this happening? Note that the condition 1+r 1 > 0 ) > 1 + r or
in other words the return on the bond is lower that the marginal product of
capital. This explains our previous result: There is an arbitrage opportunity
in this economy. The agent can borrow money on bonds and invest it in
capital to make free money.
1
The only possible equilibrium of this economy is therefore when 1+r
1 0, or the interest rate on the bond is greater than or equal to the marginal
1
product of capital. If 1+r 1 < 0 ) < 1+r then the marginal product of
capital is less than the interest rate. In this case, we might think that there
is an arbitrage opportunity from the agent selling capital and buying bonds.
However, this is not the case because the agent cannot buy less than 0 units
of capital. However, we do know that the agent will never buy capital, as
she would do better by investing in the bond. The economy therefore looks
exactly the same as if there was no capital and investment, in other words
the one we solved in part one of this question. In this case, we know that
r = 1 . So if 1 + r = 1+1 > , this will be an equilibrium with allocations
10 in each period.
There will always be another equilibrium in which = 1 + r. Here the
rate of return on bonds is exactly the same as the rate of return on capital,
so people will be indi¤erent between the two methods of investing. We can
therefore safely assume that the bonds don’t exist. In this case, we will be
in exactly the same equilibrium we calculated above with no bonds.

Problem 6.3.2 Consider the following economy. Time is t = 1; 2. There is

6.3 PROBLEMS 103

a representative consumer whose production technology is:

Yt = F (k) = Akt
where kt is the per capita capital stock at time t and A has a ”productivity
parameter” interpretation. The consumer OWNS the production technology
and thus makes the production decisions. The capital stock from one period
can be stored and re-used in the next period (with NO depreciation). The
consumer is endowed with k1 > 0 units of capital stock in t = 1: There is no
uncertainty. Write the consumer’s capital accumulation problem.
Answer. The ”story” is the following: There is one consumer (who
is assumed to represent the aggregate preferences of many consumers) who
has access to a production technology (capital can be used to produce an
output good) and a storage technology (capital is accumulated from period
to period).
The output good in each period can be used for one of two purposes
- consumption or increasing the capital stock for the next period (invest-
ment/savings).
The period by period "resource" constraints are that consumption today
plus the change in capital stock must equal production today. The change in
capital stock is the capital stock tommorrow minus the capital stock today.
This means the resource constraints are
ct + (kt+1 kt ) = y t
Remember that the production is given by the production function, yt = Akt .
Putting this into the resource constraint gives
ct + (kt+1 kt ) = Akt
The …nal thing to notice is that the consumer only lives for two periods.
Thus he/she has no incentive to build up capital stock for the next period.
As a consequence he/she will not save/invest in the …nal period. Thus the
second period resource constraint will be
c2 = Ak2
Given these constraints, the capital stock in the initial period that the con-
sumer is endowed with, the consumer needs to …nd the combination of con-
sumption in the two periods and capital stock in the second period that
maximises his/her utility. This leads to the capital accumulation problem.
104 CHAPTER 6 TIME AND UNCERTAINTY

Problem 6.3.3 The consumer needs to solve the following problem

Given fk1 g
Choose fc1 ; c2 ; k2 g to
M axfln c1 + ln c2 g s:t:
c1 + (k2 k1 ) = Ak1
c2 = Ak2

This problem can be vastly simpli…ed by substituting out c1 ; c2 and solving the
following problem

Given fk1 g
Choose fk2 g to
M ax fln[Ak1 (k2 k1 )] + ln Ak2 g

The f:o:c: for this problem is

1 1
( 1) + ( Ak2 )=0
[Ak1 (k2 k1 )] Ak2
To solve for the growth rate of consumption. The thing to notice here is
that, by the nature of using the chain rule, we can easily substitute back
the expressions for c1 and c2 : Remember from our period by period budget
constraints

c1 = Ak1 (k2 k1 )
c2 = Ak2

Putting these into the f:o:c: and rearranging

1 1
( 1) + ( Ak2 ) = 0
[c1 ] c2
1 1
( Ak2 ) =
c2 c1
1 c2
Ak2 =
c1
as was done in the lecture notes. Do notice that this is not technically a full
solution to the problem yet since k2 is the choice variable of the problem. To
6.3 PROBLEMS 105

solve for k2 , rearrange the f:o:c: and massage them:

1 1
= ( Ak2 )
[Ak1 (k2 k1 )] Ak2
1
( Ak2 )
=) 1 = [Ak1 (k2 k1 )]
Ak2
=) 1 = [Ak1 (k2 k1 )] k2 1
=) k2 = Ak1 k2 + k1
=) (1 + )k2 = (Ak1 + k1 )
=) k2 = (Ak1 + k1 )
(1 + )

The economy is exactly the same as in Question 1, except that now we in-
troduce uncertainty over the value of the productivity parameter, A, which is
now a random variable. In any time period, t, it takes the value A1 with
probability p and the value A2 with probability (1 p). Formulate the capital
accumulation problem.

Answer. As a consequence of uncertainty, the consumer does not know

what his resource constraints will be in future periods. This depends on
which "state" the world is in, i.e. whether productivity is A1 or A2 . Thus in
future periods there will be one resource constraint for each state. Speci…-
cally this will mean in period 2 we have

c2 (A1 ) = A1 k2 if A = A1
and
c2 (A2 ) = A2 k2 if A = A2

Thus the consumer is also uncertain as to his/her future consumption.

To describe the knowledge that the consumer has more fully, at any time
period, t, the consumer knows

kt
A at time t (this can be either A1 or A2 , but we will denote it A)
ct (once he/she has chosen it)
kt+1 (once he/she has chosen it)
106 CHAPTER 6 TIME AND UNCERTAINTY

The consumer does not know

A at time t+1
ct+1
The consumer is assumed to be an expected utility maximiser. Thus he
chooses fc1 ; c2 ; kt+1 g in order to maximise the following objective function
E(u(c1 ; c2 )) = E(ln c1 + ln c2 )
where E is the expectation operator and the expectation is taken over the
random variable A. Since the E is a linear operator (essentially this means
that E(aX + bY ) = aE(X) + bE(Y )) and c1 is known when the optimisation
is done (in period one)
E(u(c1 ; c2 )) = ln c1 + E(ln c2 )
Since the random variable A is discrete, the expectation of a function of the
random variable is given by
E(f (A)) = pf (A1 ) + (1 p)f (A2 )
where f (:) is a function that takes the random variable, A, as an arguement.
This means that the second term of the objective function can be ex-
pressed as
E(ln c2 ) = p ln[c2 (A1 )] + (1 p) ln[c2 (A2 )]
The consumer needs then to solve the following problem
Given fk1 ; Ag
Choose fc1 ; c2 (A1 ); c2 (A2 ); k2 g to
M axfln c1 + p ln[c2 (A1 )] + (1 p) ln[c2 (A2 )]g s:t:
c1 + (k2 k1 ) = Ak1
c2 (A1 ) = A1 k2
c2 (A2 ) = A2 k2
This problem can be vastly simpli…ed by substituting out c1 ; c2 (A1 ); c2 (A2 )
and solving the following problem
Given fk1 ; Ag
Choose fk2 g to
M ax fln[Ak1 (k2 k1 )] + p ln[A1 k2 ] + (1 p) ln[A2 k2 ]g
6.3 PROBLEMS 107

The f:o:c: for this problem is

1 p 1 (1 p) 1
( 1) + ( A1 k2 )+ ( A2 k2 )=0
[Ak1 (k2 k1 )] A1 k 2 A2 k 2
Solve for the growth rate of consumption. Notice that, by the nature of
using the chain rule, we can easily substitute back the expressions for c1 and
c2 (A1 ), c2 (A2 ). Remember from our period by period budget constraints
c1 = Ak1 (k2 k1 )
c2 (A1 ) = A1 k2
c2 (A2 ) = A2 k2
Putting these into the f:o:c: and rearranging
1 p 1 (1 p)
( 1) + ( A1 k2 )+ ( A2 k2 1 ) = 0
[c1 ] c2 (A1 ) c2 (A2 )
p 1 (1 p) 1
( A1 k2 )+ ( A2 k2 1 ) =
c2 (A1 ) c2 (A2 ) c1
1 A1 A2 1
k2 p + (1 p) =
c2 (A1 ) c2 (A2 ) c1
A 1
k2 1 E =
c2 (A) c1
c1
E Ak2 1 = 1
c2 (A)
0
u (c2 )
E (M P K) = 1
u0 (c1 )
Problem 6.3.4 Consider the following economy. There are 3 periods: t =
1; 2; 3: Population is constant population equal to L every period. Consider
a representative agent. Preferences of the representative agent are
2
u(c1 ) + u(c2 ) + u(c3 ); where u(c) = log(c)
The technology is as follows. Production is Cobb-Douglas, i.e., Yt = AKt L1t :
The capital stock from can be stored and re-used in the next period (with de-
preciation rate ). The initial capital’s endowment is K1 > 0: There is no
uncertainty. Formulate the problem for the representative consumer and solve
for consumption and investment per capita every period with full deprecia-
tion, i.e. = 1.
108 CHAPTER 6 TIME AND UNCERTAINTY

Answer. To formulate the problem for the representative agent, we …rst

de…ne the per capita variables.
Let yt and kt denote output per capita and capital per capita, respectively,
i.e. yt = LYtt and kt = K
Lt
t
: Notice that, by dividing the production function
by total labor Lt ; we express yt in terms of kt ;
1
Yt Kt Lt
= A
Lt Lt Lt
yt = Akt

In our setup there is no population growth, i.e. Lt = L for t = 1; 2; 3:

In case there was, it is necessary to make the appropiate adjustment in the
investment process by the population growth rate. For example, consider a
population growth percent rate equal to every period, i.e. 1 + = LLt+1 t
.
Then, investment per capita in period 2 is given by
K3 (1 )K2 K3 K2
i2 = = (1 )
L2 L2 L2
K3 L 3
= (1 )k2 = k3 (1 + ) (1 )k2
L3 L2
Before formulating the problem notice that, since the consumer lives for
only three periods, it makes no sense to invest in the third period, that is,
increase future capital stock. In the third period, the consumer will optimally
choose not to invest and consume all the output instead.
Hence, the capital accumulation problem for the consumer consists of
choosing consumption and investment allocations (c1 ; c2 ; c3 ; i1 ; i2 ). to

max log(c1 ) + log(c2 ) + 2 log(c3 )

subject to c1 + i1 = Ak1 with i1 = k2 (1 )k1 (6.5)
c2 + i2 = Ak2 with i2 = k3 (1 )k2 (6.6)
c3 = Ak3 (6.7)
K1
given k1 = >0
L
With = 1 investment collapses to i1 = k2 and i2 = k3 :

In our model the representative consumer decides to invest in capital

stock for two main reasons. First, it is clear from the preferences of the
6.3 PROBLEMS 109

consumer that he wants to smooth consumption over time, and the only
means to do this is by carrying over capital to the next period and using it in
the production. Second, the higher the capital stock, the higher the output
and, consequently, the higher the consumption.

Using the three resource constraints we obtain expressions for consump-

tion per capita in terms of capital per capita and plug them into the objective
function:
2
max log(Ak1 k2 ) + log(Ak2 k3 ) + log(Ak3 )

Taking FOCs with respect to the two choice variables, k2 and k3 ; we have6
1 1 1
+ Ak2 = 0
c1 c2
1 1
+ 2 Ak3 1
= 0
c2 c3
which imply respectively
c2 1
= Ak2 (6.8)
c1
c3 1
= Ak3 (6.9)
c2
Plugging (6.7) into (6.9) we obtain the following expression for k3

k3 = c2 (6.10)

Substituting k3 by expression (6.10) in resource constraint (6.6) leads to

the following condition for c2
1
c2 = Ak2 (6.11)
1+
Then, combining (6.11) and (6.8) we obtain
1 1
c1 = k2 (6.12)
1+
6
Recall the Chain Rule:
@f (g(x)) @f @g
=
@x @g @x
110 CHAPTER 6 TIME AND UNCERTAINTY

We can now pin down optimal k2 as a function of k1 by plugging (6.12)

in resource constraint (6.5)
1
1 1
k2 = Ak1
1+

Given this value for k2 ; optimal c1 ; c2 ; k3 and c3 are determined by apply-

ing equations (6.12), (6.11), (6.10) and (6.7), respectively.

Problem 6.3.5 Consider a savings problem with budget constraints

c1 + s = m1
c2 = m2 + (1 + r)s

and preferences:
ln c1 + ln c2
Construct the intertemporal budget constraint (IBC) and solve for savings
s. Find how s depends on r:Embed now the economy in a overlapping gener-
ations economy with 2 consumers, i = A; B with preferences:

U i (ci1 ; ci2 ) = ln ci1 + ln ci2

2 (0; 1);

and endowments:

(mA A
1 ; m2 ) = (0; m)
(mB B
1 ; m2 ) = (m; 0):

Answer. The intertemporal budget constraint (IBC) is constructed as

follows from the budget constraints:
c2 m2
s =
(1 + r) (1 + r)
c2 m2
=) c1 + = m1
(1 + r) (1 + r)
c2 m2
=) c1 + = m1 +
(1 + r) (1 + r)
6.3 PROBLEMS 111

The intertemporal choice problem is:

To choose fc1 ; c2 g to
M axfln c1 + ln c2 g s:t:
c1 0
c2 0
c2 m2
c1 + = m1 +
(1 + r) (1 + r)

Use Langrangian technique

c2 m2
L= ln c1 + ln c2 + c1 + m1
(1 + r) (1 + r)

The f:o:c: are

@L 1
= + =0
@c1 c1
1
=) =
c1
@L
= + =0
@c2 c2 (1 + r)
(1 + r)
=) =
c2
Eliminating

1 (1 + r)
=
c1 c2
=) c2 = (1 + r) c1

Putting this into the IBC

(1 + r) c1 m2
c1 + = m1 + =: W
(1 + r) (1 + r)
W
=) c1 =
(1 + )
(1 + r)
=) c2 = W
(1 + )
112 CHAPTER 6 TIME AND UNCERTAINTY

Solving for the …rst period savings

s = m1 c1
m2
!
m1 + (1+r)
= m1
(1 + )
m2
=) s = m1
(1 + ) (1 + r)(1 + )
Then
@s m2 d 1
= (1 + r)
@r (1 + ) dr
m2
=
(1 + )(1 + r)2
0
In the overlapping generations economy, an allocation is a list of consumption
bundles, one for each consumer
fcA ; cB g where
cA = (cA A
1 ; c2 ) 2 X
cB = (cB B
1 ; c2 ) 2 X

A feasible allocation is an allocation that satis…es

cA B A B
1 + c1 = m 1 + m1 = m
cA B A B
2 + c2 = m 2 + m2 = m

i.e. satis…es the resource constraints of the economy.

A price-allocation pair, rb; fcbA ; cc b= (1; rb) and cbi = (b
B g where r ci1 ; b
ci2 ), is
a competitive equilibrium (CE) if:

For i = A; B and given prices rb= (1; rb), the bundle cbi = (b
ci1 ; b
ci2 ) solves the
problem:
Choose fci1 ; ci2 g to maxfln ci1 + ln ci2 g s:t:
ci1 0
i
c2 0
i
c2 mi2
ci1 + = mi1 +
(1 + rb) (1 + rb)
6.3 PROBLEMS 113

Markets clear:

cA
b cB
1 +b
A B
1 = m1 + m1 = m
cA
b cB
2 +b
A B
2 = m2 + m2 = m

Therefore, for f(cA A B B

1 ; c2 ); (c1 ; c2 )g the consumer optimisation condition im-
plies:

WA m
cA
1 = =
(1 + ) (1 + )(1 + r)
(1 + r)
cA
2 = WA = m
(1 + ) (1 + )
WB m
cB
1 = =
(1 + ) (1 + )
(1 + r) (1 + r)
cB
2 = WB = m
(1 + ) (1 + )

Using the market clearing for consumption at time 1:

m m
+ = m
(1 + )(1 + r) (1 + )
=) 1 + (1 + r) = (1 + )(1 + r)
=) 2 + r = (1 + ) + r + r
=) 1 =r
1
=) 1 + r = 1 +
1
=) 1 + r =

Checking the solution with the market clearing condition for consumption at
time 2
1
m
m+ m = (1 + )
(1 + ) (1 + ) (1 + )
= m
as required.
114 CHAPTER 6 TIME AND UNCERTAINTY

The price-allocation pair that forms a Competitive Equilibrium is

1
rb =
m
cA
b1 =
(1 + )
m
cA
b2 =
(1 + )
m
cB
b1 =
(1 + )
m
cB
b2 =
(1 + )
Notice two important elements of this equilibrium:
Agent B receives more consumption in both periods and thus has a higher
utility than agent A.
Both agents perfectly smooth their consumption over the two time peri-
ods.

Problem 6.3.6 For he following intertemporal utility functions, solve the

consumer’s problem (in other words, solve for the demand of each good in each
period as a function of prices. To do this, you will have to construct a budget
constraint for the agent. Assume that the agent has an income of wi in each
period and that there is an interest rate 1+ri between period i and i+1. ). In
each case, calculate savings as a function of interests rate(s), and decide how
savings will change as the interest rate changes. i) U = u(c1 ) + u(c2 ) where
1
u(c) = c1 : ii) U = u(c1 ) + u(c2 ) + 2 u(c3 ) where u(c) = ln c: (i.e., there
are three periods). iii) U = x1 x2 + x1 x2 . Here there are 2 goods available in
each period. Assume that, in each time period, the price of x1 is p1 and the
price of good x2 is p2 : iv) U = c1 + c2 (here you should consider three cases,
where is greater than, the same as, or less than 1 + r: v) For Question i
above, assume that there is in‡ation is 10% between the two periods (i.e. there
is a price p1 , p2 for the good in the two periods, and that p2 = 1:1p1 . First
assume that the agents income is denominated in real terms. What is the equi-
librium nominal interest rate. Now assume that the agent’s income in denom-
inated in nominal terms. What is the agent’s income? vi) For the following
economies, solve for the competitive equilibrium interest rate(s) and alloca-
tions 1) UA = u(c1A ) + u(c2A ), UB = u(c1B ) + u(c2B ), where the instantaneous
1 2 1 2
utility functions are logarithmic, and with endowments (wA ; wA ), (wB ; wB ):
6.3 PROBLEMS 115

1
2) As above, but let u(c) = c1 and (wA 1 2
; wA 1
) = (1; 3), (wB 2
; wB ) = (2; 1): 3)
2 2
1 2 3 1
UA = u(cA ) + u(cA ) + u(cA ), UB = u(cB ) + u(cB ) + u(c3B ), where 2

the instantaneous utility functions are logarithmic, and with endowments

1 2 3 1 2 3
(wA ; wA ; wA ), (wB ; wB ; wB ): 4) UA = log c1A + log c2A ; UB = c1B + c2B and
1 2 1 2
endowments (wA ; wA ) = (1; 3), (wB ; wB ) = (2; 1):
c1
Answer. i) U = u(c1 ) + u(c2 ); where u(c) = 1
The consumer problem is given by
max u(c1 ) + u(c2 )
c2 w2
subject to c1 + = w1 +
1 + r1 1 + r1
Taking FOC with respect to c1 and c2 we obtain respectively
u0 (c1 ) = 0
u0 (c2 ) = 0
1 + r1
where is the Lagrange multiplier associated with the budget constraint.
Using these two equations we obtain
u0 (c2 ) 1
0
=
u (c1 ) 1 + r1
Given the utility speci…cation we assumed, this condition collapses to
c2 1
=
c1 1 + r1
implying that
c2 1
= ( (1 + r1 ))
c1
1
Let k = ( (1 + r1 )) : Then, c2 = kc1 : Substituting c2 by kc1 in the budget
constraint leads to
k w2
c1 1 + = w1 +
1 + r1 1 + r1
Then we can express c1 as a function of the endowments, ; r1 and ;
w1 (1 + r1 ) + w2
c1 = 1 (6.13)
1 + r1 + ( (1 + r1 ))
116 CHAPTER 6 TIME AND UNCERTAINTY

We can now pin down the saving function as

w1 (1 + r1 ) + w2
s1 = w1 c1 = w1 1 (6.14)
1 + r1 + ( (1 + r1 ))

ii) U = u(c1 ) + u(c2 ) + 2 u(c3 ); where u(c) = log(c)

The consumer problem is given by

max u(c1 ) + u(c2 ) + 2 u(c3 )

c2 c3 w2 w3
subject to c1 + + = w1 + +
1 + r1 (1 + r1 )(1 + r2 ) 1 + r1 (1 + r1 )(1 + r2 )
Taking FOC with respect to c1 ; c2 and c3 we obtain respectively

u0 (c1 ) = 0
u0 (c2 ) = 0
1 + r1
2 0
u (c3 ) = 0
(1 + r1 )(1 + r2 )
where is the Lagrange multiplier associated with the budget constraint.
Using the …rst two equations we obtain
u0 (c2 ) 1
0
= (6.15)
u (c1 ) 1 + r1
while using the …rst and third equation leads to
0
2u (c3 ) 1
= (6.16)
u0 (c 1) (1 + r1 )(1 + r2 )

Given that u(c) = log(c), condition 6.8 collapses to

c2 1
=
c1 1 + r1
implying that
c2
= (1 + r1 )
c1
or, equivalently,
c2 = (1 + r1 )c1 (6.17)
6.3 PROBLEMS 117

Condition 6.9 turns to be

c3 1
=
c1 (1 + r1 )(1 + r2 )
so,
c3 = (1 + r1 )(1 + r2 )c1 (6.18)
We can now plug equations 6.17 and 6.18 into the budget constraint and
derive an expression for c1 as a function of the endowments, interest rates
and :
1 w2 w3
c1 = 2 w1 + +
(1 + + ) 1 + r1 (1 + r1 )(1 + r2 )
We now pin down c2 and c3 by substituting this expression into 6.17 and 6.18
(1 + r1 ) w2 w3
c2 = 2 w1 + +
(1 + + ) 1 + r1 (1 + r1 )(1 + r2 )
(1 + r1 )(1 + r2 ) w2 w3
c1 = 2 w1 + +
(1 + + ) 1 + r1 (1 + r1 )(1 + r2 )
It follows that the saving functions in period 1 and 2 are
1 w2 w3
s1 = w1 c2 = w1 2 w1 + +
(1 + + ) 1 + r1 (1 + r1 )(1 + r2 )
s2 = w2 + s1 (1 + r1 ) c2

iii) U = x11 x12 + x21 x22 ; where xti is the consumption of good i in period t:
We can formulate this consumer problem as follows

max x11 x12 + x21 x22

x2 + p0 x22 w2 + p0 w22
subject to x11 + px12 + 2 = w11 + pw21 + 1
1 + r1 1 + r1
p1 p22
where p = p21 is the relative price in the …rst period and p0 = p21
is the relative
1
price in the second period.

This utility function is convex and we will be having corner solutions. In

particular either all the consumption today is zero or all future consumption
is zero7 .
7
We face the same problem as in the Planner problem solved in the lab.
118 CHAPTER 6 TIME AND UNCERTAINTY

w2 +p0 w2
Let W = w11 + pw21 + 11+r1 2 : Now, if we assume that consumption today
is positive, we can take FOC for x11 and x12 ;

x11 : x12 =0
x12 : x11 p =0

where is the Lagrange multiplier for the budget constraint. It follows that

px12 = x11

Hence, if consumption today is positive, the individual will be spending half

of W in good 1 and the other half in good 2. In that case the utility is

WW W2
U= + 0 0=
2 2p 4p

Now let us assume that consumption in period 1 is zero and consumption in

period 2 is positive. Taking FOC with respect to x21 and x22 we have,

x21 : x22 1+r1

=0
p0
x22 : x21 1+r1
=0

implying that
p0 x22 = x21
Now the individual is spending half of W (1 + r1 ) in good 1 and the other
half in good 2;8 getting a utility value of
2
W (1 + r1 ) W (1 + r1 ) W (1 + r1 )
U =0 0+ = 0
2 2p0 p 2

We have to decide whether the individual has zero consumption in period 1

2
W2 W (1+r1 ) p0
or 2. As long as 4p p0 2
, or equivalently, p
(1 + r1 )2 ; the
0
individual will only be consuming in period 1. In case pp (1 + r1 )2 he will
only consume in period 2. It is left to you to derive the saving function.
8
Notice that now he consumes W (1 + r1 ) rather than just W because he is consuming
in period 2.
6.3 PROBLEMS 119

1) UA = u(c1A ) + u(c2A ); UB = u(c1B ) + u(c2B );where u(c) = log(c); and

1 2 1 2
(wA ; wA ); (wB ; wB ):
In the previous question we showed that when u(c) = log(c) condition
6.17 holds. This means that in this case we have

c2A = (1 + r1 )c1A
c2B = (1 + r1 )c1B

Summing up these two equations we obtain

C 2 = (1 + r1 )C 1

where C i is the total consumption in period i: Using the resource constraints,

we know that C 1 = wA 1 1
+ wB and C 2 = wA 2
+ wB2
: Hence, the equilibrium
interest rate is given by
2 2
1 wA + wB
(1 + r1 ) = 1 1
wA + wB
The allocations can be pinned down by substituting this value for 1 + r1 in
the saving functions:
2
1 1 1 wA
sA = wA wA +
1+ 1 + r1
2
1 1 1 wB
sB = wB wB +
1+ 1 + r1
1
2) UA = u(c1A ) + u(c2A ); UB = u(c1B ) + u(c2B );where u(c) = c1 ; and
1 2 1 2
(wA ; wA ) = (1; 3); (wB ; wB ) = (2; 1):
Now we are considering an economy populated by two individuals, A and
B; where they can borrow and lend from each other, and the interest rate r1
will adjust endogenously to clear the market.
Since we know that the two individuals have same preferences as before,
it follows that

c2A = kc1A
c2B = kc1B
1
where k = ( (1 + r1 )) : If we aggregate consumption we obtain that C2 =
kC1 ; where C1 and C2 are total consumption in period 1 and 2, respectively.
120 CHAPTER 6 TIME AND UNCERTAINTY

1 1
Moreover from the resource constraints we have C1 = wA + wB = 3 and
2 2 4
C2 = wA + wB = 4: Thus, k = 3 which implies that
1
4
3
1 + r1 =

The saving functions will look the same as in the previous exercise; we just
need to plug in this expression for r1 :
Chapter 7

Growth

Robert Solow David Cass Tjalling Koopmans Robert Lucas

What determines the rate of growth of consumption or income (as mea-

sured e.g., by Gross Domestic Product) ? What determines the rate of growth
of per-capita consumption or per-capita income ?
Let variables in capital letters denote aggregate quantities, and lower case
variables denote per-capita quantities; and let L denote the number of agents
in the economy:
Y
y=
L
Let an index t denote time: yt is y at time t. Aggregate growth rates depend
on the growth rate of the population, LLt+1 t
:
Yt+1 yt+1 Lt+1 yt+1 Lt+1
= =
Yt y t Lt y t Lt

121
122 CHAPTER 7 GROWTH

As a consequence per-capita growth are a better indicator of economic

development. As an example, consider the average rate of growth of income
in Mexico and in Italy in the period 1950-2000 (get them from the Penn
World Dataset of Summers-Heston, at

http : ==pwt:econ:upenn:edu=php_site=pwt61_f orm:php)

While the growth rate of the population is in general related to the growth
rate of per-capita consumption and income of an economy (and in interesting
ways, when you endogeneize fertility and allow the agents choose how many
children to have), in …rst approximation we consider it instead exogenous
and we study the growth rate of per-capita consumption and income per-se.

7.0.1 Production
Aggregate production is described by a production function

Yt = F (Kt ; Lt )

where Yt is aggregate income at time t, Kt is aggregate capital (in the same

units as income) at t, and Lt is the population of (the total number of agents
in) the economy at t; assuming that each agent is the same (no productivity
di¤erences, e.g., related to education) and works the same number of hours,
we also refer to Lt as the aggregate labor supply in the economy.
We assume the production function is Cobb-Douglas (this works well with
data):

Yt = A (Kt ) (Lt )1 ; 0< <1

The marginal product of capital in the economy is:
@Yt 1
M PK = = A (Kt ) (Lt )1
@Kt
Note it is decreasing in capital, Kt , and increasing in labor Lt . How do
you see this ? Compute the derivative of M PK with respect to Kt and Lt ,
respectively.
The marginal product of labor in the economy is:
@Yt
M PL = = (1 ) A (Kt ) (Lt )
@Kt
 123

Note it is increasing in capital, Kt , and decreasing in labor Lt . How do

you see this ? Compute the derivative of M PL with respect to Kt and Lt ,
respectively.
The per-capita production function can be computed as follows (the
Cobb-Douglas speci…cation makes it very easy):
Pass to per capita income by dividing by Lt :

Yt (Kt ) (Lt )1 (Kt ) (Lt )1

yt = =A =A
Lt Lt (Lt ) (Lt )1
that is
yt = Ak
The marginal product of capital in per-capita terms is:
@yt 1
= Ak
@kt
it is decreasing in per-capita capital kt ; that is: rich economies have lower
marginal product of capital; where of course "rich" means with high per-
capita capital, and the marginal product of capital is in per-capita terms.

draw …gure

7.0.2 Growth and Savings

Consider an economy in which all agents are identical (we are not interested in
distributional issues, but only in per-capita variables); and live for 2 periods,
t and t + 1.
The budget constraint of the representative agent at time t is:

ct + (kt+1 kt ) = A (kt )
Notice that (kt+1 kt ) is his savings, and also his investment in capital (he
has no other ways of saving, and cannot or will not borrow). The agent’s
income is due to production at time t: A (kt ) .
The budget constraint of the representative agent at time t + 1 is:

ct+1 = A (kt+1 )
and the agent consume all his production, since he will not be around next
period.
124 CHAPTER 7 GROWTH

The representative agent’s preferences are as usual:

u(ct ) + u(ct+1 )
where we assume 0 < < 1; and also u(c) = lnc for simplicity.
The representative agent’s maximization problem (after substitution his
budget constraints) into the speci…cation of preferences) is:
max ln (A (kt ) kt+1 + kt ) + ln (A (kt+1 ) ) (7.1)
kt+1

The …rst order condition (derive it yourself) can be written as:

ct+1
= A (kt+1 ) 1
ct
1
Notice that the right-hand-side is the marginal product of capital (in
per-capita terms) at time t + 1. Notice also that this is the same equation
we derived when studying savings, only that the marginal product of capital
takes the place of (1+r). In fact this is the same problem; the agent is saving
by investing in capital and receives a return equal to the marginal product
of capital.
Finally, notice that the …rst order condition says that: rich economies
have lower rates of growth of consumption; where, again, "rich" means with
high per-capita capital, and the growth rate of consumption is in per-capita
terms.
Is this last implication true in the data ? Look at the the Penn World
Dataset of Summers-Heston, at
http : ==pwt:econ:upenn:edu=php_site=pwt61_f orm:php):
It is true in a subset of countries, Asian New Industrialized Countries
(South Korea, Hong Kong, Thailand, ..), China which is "poor" in per-capita
capital and has been growing at 8% recently, while Germany (and the rest of
Europe), the U.S. have been growing at less than 3%. It is not true however
for African and Latin American countries, who are both "poor" in per capita
income (and capital) and have been growing very slowly if at all.
1
By substituting the budget constraint equations into the …rst order condition you can
solve for the growth rate of capital in the economy, kt+1
kt . Try doing it (when you have
some free time, it’s easy but algebraically involved. The solution I got is:
kt+1 1
= 1 + A (kt )
kt 1
 125

7.0.3 Other Determinants of Growth

Write the …rst order conditions of the maximization problem in (7.1) in terms
of kt+1 and kt . You will have an equation of the form:
kt+1 = constant1 : (kt ) + constant2 : kt ; constant2 < 1 (7.2)
Study this equation mathematically as a di¤erence equation (but be careful
here: the equation is derived as the …rst order condition of the maximization
problem of an agent who lives only 2 periods; by considering it a di¤erence
equation we are implicitly assuming that the same equation would hold if the
agents lived an in…nite number of periods; in fact this is true, in the sense that
an equation of the form (7.2) holds in this case, even though the expression
for the constants is di¤erent). The equation has two steady states: k = 0
and k > 0. The steady state k is globally stable: all paths of kt starting
from any k0 > 0 converge to it. We conclude that for < 1 the economy
does not grow inde…nitely!
There are two possible modi…cations in our basic model which are such
that the economy grows inde…nitely. We can then ask ourselves in these
economy what are the determinants of the growth rate and hence attempt
some explanations which help account for the low growth rates of Africa and
Latin America. Both modi…cations require changing the production function.

Human Capital.
Suppose the production function is:
Yt = (AKt ) (ht Lt )1
where ht is an index of quality of labor, called human capital, and
ht+1 = ht ; >1
In per capita terms, the production function becomes:
yt = (kt ) (ht )1
ht+1 = ht
In this case, by proceeding exactly as in the previous section (do it as an
exercise), we can compute the growth rate of consumption:
1
ct+1 ht+1
= A
ct kt+1
126 CHAPTER 7 GROWTH

Convince yourself that in this case the economy grows. The argument is
as follows. Suppose not. Then kt+1kt
! 1 for kt large enough. hht+1 t
! >1
ht+1
instead, independently of kt . Therefore, for large kt , kt+1 increases over time,
and the economy grows. The important question then is: What does depend
on ? Partial answer for discussion:

The structure of the family

Fertility and its determinants

Schooling system

Urban Development

Technological Innovations and Knowledge.

Suppose the parameter A in the production function, which measures total

productivity (sometimes we refer to it as the general knowledge of an econ-
omy), grows over time,
At+1 = At (7.3)

Once again you can show that (think about how you would do this)
in this case the economy will grow. Once again, the important question
then is: What does depend on ? To develop a listing of partial answers
for discussion a comment is very useful: General knowledge can be used
by the whole economy without diminishing returns: think of a blueprint to
produce a medicine; it can be freely copied and used by many …rms. We say
general knowledge is non-rival. General knowledge can be private or public
depending on the institutions: think of the blueprint; we can protect it or not
with a system of patents. We say that general knowledge is excludable. (Non-
excludable goods exist: …shing in the sea has proved very hard to exclude
over the years).What is the di¤erence between institutions which guarantee
exclusion and those which do not ? Think of the following model of total
productivity or knowledge:

At = A (jt )1
 127

where jt is knowledge and

8
>
> k under excludability
>
< t
jt =
>
>
>
: k under non-excludability
t

and k t is a measure of average capital that the representative agent in the

economy cannot a¤ect with his capital investment. At equilibrium though
all agents choose the same capital (there is a single representative agent),
and hence k t = kt . Compute now for both models the marginal product of
capital in per capita terms:

@ 1 1 1
A kt (kt ) = A k t (kt ) = A at equilibrium [excludable]
@kt

@
A (kt )1 (kt ) = A [non excludable]
@kt
In the model then is endogenous and depends on the existence of in-
stitutions which guarantee excludability of general knowledge. If you are
thinking that it might not be that At+1 satis…es (7.3) in equilibrium, you are
right. But this is not so important, and we can rig the model so that it does.
Partial answers to the What does depend on ? question for discussion
(also look at the papers by Murphy-Shleifer-Vishny (1991) and Acemoglu-
Robinson (2004) posted on my website):

The structure of the legal system

Property rights

Industrial organization and the patent system

Urban Development

A great (and simple to read) book to understand all this and more is:
W. Easterly, The Elusive Quest for Growth: Economists’Adventures and
Misadventures in the Tropics, MIT Press, 2001.
128 CHAPTER 7 GROWTH

Generosity of spirit, good intentions, and great accomplishments in en-

tertainment

Ali Hewson and Bono Angelina Jolie

are not always enough:

Chapter 8

Finance

Ken Arrow Robert Merton Myron Scholes and Fisher Black Robert Luca

In this chapter we shall see how a model-driven de…nition of risk is useful

and counterintuitive. We shall also see that, an implication of this de…nition
is that,
there are no free lunches, not even in …nance!

8.1 Asset Pricing and the CAPM

Consider an agent living two periods, t and t + 1. His utility in terms of
consumption ct and ct+1 is:
u(ct ) + u(ct+1 )

129
130 CHAPTER 8 FINANCE

where < 1 is the discount rate. We assume u(c) is smooth, strictly

monotonic, and strictly concave, and the agent is an expected utility maxi-
mizer.

8.1.1 Portfolio Choice

The agent faces a wealth process wt , wt+1 . He can trade J assets; asset
j = 1; : : : ; J has payo¤ xjt+1 at t + 1.
The agent’s problem is the following:

max j
; j=1;:::J u(ct ) + Et u(ct+1 ) (8.1)

subject to:
X
J
ct = wt qtj j

j=1

X
J
ct+1 = wt+1 + xjt+1 j

j=1

where qtj is the price of asset j, and j is the amount of asset j in the agent
portfolio, the choice variable. The …rst order conditions for the maximization
problem imply the fundamental asset pricing equation:

u0 (ct+1 ) j
qtj = Et x (8.2)
u0 (ct ) t+1
If the asset is long lived, e.g., a stock, its payo¤ at t + 1 is the sum of its
j
resale price, qt+1 and its cash ‡ow at time t + 1, e.g., its dividend; we write
j j j
xt+1 = qt+1 + dt+1 . In this case, the pricing equation becomes:

u0 (ct+1 ) j
qtj = Et qt+1 + djt+1 (8.3)
u0 (ct )

8.1.2 Time and Risk Correction

Let’s examine the asset pricing equation (8.2). If there is no uncertainty,
that is if the agent is able to perfectly insure, so that ct = ct+1 , or if he is
risk neutral, so that u0 (c) is a constant,

qtj = Et xjt+1
8.1 ASSET PRICING AND THE CAPM 131

and the price of asset j is the net present value of its expected payo¤ at t + 1
discounted at the pure discount rate .
If instead the agent is risk averse and his consumption is a stochastic
u0 (ct+1 )
process, then by (8.2), the discount factor is stochastic: u0 (ct )
. In this
case, the discount contains a risk correction. To see this, note that (8.2) can
be written:1

u0 (ct+1 ) u0 (ct+1 ) j
qtj = E E xjt+1 + cov ; xt+1 (8.4)
u0 (ct ) u0 (ct )

and the covariance of the asset payo¤ with the marginal rate of substitution
of the agent is the relevant component of the risk of the asset which enters
in the price.
Because of the concavity of the utility function u(c) (that is, u00 (c) < 0),
u0 (ct+1 )
if cov u0 (ct )
; xjt+1 > 0, then cov ct+1
ct
; xjt+1 < 0. Therefore, equation
(8.4) implies that assets whose payo¤ is negatively correlated with the agent’s
consumption are more valued by the agent and hence have a higher price.
Why are they more valued, in intuitive terms? Because they allow the agent
to insure, that is to reduce the risk (roughly, variance) of his consumption
process. Remember:
Agents care about the variance of their consumption and hence about the
covariance of the asset’s payo¤ with his consumption.
Note that the consumption which enters the stochastic discount factor is
each single agent’s consumption in the economy. If …nancial markets are de-
veloped enough (they are complete, in the economist’s parlance), and utility
0
functions are well behaved, then uu(c0 (ct+1
t)
)
is equalized for any agent in equilib-
rium and we can think without loss of generality of ct+1 as of the economy’s
consumption.2

1
For any two random variables x and y;

E (x y) = Ex Ey + cov(x; y):

2
This is a fundamental result in economic theory. If you want to know more about it,
look at J. Cochrane, Asset Pricing, Princeton University Press, 2001; this is not part of
the course though.
132 CHAPTER 8 FINANCE

8.1.3 Beta Representation

Often the pricing equation is written as an equation for excess returns. The
j xj
return of asset j is written Rt+1 = qt+1j . We can write then the pricing
t
equation (8.2) as:
u0 (ct+1 j
1=E R (8.5)
u0 (ct ) t+1
f
Consider a risk free return Rt+1 , that is, a return which is is known at t. In
this case equation (8.5) can be written:

f 1
Rt+1 = (8.6)
u0 (ct+1
E u0 (ct )

Tedious algebra (but nothing more than algebra) will let us now rewrite (8.4)
as:

0 u0 (ct+1 j
10 u0 (ct+1
1
cov u0 (ct )
; Rt+1 var u0 (ct )
j
E Rt+1 f
Rt+1 =@ A@ A (8.7)
u0 (ct+1 u0 (ct+1
var u0 (ct )
E u0 (ct )

u0 (ct+1 ) j
!
cov u0 (ct )
;Rt+1
j
Think of = u0 (ct+1
as of a measure of the riskiness of
var u0 (ct )

return j, its conditional beta (not to be confused with the pure discount
u0 (ct+1
!
var u0 (ct )
factor). Think of u0 (ct+1
instead as a measure of the price of risk.
E u0 (ct )

Naturally, the price of risk is 0 if the variance of consumption is 0; and in

j f u0 (ct+1 j
this case, Rt+1 = Rt+1 . Moreover, note that if cov u0 (ct )
; Rt+1 < 0, then
Rtj > Rtf ; and the risk premium is positive.
In their 1972 study on The Capital Asset Pricing Model: Some Empirical
Tests, Fischer Black, Michael C. Jensen and Myron Scholes documented the
linear relationship between the …nancial returns of stock portfolios and their
betas, with return data from New York Stock Exchange stocks between 1931
and 1965. Here’s the …gure:
8.1 ASSET PRICING AND THE CAPM 133

A more modern …gure, from John Cochrane, Financial markets and the
Real Economy, mimeo (2007). Fitted returns are the returns implied by the
Consumption CAPM model, that is, implied by the betas. If the model is
correct the returns should be on the 45o line.
134 CHAPTER 8 FINANCE

The 25 data points are each an asset portfolio which is known not to work
very well with the theory (these are called Fama-French 25 portfolios).

8.1.4 The Term Structure of Interest Rates

The risk free rate is usually directly available in the market; it is the rate
of return of a 1 period T-bill (as risk-free as possible a bond). Let’s de…ne
a period to be one year, for simplicity. Risk free rates for many di¤erent
maturities are traded in …nancial markets, a quarter, a year, 5 years, 10 and
20 years. Typically, a longer maturity is compensated by a higher yearly
return (all our returns are yearly, for comparison). This smell of risk! But is
there risk in a 2-year risk free bond? This seems meaningless, at …rst reading
(if it’s risk free it’s risk free), but it is not. The risk is implicit in the fact
that there are two strategies to transfer income from time t to time t + 2:
f
1. Buying a 2 year risk free bond, which pays Rt;t+2 with certainty at time
t + 2 for a price of 1 at t:
f
2. Buying a 1 year risk free bond, which pays Rt;t+1 with certainty at time
f
t + 1 for a price of 1 at t; then, at time t + 1 investing Rt;t+1 in a 1
f f
year risk free bond, which pays Rt;t+1 Rt+1;t+2 with certainty at time
t + 2:
8.1 ASSET PRICING AND THE CAPM 135

Note that the risk free rate of a 1 year bond at t + 1 will be known at
t + 1, but it is not known at time t. To convince yourself of this, write the
f
asset pricing equation for Rt+1;t+2 :

f
1 u0 (ct+2 )
Rt+1;t+2 =E ;
u0 (ct+1 )

and note that at time t, ct+1 is not known. Hence the uncertainty. (Why is
strategy 1 also not riskless?)
We can then compute the value of a 2 year risk free bond:
0
2u (ct+2 )
1
f
Rt;t+2 =E : (8.8)
u0 (ct )
From (??) and (8.8) one gets an expression for the term structure of
interest rates:

f
1
f
1
f
1 u0 (ct+1 ) f
1
Rt;t+2 = Rt;t+1 E Rt+1;t+2 + cov ; Rt+1;t+2 :
u0 (ct )
(8.9)
1
u0 (c t+1 ) f
If cov u0 (ct )
; Rt+1;t+2 = 0 it is either because there is no risk in
u0 (ct+1 ) f
the economy, u0 (ct )
is constant, or because Rt+1;t+2 is known at time t:
f f
Rt+1;t+2 =E Rt+1;t+2 : In this case then we have

f f f
Rt;t+2 = Rt;t+1 Rt+1;t+2

That is, the yearly interest rates are the same at di¤erent maturities and
strategies 1 and 2 are equivalent. Otherwise, a risk correction appears.

8.1.5 The Market Rate of Return and the betas

If we knew u(c) and we could measure precisely ct+1 , ct (aggregate consump-
tion), we would have no problem; and we could compute

u0 (ct+1 )
Rtm =
u0 (ct )
136 CHAPTER 8 FINANCE

and 0 1
u0 (ct+1 ) j
covt u0 (ct )
; Rt+1
j
=@ A
u0 (ct+1
vart u0 (ct )

In fact we do not know what u(c) and aggregate consumption data is

quite bad. Therefore the standard procedure is to estimate a multi-factor
beta equation of the form:

Rtj = Rtf + j 1
f 1 ft + j 2
f 2 ft + ::: + j
t (8.10)

for any asset j; where f i are factors, i.e., proxies for the intertemporal mar-
ginal rate of substitution, like several indices of stock returns, GNP, in‡ation,
and so on. (Note that only unconditional betas can be estimated; hence un-
der the assumption that betas are constant over time).

8.1.6 Equivalent Risk Neutral Representation

Consider the random variable xjt+1 ; E xjt+1 is its expected value with respect
to our basic probability distribution. Let mt+1 be a non-negative random
variable, and let E xjt+1 = E Em t+1
xj
t mt+1 t+1
. Notice that we have implicitly
de…ned a di¤erent probability distribution with respect to the basic one.3
f 1 u0 (ct+1
Recalling that Rt+1 = u0 (ct+1
, and letting mt+1 = u0 (ct )
we now
E u0 (ct )

have:
E xjt+1
qtj = f
(8.11)
Rt+1
Notice that (8.11) is the pricing equation of an economy with risk neutral
agents with the di¤erent probability distribution over uncertainty we con-
structed; we call therefore this probability distribution "risk neutral". The
risk neutral probability distribution corrects for risk. In fact, with respect
to the basic probability distribution, the risk neutral probability distribu-
tion weights more the states with lower consumption. This is to say: a risk
averse agent is like a risk neutral agent who believes that the "bad" states of
uncertainty are more probable.
3
That is, Et (x) 0 if x 0 and Et (x) = x if x is constant.
8.2 EFFICIENT MARKET HYPOTHESIS 137

8.2 E¢ cient Market Hypothesis

Eugene Fama

The analysis of asset pricing and valuation in the previous section is often
referred to as the E¢ cient Market Hypothesis. It has several implications
that have been tested over the years. For a critical review, see A. Schleifer,
Ine¢ cient Markets, Oxford University Press, 2000.

8.2.1 Idiosyncratic Risk Does Not A¤ect Prices

An asset j is one of idiosyncratic risk if its payo¤ is independent of the econ-
omy’s aggregate consumption. You can check0 immediately that idiosyncratic
risk is not priced as its covariance with uu(c t+1
0 (c )
t
is zero (see equation 8.4).
Note that this is true no matter how large is var(xjt+1 ). What is idiosyn-
cratic risk? Typically most accident risks as well as some health risk are
idiosyncratic, that is, independent across agents. (But, Is hurricane risk
idiosyncratic?) Idiosyncratic risk is not priced if the cost of an insurance
contract on it is the expected present discounted (at the risk-free interest
rate) of its net payments.

8.2.2 Prices Adjust Immediately to All Available In-

formation
In the fundamental pricing equation (8.2), the information included in the
price at time t is the whole information available at time t, on which the
138 CHAPTER 8 FINANCE

expectation operator E(:) is conditional. The E¢ cient Market Hypothesis

does not allow for di¤erential information across agents in the market. This
is a major (and somewhat unrealistic) assumption.

8.2.3 Risk Adjusted Prices Are Not Predictable

Suppose agents are risk neutral (u(c) is linear) and do not discount the future
( close to 1). Consider a stock j which pays no dividends, xjt+1 = qt+1
j
; then,
using (8.2):
qtj = Eqt+1
j
(8.12)
and its price is a random walk (or, more precisely, a martingale). This is the
typical example of a non-predictable stochastic process. It relies on extreme
implausible assumptions.
In general prices are predictable, as are returns and excess returns. For!
u0 (ct+1
var u0 (ct )
instance, from (8.7), one sees that even if the price of risk u0 (ct+1
E u0 (ct )

is constant over time, the predictability of excess returns follows from the
predictability of the conditional beta. So it is, for instance, that stocks pay
a higher return than bonds, consistently over time, as stocks are more risky
than bonds and hence pay a higher excess return.
Let risk adjusted prices be de…ned as prices minus the risk premium.
The fact that risk adjusted stock prices are unpredictable is an immediate
consequence of the asset pricing equation for stocks, 8.3, with no dididends,
when written as:
u0 (ct )qtj = E u0 (ct+1 )qt+1
j
:
Another way to say the same thing is the following: the price of a stock
which pays no dividends is always a martingale in the risk neutral probability
measure (which adjusts for risk), see equation (8.11), but not for the basic
probability measure (which drives the uncertainty in the economy).

8.2.4 Arbitrage Opportunities Are Not Present in Fi-

nancial Markets
An arbitrage opportunity exists in …nancial markets when a portfolio can be
traded whose price is either 0 or negative and its payo¤ is always non-negative
and strictly positive with positive probability. At equilibrium, however, asset
8.2 EFFICIENT MARKET HYPOTHESIS 139

prices must be such that arbitrage opportunities cannot exist. Otherwise,

any agent, exploiting the opportunities would have in…nite wealth and would
demand an in…nite consumption allocation in some states, 0
which contradicts
market clearing. In fact, the discount factor, uu(c t+1
0 (c )
t
is strictly positive
0
(utility is strictly monotonic, and hence u (c) > 0). Any asset with a payo¤
which is non-negative and strictly positive with some probability has then a
strictly positive price, by (8.2).
To better understand the implications of No-arbitrage as a property of
equilibrium, consider an economy with two states of nature fs1 ; s2 g and two
primary assets, f1; 2g. The …rst asset is a stock with price q1 and returns
(x1s1 ; x1s2 ); x1s1 6= x1s2 . There is also a bond with price q2 and returns (x2 ; x2 ):
Taking this data as parameters, consider the problem of pricing a derivative,
asset d, with returns (xds1 ; xds2 ): Denote the price of the derivative asset as qd .
Construct a portfolio with the two primary assets which replicates the payo¤
vector of the new asset. This requires solving the following system

0 1 0 1 0 1
x2s1 x1s1 x2
@ A= 1
@ A+ 2
@ A:
x2s2 x1s2 x 2

Note that the solution to this system exists and is unique.

Now, No-arbitrage implies that the replicating portfolio and the derivative
asset must have the same price since they have the same returns vector:

qd = 1 q1 + 2 q2 :

This is called the Law of one price. Prove that the Law of one price is a
consequence of No-arbitrage, that is,
if either qd > 1 q1 + 2 q2 or qd < 1 q1 + 2 q2 an agent could construct
a traded portfolio whose price is negative and its payo¤ strictly positive with
probability 1.
Pricing derivatives by constructing a replicating portfolio and then using
the Law of one price is one of the main activities of Wall Street. Though
the derivatives are typically quite complex, conceptually this activity is a
straightforward as exposed in this notes.
140 CHAPTER 8 FINANCE

8.3 Modigliani-Miller Theorem

Franco Modigliani Merton Miller

Consider an economy for which the E¢ cient Market Hypothesis holds

and in which all …nancial streams are possibly traded (abusing words we call
this Complete Markets assumption). Then

MM1 The value of the …rm is independent of its capital structure, that is, of
the proportion of debt and equity used to …nance the …rm’s operations.

MM2 The value of the …rm is independent of the …rm’s dividend policy.

To understand the irrelevance propositions, MM1 and MM2, and the

assumptions required for them to hold, it is useful to pass to more formal
statements.

8.3.1 Modigliani-Miller More Formally

Let’s consider MM1 …rst. Consider a …rm whose cash ‡ow at t + 1 is a
random variable xjt+1 . The cash ‡ow is composed of the dividend plus the
re-sale price,
xjt+1 = djt+1 + qt+1
j

and let djt+1 d > 0.

Suppose the …rm …nances its operations by issuing equity only. The E¢ -
cient Market Hypothesis implies that any cash ‡ow x is priced by our basic
valuation equation, (8.2).
8.3 MODIGLIANI-MILLER THEOREM 141

Then the …rm’s value is

u0 (ct+1 ) j
qtj = E x :
u0 (ct+1 ) t+1
Suppose instead the …rm issues units of debt (assume, but only for simplic-
ity, that < d, so that debt is risk free) and the residual is equity. What is
the value of the …rm ? It is the value of the debt and equity issued. The cash
‡ow of the debt at t + 1 is ; while the cash ‡ow of the equity is xjt+1 .
The value of the …rm in this case is therefore:
u0 (ct+1 ) u0 (ct+1 ) j
E + xjt+1 =E x = qtj :
u0 (ct+1 ) u0 (ct+1 ) t+1
In other words, we have proved MM1. Let’s try with an intuition. Consider
the change in the capital structure of the …rm, from all equity to equity and
bonds. In equilibrium, the agents must hold the new debt and the equity,
after the change in capital structure. Will they need to be compensated for
this change? No, because the agents will not have to change their portfolios
in equilibrium: they can reproduce whatever it was their portfolio with no
change; as long as the aggregate supply of the cash ‡ows is unchanged and
all portfolios can be reproduced (this is the assumption that all …nancial
streams are possibly traded in our economy). The supply of bonds in the
economy is increased, as the …rm is issuing extra units. But the demand of
bonds is increased exactly to compensate the change in the supply, as those
agents holding equity before the change will have to buy the bonds after
the change to keep their portfolio unchanged.
The original proof of the result by Modigliani-Miller in American Eco-
nomic Review 1958 was a No-arbitrage proof. It helps develop intuition for
this result. Consider two …rms, identical in terms of cash ‡ow, but …nanced
di¤erently: one with only equity, the other with units of a bond security
and the residual as equity. Suppose they have a di¤erent price. How could
an agent trading in bonds and in the original equity security (with payo¤
xjt+1 ) make a positive arbitrage ?
Consider the arguments we just made to prove MM1. Do they depend
on the assumption that djt+1 > d > , so that the debt issued by the …rm is
risk free ? Convince yourself that the answer is no.
Let’s consider MM2 now.
Consider now a change in the …rm’s dividend policy. For instance, suppose
that the …rm’s dividends at time t + 1 go from djt+1 to djt+1 + et+1 , for some
142 CHAPTER 8 FINANCE

random variable et+1 . How does the …rm …nances its new dividend policy
without changing its investment plan ? The …rm has to add a security to its
portfolio which pays et+1 at time t+1. Such a security exists by the complete
markets’ assumption. The price at time t of a security which pays et+1 at
time t + 1 is
u0 (ct+1 )
qte = E et+1 (8.13)
u0 (ct+1 )
What is then the value of the …rm with the new dividend policy ? It is
the value of the …rm’s payo¤ at time t + 1, appropriately discounted, minus
the cost of the security that the …rm is buying at time t:

u0 (ct+1 ) j
E d + et+1 qte
u0 (ct+1 ) t+1
Combining this equation with (8.13), we have that
u0 (ct+1 ) j
qtj = E x
u0 (ct+1 ) t+1
and we have proved MM2.
Consider the cases in which the …rm …nances the new issue of dividends
with cash that it holds at time t or with cash generated by its project at time
t + 1: Show that the same MM2 result applies in this case.

8.3.2 Other Implicit Assumptions

Several implicit assumptions of practical (but not conceptual) relevance re-
quired for MM1 and MM2:

No transaction costs.
No di¤erential taxation of debt and equity.4
No costs of …nancial distress (in particular, no bankruptcy costs).
Managers maximize shareholders’wealth, that is, the value of the …rm.
4
In fact, in the U.S., interest rates are not taxed at the corporate level, but dividends
and capital gains are. This gives an advantage to debt as a form of …nancing, contrary to
the MM results. On the other hand, the personal tax on equity is higher than the personal
tex on debt. This gives an advantage to equity …nancing. The advantages roughly cancel
out.
8.4 PROBLEMS 143

8.4 Problems
Problem 8.4.1 Consider an economy with two …nancial assets: A bond,
which pays a riskless return R; and a stock, which pays a return S, a random
variable taking values S1 with probability p and S2 with probability 1 p.
Consider an agent with an arbitrary sum of money, w, choosing the share
of bond and (1 ) of stock in his portfolio to maximize the expected util-
ity of his consumption after returns are payed. i) Write down the agent’s
maximization problem, by choice of : ii)Write down the …rst order condi-
tion for the maximization problem. (Careful to corners). iii) Examine the
condition (being careful about corners once again). Under which conditions
on the parameters R; S1 ; S2 is = 1? Under which conditions is it < 1?

Problem 8.4.2 Consider an economy with 2 states of the world s1 = H;

s22 =3 L: 3 assets are traded in this economy: asset 1 has payo¤ x1 =
1
4 5(that is, payo¤ equal to 1 in state H and payo¤ equal to 0 in state
0
2 3
1
L); asset 2 is a riskfree bond and has payo¤ x2 = 4 5; asset 3 has payo¤
1
2 3
1
x3 = 4 5. The prices of the 3 assets are q1 ; q2 ; q3 : Suppose now you want
3
2 3
2
to price by no-arbitrage a derivative asset whose payo¤ is xd = 4 5 : Do
7
this as a function of prices q1 ; q2 : Do this again as a function of prices q1 ; q3 :
Does no arbitrage impose any restriction on prices q2 ; and q3 ?

Problem 8.4.3 Consider an agent with the following preferences:

1
u(c) = c1 ; >0
1

In an economy with 2 time periods (and utility as above for each period and
discount rate = :9) and 4 states of the world with associated probabilities
1; 2; 3; 1 1 2 3 ; what is the expression of the price of an asset
144 CHAPTER 8 FINANCE
2 3
1
6 7
6 7
607
whose payo¤ is x = 6
6 7?
7
627
4 5
4

Problem 8.4.4 Consider an agent living 3 periods, with logarithmic prefer-

ences, and with endowments wt in period t = 1; 2; 3. He/she does not discount
the future and faces an interest rate r1 > 0 in period 1 and an interest rate
r2 = 0 in period 2. What is the e¤ect of a change (e.g., a marginal increase)
in the interest rate in period 1 on the agent’s consumption in period 2 and
in period 3.

Problem 8.4.5 Consider an agent with the following preferences:

1
u(c) = c1 ; >0
1
How does risk aversion depend on ? (De…ne …rst risk aversion, there are
several de…nitions, make up one that makes sense to you.) Can be negative?

Problem 8.4.6 Consider a risk averse agent. He faces a health risk of D

(he/she has to go to the hospital and pay D, and he/she’ll be …ne) with
probability . He/she can buy insurance at price q (that is he/she can buy at
cost q a contract that pays 1 dollar if he/she has to go to the Hospital). How
many units of the contract will the agent buy is the price is q = ?

Problem 8.4.7 On tradesports.com a contract is traded which pays 100 dol-

lars if Bush is elected president in the next presidential election. The cost
of the contract is 56 dollars. Does this tell you something about the expected
value of Bush winning? (Argue very carefully.) And about he variance?
What is the covariance of Bush winning and Kerry winning? (Does Nader
has something to do with your last answer?)

Answer. First, consider the case where the all the people buying and
selling the contract are risk neutral. This means that the only thing they
are interested is the expected value of the contract. If the expected value of
buying the contract is greater than 0, the demand for the contract will be
+1. For each contract that an agent buys, their expected wealth goes up.
8.4 PROBLEMS 145

As they are risk neutral, they want to maximize their expected wealth, so
however many contracts an agent owns, she will always want more! Similarly,
if the expected value of the contract is less than zero, the supply of the
contract will be +1. For every contact an agent sells (remember, and agent
can have a negative number of contracts - this just means that they are acting
as the bookmaker - they accept $56, and pay out $100 if Bush wins), their
expected wealth goes up. Again, however many contracts an agent has sold,
they will always want to sell one more. It follows that, for the market to
be in equilibrium (i.e., for supply to equal demand), the expected value of
the contract must equal zero. At this point, agents are indi¤erent between
buying and not buying the contract
Let p be the probability of Bush winning. The expected value of the
contract is therefore:

p(100 56) + (1 p)( 56) = 0

56
) p= = :56
100
What if the agents are not risk neutral? We would still argue that for the
market to be in equilibrium, the agent would have to be indi¤erent between
buying and not buying the contract. But if the agent is risk averse, then the
expected value of a bet has to be greater than zero to make the expected
utility of the bet zero (remember, risk averse people don’t like gambling - to
make them take a gamble the odds have to be better than fair). We therefore
know that

p(100 56) + (1 p)( 56) > 0

56
) p> = :56
100
However, without a speci…c utility function, we cannot say more than
this.
Now, to answer questions about the expected value, variance and covari-
ance of Bush winning, we need to de…ne a random variable, X, which is equal
to 1 if bush wins and zero otherwise:
State of the World Probability Value of X
Bush wins P 1
Bush Does not win 1 p 0
146 CHAPTER 8 FINANCE

Again let p be the probability of bush winning. Employing the formula

for expected value, we see that

E(X) = p1 + (1 p)0 = p

Or, if we are in the risk neutral world, E(X) = 0:56

Employing the formula for variance, we see that

V ar(X) = E[(x E(X))2 ]

= p(1 p)2 + (1 p)( p)2
= p(1 p)(1 p + p)
= p(1 p)

Or, in the risk neutral world, V ar(X) = 0:2464

To talk about the covariance between Bush winning and Kerry winning,
we need to de…ne another random variable, Y , equal to 1 if Kerry wins
and zero otherwise. First consider the case where there is no Nader, or the
probability of Nader winning is zero. In this case, there are still only two
states of the world, one in which Bush wins and Kerry does not, and another
in which Kerry wins and Bush does not:
State of the World Probability Value of X Value of Y
Bush wins (Kerry does not) p 1 0
Bush Does not win (Kerry does) 1 p 0 1
Employing the formula for covariance, and noting that E(Y ) = 1 p, we
see that

Cov(X; Y ) = E[(X E(X))(Y E(Y ))]

= p(1 p)( (1 p)) + (1 p)( p)(1 (1 p))
= p(1 p)2 p2 (1 p)
= p(1 p)

In the risk neutral case, cov(X; Y ) = :2464

Now consider the case where there is a Nader. Here, we have to add a
third state of the world, in which Nader wins. We also need to introduce a
new probability, q of Nader winning
8.4 PROBLEMS 147

State of the World Probability Value of X Value of Y

Bush wins p 1 0
Kerry wins 1 p q 0 1
Nader wins q 0 0
Again, employing the covariance formula, and noting that E(Y ) = 1 p q
,

cov(X; Y ) = E[(X E(X))(Y E(Y ))]

= p(1 p)( (1 p q) + (1 p q)( p)(1 (1 p q) + q( p)(1 p q)

Note that even in the risk neutral case, we cannot solve explicitly for
cov(x; y) in this case, because we do not know q.

Problem 8.4.8 Consider an agent living 3 periods, with logarithmic prefer-

ences, and with endowments wt in period t = 1; 2; 3. He/she does not discount
the future and faces an interest rate r1 > 0 in period 1 and an interest rate
r2 = 0 in period 2. What is the e¤ect of a change (e.g., a marginal increase)
in the interest rate in period 1 on the agent’s consumption in period 2 and
in period 3.

Answer. This is just a standard intertemporal optimization problem.

In general, the utility function that the agent is trying to maximize would
be of the form
2
U (c1 ; c2 ; c3 ) = u(c1 ) + u(c2 ) + u(c3 )

The question tells us that the utility function in this case is

U (c1 ; c2 ; c3 ) = ln c1 + ln c2 + ln c3

We can construct the budget constraint in the usual manner. De…ne

savings between period one and two and between two and three respectively
as

s1 = w1 c1
s2 = w2 + s1 (1 + r1 ) c2
148 CHAPTER 8 FINANCE

Note that savings at the end of period 3 will be zero, as the agent ’dies’
at this point, so:
0 = w3 + s2 (1 + r2 ) c2

By using the last equation to get an expression for s2 , substituting this

into the second equation, getting an expression for s1 and substituting this
into the …rst equation, we get the present value budget constraint:

c2 c3 w2 w3
c1 + + = w1 + +
(1 + r1 ) (1 + r1 )(1 + r2 ) (1 + r1 ) (1 + r1 )(1 + r2 )

Though for this problem we are told that r2 is equal to 0.

The problem therefore becomes

maximize U (c1 ; c2 ; c3 ) = ln c1 + ln c2 + ln c3
c2 c3 w2 w3
subject to: c1 + + = w1 + +
(1 + r1 ) (1 + r1 ) (1 + r1 ) (1 + r1 )

Which gives the Lagrangian:

c2 c3 w2
L(c1 ; c2 ; c3 ; ) = ln c1 + ln c2 + ln c3 (c1 + + w1
(1 + r1 ) (1 + r1 ) (1 + r1 )
w3
)
(1 + r1 )

Which gives …rst order conditions:

@L 1
= =0
@c1 c1
@L 1
= =0
@c2 c2 (1 + r1 )
@L 1
= =0
@c3 c3 (1 + r1 )

The last two expressions combine to give c2 = c3 , while the …rst and
second combine to give
c2 = (1 + r1 )c1
8.4 PROBLEMS 149

substituting this into the budget constraint gives

c2 c3
c1 + + =
(1 + r1 ) (1 + r1 )
(1 + r1 )c1 (1 + r1 )c1
c1 + + =
(1 + r1 ) (1 + r1 )
w2 w3
c1 + c1 + c1 = 3c1 = w1 + +
(1 + r1 ) (1 + r1 )
1 w2 w3
c1 = (w1 + + )
3 (1 + r1 ) (1 + r1 )

And using the expression we have for c2 (and therefore c3 ) as a function

of c1 gives

c2 = (1 + r1 )c1
1 w2 w3
= (1 + r1 ) (w1 + + )
3 (1 + r1 ) (1 + r1 )
1
= ((1 + r1 )w1 + w2 + w3 )
3
= c3

The marginal change of consumption of period two and three with respect
to the interest rate in period one is therefore
@c2 @c3 w1
= = :
@r1 @r1 3
Problem 8.4.9 Consider an agent with the following preferences:
1
u(c) = c1
1
How does risk aversion depend on ? (De…ne …rst risk aversion, there are
several de…nitions, make up one that makes sense to you.) Can be negative?

Answer. There are many possible ways of measuring risk aversion, some
of which we will talk about in class. One popular one is the coe¢ cient of
relative risk aversion, de…ned as
cu00 (c)
r=
u0 (c)
150 CHAPTER 8 FINANCE

which is positive for risk averse people, zero for risk neutral people and
negative for risk lovers. Plugging into this formula for the utility function
above gives:
r=
or in other words, the coe¢ cient of relative risk aversion is constant. can
be negative. This just implies that the agent is a risk lover.

Problem 8.4.10 Consider a risk averse agent. He faces a risk of incurring

a cost D1 with probability :3 and of D2 with probability :4. He/she can buy
insurance at price q, that is he/she can buy at cost q a contract that pays 1
dollar if he/she either cost D1 or D2 are realized. How many units of the
contract will the agent buy is the price is q = 3:

Answer. Think about an agent who buys x units of insurance. If they

get ill, the amount of money that they will have is W D qx + x, where
W is initial wealth. If they do not get ill, their wealth will be W qx. The
expected utility of such an agent is:

U = u(W D qx + x) + (1 )u(W qx)

where u(:) is some arbitrary decreasing marginal returns utility function. The
agent wishes to choose x to maximize their expected utility. We therefore
take the derivative of the above function with respect to x and set it equal
to zero.
@U
= (1 q)u0 (W D qx + x) (1 )qu0 (W qx) = 0
@x
) (1 q)u0 (W D qx + x) = (1 )qu0 (W qx)

The question tells us that q = , so we have:

(1 )u0 (W D x + x) = (1 ) u0 (W x)
u0 (W D x + x) = u0 (W x)

We would like to conclude from this that W D x+x = W x, and

we will do. But note that we can only do that under some assumptions. We
are assuming that if the slope of the function is equal, then the argument of
the function is also equal. This is only true if the slope of the function is not
the same for any two points. One way to ensure this is to assume that the
second derivative of the function is always less than 0.
8.4 PROBLEMS 151

Bearing that in mind, we get:

W D x+x = W x
D+x = 0
D = x

This makes sense. In this case, the expected value of buying insurance is
zero, as the cost of one unit of insurance is q = , and the expected return on
the insurance is 1 +0(1 ) = . Buying insurance therefore doesn’t reduce
ones expected wealth, but does reduce risk. A risk averse agent will therefore
buy enough insurance to completely cover their risk. This is what happens
in the case above: The agent keeps buying insurance until her income is the
same in either state of the world.

Problem 8.4.11 Consider an economy under uncertainty: there are three

states of the world, s1 ; s2 ; s3 , with probability p; q; 1 p q; respectively. All
agents are identical and live for 2 periods, t, t + 1. Agents in this economy at
time t can trade at price Q an asset which pays A1 ; A2 ; A3 in the three states
s1 ; s2 ; s3 at time t + 1: i) What is the expected value of the payo¤ of the asset
at time t + 1 ? ii) Consider the case of risk neutral agents - more speci…cally,
agents with utility

u(xt ) + u(xt+1 ) = C + xt + (C + xt+1 )

for some constant C: What is the price of the asset at time t ? (What I
mean here is: write down the equation of Q as a function of all exogenous
variables in the economy - p; q; A1 ; A2 ; A3 ; C; ; ): iii) Consider the case of
risk averse agents - more speci…cally, agents with utility

u(xt ) + u(xt+1 ) = log xt + log xt+1

Assume also that xt = 1; and xt+1 is x1 ; x2 ; x3 in the three states s1 ; s2 ; s3 ;

with x1 = 2; x2 = 3; x3 = 1: What is the price of the asset at time t ? (Once
again, what I mean here is: write down the equation of Q as a function of
all exogenous variables in the economy - p; q; A1 ; A2 ; A3 ; ):

Problem 8.4.12 Consider an economy with 2 states of the world, either

"a tax cut is voted" or "a tax cut is not voted." The tax cut is voted with
probability :3. Consider a …nancial market with 2 basic assets. The …rst pays
152 CHAPTER 8 FINANCE

2 dollars if the tax cut is not voted and 3 if it is. The second asset pays
nothing if the tax cut is not voted and 1 dollar if it is. The prices of the
two assets are, respectively 3 and 1 dollars. There is also another asset, a
risk free bond that pays 10 dollars no matter if the tax cut is voted or not.
This last asset sells for 9 dollars. i) Is there an arbitrage opportunity in this
economy? ii) Does your answer depend on the probability of the tax cut?
Explain why. iii) Just for the fun of it, compute the variance of the two basic
assets, and their covariance.

Problem 8.4.13 Consider an economy with two states of nature fs1 ; s2 g

and with two primary assets, f1; 2g. The …rst asset is a stock with price
q1 and returns (x1s1 ; x1s2 ); x1s1 6= x1s2 . There is also a bond with price q2 and
returns (x2 ; x2 ):Which general form do prices q1 and q2 have for the …nancial
market to display no arbitrage opportunities ? Taking this data as para-
meters, consider the problem of pricing a derivative, asset d, with returns
(xds1 ; xds2 ): Denote the price of the derivative asset as qd . Construct a formula
for this price as a function of the parameters of the economy.

Answer. No Arbitrage implies

q1 = m1 x1s1 + m2 x1s2
q2 = m1 x2s1 + m2 x2s2

with m1 and m2 both > 0.

First solve for 1 and 2 the following linear equation system:
0 1 0 1 0 1
xds1 x1s1 x2s1
@ A= 1 @ A+ 2 @ A
d 1 2
x s2 x s2 x s2

then
qd = 1 q1 + 2 q2

Can you always (for all values of the parameters) do this ? Only if there
exists a solution for ( 1 ; 2 ), that is if the underlying assets have independent
payo¤s and hence (markets are complete and) the derivative can be generated
by trading the underlying assets.

Problem 8.4.14 Idiosyncratic risk does not a¤ect prices. Can you give an
explanation in terms of no arbitrage ? In other words, if the price of a stock
8.4 PROBLEMS 153

of a …rm had a component representing the …rm’s idiosyncratic cash ‡ow

risk, which arbitrage opportunity would be left open in the …nancial markets
?

Answer. Suppose the price of the idiosyncratic risk is negative; that is a

…rm with extra idiosyncratic risk is sold at a discount with respect to another
…rm with identical systematic risk component but with no idiosyncratic risk.
Then, buy the …rm with idiosyncratic risk, diversify this risk away by trading
an in…nite number of other …rms (assets) with identical payo¤ but indepen-
dent idiosyncratic risk (they exist by complete markets; otherwise the risk
would not be idiosyncratic, as explained in class; by another no arbitrage ar-
gument these …rms must all trade at the same price as our original …rm and
hence the diversi…cation strategy has cost zero). This portfolio has generated
the same payo¤ as the …rm with no idiosyncratic risk. Selling now the more
expensive …rm without idiosyncratic risk gives you the arbitrage.

Problem 8.4.15 Derive the expression for the two-period ahead risk free
rate in terms of marginal rates of substitution. (This is also known as the
term structure of interest rates). You might want to know the law of iterated
expectations: Et (Et+1 (xt+2 )) = Et (xt+2 ).

Answer. From the fundamental equation of asset prices:

f 1 u0 (ct+2 ) u0 (ct+1 )
(Rt+2 ) = Et
u0 (ct ) u0 (ct+1 )

which, using the Law of Iterated Expectations, becomes:

u0 (ct+1 ) u0 (ct+2 )
= Et Et+1
u0 (ct ) u0 (ct+1 )

In turn, using the fundamental equation of asset prices on the one period
f
ahead risk free rate at time t + 1; Rt+1;t+2 :

u0 (ct+1 ) f 1
= Et (Rt+1;t+2 )
u0 (ct )

Finally, using the decomposition

E(xy) = ExEy + cov(x; y)

154 CHAPTER 8 FINANCE

one gets the term structure equation:

f 1 u0 (ct+1 ) f
1 u0 (ct+1 ) f 1
(Rt+2 ) = Et Et Rt+1;t+2 + covt ; (Rt+1;t+2 )
u0 (ct ) 0
u (ct )

Under which conditions would you expect the term structure to slope up;
3 2
f f f
that is: Rt+3 > Rt+2 > Rt+1 ? (Explain the argument and discuss
its plausibility.)
f
The simplest way is to assume that Et (Rt+1;t+2 ) 1 is roughly equal to
f
(Rt+1 ) 1 . Then the sign of the covariance term drives the answer. In the
phase of the cycle in which with high probability consumption is increasing
at t + 2 if it has grown at t + 1, the covariance is positive and the term
structure slopes down. Note though there is no correct answer here; only
reasoned and unreasoned ones.

Problem 8.4.16 Imagine that there are 3 states of the world: A, B and C
which occur with probability p, q and (1 p q) respectively. What is the
expected value of the following …nancial instruments? What is the variance
of the expected payo¤ of these …nancial instruments? What is the covariance
between the …rst …nancial instrument and each of the others? If the agent
was risk neutral, how much would they pay for each of these …nancial instru-
ments? i) A share that pays o¤ 1 unit in state A: ii) A bond which pays o¤ 1
unit in states A; B and C: iii) A share that pays o¤ 1 unit in state A and C:
iv) Insurance that pays o¤ 1 unit in state B and -1 units in any other state.
1 1
v) A share that pays o¤ 3p in state A, 3q in state B and 3(1 1p q) in state C.
For each of the above assets, write down the expected utility of holding such
an asset. If the agent has a logarithmic utility function, what is the expected
utity of each asset?

Problem 8.4.17 Prove (analytically, i.e., without graphs), that an agent

with a concave utility function must be risk averse.

Problem 8.4.18 Consider an economy with two states of the world and two
agents, A and B. In state 1, agent A gets 10 units of good and agent B gets
0 units. In state 2, A gets 0 units and B gets 15 unit·s. Say that there exists
a …nancial instruments which each agent can buy or sell which promises to
pay the bearer 1 unit in state 1 at the cost of p units in state 2: Each agent
8.4 PROBLEMS 155

1
has a utility function u(c) = c1 : i) Work how many units of the …nancial
instrument each agent will demand as a function of p. (remember this can
be positive or negative as the agents are allowed to buy and sell the asset).
ii) Solve for the equilibrium value of p. (note that feasability states that the
total demand for the …nancial asset must be zero, as if one agent buys a unit
of the assent, the other must sell it to him). iii) Work out the allocations of
each agent in equilibrium. iv) Repeat steps 1-3, but this time assume that the
utility function of agent B is u(c) = c:

Problem 8.4.19 Consider an agent with the following preferences:

1
u(c) = c1 ; >0
1
How does risk aversion depend on ? (De…ne …rst risk aversion, there are
several de…nitions, make up one that makes sense to you.) Can be negative?

Problem 8.4.20 Consider an economy with 2 time periods, t and t + 1; a

representative agent with utility
1 1
(ct )1 + :9 E (ct+1 )1 ; >0
1 1
and 3 states of the world with associated probabilities 1; 2; 1 1 2: What
1
is the expression of the price of an asset whose payo¤ is x = 0 ? (Hint: the
4
expression is in terms of ct ; ct+1 ; and parameters 1; 2; ):

Problem 8.4.21 Imagine that there are 3 states of the world: A, B and C
which occur with probability p, q and (1 p q) respectively. What is the
expected value of the following …nancial instruments? What is the variance
of the expected payo¤ of these …nancial instruments? What is the covariance
between the …rst …nancial instrument and each of the others? If the agent
was risk neutral, how much would he/she pay for each of these …nancial
instruments? i) A share that pays o¤ 1 unit in state A: ii) A bond which
pays o¤ 1 unit in states A; B and C: iii) A share that pays o¤ 1 unit in state
A and C: iv) Insurance that pays o¤ 1 unit in state B and -1 units in any
156 CHAPTER 8 FINANCE

1 1
other state. v) A share that pays o¤ 3p in state A, 3q in state B and 3(1 1p q)
in state C. For each of the above assets, write down the expected utility of
holding such an asset. If the agent has a logarithmic utility function, what is
the expected utity of each asset.

Problem 8.4.22 Consider a risk averse agent with income W . He faces

a risk of incurring a cost D1 with probability :3 and of D2 with probability
:4. He/she can buy insurance at price q, that is he/she can buy at cost q
a contract that pays 1 dollar if he/she either cost D1 or D2 are realized?
(Consider D1 > D2 ). Will the agent be able to perfectly insure? (Need to
de…ne perfect insurance; take a stab at this). How many units of the contract
will the agent buy is the price is q = :7? And will the agent be able to perfectly
insure in this case (according to your de…nition of perfect insurance)?
Chapter 9

Incentives

George Akerlof Joseph Stiglitz Michael Spence

Economists are all about incentives. Facing any ine¢ ciency, economists
look for misaligned incentives as culprits.
The …nancial markets crisis? Look at incentive compensation of managers
and at Fannie and Freddie’s political incentives.
The poor educational system? Look at the incentives implicit in the
contract of unionized teachers.
The excessive cost of health insurance? Look at how doctors are com-
pensated for services and at the policitcal incentives which govern Medicare
and Medicaid (and at the market power of insurance companies, but this
does not have to do with incentives as much as with competition - another
obsession of economists).

157
158 CHAPTER 9 INCENTIVES

To discuss incentives properly, we need to introduce private information

(or, sinonimously, asymmetric information)

9.1 Principal-Agent Problems

Consider an economy with s = 1; : : : ; S states of uncertainty, and a project
(e.g., but not necessarily, a …rm) with cash ‡ow Rs > 0 in state s. Without
loss of generality rank Rs so that R1 < R2 < :::RS :
We consider two distinct (but related) cases.
Making the agent work. A principal owns the project ex-ante, but an
agent necessarily manages it, by spending e¤ort a. By increasing the
e¤ort he/she employs into managing the project, e¤ort a, the agent
increases the probability that the the project is successful, that is that
it has a high cash ‡ow Rs . Let ps (a) denote the probability that state s
occurs (and hence that the …rm’s cash ‡ow isP Rs ) when the agent’s e¤ort
is a. We assume that the expected cash ‡ow Ss=1 ps (a)Rs is increasing
in a.1 The agent’s utility function is increasing in consumption c and
is decreasing in e¤ort a:
u(c) (a)
The principal’s utility function is increasing in consumption:
v(c):

Paying the agent according to his productivity. A principal owns the

project ex-ante, but an agent necessarily manages it. The principal
hires a random agent, but agents are di¤erent in their abilities. Better
agents have a higher probability that the the project they manage is
successful, that is that it has a high cash ‡ow Rs . Let ps (q) denote the
probability that state s occurs (and hence that the …rm’s cash ‡ow is
Rs ) when the agent’s quality
PS is q = 1; :::; Q. We order quality q so that
2
the expected cash ‡ow s=1 ps (q)Rs is increasing in q. The agent’s
1
Formally, we assume that
ps (b)
for any e¤ort a < b; the ratio is increasing in s:
ps (a)
This assumption is reasonable, but it is restrictive and far from innocuous. It is called
single-crossing-property in the literature.
2
Once again this requires that the single-crossing-property be satis…ed.
9.1 PRINCIPAL-AGENT PROBLEMS 159

utility functions are increasing in consumption c

u(c);

and so is the principal’s:

v(c):

9.1.1 The symmetric information benchmarks

Suppose that the principal observes everything about the agent.

Making the agent work

In this economy the principal observes and measures, in particular, the man-
agerial e¤ort of the agent, a: In other words, the principal can dictate how
much the agent works to insure the success of the project.
We can now write the Pareto planning problem for this economy (recall
we de…ned planning problem in the chapter on Equilibrium). The solution of
the Pareto planning problem is an optimal (Pareto e¢ cient) contract between
the principal and the agent. It is convenient to write the Pareto planning
problem in the …rm of the utility maximization for the principal under the
constraint that the agent derives a minimal given parametric utility level, u.3
The principal must then choose, before the uncertainty is realized, the
best wage schedule, ws , to pay the agent. Also, the principal chooses which
managerial e¤ort a to dictate to the agent. The principal’s choices maximize
his/her utility, and are such that the agent will work for him/her; that is:

X
S
max ps (a)v (Rs ws ) (9.1)
a;fws gS
s=1 s=1
X
S
s.t. ps (a)u(ws ) (a) u (9.2)
s=1

The …rst order conditions of problem (9.1-??) include:

v 0 (Rs ws )
= (9.3)
u0 (ws )
3
The utility level u can also be interpreted as the utility the agent obtains if he does
not work for the principal (and does something else in the market).
160 CHAPTER 9 INCENTIVES

where is the Lagrange multiplier associated to (??).

Consider the important case in which the principal is risk neutral, that
is, without loss of generality,
v(c) = c:
In this case, (9.3) implies that
ws = w;
that is, the wage is constant, and the principal fully insures the agent. It’s
natural that it’d be so, as the principal is risk neutral and the agent risk
averse.4

Paying the agent according to his productivity

In this economy the principal observes and measures, in particular, the man-
agerial skills of the agent, his ability q = 1; :::; Q: In other words, the principal
can pay the agents she hires according to his productivity, his ability to in-
sure the success of the project. Assume that a fraction q of the agents are
P
born with ability q; so that Q q=1 q = 1: The principal, therefore hires an
agent with ability q; randomly, with probability q :
Let’s now write the Pareto planning (optimal) problem. It is convenient
to write the problem in the form of a weighted maximization of P the utility of
the agents of di¤erent ability types q, with weights q such that Q q=1 q = 1;
under the constraint that the principal derives a minimal given parametric
utility level, v. (Why is this convenient? Try and write the dual case in which
we maximize the utility of the principal under the constraint on the mini-
mum utility of the agents.) The planner chooses wages (contracts) (w1q ; w2q );
respectively, for agents of type q:
Q
X X
S
max q ps (q)u(wsq ) (9.4)
Q;S
wsq
( ) q=1 s=1
q=1;s=1

subject to
Q
X X
S

q ps (q)v (Rs wsq ) v:

q=1 s=1

4
Suppose instead that the agent is risk neutral and the principal risk averse. In this
case (9.3) implies that Rs ws = r. That is, the return to the principal is constant, the
agent holds all the risk and insures the pricipal. It is as if the principal would sell the
project to the agent.
9.1 PRINCIPAL-AGENT PROBLEMS 161

For simplicity let’s only consider the case in which the principal is risk
neutral. In this case, without loss of generality, v(c) = c and the planner
chooses (wsq )Q;S
q=1;s=1 which satisfy

wsq = wq ; for any q:

Once again, the risk neutral principal is fully insuring the agent, no matter
what the type of the agent is. But the agent’s wage will in general depend
on his productivity/ability/skills q:

9.1.2 Moral hazard

Consider now the …rst economy, where the incentive problem is to make the
agent work. Suppose however now that the e¤ort a is not observed, nor
can be monitored, by the principal (we say this economy is subject to moral
hazard, as the agent can hide his/her e¤ort to the principal). The agent,
given a schedule fws gs=1;:::;S chosen for her/him by the principal, chooses a
freely.
Are the solutions of the principal-agent problem we determined in the
previous section still optimal? Can the optimal contract we obtained still
be achieved if the e¤ort a is not observed, nor can be monitored, by the
principal? The answer is obviously not, in general. Consider a few examples
of what can go wrong.
In an economy with moral hazard, that is, in an economy where the e¤ort
choice a of the agent is not observable, the social planner cannot choose any
possible level of e¤ort a on the part of the agent. Instead, the social planner
must choose the wage fws gSs=1 so as to incentivate the agent to choose an
optimal e¤ort a:
Formally, the social planning problem needs to satisfy an addictional
constraint, called the incentive compatibility constraint restricting his e¤ort
choice to the ones the agent himself would choose under the wage optimal
contract:

X
S
a 2 arg max ps (a)u(ws ) (a): (9.5)
s=1

Consider the case in which the principal is risk neutral. With symmetric
information, we have shown that we have ws = w; and the principal fully
insures the agent. But with private information, in the moral hazard case, if
162 CHAPTER 9 INCENTIVES

the agent is o¤ered a contract such that ws = w; he would choose the minimal
e¤ort a = arg min (a). This is in general is not the optimal choice of the
principal (the principal will want to maximize a, in fact, under the constraint
that the agent will accept working for him). A di¤erent wage contract need
to be chosen by the social planner.
Rather than attempting a general solution (which is hard), consider a
simple the economy in which s = 1; 2, and a = L; H. Suppose R2 > R1 .
This simply means that state s = 2 is the good state. Naturally, p2 (H) =
1 p1 (H) > p2 (L) = 1 p1 (L), that is, p1 (H) < p1 (L); and furthermore,
(H) > (L). This simply means that H is the high e¤ort, which increases
the probability of the good state and has a higher utility cost.
Assume the principal is risk neutral. Rather than studying (9.1-??), we
study the equivalent problem in which the agent maximizes his/her utility
guaranteeing a minimal utility to the principal (yes, just because we can!).
By varying the minimal utility to the principal, we obtain the whole set of
optimal contracts. We choose to characterize the one in which the minimal
utility for the principal is 0. This is because, being the principal risk averse,
his/her utility e¤ectively coincides with his/her pro…ts, and in an economy
with perfect competitition pro…ts will in fact be 0 in equilibrium. If e.g.,
we interpret the principal as an insurance company (resp. a …rm), we are
studying then a perfectly competitive insurance market (resp. a perfectly
competitive labor market for managerial services). S
If it is optimal for the principal to induce the e¤ort a = H (the only
interesting case), under asymmetric information, the social planning (optimal
contracting) problem is:

max p1 (H)u(w1 ) + (1 p1 (H)) u(w2 ) (H)

w1 ;w2

subject to
0 1
p1 (L)u(w1 )+
p1 (H)u(w1 ) + (1 p1 (H)) u(w2 ) (H) @ A
(1 p1 (L)) u(w2 ) (L)
(9.6)
p1 (H) (R1 w1 ) + (1 p1 (H)) (R2 w2 ) 0 (9.7)
Notice that (9.6), the incentive compatibility constraint, must be binding
for w1 = w2 , as in this case the agent prefers a = L. But since w1 = w2
is the solution when the incentive constraint is not imposed (see before), it
9.1 PRINCIPAL-AGENT PROBLEMS 163

follows that (9.6) is in fact binding at the solution. We conclude that w1 , w2

are chosen so that the agent is indi¤erent between a = H and a = L. The
constraint (9.7) is also binding (this is a bit harder to see; if it wasn’t then
you could transfer resources to the agent, and this could be done without
signi…cant e¤ects on (9.6)).
We characterize now the optimal contract (w1 ; w2 ) which solves the social
planning (optimal contracting) problem under the assumption that u(c) =
ln(c): The wage schedule (contract) (w1 ; w2 ) is chosen to maximize the utility
of the agent under

p1 (H)w1 + (1 p1 (H)) w2 = p1 (H)R1 + (1 p1 (H)) R2 ;

and the incentive constraint (9.7). Note that p1 (H)w1 + (1 p1 (H)) w2 =

p1 (H)R1 + (1 p1 (H)) R2 can be interpreted as a budget constraint (R1 and
R2 are the endowments,..., what are the prices?). Furthermore, under log
preferences, the incentive constraint (9.7), holding with equality, can be writ-
ten as (do it yourselves; it requires some algebra combined with knowledge
of the properties of logs and exponential):
w2
=k
w1
for some constant k: If you’ve done it yourselves, as I asked you to, you have
solved for k; now checkthat k > 1.
Therefore, the solution of the social planning (optimal contracting) cor-
responds to the (w1 ; w2 ) which solves (10.1-??). It is easily checked that the
w2
solution is unique. Since the solution satis…es w 1
= k, with k > 1; it has
the property that w2 > w1 . But w1 is the payment to the agent in the low
cash ‡ow state, s = 1, and w1 w2 is a measure of the distance from full
insurance in the contract (w1 = w2 corresponds to full insurance).
You can now draw a picture (I admit in this case a picture is worth a
million algebraic equations). Do this in steps:

1. Put w2 on the vertical axis and w1 on the horizontal axis.

2. Draw the line corresponding to (budget) constraint (10.1), passing

through the (endowment) point (R1 ; R2 ).

3. Draw the 45o line through the origin which corresponds to the locus of
points with full insurance (that is, such that w1 = w2 ; obvious, right?).
164 CHAPTER 9 INCENTIVES

4. Draw the incentive constraint (??).

5. Find the unique point (w1 ; w2 ) which satis…es (10.1-??).
The result should be something like this:

Moral hazard example

The moral hazard we have just studied is rather abstract and has very
many interpretations and applications in economics. It is worth listing a few.
E¤ort and incentive compensation. Managers might shirk if their
compensation is not linked to the success of the project they manage; in-
centive compensations in the form of wages which depend on the realized
cash-‡ow Rs ; that is, on s; are necessary to at least partly avoids this. Can
you see this is the rationale for the equity (or stock) compensation of man-
agers?
Risk and insurance. The case in which the principal is risk neutral also
captures insurance markets. The principal is the insurer and the agent is the
9.1 PRINCIPAL-AGENT PROBLEMS 165

insuree. The e¤ort a is interpreted as any measure to limit the probability

of an accident, e.g., installing …re alarm devices (or at least avoiding to get
rid of their batteries).
An analogous moral to the one obtained in the previous case can be
extracted: If the agent is fully insured he/she will not invest any e¤ort in
limiting the probability of the accident he/she is insured against. Insurance
contracts therefore need to steer away from full insurance to avoid this. Can
you see this is the rationale for co-payments in insurance contracts?5

9.1.3 Adverse selection

Consider now the economy where the incentive problem is to pay the agent
according to his productivity. We restrict ourselves to a simple example not
unlike the one we studied in the moral hazard economy, in which s = 1; 2, and
q = L; H. As before, suppose R2 > R1 , and p2 (H) = 1 p1 (H) > p2 (L) =
1 p1 (L), that is, p1 (H) < p1 (L). Type H is then the high quality type,
which increases the probability of the good state. Let H be the fraction of
the agents with q = H ( 1 H with q = L):
Let’s now write the Pareto planning (optimal) problem in the form of a
weighted maximization of the utilities of the two types (q = H; L) of agents
(see the chapter on Equilibrium once again), where the weights are ( H ; 1
H ); not necessarily the population weights of the two types. The planner
chooses wages (contracts) (w1H ; w2H ); (w1L ; w2L ), respectively, for agents of type
q = H and q = L:

max H p1 (H)u(w1H ) + (1 p1 (H)) u(w2H ) +

w1H ;w2H ;wL ;w2L

(1 H) p1 (L)u(w1L ) + (1 p1 (L)) u(w2L )

5
A di¤erent interesting example can be constructed for the case (which we did not
study, but we could have) in which it is the agent to be risk neutral. With symmetric
information (and limited liability) the principal payo¤ is Rs ws = minfRs ; rg: Suppose
that the e¤ort a is now interpreted as a form of managing e¤ort with the e¤ect of reducing
the risk of the …rm (for a constant mean). If the agent is risk neutral, it is easy to construct
examples in which the agent will lose in term of expected utility by reducing the risk of
the …rm. The agent hence will not do so, charging the principal with excessive risk. Can
you see this implies that the manager of a leveraged …rm might be induced to take up too
much risky projects?
166 CHAPTER 9 INCENTIVES

subject to
0 1
p1 (H)u(w1L )+
p1 (H)u(w1H ) + (1 p1 (H)) u(w2H ) @ A (9.8)
+ (1 p1 (H)) u(w2L )
0 1
p1 (L)u(w1H )+
p1 (L)u(w1L ) + (1 p1 (L)) u(w2L ) @ A (9.9)
+ (1 p1 (L)) u(w2H )

H p1 (H) R1 w1H + (1 p1 (H)) R2 w2H + (9.10)

+(1 H) p1 (L) R1 w1L + (1 p1 (L)) R2 w2L 0:

Constraint (9.8) are the incentive compatibility constraints. They require
that, given the wages (contracts) (w1H ; w2H ); (w1L ; w2L ), selected by the planner,
respectively, for agents of type q = H and q = L:

i) agents of type q = H prefer (w1H ; w2H ) to (w1L ; w2L ) and

ii) agents of type q = L prefer (w1L ; w2L ) to (w1H ; w2H ).

These constraints are necessary because the principal, while o¤ering dif-
ferent contracts to the two di¤erent types of agents, in fact does not observe
which type is which. The optimal contract therefore must provide the agents
with the incentive to report their true type.
Constraint (9.10) is instead the condition that the principal’s pro…ts are
weakly greater than 0. (Note the principal now obtains pro…ts, in principle,
from the two di¤erent types of contracts).
Two kind of optimal contracts arise in this economy, a pooling contract or
a separating contract (this is hard to prove from the statement of the Pareto
optimal problem; I encourage you to try, and then, if necessary to believe
me). We examine them in turn.

Pooling contract
In this case, the optimal contracts o¤ered to the two di¤erent types of agents
are the same:
(w1L ; w2L ) = (wH ; w2H ) = (w1 ; w2 )
9.1 PRINCIPAL-AGENT PROBLEMS 167

As a consequence, the incentive compatibility constraints (9.8) are triv-

ially satis…ed. The Pareto planning (optimal contracting) problem, restricted
to pooling contracts, reduces to:

max 1 u(w1 ) + 2 u(w2 ) (9.11)

w1 ;w2

subject to
1 (R1 w1 ) + 2 (R2 w2 ) 0 (9.12)
where

1 = [ H p1 (H) + (1 H )p1 (L)] and

2 = [ H (1 p1 (H)) + (1 H ) (1 p1 (L))] :

In fact (9.12) can be written as a budget constraint:

1 w1 + 2 w2 1 R1 + 2 R2

and at the solution,

w1 = w2 :
Both types of agents are perfectly insured. Note that each L type receives as
much as each H type. But the H types are "more productive" (have higher
skills, q = H) in this economy. A planner with a relatively large weight H
will not want to choose this allocation. In this case, the solution will be to
select a separating contract.

Separating contract
In this case, the optimal contracts o¤ered to the two di¤erent types of agents
are not the same:
(w1L ; w2L ) 6= (wH ; w2H ):
The characterization of these contracts is not easy, and we’ll provide it here
without proof:

1. The optimal contract for agents of type q = L, (w1L ; w2L ); satis…es

1
w1L = w2L = wL = (p1 (L)R1 + (1 p1 (L)) R2 )
2
p1 (L)
so that they are fully insured at f air odds: 1 p1 (L)
:
168 CHAPTER 9 INCENTIVES

2. The optimal contract for agents of type q = H, (w1H ; w2H ); satis…es

max p1 (H)u(w1H ) + (1 p1 (H)) u(w2H ) (9.13)

w1H ;w2H

subject to

0 1
p1 (L)u(w1H )+
p1 (L)u(w1L ) + (1 p1 (L)) u(w2L ) = @ A (9.14)
+ (1 p1 (L)) u(w2H )

p1 (H) R1 w1H + (1 p1 (H)) R2 w2H = 0 (9.15)

Agents of type H are not fully insured:

w2H > w1H :

They also face f air insurance odds, 1 p1p(H)

1 (H)
: At this price the planner
would like to insure them fully. But in this case the planner could
not separate the L types, who would claim to be H types to obtain
full insurance at better odds: 1 p1p(H)
1 (H)
< 1 p1p(L)
1 (L)
: The planner therefore
chooses the best contract for agentsof type H which agents of type L do
not have incentive to acquire (this is exactly the meaning, in words, of
requiring that the planner will be restricted by (??) when maximizing
utility 9.13).

Here’s the picture worth a million words (and equations).

9.1 PRINCIPAL-AGENT PROBLEMS 169

Adverse selection: Pooling and separating contracts

9.1.4 Social security

Here’s another example where asymmetric information plays a role. Consider
an economy in which agents derive utility from consumption u(c) and dis-
utility from labor v(l): We assume u(c) is strictly increasing and strictly
170 CHAPTER 9 INCENTIVES

concave and v(l) is strictly increasing and strictly convex (so that v(l)
is strictly concave). Agents have stochastic, idiosyncratic, and permanent
labor productivity (s), where s 2 S and s has probability (s). An agent
with productivity (s) can generate income y = (s)l. The environment
is similar to the one considered by Mirrlees (1971). Individual income y is
observable, but neither (s) nor l are. Here’s the asymmetric information: a
relatively small amount y produced could be due to low productivity or low
labor e¤ort.

Symmetric information ! “Communism”

If the planner observed each agent’s productivity and labor, that is, if the
planner faced a full information problem, he would choose an allocation which
solved: X
max (s) (u(c(s)) v(l(s))) (9.16)
fc(s);l(s)gs2S
s2S

subject to X X
(s) (s)lt (s) (s)ct (s): (9.17)
s2S s2S

At a solution, c(s) = c(s0 ) c, and v 0 (l(s))=v 0 (l(s0 )) = (s)= (s0 ), 8s; s0 2 S.

Prove this!
All agents consume the same amount and the more productive agents
work more. This is the “communist” solution where each agent receives
consumption according to his needs and contributes labor according to his
abilities:

In a higher phase of communist society, after the enslaving

subordination of the individual to the division of labor, and there-
with also the antithesis between mental and physical labor, has
vanished; after labor has become not only a means of life but life’s
prime want; after the productive forces have also increased with
the all-around development of the individual, and all the springs
of co-operative wealth ‡ow more abundantly— only then can the
narrow horizon of bourgeois right be crossed in its entirety and
society inscribe on its banners: From each according to his
ability, to each according to his needs!, Karl Marx, 1875,
Critique of the Gotha Program.
9.1 PRINCIPAL-AGENT PROBLEMS 171

The phrase From each according to his ability, to each according

to his needs! summarizes the principles that, under a communist system,
every person should contribute to society to the best of his ability and con-
sume from society in proportion to his needs, regardless of how much he has
contributed.6

Asymmetric information ! compensate productivity

Suppose now that the planner observes the production y of each agent, but
not (s) nor l. How is the planning problem changed? Which other constraint
needs to be added? Can you write it?
6
The following is taken from Wikipedia on ??
Although Marx is popularly thought of as the originator of the phrase, the slogan was
common to the socialist movement and was …rst used by Louis Blanc in 1840, in "The
organization of work", as a revision of a quote by the utopian socialist Henri de Saint
Simon, who claimed that each should be rewarded according to how much he works. The
origin of this phrasing has also been attributed to the French communist Morelly, who in
his Code of Nature: Sacred and Fundamental Laws that would tear out the roots of vice
and of all the evils of a society, 1755, proposed:

I. Nothing in society will belong to anyone, either as a personal possession

or as capital goods, except the things for which the person has immediate use,
for either his needs, his pleasures, or his daily work.
II. Every citizen will be a public man, sustained by, supported by, and
occupied at the public expense.
III. Every citizen will make his particular contribution to the activities of
the community according to his capacity, his talent and his age; it is on this
basis that his duties will be determined, in conformity with the distributive
laws.

The phrase may also …nd an earlier origin in the New Testament. In Acts 4:32-35, the
Apostles lifestyle is described as communal (without individual possession), and uses the
phrase

distribution was made unto every man according as he had need.

Finally, the United Nations’Universal Declaration of Human Rights, Article 22, states:

Everyone, as a member of society, has the right to social security.

172 CHAPTER 9 INCENTIVES

9.1.5 Other examples [to be added]

1. Akerlof’s lemons
2. Spence’s signaling in education
3. Myerson-Satterwaite theorems on gains from trade
4. Milgrom-Stokey’s no-trade theorem.

9.2 Problems
Problem 9.2.1 Consider a moral hazard economy with no asymmetric in-
formation, that is, the planner can observe and hence choose the agent’s
e¤ort. Suppose however that the social planner is restricted by limited liabil-
ity, that is, he has to guarantee that neither the agent nor the principal can
receive negative amounts from the …rm:
Rs ws 0
Characterize the optimal contract in this case, when eithr the principal or the
agent is risk neutral.

Answer. Formally, the constraint

Rs ws 0
need to be added to (9.1-??) in the social planning problem. The optimal
contract requires, in this case, that:

If the principal is risk neutral, ws = minfRs ; wg and Rs ws = maxfRs

w; 0g. The principal receives the residual cash ‡ow of the project, after
having compensated the agent.
If the agent is risk neutral, ws = maxfRs r; 0g and Rs ws = minfRs ; rg.
It is the the agent who receives the residual.

Problem 9.2.2 Consider a two period economy. Agents are all identical,
that is, there is one representative agent. The representative agent is alive
at time t and t + 1, and has preferences:
lnxt + lnxt+1 ; < 1:
9.2 PROBLEMS 173

This agent is endowed with wt units of the consumption good at time t and
wt+1 at time t + 1. Write down the consumer maximization problem of the
representative agent, under the assumption that he/she can trade a bond (bor-
row or lend) at an equilibrium rate r: [It’s a real interest rate - no in‡ation].
Write down the market clearing conditions (that is, feasibility conditions) for
the whole economy. [Yes, there is only a representative agent and the port-
folio positions of the bond must clear; it’s a trivial economy]. Write down
the Pareto planning problem for this economy. [Once again, there is only a
representative agent; it’s a trivial economy]. Solve for the Pareto optimal
allocation. Solve for the equilibrium interest rate and for the representative
agent equilibrium allocation. [Help yourself with the Pareto optimal allo-
cations; the competitive equilibrium allocations are Pareto optimal in this
economy, right?]. Suppose now that the bond market is closed (does not ex-
ist). But the agent can, at time t, put goods in a bucket which he/she …nd
then at time t + 1: Any unit put in the bucket at time t depreciates to 1 at
time t + 1: Assume that the agent has no endowment at time t + 1; wt+1 = 0;
and solve for the Pareto optimal allocations [The solution will be in terms of
wt , ; and ]:Suppose now that the Government taxes the goods in the bucket,
at time t + 1, at the rate : Will agent now put more or less goods in the
bucket at time t? The utility of the Government is linear in tax revenues.
Will it tax the representative agent maximally? That is, will the Government
choose = 1? Suppose now that the Government cannot look in the bucket.
He still wants to tax the goods in the bucket, but just does not know how
many there are. Write down the Pareto optimal allocation problem for this
economy [choose the form in which the utility of the Government is maxi-
mized for a given level of utility U of the representative agent; be careful to
write down the incentive compatibility constraint]. Do not try and solve the
problem formally; but, given what you understood about this economy, can
you guess what the Pareto optimal tax rate will be in this case?
Chapter 10

Fiscal policy

What are the e¤ects of …scal policies?

10.1 Ricardian Equivalence - a.k.a. Arith-

metics

Robert Barro Tom Sargent Neil Wallace

Consider a two-period economy with a representative agent. A production

(or a …nancial) technology allows the agent to save any non-negative amount
St at t and receive (1 + r)St at t + 1. The representative agent’s budget
constraints are:

175
176 CHAPTER 10 FISCAL POLICY

ct + St = wt
ct+1 = St (1 + r) + wt+1

We have noted already that these budget constraints (solving and substi-
tuting for St ) can be reduced to

1 1
ct + ct+1 = wt + wt+1
1+r 1+r
We have also computed a solution for the consumption-saving problem
of the representative agent with preferences u(ct ) + u(ct+1 ). If we assume
that:

1
u(ct ) = ln ct ; u(ct+1 ) = ln ct+1 ; = ;
1+r
we have the following consumption allocations:

1+r 1
ct = wt + wt+1
2+r 1+r
1+r 1
ct+1 = wt + wt+1
2+r 1+r

Consider now a government which can tax/subsidize agents at either mo-

ment of time. Let a subsidy to the representative agent at time t be gt , and
a tax be t : Similarly, subsidies and taxes at time t + 1 are, respectively, gt+1
and t+1 : Note that taxes are assumed to be lump-sum, that is, not indexed
to income, savings, consumption, or any other of the agent’s choices.1 The
government borrows at the interest rate r and balances budgets dynami-
cally (not necessarily period by period). Therefore, the government dynamic
budget constraint is:
1 1
gt + gt+1 = t + t+1
1+r 1+r
1
In other words taxes are the same in absolute amount independently on the agent’s
choices. In an economy with heterogeneous agents, rather than a representative agent,
lump-sum taxes are the same in units of consumption goods for each agent. See the next
section for a relaxation of the lump-sum assumption.
10.1 RICARDIAN EQUIVALENCE - A.K.A. ARITHMETICS177

With government intervention, the representative agent’s budget con-

straints are:

ct + St = wt + gt t
ct+1 = St (1 + r) + wt+1 + gt+1 t+1

which can be reduced to

1 1 1 1
ct + ct+1 = wt + wt+1 + gt + gt+1 t t+1
1+r 1+r 1+r 1+r
The terms in parenthesis on the right-hand-side is = 0 because of the
government budget constraint.
What does this mean? That in this economy,

government expenditures have no e¤ect - because the repre-

sentative agent rationally anticipates that every expenditure at t
is …nanced through taxes at t + 1.

This proposition, forgotten by keynesian economists, has been returned

to light by Robert Barro (who called it Ricardian Equivalence - from David
Ricardo, who apparently …rst stated it informally) and by Tom Sargent and
Neil Wallace (who properly called it arithmetics).

10.1.1 Borrowing-Constraints
The analysis in the previous section assumes that the representative agent
does not face any borrowing constraints.2 Suppose on the contrary that the
agent can lend at rate r;but cannot borrow. That is,

St 0;

or, equivalently:
ct wt + gt t

In this case, obviously, positive government expenditures net of taxes at

time t, gt t > 0, have an e¤ect on consumption at time t, as long as the
borrowing constraint is binding, that is, ct = wt + gt t:
2
Similar (but less clean) results are obtained if we assume that the representative agent
can borrow but at a rate r0 > r:
178 CHAPTER 10 FISCAL POLICY

We conclude that in this case government expenditure has e¢ ciency gains,

it allows for consumption smoothing which would be impossible without it.
In other words (can you show this formally?), goverment expenditure can
support the optimal consumption allocation

1+r 1
ct = wt + wt+1
2+r 1+r
1+r 1
ct+1 = wt + wt+1
2+r 1+r

even when such allocation would not satisfy the borrowing constraint with
no government (gt = t = 0), that is, when

1+r 1
wt < wt + wt+1
2+r 1+r

or (better written), when:

wt < wt+1

Before concluding that, because agents are in fact borrowing constrained

especially during recessions, government expenditure is a desirable policy,
you should ask yourself why is it that the representative agent is constrained
and the government is not. Why don’t private banks lend to the agent? Well,
.... the government has an advantage with respect to banks, in principle, in
making sure agents re-pay debts (the government can tax and imprison
those who do not pay taxes; the bank cannot).3 But at the same time the
government has a disadvantage with respect to banks, normally, in screening
and servicing debtors.

10.1.2 Distortionary Taxes

Suppose taxes are not lump-sum but distortionary. Taxes are distortionary
when they are given as a function of an endogenous variable, a variable chosen
by the agent in the economy.
3
In fact, even this advantage is debatable. See the notes on Time Inconsistency of
Government Policies.
10.1 RICARDIAN EQUIVALENCE - A.K.A. ARITHMETICS179

Typically, income taxes are distortionary, because income depends on

labor supply which is chosen by the agent. In our economy income is ex-
ogenous, it takes the form of endowments, and hence distortionary taxes are
e.g., taxes on savings:
t+1 = (1 + r)St

The government budget constraint is

1 1
t + t+1 = gt + gt+1
1+r 1+r
Suppose also, without loss of generality, that t = 0; we have

t+1 = (1 + r)gt + gt+1 :

The representative agent’s budget constraints are

ct + St = wt + gt
ct+1 = St (1 + r)(1 ) + wt+1 + gt+1

Under the maintained assumptions that:

1
u(ct ) = ln ct ; u(ct+1 ) = ln ct+1 ; = ;
1+r
we have the following consumption allocations (Make sure you derive these
yourself!):4

1+r 1
ct = wt + wt+1
2+r 1+r
(1 + r) (1 ) 1
ct+1 = wt + wt+1
2+r 1+r

Taxes are derived from the intertemporal government budget constraint:

4
In fact, be careful. If the agent is borrowing, St < 0, he receives a subsidy form the
government, the way we have written the agent’s maximization problem (the interest rate
he faces is lower than 1 + r; it’s (1 + r)(1 ) in fact. This is not typically the case in the
…scal code. The equations we obtain in the following only hold for S 0: Suppose that
the interest rate on borrowing is 1 + r and derive the consumption equations in this case.
How does the whole consumption function look like as a function of permanent income?
Draw a …gure.
180 CHAPTER 10 FISCAL POLICY

t+1 = (1 + r)St = (1 + r)gt + gt+1

and hence
1
gt + g
1+r t+1
=
St
1 1
gt + 1+r gt+1 gt + g
1+r t+1
= = 1+r 1
wt + gt ct wt + gt 2+r
wt + w
1+r t+1

We conclude that:

- ct > ct+1 (Show this!)

- increases with gt and gt+1 (Show this as well! Be careful, it requires some
messy derivation and the use of the Implicit Function Theorem)

Let the distortion due to taxes be measured by

ct ct+1
ct

Finally, and most importantly, we conclude from the analysis above that the
distortion increases with gt and gt+1 (Try this! But don’t be surprised if you
can’t do it).
Finally, let’s derive the La¤er curve, one of the main bullets of "small
government" types:

An decrease in the tax rate could give rise to an increase in the

government tax revenues, t+1 = (1 + r)St :

How is this possible? Easy to prove: take = 1 (all savings go to the

government); show that in this case St = 0 and hence t+1 = 0. Notice
that the proof works at very high taxes; supply-side economics relies on an
empirical question: at which tax rate L does the La¤er curve e¤ect kick in?
Are tax rates now in the U.S. in the range of L ? Answer: not even close.
10.1 RICARDIAN EQUIVALENCE - A.K.A. ARITHMETICS181

10.1.3 Krugman, ....., and the army of …scal policy

hawks

?? on the New York Times on November 11th, 2008, forgets all this;
including the arithmetics:

The economic lesson is the importance of doing enough. F.D.R.

thought he was being prudent by reining in his spending plans; in
reality, he was taking big risks with the economy and with his
legacy. My advice to the Obama people is to …gure out how much
help they think the economy needs, then add 50 percent. It’s much
better, in a depressed economy, to err on the side of too much
stimulus than on the side of too little.

What’s behind this statement? Any theory? The consumption multiplier!

Let’s see what it is.

The consumption multiplier

The consumption multiplier is an analytical concept which arises in John

Hicks’ representation of Keynesian economics, the famed IS-LM model. It
works as follows:
Suppose agents’consumption is a a constant fraction c of income:
182 CHAPTER 10 FISCAL POLICY

ct = a + c (wt + gt )
ct+1 = a + c (wt+1 + gt+1 )

Also, income wt at time t is endogenous, equal to consumption

plus investment (the economy produces to satisfy its demand).
Not having investment, we have ct = wt :
a + gt
wt = a + c (wt + gt ) =) wt + gt =
1 c
The consumption multiplier is
1
> 1:
1 c
Every extra unit of government spending gt increases aggregate
income wt + gt of 1 1 c > 1 units.

I think it’s fair to say that the multiplier as a theoretical construct con-
cept has been relegated to the disrespect of macroeconomists by Bob Lucas’
"Critique." Here’s why: if agents are rational and choose consumption to
maximize present discounted expected utility, then the marginal consump-
tion out of income, c; is not a constant - consumption depends on permanent
dct
income - and dg t
will depend on income today but also on income tomorrow.
As a consequence, taxes tomorrow enter the determination of consumption
today. In a world without frictions (e.g., when nobody is borrowing con-
trained) then government expenditure has no e¤ect (Ricardian equivalence).
But even with frictions, the multiplier depends on the e¤ect of government
expenditure on the permanent income.

The dynamic multiplier

But it takes much more than this to wear Paul Krugman down. He can take
all sorts of ?? :

Even the claim that we’ll have to pay for stimulus spending
now with higher taxes later is mostly wrong. Spending more on
recovery will lead to a stronger economy, both now and in the fu-
ture — and a stronger economy means more government revenue.
10.1 RICARDIAN EQUIVALENCE - A.K.A. ARITHMETICS183

Stimulus spending probably doesn’t pay for itself, but its true cost,
even in a narrow …scal sense, is only a fraction of the headline
number.

What is behind this outlandish statement that government expenditures

now have a positive e¤ect on GDP in the future: the dynamic multiplier of
Blinder-Solow.5
Here’s the story. Suppose agent’s consumption depends on wealth, rather
than on income:
ct = cWt :
If we de…ne wealth as permanent income,
1
Wt = wt + wt+1
1+r
we have the consumption equation we have derived from …rst priciples (dy-
namic optimization). Adding a government sector, this notion of permanent
income implies that wealth is as follows:
1 1 1
WtRE = wt + wt+1 + gt + gt+1 t + t+1
1+r 1+r 1+r

We call this notion of wealth WtRE as a reminder that it implies Ricardian

equivalence. Recall in fact that the government budget balance

gt t = Bt
gt+1 + (1 + r)Bt = t+1

where Bt is government debt at t; can be written as:

1 1
gt + gt+1 t + t+1 =0
1+r 1+r
1 1
Recognizing that government balance implies that gt + 1+r gt+1 t + 1+r t+1 =
0, the representative agent permanent income is not a¤ected by government
expenditures, which have then has no e¤ects.
5
Though the …rst formulation apparently is in Carl Christ (1968). Thanks to Alessandro
Lizzeri for reminding me of this concept - which I had not heard of anymore since my
undergraduate times of Dornbush-Fisher economics).
184 CHAPTER 10 FISCAL POLICY

Naturally, the notion of wealth in Blinder-Solow’s formulation of the dy-

namic multiplier, let is be denoted WtBS must be di¤erent.6 They assume
the following:

1
WtBS = wt + wt+1 + Bt :
1+r
The rationale for WtBS is that agents in equilibrium must hold the government
debt, Bt , in their portfolios, and hence consider it as part of their wealth.
Once again, however, they do not anticipate the taxes at time t + 1 necessary
to balance the budget.
Suppose that t = 0 for simplicity. Then

1
ct = c wt + wt+1 + gt
1+r
and government expenditures have an e¤ect at time t.

Kevin Murphy’s static set-up

Kevin Murphy

6
Blinder-Solow’s formulation also accounts for changes in the interest rate at equilib-
rium - which we disregard in the analysis here without losing the thrust of the argument.
10.1 RICARDIAN EQUIVALENCE - A.K.A. ARITHMETICS185

Kevin Murphy has a very clean set-up to collect all the costs and bene…ts
of government expenditures. We reproduce it here.
Let’s set some notation:
g = increase in government spending with respect to the steady state spend-
ing (the level of spending which would have obtained were there not a
recession); that is, size of the …scal stimulus;
1 a = value of a dollar of government spending; that is, a measures the
resource allocation ine¢ ciency due to the fact that the government
chooses what to spend on rather than households;
f = fraction of government spending obtained by borrowing constrained
agents;
l = relative value of relaxing the borrowing constraint;
d = deadweight cost per dollar of the revenue from future taxation required
to …nance G.
Then, the bene…ts in value due to a …scal stimulus of size g are:

(1 a)g + lf g
The costs are:

(1 f )g + dg
The net gains (bene…ts minus costs) of a …scal stimulus g are then:

N et Gain = (1 a)g + lf g [(1 f )g + dg] = [f (1 + l) (a + d)]g

Paul Krugman, Brad De Long, and the economist who are more openly
in favor of the stimulus tacitly assume:
a = 0, that is, the government does what’s best for agents, in terms of their
own (the agents’) preferences;
l is high, that is, borrowing constrained agents have a very high marginal
utility of consumption;
d = 0, that is, taxes are not distortionary
They also assume f is large –close to 1.
186 CHAPTER 10 FISCAL POLICY

10.2 Optimal taxation

James Mirlees

We study optimal taxation in this Chapter. By "optimal" we mean that

the government raising taxes does this in the interest of the agents of the
economy and not in the personal interest of the bureaucracies that constitute
the government itself. This is a strong assumption, but we will show that,
even in this case, we need to …nd ways to control the sovereign (Ed Prescott,
who has been awarded the Nobel Prize in 2005 for his work with F. Kydland
on Time Inconsistency).

10.2.1 Distortionary Taxation

Suppose a government has to raise an amount g to provide the economy
with a public good (e.g., military defence, police protection, communication
infrastructures, education and health care,....). The public good consists of
units of the single consumption good in the economy at time t + 1. The
government raises taxes. We consider two di¤erent tax systems:

taxes on savings, and

lump-sum taxes.

We will show that lump-sum taxes are to be preferred in terms of Pareto

e¢ ciency. Taxes on savings are most commonly used though because they
are easier to implement in real economies.
10.2 OPTIMAL TAXATION 187

Consider a two-period economy with a representative agent. A production

technology allows the agent to save any non-negative amount St at t and
receive (1 + r)St at t + 1. With lump-sum taxes the representative agent’s
budget constraints are:

ct + St = wt ; ct+1 = St (1 + r)

=g
With taxes on savings, instead:

ct + St = wt ; ct+1 = St (1 + r) (1 )

St (1 + r) = g
If the representative agent has period utility u(c), and discounting , the
…rst order conditions of his/her maximization problem with taxes include:

u0 (ct+1 )
= (1 + r)[with lump sum taxes]
u0 (ct )
u0 (ct+1 )
= (1 + r)(1 )[with taxes on savings]
u0 (ct )
We say that the saving choice of the agent is distorted with taxes on
savings but not with lump-sum taxes. What we mean is that the …rst order
condition which determines savings at the margin is changed (with respect to
the benchmark with no taxes) with taxes on savings but not with lump-sum
taxes. This is the source of the ine¢ ciency of taxes on savings: A planner
with a constraint that g must be raised faces aggregate endowment yt and
yt+1 g at t + 1 (and a saving technology with return 1 + r) and chooses the
consumption allocation which results from lump-sum taxes. [Make sure you
understood this point!]

10.2.2 Public Goods Provision

Consider the economy in the previous section. Assume that the government
must now choose an amount g of public good provision. The public good
has utility for the representative agent in the amount v(g). The government
188 CHAPTER 10 FISCAL POLICY

is aware that taxes on savings distort the margin (as we have shown in the
previous section), but has no other …nancing mean available.
The government will have to choose public good provision g and tax rate
to maximize the representative agent’s utility while satisfying the constraint
that the public good needs to be …nanced by taxes on capital: g = (1 + r) St .
The government, on the other hand, cannot prescribe a saving choice to
the representative agent. (This is what distinguishes a government from a
planner: The government has power of raising taxes, but agents make their
own economic choices; the planner is instead a conceptual construct we use
as a benchmark to de…ne Pareto optimality, who chooses all of the agents’
allocations, including savings). Write down the planner’s problem and its
…rst order conditions; recall that there is a representative agent and hence
the objective function of the planner does not depend on parametric weights.
Recall that the public good is in units of the consumption good at time
t+1. Consider the case in which at time t the government decides the amount
of the public good to be provided and the tax rate to …nance it. We solve
this problem in two steps: …rst we solve for the amount of savings that the
agent chooses given any g and hence any (given g, is determined by the
…nancing constraint g = (1 + r) St ); then we solve for the optimal level of
g chosen by the government.
The …rst step, that is, the choice problem of the representative agent, is
the following:

maxct ;ct+1 ;St u(ct ) + (u(ct+1 ) + v(g))

subject to:

ct + St = wt ; ct+1 = St (1 + r) (1 ) (10.1)
(1 + r) St = g (10.2)
Assume u(c) = 1 1 c1 , with < 1; and assume wt+1 = 0. Then the
…rst order condition with respect to savings St (substitute (10.1), not (10.2),
in the utility function of the agent) is (derive it; it takes a bit of algebraic
work):
St 1
= ( (1 + r) (1 )) (10.3)
wt St
For this equation it is relatively easy to see that @S
@
t
< 0. [To prove it either
use the Implicit Function Theorem or reason as follows: the right-hand-side
10.2 OPTIMAL TAXATION 189

is increasing with St ; the left-hand-side is decreasing in if < 1, as we

assumed; therefore, if increases, the left-hand-side decreases and hence the
right-hand-side must also decrease, which can only be if St decreases.] We
can also similarly establish, by substituting (10.2) into (10.3) and re-doing
the argument, that @S @g
t
< 0 (do this; St appears in (10.2) and hence the
argument needs be modi…ed a little bit).
We conclude therefore that, for given public good provision g …nanced
through taxes on savings, the representative agent saves an amount S(g),
with S 0 (g) = @S
@g
t
< 0.
We can now solve for the second step, the optimal level of public good
provision g of the government. It is the solution of the following maximization
problem (note that we have substituted the expression for consumption into
the utility function, as well as the expression for savings, S(g), and for taxes,
g
(1+r)S(g)
, from the …nancing constraint):

g
maxg u(wt S(g)) + u S(g) (1 + r) 1 + v(g)
(1 + r) S(g)

The …rst order condition can be written as follows:

v 0 (g) = u0 (ct+1 ) S 0 (g) ( u0 (ct+1 )(1 + r) u0 (ct )) (10.4)

Let’s analyze this condition carefully. Note the following:

the left-hand-side is the marginal bene…t of g and the right-hand-side is its

marginal cost;

the term S 0 (g) ( u0 (ct+1 )(1 + r) u0 (ct )) is positive if > 0:

S 0 (g) < 0;
u0 (ct+1 )(1 + r) u0 (ct ) > 0 if > 0, from the …rst order condition of
0
the representative agent problem, which include uu0 (c(ct+1
t)
)
= (1 +
r)(1 );

it increases with the absolute value of S 0 (g), and with u0 (ct+1 )(1 +
r) u0 (ct ).

We can now compare the solution of the government problem with the
solution of the planner’s problem (that I have asked you to write down at the
190 CHAPTER 10 FISCAL POLICY

beginning of this section). The …rst order conditions of the planning problem
(or of the government problem with lump-sum taxes) include the following:

v 0 (g) = u0 (ct+1 )

(check this! convince yourself, this is important). In this equation, also, the
left-hand-side represents the marginal bene…t of g and the right-hand-side
its marginal cost. Therefore we have shown above that the marginal cost
of g in the government problem is greater than the marginal cost associated
to the planner’s problem, u0 (ct+1 ), if S 0 (g) < 0 and > 0. We conclude
that that the provision of the public good in the government problem is less
than the e¢ cient amount (resulting from the planner’s problem). [This is
not straightforward, because the levels of consumption ct and ct+1 in the two
problems are also di¤erent; in other words, it is not enough to compare the
…rst order conditions we did compare, but we need to look at the whole set of
…rst order conditions. The result is true, however, because savings is smaller
in government’s problem, and hence ct+1 smaller. Try to prove this if you
feel strong.] Equivalently, the choice of public provision g by the government
with taxes on savings is smaller than the choice that would derive from a
government problem with lump-sum taxes. [Perhaps it is worth convincing
yourself of this; write down the problem with lump-sum taxes and solve it;
show the solution coincides with the solution of the planning problem.]

Note that u0 (ct+1 )(1 + r) u0 (ct ) is a measure of the distortion due to

taxes on savings; that is, when taxes are lump-sum it is always true that
u0 (ct+1 )(1 + r) = u0 (ct ), while with taxes on savings u0 (ct+1 )(1 + r) =
0
u (ct )
1
> u0 (ct ), and hence u0 (ct+1 )(1 + r) u0 (ct ) increases with .
10.2 OPTIMAL TAXATION 191

10.2.3 Time Inconsistency

Ed Prescott and Finn Kydland

Remember we considered the case in which at time t the government

decides the amount of the public good to be provided and the tax rate to
…nance it. Suppose now we consider the case in which at time t + 1, after
savings decisions have been made by the representative agent on the basis of
the levels of g and which we derived in the previous section, the government
can choose possibly a di¤erent level of public good provision G and associated
tax T to satisfy the …nancing constraint. What is the government problem
now?
The government at time t + 1 takes as given the amount of saving S(g),
and chooses G as the solution to the following problem:

G
maxG u S(g) (1 + r) 1 + v(G)
(1 + r) S(g)
The …rst order condition of this problem is:

u0 (ct+1 ) = v 0 (g)
Convince yourself that the G which solves this problem is greater than G.
(Now, this is easy, because S(g) does not change, and hence ct+1 only changes
in account of T .)
We conclude that, if the government would be allowed to change its deci-
sion at time t+1, it would. Even though the government is always maximizing
192 CHAPTER 10 FISCAL POLICY

the utility of the representative agent. The intuition is that at time t the
government limits the provision of the public good to limit the distortions
that taxes impose on savings. At time t + 1 instead savings has already
happened, and hence raising the public good provision (and taxes) does not
distort savings.
Of course the agents at time t might not trust the government not to
raise taxes at time t + 1, and hence might reduce savings nonetheless. What
do you expect it will happen? You should be able to answer this informally.
For a formal answer we need Game Theory - coming in the next classes.

10.3 Problems
Problem 10.3.1 What do we (beginner economists) expect the e¤ects of
President-elect Barack Obama’s proposed government expenditure plan to be?
In particular, what will the e¤ect on consumption be? IMPORTANT: Your
answer should be no longer than 10 lines (and do not write tiny letters);
Negative points are taken away for answers of the sorts I can hear on the
street (-15 points is the most I will take away); I want a coherent argument
- as an economist would do - there is no "correct answer", just good and bad
arguments.
Chapter 11

Empirical economics

Karl Pearson Jacob Marschak Tjalling Koopmans James Heckman

Understanding how economists test their models and how they measure
fundamental economic quantities is a fundamental issue. As we noted in the
Introduction, in fact, economics as a discipline is largely empirical, much
more than it typically gets credit for.
The fundamental question of empirical analysis, in economics as in social
sciences in general, consists in identifying causal relationships. It is a di¢ cult
task even to de…ne what is a causal relationships, let alone identifying one in
data.
The issue of causation is central to phylosophical thought. While this is
not the appropriate place for a discussion of the history of the phylosophical
thought on this issue, it is important at least to notice that the issue of

193
194 CHAPTER 11 EMPIRICAL ECONOMICS

causation has been framed conceptually in a modern way by David Hume,

An Enquiry concerning Human Understanding and Other Writings, written
in 1748.1 Cambridge: Cambridge UP, 2007. Print. (Section VII).

David Hume

Hume distinguishes "relations of ideas," which we would now call deduc-

tive arguments, from "matters of fact," inductive arguments. Hume also
argues that "matters of fact" result from a mere habit or, in other words,
form a subject’s psychological mechanism, his idea of a causal relation de-
rived from a habitual transition in the mind. In Hume’s view a

matter of fact [is] an object, followed by another, and whose

appearance always conveys [by habit] the thought to that other.

We facing a "matter of fact" then, according to Hume, we should not

properly talk of causation, which is instead a property of "relation of ideas."
Economists refer instead to causality in the context of inductive reasoning.
If Hume would argue that the relationship between the ingestion of food and
the feeling of nourishment is not causal, economists would on the contrary
accept this relationship as a causal one. Economists and statisticians make
causal inferences from correlations in the data, while being aware of the
limitations of inference methods.
1
See Hume, David, and Stephen Buckle, An Enquiry concerning Human Understanding
and Other Writings, Cambridge: Cambridge University Press, 2007.
11.1 CORRELATION AND REGRESSION ANALYSIS 195

Economists de…ne causation as the e¤ects of changes in exogenous vari-

ables, parameters, and constraints on outcomes. Within a model, the e¤ects
on outcomes of variation in exogenous variables, parameters, and constraints
is formalized by the Marshallian notion of comparative statics, i.e., the no-
tion of a ceteris paribus change. Let e.g., a model represent the relationship
between an outcome y and exogenous variables x1 and x2 as a functional
relationship
y = f (x; z) :

We say then that x (positively) causes y at (x; z) if @f@x

(x;z)
> 0: Note that the
partial derivative captures the notion of coeteris paribus: @f@x(x;z)
encodes the
e¤ect of an (in…nitesimal) change in x on y, keeping the variable z constant.
The requirement that x and z be exogenous variables is tantamount to
a requirement that x and z can be controlled independently from y. Sup-
pose instead that x and z were determined (in the model) jointly with y.
They would then represent endogenous variables; the partial derivative @f@x (x;z)

would then be well-de…ned mathematically, but would not represent a causal

relationship.

11.1 Correlation and regression analysis

How can we then identify causal relationships in the data?2 How can we
observe independent changes of one variable x on an outcome y?
Suppose we observe a series of observations of triples (x; z; y) e.g., over
time or over space, or for di¤erent economic agents. Let observations of
(x; z; y) be indexed by i = 1; :::; N : the data are then a list (xi ; zi ; yi )N
i=1 : We
can then de…ne the correlation between x and y in the data by means of the
sample correlation coe¢ cient,

cov(x; y)
x;y = ;
x y

where x and y are sample means, cov(x; y) is the sample covariance, and x

2
This is the problem that the Cowles Commission for Research in Economics set out
to explore in the late 30’s in Chicago under the directorship of Jacob Marschak, …rst,
and Tjalling C. Koopmans, later. The Commission’s motto is illustrative of its purposes:
Science is Measurement.
196 CHAPTER 11 EMPIRICAL ECONOMICS

and y are sample standard deviations:

1 X 1 X
N N
x = xi ; y = yi ;
N i=1 N i=1

1 X
N
cov(x; y) = (xi x) (yi y) ;
N i=1
v v
u N u N
1tu X 1 uX
x = (xi x)2 ; y = t (yi y)2 :
N i=1 N i=1

A positive correlation between x and y, keeping z constant is an indication

that on average, observations of xi above its sample mean x are on average
associated to observations of yi above its sample mean y. Is this evidence
of causation? Can we associate a positive correlation between x and y with
a causal relationship between the two variables? The answer is negative, in
general, because a positive correlation is silent about the exogeneity of either
x or y; in exactly the same sense a positive partial derivative @f@x
(x;z)
is.
In the words of Karl Pearson, one of the fathers of mathematical statistics
and in particular of the notion of correlation,

correlation does not imply causation.

Then again, in some speci…c empirical contexts, we can guarantee that

some variables are exogenous. In this case, regression analysis, a generaliza-
tion and an application of the notion of correlation is useful. Consider again
data in the form of a list (xi ; zi ; yi )N
i=1 and an underlying model which implies
that y = f (x; z): Suppose we are willing to assume that the underlying model
is (closely approximated to be) linear, modulo some non-systematic errors,
e.g., measurement errors. That is,

yi = + x xi + z zi + ui ;

where ui is the residual error. The requirement that errors are non systematic
implies tha ui assumed to be a random variable with zero mean. De…ne
the best linear model as the one which minimizes the implied errors. More
precisely, the implied errors are measured by the sum (across realizations)
of squared errors, to
Phave a measure which is independent of the sign of the
N
errors themselves: i=1 (ui )2 :
11.1 CORRELATION AND REGRESSION ANALYSIS 197

In this context, regression analysis is the characterization of the best

linear model for the given data, that is, the linear function which minimizes
the the sum of squared errors in the data. It results as the solution of the
following minimization problem:

X
N
2
min [yi ( + x xi + z zi )] :
; x; z
i=1

Theorem 11.1.1 Suppose we observe data (xi ; zi ; yi )N

i=1 : The best linear
model for the data is

yi = + x xi + z zi + ui ;

with

= y xx zz
X
N
1
N
(xi x) (yi y)
cov(x; y) i=1
x = 2
=
x X
N
1
N
(xi x)2
i=1
X
N
1
N
(zi z) (yi y)
cov(z; y) i=1
z = 2
=
z X
N
1
N
(zi z)2
i=1

Proof. The f.o.c. of the minimization problem with respect to ; x; z are:

X
N
2 (yi ( + x xi + z zi )) = 0
i=1
X
N
2 (yi ( + x xi + z zi )) xi = 0
i=1
XN
2 (yi ( + x xi + z zi )) zi = 0
i=1
198 CHAPTER 11 EMPIRICAL ECONOMICS

Some simple (and somewhat tedious) arithmetics allows now to obtain the
expressions in the statement. Finally, second order conditions guarantee that
the solution of this system represents a minimum of the problem (not a maxi-
XN
mum), as the objective function [yi ( + x xi + z zi )]2 is (easily shown
i=1
to be) convex in ( ; x; z ):

11.2 Controlled experiments

A controlled experiment (or Randomized Controlled Trial - RCTs) compares
the results obtained from an experimental group of subjects against a control
group, which is identical to the experimental grou except for the one aspect
whose e¤ect is being tested (the exogenous variable). A good example would
be a drug trial. The group receiving the drug would be the experimental one;
and the one receiving the placebo would be the control one.
What is essential for a controlled experiment is that

1. the sample of subjects be representative of the population under study,

and

2. the distinction between the experimental and the control group be ran-
dom, so that the two groups are statistically equivalent.

In general, a sample is not representative of the population if subjects

are allowed to choose to participate in the experiment depending on some
relevant characteristic of the experiment itself; e.g. in a drug test, if only
subjects with some speci…c (intense) manifestations of a disease are choosing
to participate.
The earliest controlled experiments were due to Charles Darwin who, with
the help of Francis Galton, compared the growth of cross versus self- fertilized
plants (Darwin, 1876).3 Controlled experiments are now common practice
in medicine and other sciences: as of 2004, more than 150,000 RCTs were
recorded in the Cochrane Library, an archive created to prepare, maintain,
and disseminate systematic reviews of studies on the e¤ects of health care
programs.
3
It is said that Darwin was in‡uenced to some degree by path-breaking experiments in
physical sciences due to Michael Faraday.
11.3 NATURAL EXPERIMENTS 199

Controlled experiments are also typically used in what is called exper-

imental economics, an important sub-discipline in the academic economic
profession. More recently, controlled experiments have also been used in de-
velopment economics, notably e.g., by the Mexican government in the PRO-
GRESA (now called Oportunidades) project and at the Poverty Action Lab
at MIT. We shall feel content with brie‡y introducing just a few examples
illustrating the methodology.
Experimental economics. [...] Forsythe, Horowitz, Savin, and Sefton
(1994). [...]
Development economics: The PROGRESA program in Mexico o¤ers
grants, distributed to women, conditional on children’s school attendance and
preventative health measures (nutrition supplementation, health care visits,
and participation in health education programs). In 1998, when the program
was launched, the Mexican government decided to start with a randomized
pilot program in 506 of the 5,000 communities identi…ed as potential ben-
e…ciaries of the program. Half of those were randomly selected to receive
the program, and baseline and subsequent data were collected also in the
remaining communities. Many researchers have studied the resulting data.
Du‡o, Kremer, and Robinson (2006) evaluated a series of di¤erent in-
terventions to understand the determinants of the adoption of fertilizer in
Western Kenya. The design of the interventions made it possible to test some
of the standard hypotheses of the hindrance to the adoption of new technol-
ogy. Field demonstrations with treatment and control plots were conducted
to evaluate the probatability of fertilizer in the local conditions. Because
the farmers were randomly selected, those …eld trials also allowed them to
study the impact of information provision and the channels of information
transmission; other ways to provide information (starter kits, school based
demonstrations) were also examined; …nally, …nancing constraints and dif-
…culties in saving were also explored, with interventions helping farmers to
buy fertilizer.4

11.3 Natural experiments

Controlled experiments are a useful inference technique because, by con-
trolling the sampling procedures, the experimenter can guarantee that the
4
This paragraph is taken verbatim from Du‡o, Glennersterzand, Kremer (2006), Using
Randomization in Development Economics Research: A Toolkit, mimeo MIT.
200 CHAPTER 11 EMPIRICAL ECONOMICS

sample of subjects be representative of the population under study and that

the distinction between the experimental and the control group be random.
Even though an experiment is not controlled, however, it might nonetheless
be that the sample satis…es these properties. In this case we say that we have
discovered a natural experiment.

One of the earliest natural experiments was the 1854 Broad Street cholera
outbreak. On August 1854, a major outbreak of cholera struck London.
Deaths appeared concentrated near Broad Street. The physician John Snow
identi…ed the source of the outbreak as the polluted public water o¤ered by
pumps managed by one of the several companies active in London. Given the
near random patchwork development of the water supply in mid-Nineteenth
Century London, Snow considered the circumstance as a (natural) "experi-
ment...on the grandest scale" on the public health e¤ects of the exposure to
the polluted water. The treatment group, in the experiment, is composed of
the citizens living in a neighborhood whose water is supplied by the com-
pany which supplies Broad Street, while the control group is composed of all
citizens whose water is supplied by a di¤erent company.

A more recent famous natural experiment regarding the e¤ect of family

size on the labor market outcomes of the mother has been studied by An-
grist and Evans (1998). The correlations between family size and various
labor market outcomes do not tell us how family size causally a¤ects labor
market outcomes because both labor market outcomes and family size may
be a¤ected by unobserved variables such as preferences and because labor
market outcomes may itself a¤ect family size. But two-children families with
either two boys or two girls are substantially more likely to have a third child
than two-children families with one boy and one girl. The gender of the …rst
two children, then, forms a natural experiment: it is as if an experimenter
has randomly assigned some families to have two children and others to have
three or more. The authors are then able to estimate the causal e¤ect of hav-
ing a third child on labor market outcomes by comparing the labor market
outcome of the mothers in the treatment group (three-children families with
the …rst two of the same gender) with those of mothers in the control group
(two-children-of-di¤erent-gender families).
11.3 NATURAL EXPERIMENTS 201

Figure 11.1: Original map by John Snow showing the clusters of cholera cases
in the London epidemic of 1854
202 CHAPTER 11 EMPIRICAL ECONOMICS

11.4 Structural econometrics

What then of an empirical context in which controlled experiments are not
possible and the data has not been generated by what could be considered as
a natural experiment? As we noted above, correlation and regression analysis
are typically not su¢ cient to identify causal relationships in the data, as we
have no guarantee that the the necessary exogeneity assumptions apply. In
particular, in this case several classical problems plague regression analysis.
We …rst illustrate the most important of them. We shall then argue that
the combination of models and statistical methods, which is called structural
econometrics, can be used to produce a useful inference methodology.

11.4.1 Problems with regression analysis

What happens if we run a regression on data (xi ; zi ; yi )N
i=1 even though (xi ; zi )
are not exogenous? Here are some classic examples.

Identi…cation
Suppose you run a regression

yi = + x xi + ui
where yi is the aggregate quantity demanded of a speci…c commodity - say,
oil - at time i, and xi is the merket price of oil at time i: Suppose you …nd
that the parameter x is positive. Do you conclude that the demand for oil
is positively sloped, that consumers are irrational (or al least have strange
preferences), that economics is useless, and you pass to sociology? No. The
fact is that the parameter x is not identi…ed, you might be picking up supply
rather than demand. Here’s the explanation.
The observed relationship between yi and xi in the data is an equilibrium
relationship. It’s the outcome of

yid = d + dx xi + udi
yis = s + sx xi + usi
yi = yid = yis

where the apex d (resp. s) indicates demand (resp. supply). As a conse-

quence, we do not know if variation in (xi ; yi ) over time i is due to variation
11.4 STRUCTURAL ECONOMETRICS 203

of udi or usi or both. To make the contrast as stark as possible, consider the
case in which usi = 0, for any i = 1; :::; N: In this case the variation is due to
demand and the regression we have run estimates the supply function, which
is in fact naturally positively sloped. The …gure should illustrate.
Formally, the solution of the demand and supply system is:

d s
+ udi usi
xi = s d
x x
s s
yi = + x xi + usi :

Suppose that sx > 0 and dx < 0, so that demand as a negative slope and
supply a positive one, as to be expected. But suppose that usi = 0, for
d s +ud
any i = 1; :::; N: Then xi = s d
i
and yi = s + sx xi : It follows that
x x
@xi @yi s @xi d
@udi
> 0 and @u d = x @ud > 0: As a consequence positive demand shocks ui
i i
are associated with higher prices xi and higher quantities yi ; resulting in a
positive slope in the regression, as in the …gure. Convince yourself that, if
udi = 0, for any i = 1; :::; N; we would …nd a negative slope in the regression.
In fact, nothing can be said about the parameters of interest, the slope of the
demand and the supply unless we observe udi ; usi : We say that the slope of
the demand and the supply are not identi…ed.

Selection

Suppose you run a regression

yi = + x xi + ui

where yi is the monthly wage of individual i and xi is a dummy variable

equal to 1 if the individual has a college degree and 0 otherwise. Suppose you
estimate x = 2; 347: Can you conclude that the return of going to college
is on average 2; 347 dollars a month? The answer is once again negative.
The reason is that the population of those who attend college is self-selected:
they are the children of relatively rich parents who have connections and …nd
them better jobs. (Ok, another possible selection argument - just in theory
- is that those who go to college are smarter and this is why they get higher
wages).
204 CHAPTER 11 EMPIRICAL ECONOMICS
11.5 PROBLEMS 205

Reverse causality
Suppose you run a regression

yi = + x xi + ui
where yi is the growth rate of country i and xi is the average income tax-
rate in country i: Suppose you …nd a positive estimate for x : Can you
conclude that lowering taxes induces higher growth rates? The answer is
again negative. The reason is that causation could be reversed, it could well
be that it is growth rates which causes taxes, e.g., because a state government
whose economy grows fast might a¤ord lower taxes. Remember: regressions
only document correlation, not causation!

11.4.2 Back to models

Well, what do we do then if we do not have a natural (nor a controlled)
experiment and if regressions are problematic? Well, we go back to models.
Exogeneity is a property of variables in the model, never independently of the
model. Empirical analysis is only as good as the model is: empirical results
are conditional to the model. Statistical inference procedures (including re-
gressions) assume validity (or lose it) in the context of a model. We cannot
pronounce statements like,
@y @y
x and z cause y and in our data @x
= 1:3; @z
= 2:9:

We can only say,

conditionally on the model yi = + x xi + z zi + ui ; in our data

@y
@x
= 1:3; @y
@z
= 2:9:

Nothing we can do. We ought to recognize our limitations (in statistical

inference) and live with them.

11.5 Problems
Problem 11.5.1 Acemoglu, Cantoni, Robinson (2009) ask if Napoleon can
be considered a modernizing force in the territories he conquered. They use
206 CHAPTER 11 EMPIRICAL ECONOMICS

Germany as a test case, asking how regions controlled by Napoleon fared com-
pared to those that were never under French rule. Occupied areas that kept
the reforms Napoleon put in place (a written civil code; abolition of serfdom
and, in the cities, of guilds that hampered economic freedom) subsequently
experienced the most rapid economic development. Areas that the French
never controlled developed less quickly (measured in terms of urbanization),
and those regions that were invaded, but returned to their old ways after lib-
eration, developed most slowly of all. Is this a natural experiment? Discuss.
Chapter 12

Game theory

Oskar Morgernstern
and John von
Neumann John Nash John Harsanyi Reinhard Selten Robert Aumann

Game theory is the study of interacting decision-makers. In earlier sec-

tions we studied the theory of the consumer. That is, the optimal decision by
a single decision-maker, who is called the consumer, in a simple environment
in which she is a price taker. Thus, Game theory, is a natural generalization
of the single decision-maker theory which deals with how a utility maximizer
behave in a situation in which her payo¤ depends on the choices of another
utility maximizer. Many …elds, such as sociology, psychology and biology,
study interacting decision-makers. Game theory focus on rational decision-
making, which is the most appropriate model for a wide variety of economic
contexts.

207
208 CHAPTER 12 GAME THEORY

12.1 Strategic Games

A strategic game is a model of interactive decision-making in which each
decision-maker chooses his plan of action, his strategy.
Any strategic game is de…ned by:

(i) a (…nite) set of players, I = f1; :::; mg;

(ii) a (…nite) set of (pure) strategies which are the choices that each player
can make: si 2 S i is a pure strategy of player i; S = S 1 S m is
the strategy space and s 2 S is a strategy pro…le.

(iii) a set of payo¤s that indicates the utility that each player has in any
combination of strategies, ui : S ! R.

Consider 2 2 games - 2 players and 2 pure strategies to each player. For

such a game, the following bi-matrix representation is very convenient.

Representation of a 2x2 game

12.1 STRATEGIC GAMES 209

Player 1’s actions are identi…ed with the rows and the other player by the
columns. The two numbers in a box formed by a speci…c row and column
are the players’payo¤s given that these actions were chosen. For example, in
the game above A and B are the payo¤s of player 1 and player 2 respectively
when player 1 is choosing strategy T and player 2 strategy L.
Applying the de…nition of a strategic game to the game above yields:
I = 1; 2 - the set of players
S 1 = fT ,Bg, S 2 = fL,Rg and S = S 1 S 2 = f(T ,L), (T ,R), (B,L), (B,R)g
Payo¤s are given by the bi-matrix.

12.1.1 Nash equilibrium

A strategy pro…le s is Nash equilibrium if, for any player i 2 I,

ui (si ; s i ) ui (si ; s i )

for all si 2 S i .
In other words, at a Nash equilibrium no player can pro…tably deviate
from his strategy, given all other players strategies. The notion of a best
response is most useful in …nding the set of Nash equilibria.
Player i’s best response to a strategy sj of player j is de…ned as

BRi sj 2 S j = (si 2 S i ) : ui (si ; sj ) ui (si0 ; sj )

A Nash equilibrium has the property that

sj = BRj BRi sj 2 S j

12.1.2 Examples
The following are classical games representing a variety of strategic situations.
Two people wish to go out together to a ball game or to a concert. Their
main concern is to be together, but one of them prefers the game and the
other the concert.
210 CHAPTER 12 GAME THEORY

Battle of the sexes

The game has two Nash equilibriums in pure strategies: (G, G) and (C,C)
which is written as

N (G) = f(G; G); (C; C)g

Coordination game

As in the BoS, the two players wish to go together but this time they agree
that to see a ball game is much better.
12.1 STRATEGIC GAMES 211

Coordination

Find the set of Nash equilibrium N (G). In addition, discuss its reason-
ableness.

The prisoner dilemma

Two suspects in a crime are under a police investigation (in di¤erent cells).
If they both confess, each will be sentenced for three years. If one of them
doesn’t confess and the other does, the …rst one will be used as a witness
against the other, who will receive a sentence of four years. If both don’t
confess, they will be convicted in minor o¤ense and spend only one year in
prison. As it is standard, we shall encode actions in a confusing way. Action
C is not "confess" but rather "cooperate" with the other prisoner, that is,
"don’t confess". It gets worse: action D is "defect", that is "confess".
212 CHAPTER 12 GAME THEORY

Prisoner Dilemma

Find the set of Nash equilibrium N (G).

Cournot Oligopoly
Two …rms compete on the quantity they produce of good a single commodity.
They face a demand function
p = f (x)
where p is the price at which they will sell the good, which depends on the
total quantity produced, x = x1 + x2 (xi is the quantity produced by …rm
i = 1; 2). Let the demand be linear:
p= (x1 + x2 )
The production cost for …rm i is
c(xi ) = (xi )2
and its revenues are
pxi :
12.1 STRATEGIC GAMES 213

12.1.3 Mixed Strategies

Some games do not have a Nash equilibrium in pure strategies, as we de…ned

it up to now. In those games we study mixed strategies, constructed as
follows. Let S i = fsi1 ; : : : ; siN g denote the pure strategies of agent i. The
corresponding set of mixed strategies is the set of all probability vectors

X
N
i i i i
=f 1; : : : ; N g; n =1
n=1

where in takes the interpretation of the probability that the pure strategy
sin is played by player i.

Matching pennies

Each of the two players chose either head or tail. If they choose di¤erently
player 1 pays $1 to player 2 and if they choose the same player 2 pays $1 to
player 1. This kind of a game is also called strictly competitive. Moreover,
this game is also a zero sum game.
214 CHAPTER 12 GAME THEORY

12.2 Dynamic Games

Consider a game in which decisions occur over time. A typical example is

the entry game, in the …gure.

Entry
12.3 BARGAINING 215

12.3 Bargaining

Ariel Rubinstein

12.4 Reputation and repeated games

Drew Fudenberg David Levine

216 CHAPTER 12 GAME THEORY

12.5 Problems
Problem 12.5.1 Consider the following 3 2 game. Find …rst all (if any)
pure strategy Nash equilibria of this game. Then look for all mixed strategy
equilibria.

Problem 12.5.2 Consider a duopoly (an oligopoly with two …rms), i = 1; 2:

They are symmetric and produce the same good. Each …rms i chooses the
dollar price she charges, pi ; and then produces to satisfy the demand at this
price. Production of xi units involve dollar costs equal to ci = xi : Demand
d comes from a representative agent and is a declining function of the price
p the agent is charged:
d = f (p):
The agent chooses the …rm he patronizes. Write down the pro…t function of
…rm 1 as a function of her own and …rm 2 price. Write down the pro…t max-
imizing price p1 as a function of p2 (use also your economic understanding
of the problem; the mathematical maximization problem is not standard!)
Exploiting symmetry, write down the pro…t maximizing price p2 as a function
of p1 . Compute the unique Nash equilibrium of the game.
Chapter 13

Mathematical Appendix

We collect here some useful mathematical preliminaries.

13.1 Sets
A set is any well-speci…ed collection of elements. A set may contain …nitely
many or in…nitely many elements. For example

A = f1; 2; 3g

B = f1; 2; 3; :::g
The term element is left as an abstract concept. For our purposes an element
will usually be a number, but there is no need to restrict it to be so in its
de…nition. The notation a 2 A means a is a member of the set A. Conversely,
a2= A means a is not a member of the set A.
The following is an example of standard notation used to de…ne sets:
C := fx 2 A : x > 1g. This means the set C is de…ned as the elements in A
that are larger than 1 (i.e. C = f2; 3g). A set that contains no elements is
called the empty set or null set and is denoted by ?. Let A, B, be the sets
de…ned above. The set A is contained in B (i.e. all the elements of the set
A are also in the set B) and is called a subset of B. This is written A B.
This can also be written the other way round, B A and reads "B is a
superset of A".
Using this de…nition two set, say A and A0 , are equal if A A0 and
A0 A. A set A is a proper subset of a set A0 if A A0 and A0 * A. This is

217
218 CHAPTER 13 MATHEMATICAL APPENDIX

denoted A A0 . Given two sets, say A and B, new sets can be formed using
the following operations
Union: A [ B := fx : x 2 A or x 2 Bg
Intersect: A \ B := fx : x 2 A and x 2 Bg
Set subtraction: AnB := fx : x 2 A and x 2 = Bg
If it is clear that the sets under discussion are subsets of some ("univer-
sal") set, say U, then it is possible to de…ne the operation complement:
Ac := fx 2 U : x 2 = Ag

13.2 Numbers
The most basic numbers are the counting numbers f1; 2; 3; :::g, usually called
the natural numbers and denoted by the script letter N. The natural numbers
augmented with the number zero and all the negatives of the natural numbers
is called the set of integers, and is denoted Z = f:::; 3; 2; 1; 0; 1; 2; 3; :::g.
Since the sum, di¤erence and product of two integers is another integer but
the quotient of two integers is not in general, the natural extension is to the
set that includes such values. This leads to the de…nition of the rational
numbers. This is the set of all numbers that can be formed by the quotient
of an element of the integers with an element of the natural numbers, i.e.
na o
Q= : a 2 Z; b 2 N
b
Things do not end at the rationals. There is a sense in which the set of
rationals has a great deal of "holes", numbers that p cannot be written as a
quotient of an integer and a natural number (e.g. 2; 2 = Q). This leads to
the set of real numbers, denoted R. The construction of the real numbers is
not for this class and as such there will be no formal de…nition of the reals.
Some commonly used sets of numbers:
R+ = fx 2 R : x 0g, called the non-negative real number;
R++ = fx 2 R : x > 0g, called the positive real numbers;
[0; 1] = fx 2 R : 0 x 1g, called the closed interval between zero and
one;
(0; 1) = fx 2 R : 0 < x < 1g, called the open interval between zero and
one;
The sets R ; R ; (a; b); [a; b]; [a; b); (a; b] are similarly de…ned, where a; b 2
R such that a > b:
13.3 VECTORS 219

13.3 Vectors
0 1
x
B 1C
B C
B x2 C
A vector of size n is an n-tuple which in general is written asB
B
C.
C
B ::: C
@ A
xn
0 1
x1
For example, if n = 2 the general form of the vector is @ A. If x1 2 R
x2
0 1
x1
and x2 2 R, then @ A 2 R2 where R2 = R R. The " " operator is the
x2
Cartesian Product.
Set notation, vectors and graphs: Here are some examples of sets of
vectors (of size 2) that can be represented on a 2-D graph
R+
R2++
f(x1 ; x2 ) 2 R2+ : x2 m + bx2 g

13.4 Functions
A function is a rule that describes a relationship between numbers. For each
number x a function assigns a unique number y according to some rule. The
general notation, when the speci…c algebraic rule is not known, is y = f (x)
where the number x is called the independent variable and the number y is
called the dependent variable.
Often some variable y depends on several other variables, say x1 ; x2 . In
this case, we write y = f (x1 ; x2 ) to indicate that both variables together
determine the value of y. To formalize matters, a function maps each element
of a set, called the domain of the function, into a unique element of another
set, called the co-domain. This is written f : D ! C, where D is the domain
and C is the co-domain. The function is only de…ned for the elements of the
domain. Related to the co-domain is the range. This is the set of all values
that the function takes, fy 2 C : 9x 2 D such that f (x) = yg (NB: the
symbol "9" is common short hand for "there exists"). The range is always a
220 CHAPTER 13 MATHEMATICAL APPENDIX

subset of the co-domain.

The functions that are used in this book are real valued functions. These
are functions whose co-domain is a subset of the real numbers, i.e. C R.
This hand-out will focus at …rst on real valued functions on the real line,
referred to as univariate functions. Here the domain is a subset of the real
numbers, i.e. D R. Later we will deal with multivariate functions, where
D Rn for n 2 N.
Some examples of real valued functions on the real line (for each of the
functions below you should think of suitable domains, co-domains and the
corresponding ranges):

- The simplest function is called a monomial and is written y = xk where

2 R and k 2 N is called the degree of the monomial. A polynomial
is a function which is formed by adding monomials. For example a
polynomial of degree k = 2 is written y = a2 x2 + a1 x + a0 . More
Xk
generally a polynomial of degree k is y = ai x i :
i=0

x5 +1
- A rational function is a ratio of polynomials. For example, y = x 5
:

- A polynomial of degree k = 1 is a linear function. It has the form

f (x) = mx + b

where m; b 2 R. Note that a polynomial of degree k = 0 is also a linear

function which assigns the same number for any choice of x. It is called
a constant function.

13.4.1 Some Properties of Functions

Increasing and decreasing functions:
A function f (x) is an increasing function if

x1 > x2 =) f (x1 ) f (x2 )

and a strictly increasing function if

x1 > x2 =) f (x1 ) > f (x2 )

13.4 FUNCTIONS 221

A function f (x) is a is a decreasing function if

x1 > x2 =) f (x1 ) f (x2 )

and a strictly decreasing function if

x1 > x2 =) f (x1 ) < f (x2 )

Concavity and convexity:

A function of one variable is concave if

f (tx + (1 t)x0 ) tf (x) + (1 t) f (x0 )

for all x and x0 and all 0 t 1.

A function of one variable is strictly concave if

f (tx + (1 t)x0 ) > tf (x) + (1 t) f (x0 )

for all x and x0 such that x 6= x0 and all 0 < t < 1.

A function of one variable is convex if

f (tx + (1 t)x0 ) tf (x) + (1 t)f (x0 )

for all x and x0 and all 0 t 1.

A function of one variable is strictly convex if

f (tx + (1 t)x0 ) < tf (x) + (1 t)f (x0 )

for all x and x0 such that x 6= x0 and all 0 < t < 1.

A function f : Rn+ ! R is homogeneous of degree k if

f (tx) = tk f (x) for all t > 0:

Note that if we double all the arguments of a function that is homogeneous

of degree zero (k = 0), the value of the function doesn’t change.
A function f : Rn+ ! R is continuous when "the graph can be drawn
without lifting a pencil from the paper".
222 CHAPTER 13 MATHEMATICAL APPENDIX

The slope
The slope of a linear function f (x) = mx + b is the ratio xy11 yx00 where the
points (x0 ; y0 ) and (x1 ; y1 ), are any arbitrary points. Note that a linear
function has a constant slope at all points along its graph. Moreover, f (x) =
mx + b the line has the slope m and its y-intersect is at (0; b).
The slope of any function , linear or non-linear, is its derivative, written
f (x0 ) or dfdx
0 (x)
. Note that only the derivative of a linear function is the same
for any x. On the other hand, a nonlinear function has the property that its
slope changes as x changes. A tangent to a function at some point x0 is a
linear function that has the same slope.

Derivatives
If a function is di¤erentiable at a point x0 , the derivative of a function y =
f (x) at the point x0 is de…ned to be

df (x0 ) f (x0 + 4x) f (x0 )

= lim
dx 4x!0 4x
In words, the derivative is the rate of change of y with respect to x as
the change in x goes to zero. The derivative gives the precise meaning to the
phrase ”the rate of change of y with respect to x for small changes in x”.
For any integer k the derivative of f (x) = xk at a point x = x0 is f 0 (x0 ) =
k(x0 )k 1 . Now, assume that k is an arbitrary constant and f and g are
continuous di¤erentiable functions at x = x0 . Then,

(f g)0 (x0 ) = f 0 (x0 ) g 0 (x0 )

(kf )0 (x0 ) = kf 0 (x0 )
(f g)0 (x0 ) = f 0 (x0 )g(x0 ) + g 0 (x0 )f (x0 )
f 0 (x0 )g(x0 ) g 0 (x0 )f (x0 )
(f =g)0 (x0 ) = g(x0 )2
0
((f (x0 ))n ) = n(f (x0 ))n 1 f 0 (x)
(f g)(x0 ) = f 0 (g(x0 ))g 0 (x0 )

Examples
(i) f (x) = 5x3
3=2
(ii) f (x) = 3x
13.4 FUNCTIONS 223

(iii) f (x) = 3x2 9x + 7x2=5

(iv) f (x) = x+1
xp1
(v) f (x) = 12 x
(vi) f (x) = (x2 + 3x + 1)(x3 2x2 + 5)
A second derivative of a function of one variable is just the derivative of
the derivative. It is denoted as follows

d2 f (x) d dfdx
(x)
=
dx2 dx
Examples
(i) f (x) = 5x3
(ii) f (x) = 3x 3=2
If a function has more than one argument, then we can take the partial
derivative. This measures how the value of a function changes as one of
its arguments changes, keeping all other arguments constant. Formally the
partial derivative of the function y = f (x1 ::::xn ) with respect to xi at point
y = f (x1 ::::xn )is de…ned as

@f (x1 ::::xn ) f (x1 ; :::; xi + 4xi ; :::; xn ) f (x1 ::::xn )

= lim
@xi 4xi !0 4xi
In calculating partial derivatives, the key is to remember to treat the
other arguments as constants, then use the rules above
Examples
(i) f (x; w) = 4x2 w 3xw3 + 6x
(ii) f (x; w; z) = 3x 3=2 z 4 + w2 z 2
One important concept that uses the notion of partial derivatives is that
of the total derivative, which relates changes in the arguments of a function
to changes in the value of the function. For the di¤erentiable function y =
f (x; z)

@f (x; z) @f (x; z)
dy = dx + dz
@x @z

The Isoquant
An isoquant is like a contour on a map. It is the set of points in the domain
of a function for which the function is equal to some constant value. There
will be an isoquant associated with each value in the range of the function.
224 CHAPTER 13 MATHEMATICAL APPENDIX

Formally, the isoquant associated with the element y in the range of the
function f (x1 ; :::; xn ) we will denote as I(y)

I(y) = fx1 :::xn : f (x1 ::::::xn ) = yg

We can use the total derivative to calculate the slope of the isoquant. We
can do this because we know that, on the isoquant, the value of the function
does not change. Consider a function with two arguments f (x1 ; x2 ), and the
isoquant f (x1 ; x2 ) = y. We can write

@f (x1 ; x2 ) @f (x1 ; x2 )
dy = dx1 + dx2 = 0
@x1 @x2
which we can rearrange to give
@f (x1 ;x2 )
@x1 dx2
@f (x1 ;x2 )
=
dx1
@x2

13.4.2 Optimization
Say we have a function f : R ! R, such that y = f (x). How can we …nd out
which value of x will maximize y? First we ask the question: what properties
will such a maximum have? First some notation. We describe the above
problem as maximising y with resepect to x. If y is the maximum value of
f (x) then we write y = maxx f (x). If x is the value of x that maximises
f (x) we write x = arg maxx f (x)
Not all functions will have a maximum. For example, f (x) = x2 does
have a maximum, while f (x) = x does not (see …gure 1). However, if the
function has an interior maximum, then the slope of the function at that
point will be zero (see …gure 2). This is called the …rst order condition of
the maximization problem
Let f : R ! R be a di¤erentiable function. If x is the argmax of f , then
f (x)
=0
x

We have stated that if x is the argmax of f the slope of f at x must by

0. In other words, it is a necessary condition of x being the argmax that the
slope is zero at that point. Is it also true that, if the slope of f at some point
13.4 FUNCTIONS 225

x is zero, then x is the argmax of f ? In other words, is this a su¢ cient

condition? The answer is no. If the slope is zero, then the point we are
looking at could be a maximum, minimum or neither - see …gure 3. We can
di¤erentiate between these cases by looking at the second derivative of the
function. This measures the ’slope of the slope’of the function.
Let f : R ! R be a twice-di¤erentiable function. Let x be a critical point
of f , (i.e. f (x)
x
= 0 ) then if:
d2 f (x)
dx2
> 0 then x is a minimum point
d2 f (x)
dx2
< 0 then x is a maximum point
d2 f (x)
dx2
= 0 then x can be a maximum point, minimum point or a point
of in‡ection
2
Note that, for twice di¤erentiable functions, d dx
f (x)
2 > 0 implies that the
2
d f (x)
function is convex, and dx2 < 0 implies that the function is concave (at
that point). This leads to a re…nement of theorem 1.
Let f : R ! R be a twice di¤erentiable function if x is the argmax of f ,
then
f (x)
a: x
=0
d2 f (x)
b: dx2
0
Condition b is know as the second order condition of the maximization
problem. This is a stronger set of necessary conditions, but are they now
su¢ cient conditions? Again the answer is no. The reason is that the condi-
tions are enough to …nd a maximum of the function, but not necessarily the
maximum of the function. Figure 4 highlights the di¤erence between a local
and a global maximum.
To get su¢ cient conditions, we need something stronger. For example,
we could demand that our candidate point is the only critical point.
Let f : R ! R be a twice di¤erentiable function if there exists a point x
such that
f (x)
a: x
=0
d2 f (x)
b: dx2
0
f (x)
c: x
6= 0 8 x 6= x
Then x is the global argmax of f
226 CHAPTER 13 MATHEMATICAL APPENDIX

Unconstrained optimization - many variables

We can …nd analogous results to those above when maximizing a function of

many variables. Just as in the single variable case, the slope of the function
must be ‡at at an interior maximum. However, now it has to be ‡at with
respect to all the arguments of the function. In other words, all the partial
derivatives of the function will be zero at the interior maximum.
Let f : Rn ! R be a di¤erentiable function. Let x1 ::::xn be the argmax
of f then
@f (x1 ;::::;xn )
@xi

Often in economics we will be interested in optimizing a function subject

to some constraints. For example, we might want to know how a consumer
can optimize utility within a given budget set, or how a …rm can maximize
output with given labour and capital inputs. While such problems are many
and varied, they all have the same setup.
A constrained optimization problem is a problem of the following form:
Choose x1 :::::xn (arguments)
to maximize f (x1 :::::x) (objective function)
subject to h1 (x1 :::::x) = 0;
..
.
hI (x1 :::::x) = 0 (equality constraints)
g1 (x1 :::::x) 0;
..
.
gJ (x1 :::::x) 0 (inequality constraints)
How do we solve such problems? We can gain some intuition from looking
at the graph of a simple problem. Say we have only two variables and one
equality constraint. In other words, we want to solve the problem:
choose x1 ; x2 (arguments)
to maximize x1 x2 (objective function)
subject to x1 + x2 1 = 0:
Because this problem is in two dimensions, we can graph it, which we do
in …gure 5. The solid line shows the constraint, while the dotted line shows
the isoquants of the objective function
It is clear that the maximum occurs at the point where the slope of the
isoquant of the objective function is equal to the slope of the constraint
function. As we know from class 1, we can write this in terms of partial
13.4 FUNCTIONS 227

derivatives.
@f (x1 ;x2 ) @h(x1 ;x2 )
@x2 @x2
@f (x1 ;x2 )
= @h(x1 ;x2 )
@x1 @x1

The trick we use to …nd these points is to rewrite the problem as a La-
grangian Function.

L(x1 ; x2 ; ) = f (x1 ; x2 ) h(x1 ; x2 )

Under the right conditions,…nding the unconstrained optimum of this func-

tion is the same as solving the constrained optimum.
To solve the unconstrained optimization problem, we take …rst derivatives
with respect to the three arguments and set them equal to zero:
@L(x1 ; x2 ; ) @f (x1 ; x2 ) @h(x1 ; x2 )
= =0
@x1 @x1 @x1
@L(x1 ; x2 ; ) @f (x1 ; x2 ) @h(x1 ; x2 )
= =0
@x2 @x2 @x2
@L(x1 ; x2 ; )
= h(x1 ; x2 ) = 0
@
Where the third equation simply repeats our constraint. These are the …rst
order conditions of the constrained optimization problem. We can see that
this uses the …nding we made above about the ratio of the partial derivatives.
Taking the ratio of the …rst two …rst order conditions gives us the condition
that we stated above.
In the case of our example, our …rst order conditions become

@L(x1 ; x2 ; )
= x2 =0
@x1
@L(x1 ; x2 ; )
= x1 =0
@x2
@L(x1 ; x2 ; )
= x1 + x2 1 = 0
@
Using the …rst two conditions tells us that x2 = x1 , and substituting this
into the third condition gives x2 = x1 = 12 . This is the solution to the
constrained optimization problem. If we have more constraints, we can add
more Legrange multipliers. So consider the problem:
228 CHAPTER 13 MATHEMATICAL APPENDIX

Choose x1 :::::xn (arguments)

to maximize f (x1 :::::x) (objective function)
subject to h1 (x1 :::::x) = 0;
h2 (x1 :::::x) = 0
h3 (x1 :::::x) = 0 (equality constraints)
We would form the Lagrangian

L(x1 ::xn ; 1; 2; 3) = f (x1 :::xn ) 1 h1 (x1 :::; xn ) 2 h2 (x1 :::; xn ) 3 h3 (x1 :::; xn )

and have the …rst order conditions

@L(x1 ::xn ; 1 ; 2 ; 3 ) @f (x1 :::xn ) @h1 (x1 :::xn ) @h2 (x1 :::xn )
= 1 2 +
@xi @xi @xi @xi
@h3 (x1 :::xn )
3 = 0
@xi
8i 2 1::::n
@L(x1 ::xn ; 1 ; 2 ; 3 )
= h1 (x1 ; x2 ) = 0
@ 1
@L(x1 ::xn ; 1 ; 2 ; 3 )
= h2 (x1 ; x2 ) = 0
@ 2
@L(x1 ::xn ; 1 ; 2 ; 3 )
= h3 (x1 ; x2 ) = 0
@ 3
This leads us to the following theorem.
Let x1 :::::xn maximize f (x1 :::::xn ) subject to h1 (x1 :::::xn ) = 0; :::::; hj (x1 :::::xn ) =
1
0 .Then there exists a set of j numbers 1 ::: j such that fx1 :::::xn ; 1 ::: j gis
a critical point of the Lagrangian

L(x1 ::xn ; 1 ::: j ) = f (x1 :::xn ) 1 h1 (x1 :::; xn ) ::: j hj (x1 :::; xn )

As in the case of unconstrained optimization, the is a necessary but not

su¢ cient condition for …nding the constrained optimum. We have similar
worries about …nding minima instead of maxima, and local rather than global
extreme points. There are second order conditions for constrained optimiza-
tion problems. However, we will not cover them here.
What about inequality constraints? This makes the problem more dif-
…cult, as we do not know in advance whether the constraint will bind or
1
This theorem should really have another condition, that x1 :::xn satisfy the non-
degenerate constraint quali…cation. However, we will skip over this for now.
13.4 FUNCTIONS 229

not at the maximum. If the constraint does not bind, then it is not really
a constraint at all. If it does bind, then we can treat it as we would an
equality constraint. While there are techniques for dealing with inequality
constraints, in some cases it may be obvious whether or not a constraint will
bind. For the purposes of this course, we will be able to use these ’common
sense’methods, rather than grinding through the tedious maths.

13.4.3 Solving the Cobb-Douglas Consumer’s Problem

The second problem of section 1.4 of the second maths refresher homework
is a classic example of a consumer maximization problem. You will see this
problem, or a version of it, time and time again. It is therefore worth getting
familiar with it now, as it will save you lots of time later on. We will also
use it here to demonstrate the 5 steps that, quite often, will lead you to the
solution of a constrained optimization problem. While its not advisable to
think of maths in a cookbook fashion, you can treat this as such as a last
resort. First, to restate the problem:
Maximize x y (1 ) subject to px x + py y = I.(Note that ; px ; py I are
parameters for this problem your answer should be a function of x and y in
terms of these values)
The interpretation of this problem is as follows: A consumer is has to
choose between bundles of good x and good y in order to maximize her
utility function x y (1 ) . However, she only has a certain amount of income
to spend, I. The price of the two goods is denoted by px and py . So the
consumer is trying to choose the bundles to maximize her utility function
subject to the bundles being those that she can a¤ord. As stated above,
; px ; py I are what we call parameters of the problem. In other words, they
are the ’rules of the game’, or things that the consumer cannot alter. Your
solution to the problem will be a function of these parameters.
The cookbook method of solving these problems has 5 steps, described
below:

Form the Lagrangian function

This always takes the form

L(x1 ; x2 ; ) = f (x1 ; x2 ) h(x1 ; x2 )

230 CHAPTER 13 MATHEMATICAL APPENDIX

Where is the lagrangian multiplier, f (x1 ; x2 ) is the objective function

and h(x1 ; x2 ) is the constraint. In this case, we get:

L(x; y; ) = x y (1 )
(px x + py y I)

Take partial derivatives and set them equal to zero

I’ve tried to convince you that solving the constrained optimization problem
is the same as solving the unconstrained problem for the lagrangian. If you
are solving an unconstrained optimization problem, then the …rst thing we
do is take the partial derivatives and set them equal to zero. In general:

@L(x1 ; x2 ; ) @f (x1 ; x2 ) @h(x1 ; x2 )

= =0
@x1 @x1 @x1
@L(x1 ; x2 ; ) @f (x1 ; x2 ) @h(x1 ; x2 )
= =0
@x2 @x2 @x2
@L(x1 ; x2 ; )
= h(x1 ; x2 ) = 0
@

And in this speci…c case:

@L(x; y; )
= x 1y1 px = 0
@x
@L(x; y; )
= (1 )x y py = 0
@y
@L(x; y; )
= px x + py y I = 0
@

Divide the FOC from the choice variables by each other to get rid
of
We can always rewrite our partial derivatives from the choice variables by
taking one term over to the other side of the equality. In general:

@f (x1 ; x2 ) @h(x1 ; x2 ) @f (x1 ; x2 ) @h(x1 ; x2 )

= 0) =
@x1 @x1 @x1 @x1
@f (x1 ; x2 ) @h(x1 ; x2 ) @f (x1 ; x2 ) @h(x1 ; x2 )
= 0) =
@x2 @x2 @x2 @x2
13.4 FUNCTIONS 231

We can then divide these equations by each other, giving

@f (x1 ;x2 ) @h(x1 ;x2 ) @f (x1 ;x2 ) @h(x1 ;x2 )
@x1 @x1 @x1 @x1
@f (x1 ;x2 )
= @h(x1 ;x2 )
) @f (x1 ;x2 )
= @h(x1 ;x2 )
@x2 @x2 @x2 @x2

This gets rid of the from the top and bottom of our equation.
In this speci…c case, we get

1 1 1 1
x y px = 0 ) x y = px
(1 )x y py = 0 ) (1 )x y = py

And so
1 1
x y px x 1y1 px
= ) =
(1 )x y py (1 )x y py

Solve for x in terms of y (or viceversa)

This equation should now be in terms only of x; y and the parameters of
the model, as we have got rid of . In the case of Cobb-Douglas Utilities,
something magical happens. Note that, on the left hand side of the equation,
the powers of on x and y cancel, leaving a linear model!
1 1
x y y
=
(1 )x y (1 )x
px
y
) =
(1 )x py
(1 )x px
) y=
py

Substitute into the constraint and solve

At present we have two unknowns. That means that we have to …nd another
equation so we can solve the system. Luckily, we have such and equation -
the constraint. We can now substitute the above into this equation

px x + py y = I
(1 )x px
px x + py ( ) = I
py
232 CHAPTER 13 MATHEMATICAL APPENDIX

but here we can cancel the py terms, to give

(1 )
px x + px x = I

We can neaten this up a bit by multiplying the left most term top and
bottom by

(1 )
px x + px x = I
px x + (1 )px x
= I
px x
= I
I
x =
px

Similar work will tell you that

I(1 )
y=
py

We now have an expression for x and y in terms only of the parameters,

so we have solved the model. Victory!. Note that this can be rewritten as
px x = I where the left hand side of the equation is just the amount spent
on good x. This tells us that, with these preferences, the amount spent on
good x will be a constant fraction of income, whatever the price!

13.5 Linear algebra

Linear algebra is about the solution of

Ax = b
13.5 LINEAR ALGEBRA 233
2 3 2 3
a11 a1n x1
6 7 6 7
6 7 6 7
6 7 6 7
6 7 6 7
6 .. . . .. 7 6 .. 7
where A = 6 . . . 7 is an (m n) matrix, x = 6 . 7 is an (n 1)
6 7 6 7
6 7 6 7
6 7 6 7
4 5 4 5
am1 amn xn
2 3
b
6 1 7
6 7
6 7
6 7
6 .. 7
vector, and b = 6 . 7 is an (m 1) vector.
6 7
6 7
6 7
4 5
bm
The equation
Ax = b

can also be represented by the system of equations

a11 x1 + ::::: + a1n xn = b1

=
am1 x1 + ::::: + amn xn = bm

and by the equation

a1 x 1 + + an x n = b

where a1 ; :::; an are the columns of the matrix A:

13.5.1 Reduction to echelon

The augmented matrix [A j b] can be reduced to echelon and reduced echelon
form.
Every row [0 0 j 0] in the echelon form corresponds to a free variable.
A row [0 0 j bm > 0] in the echelon form implies inconsistency.
234 CHAPTER 13 MATHEMATICAL APPENDIX

13.5.2 Independent columns

The study of
Ax = b; for any b

depends crucially on the number of linear independent column of the matrix

A: Let span(A) denote the set of possible vectors b such that Ax = b for some
vector x: Then span(A) is a subspace of dimension equal to the number of
independent columns of A: The number of independent columns of A cannot
be larger than min fm; ng :

i) If span(A) = Rm , then
Ax = b

has a solution for any b: In this case,

- if n = m; the solution has no free variables and the linear transformation

x ! T (x) is one-to-one and onto; equivalently, Ax = 0 has only the
trivial solution x = 0;

- if n > m; the solution has n m free variables and the linear transformation
x ! T (x) is onto but NOT one-to-one; equivalently, Ax = 0 has an
in…nity of solutions.

ii) If span(A) = Rk , k < m, then

Ax = b

does not have a solution for some b - not enough independent variables
to satisfy the m equations. In this case, the linear transformation
x ! T (x) is NOT onto, but

- if k = n it is one-to-one (the solution has NO free variables); Ax = 0 has

only the trivial solution x = 0;

- if k < n, it is NOT one-to-one (the solution has free variables); Ax = 0

has an in…nity of solutions.
13.6 PROBLEMS 235

13.5.3 Inverse matrix

1
A square (n n) matrix A has an inverse A such that
1
A A = In

if and only if span(A) = Rn , that is, the matrix A has n independent columns.
If A has an inverse then, Ax = b has a unique solution for any b, only
the the trivial solution for b = 0; equivalently, the linear transformation
x ! T (x) is one-to-one and onto.
The general solution of
Ax = b
in this case is
x = A 1 b:

13.5.4 Determinant
A square (n n) matrix A has an inverse A 1 if and only if det(A) 6= 0:

13.6 Problems
13.6.1 Sets and numbers
p
Let A = f1; 2; 3::::::10g, B = f0; 0:5; 2; 2; 10g; R be the real numbers, N be
the natural numbers, Z be the integers and Q be the rationals. In each case
below, describe the content of the set C

1. C = fx 2 A : x > 6g
This is all the numbers in set A that are strictly greater than 6. In
other words C = f7; 8; 9; 10g

2. C = A [ B
All the elements
p that are either in set A or set B (or both). C =
f0; 0:5; 1; 2; ; 2; 3; 4; 5; 6; 7; 8; 9; 10g

3. C = A \ B
All the elements that are in both set A and set B. C = f2; 10g
236 CHAPTER 13 MATHEMATICAL APPENDIX

4. C = AnB
All the elements that are in set A but not in set B:C = f1; 3; 4; 5; 6; 7; 8; 9g

5. C = B \ R
All the elements of set B that are also real numbers. As all the numbers
that we will come across in this course
p are real numbers, everything in
set B is also in R. So C = f0; 0:5; 2; 2; 10g = B

6. C = B \ N
All the elements of set B that are also natural numbers. Remember
that the natural numbers are N = f1; 2; 3:::::g so C = f2; 10g

7. C = B \ Z
All the elements of set B that are also integers. Remember that the
natural numbers are Z = f::: 3; 2; 1; 0; 1; 2; 3:::::g so C = f0; 2; 10g

8. C = B \ Q
All the elements of set B that are also rational numbers. Remember
that the rational numbers are all the numbers that can be p expressed
as a ratio of an integer and a natural number, and that 2 is not a
rational number but 0:5 is, as it is equal to 12 . So C = f0; 0:5; 2; 10g

9. C = ZnN
All the numbers which are integers but not natural numbers. As the
integers are de…ned as Z = f::: 3; 2; 1; 0; 1; 2; 3:::::g and the natural
numbers as N = f1; 2; 3:::::g, C = f::: 3; 2; 1; 0g, or all the integers
which are not strictly positive.

10. C = NnZ
All the natural numbers that are not integers. We can see from above
that any natural must also be an integer. The set of natural numbers
which are not integers is therefore empty. We write this as C = ?

13.6.2 Derivatives
Take the …rst derivatives of the following functions
13.6 PROBLEMS 237

1. f (x) = (x2 + x)(x3 3x2 )

Product rule. Let g(x) = (x2 + x) and h(x) = (x3 3x2 ). Then
g 0 (x) = (2x + 1) and h0 (x) = (3x2 6x): The product rule tells us that
if f (x) = g(x)h(x) then f 0 (x) = g 0 (x)h(x) + h0 (x)g(x). In this case,
f 0 (x) = (2x + 1)(x3 3x2 ) + (3x2 6x)(x2 + x)
(x2 +x)
2. f (x) = (x3 3x2 )

Quotient rule. Let g(x) = (x2 + x) and h(x) = (x3 3x2 ). Then
g 0 (x) = (2x + 1) and h0 (x) = (3x2 6x):The quotient rule tells us that
g(x) 0 g(x)h0 (x)
if f (x) = h(x) then f 0 (x) = g (x)h(x)
(h(x))2
. In this case
(2x+1)(x3 3x2 )+(x2 +x)(3x2 6x)
f 0 (x) = (x3 3x2 )2

3. f (x) = (2x2 + 3x)3=2

Chain rule. let g(x) = (2x2 + 3x) and h(y) = y 3=2 then f (x) = h(g(x)).
1
Then g 0 (x) = (4x + 3) and h0 (y) = 32 y 2 : By chain rule, f 0 (x) =
h0 (g(x))g 0 (x) in this case
1
f 0 (x) = 32 (2x2 + 3x) 2 (4x + 3)

4. f (x) = log(x2 )
Two di¤erent methods. First, by chain rule, let g(x) = x2 and h(y) =
log y so that f (x) = h(g(x)). Then g 0 (x) = 2x and h0 (y) = y1 By chain
rule, f 0 (x) = h0 (g(x))g 0 (x) in this case
1 2
f 0 (x) = x2
2x = x

Alternatively, we can always rewrite log x2 as 2 log x, the di¤erential of

which is x2
2
5. f (x) = ex
Again, chain rule. let g(x) = x2 and h(y) = ey then f (x) = h(g(x)).
Then g 0 (x) = 2x and h0 (y) = ey : By chain rule, f 0 (x) = h0 (g(x))g 0 (x)
in this case
2
f 0 (x) = ex 2x

Take the …rst and second derivatives of these functions

238 CHAPTER 13 MATHEMATICAL APPENDIX

1. f (x) = 6x3 + 4x2 + x

First derivative f 0 (x) = 18x2 + 8x + 1
Second derivative f 00 (x) = 36x + 8
2. f (x) = log(x)
1
First derivative f 0 (x) = x
1
Second derivative f 00 (x) = x2

3. f (x) = ex x2
First derivative f 0 (x) = 2xex + x2 ex (product rule)
Second derivative f 00 (x) = 2xex + 2ex + 2xex + x2 ex = (2 + 4x + x2 )ex

Take all the partial derivatives of the following functions

Two things to remember:
1: There are the same number of partial derivatives as there are arguments
in a function. If we have f (x; y; z), then we will have three partial derivatives
2: When we are taking the partial derivatives, we take all the other ar-
guments as constant. So in the case of f (x; y; z) = xyz + x2 y + z 3 , if we are
taking the derivative with respect to x, we can write
f (x; y; z) = ax + bx2 + c, where a = yz; b = y and c = z 3 . The derivative
of this function is @f
@x
= a + 2bx = yz + 2yx

1. f (x; y; z) = xyz + x2 y + z 3
@f
@x
= yz + 2xy
@f
@y
= xz + x2
@f
@z
= xy + 3z 2
2. f (x; y) = xy(y 2 + x2 )
= xy 3 + yx3
@f
@x
= y 3 + 3yx2
@f
@y
= 3xy 2 + x3

3. f (x; y) = x log y
@f
@x
= log y
@f x
@y
= y
13.6 PROBLEMS 239

For the following two functions, sketch a graph of the isoquants for
f (x1 ; x2 ) equals 1, 2 and 5 in x1 ; x2 space. Write down an equation for
the slope of the isoquants
x2
1. f (x1 ; x2 ) = x1 + 2
For the case of f (x1 ; x2 ) = 1, we are looking for the set of x1 ; x2 such
that x1 + x22 = 2: Rearranging this gives x2 = 4 2x1 . This is clearly
a linear function with a slope of 2: We can also calculate the slope of
the isoquant using the result that
@f (x1 ;x2 )
@x1 dx2
@f (x1 ;x2 )
=
dx1
@x2

as @f (x
@x1
1 ;x2 )
= 1 and @f (x
@x2
1 ;x2 )
= 12 , this con…rms our result. Note that
the slope is constant along the graph, and doesn’t vary between the
isoquants 1, 2 and 5.

2. f (x1 ; x2 ) = x1 x2
For the case of f (x1 ; x2 ) = 1; we are looking for the set of points
such that x1 x2 = 1. Rearranging, this gives x2 = x11 . This shape is a
hyperbola. While you may not know this, you should be able to work
out that, as x1 gets very large, x2 gets very small (but not negative)
and as x1 gets close to zero, x2 gets very large.
We can work out the slope of the isoquant either directly, or using the
ratio of the partial derivatives. As @f (x 1 ;x2 )
@x1
= x2 and @f (x
@x2
1 ;x2 )
= x1 ; the
dx2 x2
above formula tells us that dx1 = x1 : But we can do more. Note that
dx2
x2 = x11 in this case, so we can substitute back in to give dx 1
= x12 :
1
Note that the slope of the isoquant changes both along the curve and
between isoquants.

Partial Derivatives - one more try

Take all the partial derivatives of the following functions

1. f (x; y; z) = x3 y 2 z + z 2 y + y 2 x
fx (x; y; z) = 3x2 y 2 z + y 2
240 CHAPTER 13 MATHEMATICAL APPENDIX

fy (x; y; z) = 2x3 yz + z 2 + 2yx

fz (x; y; z) = x2 y 2 + 2zy

2. f (x; y; z) = (xy + z)(x + y)3

fx (x; y; z) = y(x + y)3 + 3(x + y)2 (xy + z)
fy (x; y; z) = x(x + y)3 + 3(x + y)2 (xy + z)
fz (x; y; z) = (x + y)3

3. f (x; y) = log xy
There are two ways to solve this. One way is to note that
f (x; y) = log xy = log x + log y
1
=) fx (x; y) = x
1
fy (x; y) = y

Or by chain rule
1 1
fx (x; y) = y: yx = x
1 1
fy (x; y) = x: yx = y

4. f (x; y) = log xy
As above
1
fx (x; y) = x
1
fy (x; y) = y

3
5. f (x; y) = (x2 + (xy + y 2 )) 2
1
fx (x; y) = 32 (x2 + (xy + y 2 )) 2 (2x + y)
1
fy (x; y) = 32 (x2 + (xy + y 2 )) 2 (2y + x)

13.6.3 Maximization
We start with functions of one variable and we move to functions of two
variables.
13.6 PROBLEMS 241

Maximizing functions of one variable

For each of the following functions:

Find the critical points (i.e. the points where the function is ‡at);
Categorize these points as local maxima or minima (if possible);
Explain whether any of these points are global maxima or minima.

Take the …rst derivatives of the following functions

1. f (x) = 3x2 + 9x + 8
FOC:
f 0 (x) = 6x + 9 = 0
3
x= 2
This is the only critical point
SOC:
f 00 (x) = 6<0
This is therefore a local maximum
As we only have one critical point, and it is a local maximum, then we
know from theorem 4 from the Prelim handout 2 that we have a global
maximum.
2. f (x) = 7x2 + 2x + 3
FOC:
f 0 (x) = 14x + 2 = 0
1
x= 7
This is the only critical point
SOC:
f 00 (x) = 14 > 0
This is therefore a local minimum
As we only have one critical point, and it is a local minimum, then we
know from theorem 4 from the Prelim handout 2 that we have a global
minimum.
242 CHAPTER 13 MATHEMATICAL APPENDIX

x3
3. f (x) = 3
+ x2 8x + 10
FOC:
f 0 (x) = x2 + 2x 8 = (x + 4)(x 2) = 0
x= 4; 2
There are two critical points
SOC:
f 00 (x) = 2x + 2
f 00 ( 4) = 6<0
This is a local maximum
f 00 (2) = 6 > 0
This is therefore a local minimum
We can see that neither of these points are global extrema. The easiest
way to see this is to note that, as x gets very large, f (x) goes to +1
and as x gets very small, f (x) goes to 1

Maximizing functions of two variables

For the following functions, …nd the critical points (i.e. the point where the
function is ‡at with regard to all its arguments)

1. f (x; y) = x4 + x2 6xy + 3y 2
FOC
fy (x; y) = 6x + 6y = 0
)x=y
fx (x; y) = 4x3 + 2x 6y = 0
but as x=y
) 4x3 + 2x 6x = 0
2
) 4x(x 1) = 0
) x = 0; x = 1; x = 1
So critical points lie at
(0; 0); (1; 1); ( 1; 1)
13.6 PROBLEMS 243

2. f (x; y) = xy 2 + x3 y xy
FOC
fx (x; y) = y 2 + 3x2 y y=0
) y(y + 3x2 1) = 0
fy (x; y) = 2xy + x3 x=0
) x(2y + x2 1) = 0
The FOC with respect to x tells us that either y = 0 or y +3x2 1 = 0.
Similarly, the second equation tells us that either x = 0 or 2y +x2 1 =
0.
This gives us 4 di¤erent cases
(i) x = 0; y = 0
(ii) y = 0; 2y + x2 1=0
) x2 = 1
) x = 1; x = 1
(iii) y + 3x2 1 = 0; x = 0
)y=1
(iv) y + 3x2 1 = 0; 2y + x2 1=0
) x2 = 1 2y
) y + 3(1 2y) 1=0
2
) 5y = 2 ) y = 5

) x2 = 1 2: 25 = 1
5

)x= p1 ; x = p1
5 5
So, collecting all the di¤erent cases, we get the following critical points
(0; 0); ( 1; 1); (1; 1); (0; 1); ( p1 ; 2 ); ( p1 ; 2 )
5 5 5 5

Constrained optimization
In the following two problems there is only one critical point, which I shall
tell you is the maximum. Find it.
244 CHAPTER 13 MATHEMATICAL APPENDIX

1. Maximize 3 log x + 2 log y subject to 3x + 4y = 10

(i) Form the Lagrangian
L(x; y; ) = 3 log x + 2 log y (3x + 4y 10)
(ii) Take the …rst order conditions
3 3
Lx (x; y; ) = x
3 =0) x
=3
2 2
Ly (x; y; ) = y
4 =0) y
=4
L (x; y; ) = 3x + 4y 10 = 0
(iii) Divide the x; y FOC by each other to get rid of
3
3 3y 3
x
2 = 4
) 2x
= 4
y

(iv) Get x in terms of y (or visa versa)

3y 3 x
2x
= 4
)y= 2
(v) Substitute into the constraint and solve
3x + 4( x2 ) = 10 ) 5x = 10 ) x = 2 ) y = 1

2. Maximize x y (1 ) subject to px x + py y = I.(Note that a; px ; py I are

parameters for this problem your answer should be a function of x and
y in terms of these values)
See separate sheet

If you use the Lagrangian method on the following problem, you will get
two solutions. One will be a maximum and the other will be a minimum.
Find them both and decide which is which.
First, a quick apology - there are actually 4 solutions to this problem, but
as we shall see below, they come in two very similar pairs of two.

1. Maximize x2 + y 2 subject to x2 + xy + y 2 = 3
(i) Form the Lagrangian
L(x; y; ) = x2 + y 2 (x2 + xy + y 2 3)
(ii) Take the …rst order conditions
Lx (x; y; ) = 2x (2x + y) = 0 ) 2x = (2x + y)
Ly (x; y; ) = 2y (2y + x) = 0 ) 2y = (2y + x)
13.6 PROBLEMS 245

L (x; y; ) = x2 + xy + y 2 3=0
(iii) Divide the x; y FOC by each other to get rid of
2x (2x+y) x (2x+y)
2y
= (2x+y)
) y
= (2y+x)

(iv) Get x in terms of y (or visa versa)

x (2x+y)
y
= (2y+x)
) x(2y + x) = y(2x + y) ) 2xy + x2 = 2xy + y 2
) x2 = y 2 ) x = y or x = y
(v) Substitute into the constraint and solve
Case 1: x = y
x2 + xy + y 2 = 3 ) x2 + x2 + x2 = 3 ) 3x2 = 3 ) x2 = 1
Which gives 2 solutions: (1; 1) and ( 1; 1)
Case 2: x = y
x + xy + y = 3 ) x2
2 2
x2 + x2 = 3 ) x2 = 3
p p p p
Which gives 2 solutions: ( 3; 3) and ( 3; 3)
To …nd out which of these are the max and which are the minima,
substitute the solutions back into the objective function. This tells you
that the two solutions from case 1 give you f (x; y) = x2 +y 2 = 1+1 = 2,
whereas the solutions from case 2 are (x; y) = x2 + y 2 = 3 + 3 = 6.
The two solutions from case 1 are therefore minima whereas the two
solutions from case 2 are maxima.

13.6.4 Linear algebra

[... To be added...]