Studies in Economic Theory

Editors
Charalambos D. Aliprantis
Purdue University
Department of Economics
West Lafayette, in 47907-2076
USA
Nicholas C. Yannelis
University of Illinois
Department of Economics
Champaign, il 61820
USA
Titles in the Series

M. A. Khan and N. C. Yannelis (Eds.) C. D. Aliprantis, K. J. Arrow, P. Hammond,

Equilibrium Theory F. Kubler, H.-M. Wu and N. C. Yannelis (Eds.)
in Infinite Dimensional Spaces Assets, Beliefs, and Equilibria
in Economic Dynamics
C. D. Aliprantis, K. C. Border
and W. A. J. Luxemburg (Eds.) D. Glycopantis and N. C. Yannelis (Eds.)
Positive Operators, Riesz Spaces, Differential Information Economies
and Economics
A. Citanna, J. Donaldson, H. M. Polemarchakis,
D. G. Saari P. Siconolfi and S. E. Spear (Eds.)
Geometry of Voting Essays in Dynamic
General Equilibrium Theory
C. D. Aliprantis and K. C. Border
Infinite Dimensional Analysis M. Kaneko
Game Theory and Mutual Misunderstanding
J.-P. Aubin
Dynamic Economic Theory
M. Kurz (Ed.)
Endogenous Economic Fluctuations
J.-F. Laslier
Tournament Solutions and Majority Voting
A. Alkan, C. D. Aliprantis and N. C. Yannelis
(Eds.)
Theory and Applications
J. C. Moore
Mathematical Methods
for Economic Theory 1
J. C. Moore
Mathematical Methods
for Economic Theory 2
M. Majumdar, T. Mitra and K. Nishimura
Optimization and Chaos
K. K. Sieberg
Criminal Dilemmas
M. Florenzano and C. Le Van
Finite Dimensional Convexity
and Optimization
K. Vind
Independence, Additivity, Uncertainty
T. Cason and C. Noussair (Eds.)
Advances in Experimental Markets
F. Aleskerov and B. Monjardet
Utility Maximization. Choice and Preference
N. Schofield
Mathematical Methods in Economics
and Social Choice
Alessandro Citanna · John Donaldson
Herakles Polemarchakis · Paolo Siconolfi
Stephen E. Spear
Editors

Essays in Dynamic
General Equilibrium Theory
Festschrift for David Cass

With 23 Figures
and 3 Tables

123
Professor Alessandro Citanna Professor Paolo Siconolfi
Department of Economics and Finance Graduate School of Business
HEC, Paris Columbia University
France Finance and Economics Division
E-mail: [email protected] New York, NY 10027
and Finance and Economics Division USA
Graduate School of Business E-mail: [email protected]
Columbia University
New York, NY 10027 Professor Stephan E. Spear
USA Tepper School of Business
E-mail: [email protected] Carnegie Mellon University
Pittsburgh, PA 15213
Professor John Donaldson USA
Graduate School of Business E-mail: [email protected]
Columbia University
Finance and Economics Division
New York, NY 10027
USA
E-mail: [email protected]

Professor Herakles Polemarchakis

Brown University
Department of Economics
Providence, RI 02912
USA
E-mail: [email protected]

Library of Congress Control Number: 2004114983

ISBN 3-540-22267-7 Springer Berlin Heidelberg New York

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned,
specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on
microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is
permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version,
and permission for use must always be obtained from Springer-Verlag.Violations are liable for prosecution under
the German Copyright Law.
Springer is a part of Springer Science+Business Media
springeronline.com
© Springer-Verlag Berlin Heidelberg 2005
Printed in Germany
The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even
in the absence of a specific statement, that such names are exempt from the relevant protective laws and
regulations and therefore free for general use.
Cover design: Erich Kirchner, Heidelberg
Production: Helmut Petri
Printing: betz-druck
SPIN 11015390 Printed on acid-free paper – 42/3130 – 5 4 3 2 1 0
Essays in General Equilibrium Theory
Festschrift for David Cass

Guest Editors’ Introduction

Alessandro Citanna1 , John Donaldson2 , Herakles Polemarchakis3 ,

Paolo Siconolfi4 , and Stephen Spear5
1
Department of Economics and Finance, HEC, Paris, France
(email: [email protected])
and Finance and Economics Division, Graduate School of Business, Columbia
University, New York, NY 10027 (email: [email protected])
2
Finance and Economics Division, Graduate School of Business, Columbia
University, New York, NY 10027 (email: [email protected])
3
Department of Economics, Brown University, Providence, RI 02912
(email: Herakles [email protected])
4
Finance and Economics Division, Graduate School of Business, Columbia
University, New York, NY 10027 (email: [email protected])
5
Tepper School of Business, Carnegie Mellon University, Pittsburgh, PA 15213
(email: [email protected])

This book is a supplement to the special edition of Economic Theory hon-

oring David Cass on the 30th anniversary of his joining the faculty at the
University of Pennsylvania’s Economics Department. As with the ET is-
sue, the contributions to this volume are, for the most part, from Dave’s
students or co-authors, and we hope they communicate both to Dave and
to the economics profession generally the high regard those of us who have
trained under Dave’s tutelage or worked with him on research have for him.
Most scientists would be happy to have had one major, influential idea
over the course of their careers. Dave Cass has had three, and is still going
strong.
His first major contribution to economics was the characterization of
optimal growth trajectories in his thesis work under Hirofumi Uzawa’s su-
pervision. The celebrated Cass criterion for optimal time paths in the one
good growth model quickly followed. The essence of this work is the search
for price characterizations of efficiency for dynamic time paths, an effort
that directly pointed the way to the subsequent full dynamic decentraliza-
tion of the neoclassical optimal growth model, a fact that permits its use
VI Guest Editors’ Introduction

for modeling a wide range of business cycle and other macroeconomic phe-
nomena. Accordingly, Dave is rightly honored, in conjunction with Tjalling
Koopmans, as one of the fathers of dynamic macroeconomic analysis.
Dave’s second contribution – the notion of a so-called sunspot equilib-
rium in dynamic economies which he developed jointly with Karl Shell – is
also the stuff of legend, and grew out of his long and productive collabora-
tion with Karl at Penn. The early impetus for Dave’s interest in this topic
stemmed from work he did with Manny Yaari on overlapping generations
models, and from his early acquaintance with Bob Lucas at Carnegie Mel-
lon and Lucas’s seminal work on rational expectations in dynamic economic
models. To quote from the interview with Dave by Spear and Wright in
Macroeconomic Dynamics
I wasn’t so interested in macro, but what struck me, and this is related
to some of my later work, was the assumption that [Lucas] made to solve for
equilibrium, that the state variables were obvious. . . . Bob and I had some
long discussions, and I would say, “Well Bob, why is this the actual state
space in this model?” That question came up . . . after I came to Penn. At
some point Karl [Shell] and I started talking about that and we developed
what we called the idea of sunspots. (Spear and Wright [8])
In addition to raising troubling questions about what the right state
space was for dynamic stochastic economies, the notion of sunspot equilib-
rium raised a number of deep questions about the overall determinacy of
economic equilibria and the role of the welfare theorems in the occurrence
or non-occurrence of sunspot equilibria. These questions spawned a large
literature on determinacy in dynamic economies in which the welfare the-
orems broke down. These include overlapping generations models, growth
models with externalities or taxes, and models in which asset markets were
incomplete. All were shown to allow the existence of sunspot equilibria.
And, in a suitable twist of intellectual fate, macroeconomists have more
recently begun to explore the question of whether sunspots can provide a
more plausible source of fluctuations in dynamic equilibrium models than
the conventional aggregate productivity disturbances.
Dave’s third major contribution to economic theory was his work on
general equilibrium with incomplete markets, work which grew out of his
exploration of the question of existence of sunspot equilibria in models with
incomplete asset markets. Dave’s follow-on work on existence and deter-
minacy of general equilibrium in models with incomplete asset markets
spawned another large literature which has come to be known simply as
GEI.
The earliest work on market incompleteness goes back to Arrow in the
1950’s, Diamond in the mid-‘60’s and a number of related papers in the
finance literature between the late 1950’s and early ‘70’s (Geanakoplos [4]
provides an excellent survey of this literature). The canonical GEI model was
formulated by Radner in the early 1970’s (Radner [7]) in a paper which also
 Guest Editors’ Introduction VII

pointed up one of the fundamental puzzles about models with incomplete

markets: the possible loss of dimensionality in the span of the asset payoffs
as prices vary.
This potential for non-existence of equilibrium (which was formally de-
veloped in Hart’s [6] counterexamples to existence of equilibrium) left the
literature in limbo for almost a decade, until Dave’s work on existence in
economies with purely financial assets pointed the way out. As Geanakoplos
notes

Suddenly in the middle 1980s the pure theory of GEI fell into place.
In two provocative and influential papers, Cass [1,2] showed that the
existence of equilibrium could be guaranteed if all the assets promise
delivery in fiat money, and he gave an example showing that with
such financial assets there could be a multiplicity of equilibrium. Al-
most simultaneously Werner [9] also gave a proof of existence of equi-
librium with financial assets, and Geanakoplos and Polemarchakis
[5] showed the same for economies with real assets that promise
delivery in the same consumption good. (Geanakoplos [4])

This work was followed very quickly by results showing that the non-
existence problem pointed out by Hart was not generic, and led ultimately
to the generic existence results of Duffie and Shafer [3], and again spawned
a new literature looking positively at the welfare implication of market in-
completeness, and normatively at issues of asset engineering.
In the course of making these contributions, Dave has worked with a
large group of coauthors, including (to date): Y. Balasko, L. Benveniste,
G. Chichilnisky, A. Citanna, R. Green, M. Majumdar, T. Mitra, M. Okuno,
A. Pavlova, H. Polemarchakis, K. Shell, P. Siconolfi, S. Spear, J. Stiglitz,
A. Villanacci, H.-M. Wu, M. Yaari, and I. Zilcha. Dave’s graduate students
(to date) include S. Chae, A. Citanna, J. Donaldson, R. Forsythe, F. Kyd-
land, Y. W. Lee, M. Lisboa, A. Pavlova, T. Pietra, P. Siconolfi, S. Spear,
S. Suda, J-M. Tallon, and A. Villanacci. Those of us who have worked
with Dave and/or under his tutelage as graduate students have benefited
tremendously from his razor-sharp analytic mind, from his willingness to
work at understanding problems we have posed to him, or new methodolog-
ical techniques we have discovered, and (perhaps most importantly) from
his no-nonsense approach to doing science.

References
1. Cass, D., Competitive equilibria with incomplete financial markets, CARESS
Working Paper, University of Pennsylvania, April 1984
2. Cass D., On the ‘number’ of equilibrium allocations with incomplete financial
markets, CARESS Working Paper, University of Pennsylvania, May 1985
VIII Guest Editors’ Introduction

3. Duffie, D. and W. Shafer, Equilibrium in incomplete markets, I, Journal of

Mathematical Economics, 14, 285-300, 1985
4. Geanakoplos, J., An introduction to general equilibrium with incomplete as-
set markets, Journal of Mathematical Economics, 19, 1990
5. Geanakoplos, J. and H. Polemarchakis, Existence, regularity and constrained
suboptimality of competitive equilibrium when markets are incomplete, in W.
Heller, R. Starr and D. Starrett, eds., Essays in Honor of Kenneth Arrow,
Vol. 3, Cambridge University Press, Cambridge, MA, 1986
6. Hart, O., On the optimality of equilibrium when the market structure is
incomplete, Journal of Economic Theory, 11,3, 418-443, 1975
7. Radner, R., Existence of equilibrium of plans, prices and price expectations,
Econometrica, 40, 2, 289-303, 1972
8. Spear, S. and R. Wright, An interview with David Cass, Macroeconomic
Dynamics, 2, 533-558, 1998
9. Werner, J., Equilibrium in economies with incomplete financial markets,
Journal of Economic Theory, 36, 110-119, 1985.
Table of Contents

Suleyman Basak and Anna Pavlova

Monopoly Power and the Firm’s Valuation: A Dynamic Analysis
of Short versus Long-Term Policies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

Christian Bidard
Revealed Intertemporal Inefficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

Graciela Chichilnisky and Olga Gorbachev

Volatility and Job Creation in the Knowledge Economy . . . . . . . . . . . . . 45
Thorsten Hens, János Mayer, and Beate Pilgrim
Existence of Sunspot Equilibria and Uniqueness of Spot Market
Equilibria: The Case of Intrinsically Complete Markets . . . . . . . . . . . . . 75

Mukul Majumdar and Nigar Hashimzade

Survival, Uncertainty, and Equilibrium Theory: An Exposition . . . . . . . 107

Richard McLean, James Peck, and Andrew Postlewaite

On Price-Taking Behavior in Asymmetric Information Economies . . . . 129
Sujoy Mukerji and Jean-Marc Tallon
Ambiguity Aversion and the Absence of Indexed Debt . . . . . . . . . . . . . . 143

Tito Pietra
Sunspots, Indeterminacy and Pareto Inefficiency in Economies
With Incomplete Markets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

Aldo Rustichini and Paolo Siconolfi

Growth in Economies With Non Convexities:
Sunspots and Lottery Equilibria, Theory and Examples . . . . . . . . . . . . . 203
Jamsheed Shorish and Stephen Spear
Shaking the Tree:
An Agency-Theoretic Model of Asset Pricing . . . . . . . . . . . . . . . . . . . . . . 243

Neil Wallace
Central-Bank Interest-Rate Control
in a Cashless, Arrow-Debreu Economy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
Monopoly Power and the Firm’s Valuation:
A Dynamic Analysis of Short versus
Long-Term Policies⋆

Suleyman Basak1 and Anna Pavlova2

1
London Business School and CEPR, Institute of Finance and Accounting,
London NW1 4SA, UK [email protected]
2
Sloan School of Management, Massachusetts Institute of Technology, Cambridge,
MA 02142-1347, USA [email protected]

Summary. Recent anti-trust cases exacerbated the concerns of investors regard-

ing the effects of a firm’s monopoly power on its production choice, shareholder
value, and the overall economy. We address this issue within a dynamic equilibrium
model featuring a large monopolistic firm whose actions not only affect the price
of its output, but also effectively influence the valuation of its stock. The latter
renders time-inconsistency to the firm’s dynamic production choice. When the firm
is required to pre-commit to its strategy, the ensuing equilibrium is largely in line
with the predictions of the textbook monopoly model. When the firm behaves in a
time-consistent manner, however, the predictions are strikingly at odds. The trade-
off between current profits and the valuation of future profits induces the firm to
increase production beyond the competitive benchmark and cut prices. This policy
may result in destroying shareholder value, and does indeed fully wipe out the firm’s
profit in the limit of the decision-making interval shrinking to zero, in line with the
Coase conjecture.

1 Introduction

The pervasiveness of monopoly power, across many product categories in

almost every part of the world economy, has long pre-occupied economists
and lawmakers. There is also undisputed evidence that monopoly power is
widespread amongst firms dominant enough to matter at an economy-wide

⋆
We thank Steve Spear and the anonymous referees for helpful suggestions.
We are also grateful to Franklin Allen, Dave Cass, Peter DeMarzo, Bernard Du-
mas, Ron Giammarino, Rich Kihlstrom, Leonid Kogan, Branko Urosevic, Dimitri
Vayanos, seminar participants at Boston University, University of Colorado at Boul-
der, Columbia University, MIT, University of Pennsylvania, Princeton University,
American Finance Association Meetings, and European Finance Association Meet-
ings for valuable comments. All errors are solely our responsibility.
2 Suleyman Basak and Anna Pavlova

level, the so-called “bellwethers.” This is evidenced by the series of ongoing

anti-trust cases brought about by the U.S. government against, for exam-
ple, IBM (1969 - 1982), AT&T (1974 - 1982), Microsoft (1994 - 1995, 1998 -
present), and Bell Atlantic (1996 - 1997). Apparently, there is concern that
these large monopolists may not just exert power in their own product mar-
kets, but their actions may impact the overall economy. For example, as argued
in the Microsoft case, US consumers and businesses alike feared to become
critically dependent on Microsoft products. There are well-known dominant
players outside the US, too. For instance, OPEC’s production decisions appear
to affect not only the oil producing countries’ economies, but also the overall
world economy. While the surveys of imperfect competition by Hart (1985)
and Bonanno (1990) have highlighted the importance of a general equilibrium
analysis of such large monopolists, most studies of monopoly behavior have
been undertaken at a partial equilibrium level assuming no impact on other
markets. A true general equilibrium approach must face up to the challenge
of accounting for the so-called “feedback” or “Ford effect,” as once argued by
Henry Ford: through the influence of its own actions, a non-price-taking firm
may affect the wealth of its customers, and hence the demand for the firm’s
product. Finally, although there is growing work on general equilibrium asset
pricing with market imperfections (e.g., see the survey by Sundaresan (2000)),
the consideration of monopoly power is still missing in this literature.
Our primary objective in this paper is to investigate the optimal behavior
of a monopolist who has sufficient power to impact economy-wide pricing.
We model the extreme case of an economy containing a single monopolistic
firm who then, at a general equilibrium level, impacts the overall price of con-
sumption in the economy. Beyond the traditional assumption that its actions
impact the price of its own good, the firm’s actions also influence the valuation
of its stock. As will become evident from our analysis, the dynamic setting we
employ leads to a distinction between short-term and long-term policies of the
monopolistic firm. (For a competitive firm, there is no such distinction.) Part
of our emphasis will be to study the differences between the ensuing equilibria
under short versus long-term policies.3
We adopt a familiar Robinson-Crusoe formulation of a discrete-time
production finite-horizon economy populated by a representative consumer-
investor-worker and a representative firm. The consumer derives utility from
3
Firms’ short-termism is receiving increasing attention from both academics and
practitioners. For example, Business Week (09/13/1999) voices the often-quoted
complaint that Wall Street exerts increasing pressure on many firms for short-term
results due to a shrinking of investors’ time horizons. The article points out that
a share in AT&T is held for an average of 1.1 years, down from 3 years in 1990;
a share in General Motors is held for an average of 1 year, down from 2 years in
1990. Firms’ short-termism is also frequently blamed on ever increasing managerial
turnover. Anecdotal evidence and recent academic literature (Allen and Gale (2000),
Palley (1997)) seem to agree that revolving management may be damaging for a
long-lived firm.
 Monopoly Power and the Firm’s Valuation 3

consumption and leisure, simultaneously invests in financial markets (includ-

ing the firm’s stock), and earns a labor income. The consumer’s labor is de-
manded by the firm as the sole input to a stochastic non-constant-returns-
to-scale production technology, producing the only good in the economy. We
assume the objective of this firm is to maximize its market value, or the present
value of its expected profits. The firm has monopoly power in the good market,
in that it takes account of the impact (via market clearing) of its production
plan on the price of its output. In our setting, the monopoly power manifests
itself as an impact of the firm’s labor demand on the state prices (or the pric-
ing kernel). This manipulation of state prices results in time-inconsistency of
the firm’s production strategy, in that it has an incentive to deviate from the
initial plan at a later date. To rule out time-inconsistency, we focus on two dis-
tinct types of monopolistic strategies: a “pre-commitment” strategy in which
the firm initially chooses a plan to maximize its initial value, and thereafter
cannot deviate from that plan; and a “time-consistent” strategy in which the
firm chooses a plan each period, maximizing the value at that period, taking
into account the re-adjustments that it will make in the future. Consistently
with the literature (e.g., Blanchard and Fischer (1989, §11.4)), we interpret
the time-consistent strategy as a short-sighted or short-term strategy, and the
pre-commitment one as long-term.
Solving for the pre-committed monopolist’s strategy reveals his optimal
plan and the extent of his monopoly power to be driven by the concurrent
profit, the marginal product of labor, and the consumer’s attitude towards risk
over consumption. The most immediate implications on the ensuing dynamic
equilibrium are consistent with the predictions of textbook static monopoly
models: lower good output and lower labor input, and higher price of consump-
tion than the competitive counterpart. However, in contrast to the textbook
case, profits and the firm’s value can be either higher or lower. This arises
because it is the firm’s initial value the monopolist maximizes, not the profits
nor the value at later times. By restricting production, he moves away from
the competitive, profit maximizing, production plan.
The optimal behavior of the time-consistent monopolist contrasts sharply
with that of the pre-committed monopolist. His optimal production plan and
his monopoly power are driven by the negative of the ex-dividend stock price,
in place of concurrent profits. In attempting to maximize the current stock val-
uation, this monopolist trades off between today’s profit and the ex-dividend
stock price; hence the appearance of the ex-dividend price. In direct oppo-
sition to the pre-commitment case, the equilibrium good output and labor
demand are higher than the competitive case, while the price of consumption
is lower. It is in this short-term monopolist’s interest to depress the current
price of consumption, so as to boost today’s stock price. Yet more strikingly,
the profits in every period are decreased in the monopolist’s presence, and may
even go negative, while the firm’s value can be either lower or higher than in
the competitive benchmark. We also argue that our main conclusions regard-
ing the time-consistent equilibrium hold true in the infinite horizon limit of
4 Suleyman Basak and Anna Pavlova

our economy within a parametric example. To further explore the robustness

of these results, we provide an alternative, albeit extreme, form of a produc-
tion opportunity (constant-returns-to-scale) under which the monopoly power
vanishes, and the solution coincides with the competitive.
It is well-recognized in monopolistic models of durable goods, that absent
commitment the monopolistic firm may in a sense “compete with itself” across
different time periods, and weaken its monopoly power. The Coase (1972)
conjecture proposes that as the time interval between successive decisions is
reduced to an infinitesimal length, this intertemporal competition will drive
the firm’s profits to zero. Since our long-lived firm’s stock resembles a durable
good, it is of interest to explore the limiting case of our economies where
the decision-making time interval shrinks to zero. While the competitive and
the pre-commitment monopolistic equilibrium retain their basic discrete-time
structure and implications, we find the time-consistent equilibrium to tend to
the limit of zero profits and hence zero firm’s value at all times. Hence, un-
der the time-consistent scenario, monopoly power destroys shareholder value.
However, within our framework, this zero profit limit does not coincide with
the competitive solution.
The importance of imperfect competition is, of course, well-recognized in
many areas of theoretical and applied research. The standard textbook treat-
ment of monopoly (Mas-Colell, Whinston and Green (1995, Chapter 12), Ti-
role (1988, Chapter 1)) considers a profit-maximizing firm which is the only
producer of a good (and has no influence on other markets), in a static partial
equilibrium setting. The main implications of the textbook monopolist are
restriction of production and a raising of the price of output in the market
for its product. While this behavior is consistent with the equilibrium im-
plications of our pre-committed monopolist, it is at odds with those of the
time-consistent monopolist. The distinction is due to our accounting for the
economy-wide impact of the firm’s decision and in particular for the impact
on its stock price valuation.
In that regard, the most closely related work in a general equilibrium set-
ting where the valuation of financial securities is affected by market power
are the works of Basak (1997) and Kihlstrom (1998). These authors consider
a single large consumer-investor acting as a non-price-taker in securities mar-
kets, in a Lucas (1978)-type pure-exchange setting.4 While the monopolists in
Kihlstrom and Basak select a consumption-investment plan to maximize util-
ity, our monopolist selects a production plan to maximize the stock price of
his firm. In this respect, ours is much closer to the standard textbook monop-
olist. Basak demonstrates that the non-price-taker acting as a price-leader in
all markets manifests itself as a dependence of the state prices on the agent’s
consumption choice. Kihlstrom relates the dynamic security price choice of
the monopolist to the Coase (1972) conjecture, and shows that the inability

4
See also Lindenberg (1979) and Grinblatt and Ross (1985) for related analysis
within a static mean-variance framework.
 Monopoly Power and the Firm’s Valuation 5

of the monopolist to commit to the future (second period) price quote reduces
his monopoly rents. Consequently, the first period security price is less than
in the commitment scenario (but is still higher than the competitive price).
In the spirit of Kihlstrom, but with an additional moral hazard problem of
the monopolist, is the dynamic model of DeMarzo and Urosevic (2001).5 Ab-
sent commitment, they demonstrate an analog of the Coase conjecture in the
continuous-time limit of their economy.
The rest of the paper is organized as follows. Section 2 describes the econ-
omy. Section 3 characterizes equilibrium in the economy with a monopolistic
firm for the cases when the firm can commit to its future production plan
and when it cannot. It also presents comparison of the resulting equilibrium
quantities to those of a benchmark competitive economy. In Section 4, we take
the economy to its continuous-time limit and explore the Coase conjecture in
the context of our economy. Section 5 concludes and the Appendix provides
all proofs as well as discussions of alternative choices of the numeraire and
the case of a monopolistic-monopsonistic firm.

2 The Economy
We consider a simple Robinson-Crusoe production economy with a repre-
sentative firm and a representative consumer-investor-worker. We make the
standard assumption that the consumer-investor-worker represents a con-
tinuum of identical atomistic agents who take prices as given and cannot
act strategically. The economy has a finite horizon [0, T ], in which trad-
ing takes place at discrete times t = 0, . . . , T . There is a single consump-
tion good serving as the numeraire (other choices of the numeraire are dis-
cussed in Remark 1). Uncertainty is represented by a filtered probability space
(Ω, F, {Ft ; t = 0, 1, . . . , T }, P) generated by a production shock process ε. All
stochastic processes are assumed adapted to {Ft ; t = 0, 1, . . . , T }, all stated
(in)equalities involving random variables hold P-almost surely. We assume
all processes and expectations are well-defined, without explicitly stating the
required regularity conditions.
The financial investment opportunities are represented by: a risky stock of
the firm in constant net supply of 1 share that pays out dividends π(t), t =
1, . . . , T ; and enough zero net supply securities to dynamically complete the
market. π is endogenously determined via the firm’s optimization problem.
Dynamic market completeness allows the construction of a unique system of
Arrow-Debreu securities consistent with no arbitrage. Accordingly, we may
define the state price density process ξ (or the pricing kernel ξ(s)/ξ(t), s ≥ t),
where ξ(t, ω) is interpreted as the Arrow-Debreu price (per unit probability

5
See also DeMarzo and Bizer (1993), who were the first to point out the connec-
tion between durable goods and securities markets. DeMarzo and Urosevic (2001)
also provide a comprehensive review and classification of the literature.
6 Suleyman Basak and Anna Pavlova

P) of a unit of consumption good in state ω ∈ Ω at time t. The process ξ is

to be determined in equilibrium. The time-t (cum-dividend) value V (t) of the
stock (of the firm) is then given by

T 
 ξ(s) 
V (t) = E π(s) Ft . (1)

s=t
ξ(t)

As evident from the subsequent analysis, our main results are valid in a de-
terministic version of our model, where ξ has only one value in each period.
We introduce uncertainty to make our formulation comparable to financial
markets models, standard in the literature.

2.1 The Consumer’s Preferences and Optimization Problem

A representative consumer-investor-worker is endowed at time 0 with 1 share

of the stock, and at each time t with ℓ̄ units of available labor, to allocate
between leisure h(t), and labor ℓ(t), for which he is paid a wage at rate w(t).
The consumer intertemporally chooses a non-negative consumption process c,
labor process ℓ, and portfolio (of securities) process so as to maximize his life-
time utility.6 The consumer derives a separable von Neumann-Morgenstern
time-additive, state-independent utility u(c(t)) + v(h(t)) from consumption
and leisure in [1, T ]. The functions u and v are assumed three times contin-
uously differentiable, strictly increasing and strictly concave, and to satisfy
limc→0 u′ (c) = ∞, limh→0 v ′ (h) = ∞.7
Under the complete markets assumption, the consumer-worker’s dynamic
optimization problem can be cast in its Arrow-Debreu formulation as a static
variational problem with a single budget constraint:

T 

max E u(c(t)) + v(ℓ̄ − ℓ(t)) (2)
c, ℓ
t=1

T T



   
subject to E ξ(t) c(t) − w(t) ℓ(t) ≤E ξ(t)π(t) . (3)
t=1 t=1

We do not explicitly apply the nonnegativity constraints c(t) ≥ 0, ℓ(t) ≤

ℓ̄, ℓ(t) ≥ 0, because the conditions limc→0 u′ (c) = ∞ and limh→0 v ′ (h) = ∞
6
We introduce labor as one of the simplest, most commonly-adopted choices of
input to the firm’s production function; employing labor does not introduce the
intertemporal complexity of employing capital.
7
We make the assumption of separable utility for simplicity, given our intention
is to focus on a comparison of monopolistic behavior in a particular good market.
Our analysis readily extends to the general non-separable case u(c, h). Our major
comparative statics results (Propositions 2 and 4) remain valid under the assumption
uch ≥ 0 (which includes separable utility and Cobb-Douglas u(c, h) = γ1 (cρ h1−ρ )γ ,
for γ ∈ (0, 1).)
 Monopoly Power and the Firm’s Valuation 7

guarantee c(t) > 0, ℓ(t) < ℓ̄, while (in the equilibrium provided) the firm’s
production technology (Section 2.2) guarantees ℓ(t) ≥ 0.
The first-order conditions of the static problem (2)–(3) are

u′ (c(t)) = y ξ(t) , (4)

v ′ (ℓ̄ − ℓ(t)) = y ξ(t) w(t) , (5)

where the Lagrangian multiplier y satisfies

T 
T 
   
E ξ(t) c(t) − w(t) ℓ(t) = E ξ(t)π(t) . (6)
t=1 t=1

Consequently,
v ′ (ℓ̄ − ℓ(t))
= w(t) . (7)
u′ (c(t))

2.2 The Firm’s Production and Optimization

The representative firm in this economy faces the same information structure
and set of securities as the consumer. At each time t = 1, . . . , T , the firm uses
labor, ℓD (t), as its only input to a production technology, f , which provides
consumption good as output.8 The technology is stochastic, driven by a shock
process ε, assumed (without loss of generality) to be strictly positive. The
firm’s output at time t is given by f (ℓD (t), ε(t)). We assume f is increasing
and strictly concave in its first argument and that limℓD →0 fℓ (ℓD , ε) = ∞ and
limℓD →0 f (ℓD , ε) ≥ 0. The firm pays out a wage w(t) for each unit of labor it
utilizes, so its time-t profit is

π(t) = f (ℓD (t), ε(t)) − w(t) ℓD (t), (8)

all of which it pays out as dividends to its shareholders. The firm’s objective
is to maximize its market value, or the present value of its expected profits
under various market structures.
The firm’s behavior is the main focus of this work. For the purpose of
comparison, we first consider the optimal choice of a competitive firm, and the
resulting competitive equilibrium (Section 2.3), and then turn to an economy
where a firm exercises monopoly power in the market for the consumption
good (Section 3). In the latter set-up our assumption of the firm’s maximizing
its value is prone to the criticism that applies to all equilibrium models with
imperfect competition. To this day, the issue whether value maximization
is the appropriate objective is still open (see Remark 1). Our viewpoint in
this paper is to simply adopt the most well-understood equilibrium concept,
8
For simplicity, we do not model the time-0 consumption/leisure and production
choice. All our results for t = 1, . . . , T quantities, in the propositions of the paper,
would remain valid if the time-0 choice were additionally modeled.
8 Suleyman Basak and Anna Pavlova

the Cournot-Walrasian equilibrium, and despite its possible criticisms, focus

on the implications. In particular, we interpret our representative consumer-
shareholder as a standard Walrasian price-taking agent. As such, he perceives
his decisions as having no effect on the valuation of his consumption/labor
stream, given by the left-hand side of the budget constraint (3). On the other
hand, as a shareholder, he can affect the quantity on the right-hand side of
(3), his initial wealth, which is equal to the value of the firm. Monotonicity of
his indirect utility of wealth, then, justifies the firm’s value maximization as
a desired objective of its shareholder.

2.3 Benchmark Competitive Equilibrium

The objective of the competitive firm is choose the input ℓD so as to maximize
its time-0 value, V (0), taking the state prices and wage as given. The first-
order conditions for a competitive firm’s problem are given by
fℓ (ℓD (t), ε(t)) − w(t) = 0 t = 1, . . . , T. (9)
Definition 1 (Competitive Equilibrium). An equilibrium in an economy
of one competitive firm and one representative consumer-worker is a set of
c
prices (ξ c , wc ) and choices (cc , ℓc , ℓD ) such that (i) the consumer chooses
his optimal consumption/labor policy given the state price and wage processes,
(ii) the firm makes its optimal labor choice given the state prices and wage,
and (iii) the consumption and labor markets clear at all times:
c c
cc (t) = f (ℓD (t), ε(t)) and ℓc (t) = ℓD (t) . (10)
It is straightforward to show from (7)–(10) that in the competitive equi-
c
librium, the equilibrium labor ℓc = ℓD is given by
v ′ (ℓ̄ − ℓc (t))
fℓ (ℓc (t), ε(t)) − =0 (11)
u′ (f (ℓc (t), ε(t)))
and the equilibrium consumption and profit processes by
cc (t) = f (ℓc (t), ε(t)) , π c (t) = f (ℓc (t), ε(t)) − fℓ (ℓc (t), ε(t)) ℓc (t) . (12)
The equilibrium state price density, wage and firm value processes are given
by
ξ c (t) = u′ (f (ℓc (t), ε(t))) , wc (t) = fℓ (ℓc (t), ε(t)) , (13)
c
V (t) =
T
u′ (f (ℓc (s), ε(s)))  c
 

c c
E f (ℓ (s), ε(s)) − fℓ (ℓ (s), ε(s))ℓ (s) Ft . (14)

u ′ (f (ℓc (t), ε(t)))
s=t

The above conditions present a fully analytical characterization of the

equilibrium, with (11) determining the labor as a function of the shock ε and
then (12)–(14) determining all other quantities. Equation (11) states that the
marginal rate of substitution between consumption and leisure is equated to
the marginal product of labor.
 Monopoly Power and the Firm’s Valuation 9

3 Monopolistic Equilibrium
In this section, we assume the consumption good market to be imperfect,
in that the firm has monopoly power therein. We take the firm to act as a
non-price-taker in its output market, taking into account the impact of its
production plan choice on the price of output. The firm is still a price-taker
in its input/labor market, taking the wages w as given. (The extension to the
case where the firm is additionally a non-price-taker in the labor market is
straightforward and is discussed in Appendix C.)
We will observe that the firm’s production strategy is time-inconsistent, in
the sense that a monopolist has an incentive to deviate from his time-0 plan at
a later date. When the monopolist gets to time t, he no longer cares about the
time-0 value of the firm; rather, he would like to revise the production plan so
as to instead maximize the time-t value of the firm. In Section 3.1, we consider
the optimal choice of a monopolistic firm that chooses an initial strategy
so as to maximize its time-0 value and “pre-commits” to that strategy, not
deviating at subsequent times. In Section 3.2, we solve for the time-consistent
strategy of a monopolist who re-optimizes his production plan to maximize
the firm’s current value at each date t, taking into account the fact that he is
not restricted from revising this plan at the future dates s = t + 1, . . . , T . The
former can be thought of as a “long-term” strategy, providing the first-best
solution to the problem of maximizing the firm’s initial value. The latter can
be interpreted as a “short-term” or “short-sighted” strategy since the firm
continually re-optimizes every period to boost current value. We present the
equilibrium for both cases.

3.1 The Pre-Commitment Case

We now formulate the monopolist’s optimization problem. Recalling the price-

taking consumer’s demand (4), clearing in the consumption good market im-
plies  
ξ(t) = u′ f (ℓD (t), ε(t)) /y . (15)

The monopolist’s influence in the good market manifests itself, via (15), as
a (non-linear) impact of its input demand on state prices. Accordingly, the
pre-committed monopolist solves the following optimization problem at time
0:
 
max V (0) subject to ξ(t) = u′ f (ℓD (t), ε(t)) /y , ∀ t = 1, . . . , T , (16)
ℓD , ξ

where y satisfies (6).

Proposition 1 presents the optimal solution to this problem, assuming it
exists.9
9
It is well-known that the objective function in models with monopolistic
firms may not necessarily be concave and hence not satisfy the second-order
10 Suleyman Basak and Anna Pavlova

Proposition 1. The pre-commitment monopoly optimal labor demand ℓD ,

t = 1, . . . , T satisfies

fℓ (t) − w(t) = A(t) fℓ (t)π(t) > 0 , (17)

where u′′ (t)
A(t) ≡ − ′ ,
u (t)

and f (t), u(t) and their derivatives are shorthand for f (ℓD (t), ε(t)),
u(f (ℓD (t), ε(t))) and their derivatives.
The structure of the first-order conditions bears resemblance to that of
the textbook single-period monopolist. Any direct increase in profit due to an
extra unit of input must be counterbalanced by an indirect decrease via the
impact of that extra unit on the concurrent price system. In our set-up, the
extent of monopoly power is driven by the current profit, the marginal product
of labor and the (positive) quantity A, which captures the consumer’s attitude
toward risk over consumption. (The quantity A can also be restated in terms
of the textbook “monopoly markup.”) The higher the marginal product the
more responsive is output to an extra unit of input. The more “risk-averse”
the consumer, the less his consumption reacts to changes in the state price,
or conversely, the more the state price reacts to changes in his consumption,
and so the more incentive the monopolist has to deviate from competitive
behavior. In the limit of a risk-neutral investor (preferences quasi-linear with
respect to consumption), the monopolist cannot affect the state price at all
and so the best he can do is behave competitively.
We now turn to an analysis of equilibrium in this economy.
Definition 2 (Monopolistic Pre-Commitment Equilibrium). An equi-
librium in an economy of one monopolistic firm and one representative
consumer-worker is a set of prices (ξ ∗ , w∗ ) and choices (c∗ , ℓ∗ , ℓD∗ ) such
that (i) the consumer chooses his optimal consumption/labor policy given the
state price and wage processes, (ii) the firm makes its optimal labor choice in
(16) given the wage, and taking into account that the price system responds
to clear the consumption good market, and (iii) the price system is such that
the consumption and labor markets clear at all times:
c∗ (t) = f (ℓD∗ (t), ε(t)) and ℓ∗ (t) = ℓD∗ (t).
The fully analytical characterization of the equilibrium in the monopolistic
pre-commitment economy is given by (18)–(21). Equilibrium is determined by
conditions without additional assumptions (e.g., Tirole (1988, Chapter 1)). In
the pre-commitment case a sufficient condition for concavity is u′ (c)fℓℓ (ℓ, ε) +
2u′′ (c)fℓ (ℓ, ε)(fℓ (ℓ, ε)−w)+u′′ (c)fℓℓ (ℓ, ε)(f (ℓ, ε)−wℓ)+u′′′ (c)fℓ2 (ℓ, ε)(f (ℓ, ε)−wℓ) <
0, c = f (ℓ, ε); ∀ε, ℓ, w. Examples of utilities and production functions that satisfy
this condition include the commonly employed power preferences over consumption
u(c) = cγ /γ with γ ∈ (0, 1) and power production f (ℓ, ε) = εℓν , ν ∈ (0, 1) (no
additional restriction on v(h) is required).
 Monopoly Power and the Firm’s Valuation 11

computing the supply and demand for labor from the consumer’s and firm’s
first-order conditions, and then applying labor market clearing ℓ∗ = ℓD to
∗

10
yield the labor as a function of the shock ε (18). The remaining quantities are
then straightforward to determine; we list them in (19)–(21). The equilibrium
labor is given by

v ′ (ℓ̄ − ℓ∗ (t))
fℓ (ℓ∗ (t), ε(t)) −
u′ (f (ℓ∗ (t), ε(t)))
v ′ (ℓ̄ − ℓ∗ (t))


= A(t)fℓ (ℓ∗ (t), ε(t)) f (ℓ∗ (t), ε(t)) − ℓ∗

(t)
u′ (f (ℓ∗ (t), ε(t)))
(18)

and the equilibrium consumption and profit processes by

v ′ (ℓ̄ − ℓ∗ (t))
c∗ (t) = f (ℓ∗ (t), ε(t)) , π ∗ (t) = f (ℓ∗ (t), ε(t)) − ℓ∗ (t) . (19)
u′ (f (ℓ∗ (t), ε(t)))

The equilibrium state price density, wage and firm value processes are given
by

v ′ (ℓ̄ − ℓ∗ (t))
ξ ∗ (t) = u′ (f (ℓ∗ (t), ε(t))) , w∗ (t) = , (20)
u′ (f (ℓ∗ (t), ε(t)))
V ∗ (t) =
T
u′ (f (ℓ∗ (s), ε(s)))  ∗ v ′ (ℓ̄ − ℓ∗ (s))
 

∗
E f (ℓ (s), ε(s)) − ′ ℓ (s) Ft . (21)

u ′ (f (ℓ∗ (t), ε(t))) u (f (ℓ ∗ (s), ε(s)))
s=t

Proposition 2 summarizes the comparison of pertinent quantities across

the monopolistic and competitive economies.
Proposition 2. The equilibrium labor, output, state price and wage in the
pre-commitment monopoly economy and competitive economy, satisfy for all
t = 1, 2, . . . T , all ε(t):

ℓ∗ (t) < ℓc (t), f (ℓ∗ (t), ε(t)) < f (ℓc (t), ε(t)), c∗ (t) < cc (t) (22)
ξ ∗ (t) > ξ c (t), ∗
w (t) < w (t).c
(23)

The firm’s initial value satisfies,

V ∗ (0) > V c (0).

However, the time-t profit π ∗ (t) > 0 and firm’s value V ∗ (t) can be either
higher or lower than π c (t) and V c (t), respectively, ∀ t = 1, . . . , T.

10
See Lemma A.1 of Appendix A for the existence of an interior solution ℓ∗ (t) ∈
(0, ℓ̄) to equation (18).
12 Suleyman Basak and Anna Pavlova

The monopolist has an incentive to manipulate the price of consumption by

producing less than his competitive counterpart. Reducing the supply raises
the price at which clearing occurs, thereby increasing the valuation of the
monopolist’s intertemporal profits along with his stock price. The firm needs
to use less input and, accordingly, the wage rate decreases. These results are
consistent with the textbook static monopoly analysis. A difference in our
model is that the flow of profits, π(t), in the monopolistic equilibrium may be
lower than that in the competitive benchmark (as illustrated in Example 1 of
Section 3.2); this is because it is the present value of profits that our monop-
olist is striving to maximize, not profits per se. By restricting production, the
monopolist moves away from the competitive, profit maximizing, production
plan. Since the monopolist is maximizing the firm’s initial value and since he
has market power, it is intuitive that the firm’s initial value will be higher
than in the competitive case. However, we have not forced him to maximize
later values of the firm, so it is not surprising that these later values may lie
higher or lower than the competitive case (as illustrated in Example 1).
The optimal policies and equilibrium in this subsection rely on the ability
of the monopolist to pre-commit to his time-0 plan. Without this commitment,
the monopolist would later want to deviate from his initial plan; that is, at
any date t > 0, the solution to the firm’s problem
 
max V (t) subject to ξ(s) = u′ f (ℓD (s), ε(s)) /y , ∀ s = t, . . . , T
ℓD (s), ξ(s), s≥t

is different from his time-0 plan, unless V (t) = 0. Due to the intertemporal
dependence of V on the price of consumption, if the monopolist solves his prob-
lem at the initial period, he would change his mind at a later stage about the
optimal ξ process. In other words, his production plan is not time-consistent.
A similar problem arises in the context of a durable good monopoly (e.g., Ti-
role (1988, Chapter 1)), or in the context of dynamic securities markets with
non-price-taking investors (Basak (1997), Kihlstrom (1998)). The firm’s stock
in our model is similar to a durable good since it provides value over many
periods. If there were a credible mechanism for the monopolist to commit to
his time-0 plan, the time-0 share price of the firm that he owns would be ce-
teris paribus the highest possible; if however such commitment is impossible,
we will show that the monopolist might be better off behaving competitively
(Example 1). Possible pre-commitment mechanisms may include (i) employ-
ing a production technology with a time-to-build feature, and following Tirole
(1988), (ii) handing the firm over to a third party instructed to implement
the optimal strategy with a penalty for deviating from it (although renegoti-
ation would be a potential issue), (iii) long-term relationships/contracts, (iv)
“money-back guarantee” penalizing deviations from a targeted labor demand.
Modeling additionally a pre-commitment mechanism is beyond the scope of
our analysis. However, we provide the pre-commitment solution since we view
it as a valuable yardstick against which we compare the time-consistent solu-
 Monopoly Power and the Firm’s Valuation 13

tion, reported below. Moreover, the pre-commitment solution also resembles

the standard textbook monopoly behavior.

3.2 The Time-Consistent Case

We now turn to the time-consistent monopolist’s optimization problem. For-

mally, the time-consistency requirement imposes an additional restriction on
the firm’s production choice: at no time t can the monopolist be willing to
deviate from his time-0 strategy. If the firm is not restricted from revising
its strategy at dates t = 1, . . . , T , it should optimally choose the intertempo-
ral production plan taking into account that it will always act optimally in
the future. We proceed to find the monopolist’s subgame perfect strategy by
backward induction; this is a time-consistent strategy. Specifically, the mo-
nopolist chooses the current input and price of consumption to maximize the
firm’s current value given the firm’s future maximized value ∀ t = 0, . . . , T,
i.e., solves the dynamic program:

V (t) ≡max V (t)

ℓD (t), ξ(t)
 
subject to ξ(t) = u′ f (ℓD (t), ε(t)) /y , (24)
V (s) = V (s) , (ℓD (s), ξ(s)) = argmax V (s) , ∀ s = t + 1, . . . , T,

where y satisfies (6).

Proposition 3 presents the optimal solution to this problem, assuming it
exists.11
Proposition 3. The time-consistent monopoly optimal labor demand ℓD , t =
1, . . . , T − 1 satisfies

fℓ (t) − w(t) = −A(t) fℓ (t) Vex (t) < 0 , (25)

where Vex (t) denotes the ex-dividend value of the firm given by

T 
 u′ (s) 
Vex (t) ≡ E [f (s) − w(s) ℓ(s)] Ft ∀ t = 1, . . . , T − 1;

s=t+1
u′ (t)
Vex (T ) = 0.

At time T , the time-consistent monopoly optimal labor demand ℓD satisfies

fℓ (T ) − w(T ) = 0.

11
In the time-consistent case a sufficient condition for concavity of the objective
′′ ′′′
function is −2 uu′ (c)
(c)
+ uu′′ (c)
(c)
< − ffℓℓ2 (ℓ,ε)
(ℓ,ε)
, c = f (ℓ, ε); ∀ε, ℓ, satisfied, for example,
ℓ
by power preferences u(c) = cγ /γ with γ ∈ [0, 1), which includes u(c) = log(c) (no
additional restriction on the production function f (ℓ, ε) is required).
14 Suleyman Basak and Anna Pavlova

The ex-dividend value of the firm’s stock at time T is zero; hence the
terminal first-order condition and optimal choice of the firm coincide with
those of the competitive firm. The first-order condition at time T is the ter-
minal condition for the backward induction: the remaining labor choices are
determined by solving (25) backwards.
Similarly to the pre-commitment case, at the optimum for the time-
consistent monopolist the increase in profit due to an extra unit of input used
must counteract the indirect decrease via the impact of that extra unit on the
price system. However, the extent of this monopoly power is now driven by
the negative of the ex-dividend stock price in place of the current profit. (The
consumer’s attitude toward risk for consumption and the marginal product of
labor appear as in the pre-commitment case.) The short-sighted monopolist
only cares about (and tries to manipulate) the current valuation of the stock,
which is made up of current profit plus the ex-dividend value of the firm.
Given the time-consistency restriction, this monopolist takes the future value
of profits as given and can only boost the firm’s ex-dividend value by depress-
ing the current price of consumption. So, while it was in the pre-committed
monopolist’s interest to boost the current price of consumption, it is in the
time-consistent monopolist’s interest to depress it; hence the negative term
in (25). The ex-dividend value of the firm appears because this is what he
manipulates; the higher is Vex (t), the stronger the incentive to cut concurrent
profit to manipulate the firm’s valuation. Hence, the firm’s value serves as an
extra variable in determining the optimal policy of the firm.
We now define an equilibrium in a monopolistic economy containing the
time-consistent firm.
Definition 3 (Monopolistic Time-Consistent Equilibrium). An equi-
librium in an economy of one monopolistic firm and one representative
consumer-worker is a set of prices (ξ, ˆ ŵ) and choices (ĉ, ℓ̂, ℓ̂D ) such that
(i) the consumer chooses his optimal consumption/labor policy given the state
price and wage processes, (ii) the firm chooses a time-consistent strategy so
as to maximize its objective in (24) given the wage, and taking into account
that the price system responds to clear the consumption good market, and (iii)
the consumption and labor markets clear at all times:
ĉ(t) = f (ℓ̂D (t), ε(t)) and ℓ̂(t) = ℓ̂D (t).
The fully analytical characterization of the monopolistic time-consistent
equilibrium is presented in (26)–(29). Again, from labor market clearing, we
first determine the labor as a function of the shock ε (26).12 Then, (27)–(29)
give the remaining equilibrium quantities. The equilibrium labor is given by

v ′ (ℓ̄ − ℓ̂(t))
fℓ (ℓ̂(t), ε(t)) − = −A(t) fℓ (ℓ̂(t), ε(t)) Vex (t) (26)
u′ (f (ℓ̂(t), ε(t)))
12
See Lemma A.3 of the Appendix for the existence of an interior solution ℓ̂(t) ∈
(0, ℓ̄) to equation (18).
 Monopoly Power and the Firm’s Valuation 15

and the equilibrium consumption and profit processes by

v ′ (ℓ̄ − ℓ̂(t))
ĉ(t) = f (ℓ̂(t), ε(t)) , π̂(t) = f (ℓ̂(t), ε(t)) − ℓ̂(t) . (27)
u′ (f (ℓ̂(t), ε(t)))
The equilibrium state price density, wage and firm value processes are given
by

ˆ = u′ (f (ℓ̂(t), ε(t))) , v ′ (ℓ̄ − ℓ̂(t))

ξ(t) ŵ(t) = , (28)
u′ (f (ℓ̂(t), ε(t)))
T


 u′ (f (ℓ̂(s), ε(s)))  v ′ (ℓ̄ − ℓ̂(s)) 
V̂ (t) = E f (ℓ̂(s), ε(s)) − ℓ̂(s) Ft (29)
.

′ u′ (f (ℓ̂(s), ε(s)))
s=t u (f (ℓ̂(t), ε(t)))

Proposition 4 presents a comparison of the monopolistic time-consistent

and the competitive economies.

Proposition 4. The equilibrium labor, output, state price and profit in the
time-consistent monopolistic and the competitive economies satisfy, for all
t = 1, . . . T − 1, all ε(t):

ℓ̂(t) > ℓc (t), f (ℓ̂(t), ε(t)) > f (ℓc (t), ε(t)), ĉ(t) > cc (t) (30)
ˆ < ξ c (t),
ξ(t) ˆ
ŵ(t) > wc (t), ξ(t)π̂(t) < ξ c (t)π c (t), π̂(t) < π c (t). (31)

The firm’s initial value satisfies,

V̂ (0) < V c (0).

At time T , all the equilibrium quantities coincide with those of the competitive
benchmark. The firm’s value, V̂ (t), can be either higher or lower than in the
competitive economy, V c (t), t = 0, . . . , T − 1.

The short-sighted monopolist’s behavior is exactly opposite to that of the

monopolist who pre-commits. He desires to depress the current price of con-
sumption (to boost the current stock price); hence the lower level of state
price density. He achieves this by increasing output, hence demanding more
labor input, and in return pushing up the wage rate. Although this monop-
olist moves his labor demand in the opposite direction to the pre-committed
monopolist, he also moves away from the maximum profit point and so prof-
its are reduced relative to the competitive case. In his attempt to manipulate
the stock price, concurrent profit and its value are cut. This type of behav-
ior may overall result in a higher or lower value of the firm’s stock price. By
consistently reducing the value of concurrent profit, the short-sighted owners
of the monopolistic firm may, then, cause damage to the firm and its stock
price. This is so even though the time-consistent monopolist is restricted to
maximize the firm’s value at each point in time. For the pre-committed mo-
nopolist, we could explain this result by his not attempting to maximize this
16 Suleyman Basak and Anna Pavlova

quantity, but here the explanation must be more complex (see Examples 1
and 2).
We now present three examples which deliver additional insights into the
equilibrium quantities. The first example demonstrates that the monopolistic
firm’s profits can go negative in equilibrium. The second example allows an
investigation of the relationship between the monopolistic and the competitive
firm values, as well as the infinite horizon limit. The third example shows
how under an alternative, albeit extreme, form of a production opportunity
the time-consistent monopolist’s profits go to zero, while the pre-committed
monopolist’s profits do not.
Example 1 (Negative Profits) Here, we provide a numerical example in
which the time-consistent monopolistic firm’s equilibrium profits π̂(t) are
sometimes negative, and its stock price is always lower than that of a com-
petitive firm. Consider the following parameterization:
cγ (ℓ̄ − ℓ)ρ
u(c) + v(ℓ̄ − ℓ) = + , γ ∈ (0, 1), ρ < 1;
γ ρ
f (ℓ, ε) = ε + ℓν , ν ∈ (0, 1); T = 2.
The productivity shock enters the technology additively; this type of shock
can be interpreted as providing an additional endowment as in a Lucas (1978)-
type pure-exchange economy. We set ℓ̄ = 1, γ = 0.5, ρ = ν = 0.9, ε(1) =
0.1, ε(2) = 0.5 (deterministic for simplicity of exposition). The resulting
equilibrium labor choices, profits and stock prices are reported in Table 1.

Table 1. Equilibrium labor choice, profits and firm value in the com-
petitive, monopolistic pre-commitment and monopolistic time-consistent
economies. The reported values are for the economies parameterized by u(c) +
v(ℓ̄ − ℓ) = cγ /γ + (ℓ̄ − ℓ)ρ /ρ, f (ℓ, ε) = ε + ℓν , T = 2, with ℓ̄ = 1, γ = 0.5, ρ = ν =
0.9, ε(1) = 0.1, ε(2) = 0.5

Monopolistic Monopolistic
Competitive
Pre-Commitment Time-Consistent
Equilibrium
Equilibrium Equilibrium

ℓ(1) 0.60 0.41 0.84

ℓ(2) 0.37 0.01 0.37
π(1) 0.16 0.22 -0.27
π(2) 0.54 0.51 0.54
V (0) 0.75 1.02 0.54
V (1) 0.65 0.75 0.52
V (2) 0.54 0.51 0.54

At time 1 the time-consistent monopolist sacrifices his concurrent profit in

order to boost the time-1 value of the stock. Indeed, consistent with Proposi-
 Monopoly Power and the Firm’s Valuation 17

tion 4, depending on the parameterization, nothing prevents the sign of π̂(1)

from becoming negative. This is a notable feature of our solution. Although
real-world firms frequently announce losses, this phenomenon cannot be gen-
erated in a competitive model where production occurs within the decision
period (provided the firm has an option to costlessly shut down during that
period). Negative profits can obtain in models of predation, where rivalrous
producers are competing for market share (e.g., Tirole (1988, Chapter 9)); in
our model, in contrast, the monopolist controls the entire market and does
not fear entry, yet his profit may still go negative.
A further striking conclusion revealed by Table 1 is that non-competitive
behavior of the firm does not necessarily result in a higher value of the firm;
in fact, the time-consistent monopolistic firm’s stock price is lower than its
competitive counterpart at all times. The recognition of his monopoly power
actually damages the time-consistent monopolist at all points in time, and
he would be better off behaving competitively. For completeness and com-
parison we also provide the equilibrium quantities for the monopolistic pre-
commitment economy. Example 2, having a longer time horizon, and yielding
an explicit solution, allows us to investigate this stock price effect more closely.
Example 2 (Firm Value Comparison and Time Horizon Effects)
Since it appears that the intertemporal nature of the problem is critical, in
this example we extend the horizon beyond T = 2. It would also be important
to discuss the limiting case of our model as T → ∞, since in some models
(e.g., repeated principal-agent), the equilibrium changes dramatically on tran-
sitioning from a finite to infinite horizon. In order to obtain explicit formulae
for all equilibrium quantities, we simplify beyond the general specification
of the utility function employed so far. The utility we adopt has the form
u(c(t), t) − v(ℓ(t), t) = β t (log c(t) − ℓ(t)ρ /ρ), β ∈ (0, 1), ρ > 1, where β is
the time discount factor, required to make pertinent quantities well-defined
in the limiting case of T → ∞, and the second term is the disutility of labor.
We show equilibrium labor to still be bounded from above by ℓ̄ = 1. The
firm’s technology is represented by f (ℓ(t), ε(t)) = ε(t) ℓ(t)ν , ν ∈ (0, 1). To
ensure that the equilibrium quantities are well-defined in the infinite horizon
limit, we assume that ε is bounded, as well as assume away speculative bub-
bles. The resulting exact formulae for the equilibrium labor choices, profits
and stock prices are reported in Table 2, with the comparative statics as in
Proposition 4.13 The time-consistent labor choice monotonically increases as
T − t increases, converging to a limit (and diverging away from the competi-
tive labor, which is constant). As T − t increases, time-consistent profits fall,
13
We do not report the monopolistic pre-commitment equilibrium quantities since
for logarithmic utility over consumption, the monopolist’s problem is not well-
defined, as is the case in the standard textbook treatment of monopoly (Mas-Colell,
Whinston and Green (1995, p.429)). Moreover, one may extend our earlier analy-
sis of the monopolistic pre-commitment and competitive equilibria to the case of
infinite horizon without major difficulties.
18 Suleyman Basak and Anna Pavlova

although do not vanish. At this point, the parallel with to the Coase (1972)
conjecture is unavoidable: ceteris paribus, the more future production deci-
sions the time-consistent firm can make, the lower its profits today. Does the
firm erode its profits by allowing for more future decisions? We revisit this
issue in Section 4.

Table 2. Equilibrium labor choice, profits, and firm value and their lim-
its as T − t → ∞ in the competitive and monopolistic time-consistent
economies. The reported formulae are for the economies parameterized by
u(c(t), t) − v(ℓ(t), t) = β t (log c(t) − ℓ(t)ρ /ρ), β ∈ (0, 1), ρ > 1; f (ℓ(t), ε(t)) =
ε(t) ℓ(t)ν , ν ∈ (0, 1)

Competitive Monopolistic Time-Consistent

Equilibrium Equilibrium

 T −t+1
1/ρ
ℓ(t) ν 1/ρ ν 1−(β(1−ν))
1−β(1−ν)
 T −t+1
ν/ρ
π(t) ε(t) ν ν/ρ (1 − ν) ε(t) ν 1−(β(1−ν))
1−β(1−ν)
T −t+1
× (1−ν)(1−β)+ν(β(1−ν))
1−β(1−ν)
 ν/ρ+1
1−(β(1−ν))T −t+1
V (t) ε(t) ν ν/ρ (1 − ν) ε(t) ν ν/ρ (1 − ν) 1−β(1−ν)
T −t+1
× 1−β1−β
 1/ρ
lim ℓ(t) ν 1/ρ ν
1−β(1−ν)
T −t→∞
 ν/ρ
(1−ν)(1−β)
lim π(t) ε(t) ν ν/ρ (1 − ν) ε(t) ν
1−β(1−ν) 1−β(1−ν)
T −t→∞
ε(t) ν ν/ρ (1−ν) ε(t) ν ν/ρ (1−ν)
lim V (t) 1−β [1−β(1−ν)]ν/ρ+1
T −t→∞

Of special interest is the value of the firm’s stock. In Figure 1, we plot

the firm value in the competitive and monopolistic equilibria as a function of
time. The monopolistic firm value can be either higher or lower than that of
the competitive firm, depending on the age of the firm. At first blush, this
result seems surprising, because the time-consistent monopolist is optimizing
the firm’s value at each point in time. Since he has more market power than a
competitive firm, how can his firm’s value come out lower? At time t he solves
the problem

V̂ (t) = max f (ℓD (t), ε(t)) −ŵ(t)ℓD (t)

ℓD (t),ξ(t)
1  
ˆ + 1) V̂ (t + 1) |Ft .
+ E ξ(t
u′ (f (ℓD (t), ε(t)))
The competitive firm solves the same problem except that: it has no power
ˆ + 1) are replaced by their respective
over ξ(t); and that ŵ(t), V̂ (t + 1), ξ(t
 Monopoly Power and the Firm’s Valuation 19

values in the competitive equilibrium. This latter distinction explains why

the monopolist’s firm value may come out lower: he faces both a different
equilibrium wage and different choices of the future state prices and firm
value.

V (t)

2.5
V c (t)

2 V̂ (t)

1.5

0.5

0 1 2 3 4 5
t

Fig. 1. Equilibrium firm value versus time in the competitive and mo-
nopolistic time-consistent economies. The dotted plot is for the competitive
economy and the solid plot is for the monopolistic time-consistent. The economies
are parameterized by u(c(t), t) − v(ℓ(t), t) = β t (log c(t) − ℓ(t)ρ /ρ), β = 0.9, ρ =
1.05; f (ℓ(t), ε(t)) = ε(t) ℓ(t)ν , ν = 0.2, ε(t) = 1, ∀t; T = 5

Figure 1 also reveals that in this example the monopolistic firm increases
the firm’s value later in its lifetime and decreases it earlier in life. This is
because the monopolistic firm competes with itself in future time periods;
earlier in life it faces more competition, which adversely affects its value. In
the limit of T − t → ∞, the competitive firm value is unambiguously higher
(Table 2).
Example 3 (Constant Returns to Scale Technology) A constant ret-
urns to scale technology is widely employed in both the monopoly literature
(e.g., Tirole (1988)), and asset pricing literature (e.g., the workhorse produc-
tion asset pricing model of Cox, Ingersoll and Ross (1985)). Thus far, we have
been assuming strictly decreasing returns to labor; in this example, we extend
our analysis to the case of constant returns, f (ℓ, ε) = εℓ. The firm’s profit is
now given by π(t) = ε(t)ℓ(t) − w(t)ℓ(t).
20 Suleyman Basak and Anna Pavlova

For the competitive firm to demand a finite positive amount of labor, the
equilibrium wage wc (t) must equal ε(t); the equilibrium labor ℓc (t) is then
determined from the consumer’s optimization (7):

v ′ (ℓ̄ − ℓc (t))
= ε(t). (32)
u′ (f (ε(t), ℓc (t)))

The remaining equilibrium quantities, for all t = 1, . . . , T , are

cc (t) = f (ℓc (t), ε(t)), π c (t) = 0,

ξ c (t) = u′ (f (ℓc (t), ε(t)), V c (0) = V c (t) = 0.

Adaptation of our analysis in Section 3.1 shows that the pre-commitment

monopolistic equilibrium labor demand ℓ∗ (t) (assuming it exists and yields a
maximum in the firm’s problem) solves

u′′ (f (ℓ∗ (t), ε(t)))

− ε(t)ℓ∗ (t) = 1 , (33)
u′ (f (ℓ∗ (t), ε(t)))

implying ℓ∗ (t) < ℓc (t). The remaining equilibrium quantities are obtained
from (19)–(21). The pre-commitment equilibrium, then, retains the main
implications derived in Section 3.1; in particular, the firm cuts production
and raises the price of output. The main difference from Section 3.1 is that
monopoly profits are now always higher than the competitive ones (which are
zero), consistent with the prediction of the textbook monopoly model. The
firm’s value is also higher.
The monopolistic time-consistent equilibrium coincides with the compet-
itive one. To see why, recall from Proposition 4, that at the final time,
π̂(T ) = π c (T ). Since π c (T ) = 0, the time-(T -1) ex-dividend value of the
monopolistic firm, V̂ex (T − 1), is zero; hence by backward induction it can be
shown that ℓ̂(t) = ℓc (t) ∀t, and the competitive equilibrium obtains:

ĉ(t) = f (ℓc (t), ε(t)), π̂(t) = 0,

ˆ = u′ (f (ℓc (t), ε(t)), V̂ (0) = V̂ (t) = 0,
ξ(t) t = 1, . . . T.

In contrast to the decreasing returns to scale case, the time-consistent mo-

nopolist loses all his monopoly power. Accordingly, the value of his stock, as
well as profit, are strictly lower than in the pre-commitment equilibrium. It
is quite clear that the short-sighted behavior has a devastating effect on the
firm’s valuation. Ironically, it benefits the representative consumer, who, in
effect, sets the labor demand of the firm so as to maximize his own expected
lifetime utility.
 Monopoly Power and the Firm’s Valuation 21

Remark 1 (Price Normalization in Monopolistic Models).

Ours is a general equilibrium model with imperfect competition; accord-
ingly, it is not immune to the general criticism of all such models initiated by
Gabszewicz and Vial (1972) (see also Mas-Colell (1982)). The criticism has
to do with the so-called “price-normalization problem” which arises in the
presence of multiple goods because Walrasian (and Cournot-Walrasian) equi-
librium theory determines relative prices but has little to say about nominal
price formation. In the perfectly competitive benchmark, this deficiency is not
an issue, since the nominal price indeterminacy does not affect real quantities.
In the monopolistic (imperfectly competitive) equilibrium, however, the real
quantities become dependent on the choice of price normalization. Indeed, for
some choices of normalization, Dierker and Grodal (1986) show that no equi-
librium exists, while for others it does exist. This prompted us to consider
the robustness of our results to alternative specifications of the numeraire.
Our model contains two commodities, the consumption good and labor, at
each time and state. Instead of the consumption good, we could have speci-
fied a basket of commodities consisting of α units of the consumption good
and (1 − α) units of labor as the numeraire, α ∈ (0, 1).14 Our main quali-
tative results and corresponding intuitions would then still remain valid, as
demonstrated in Appendix B.
Related to the price-normalization problem, is the issue of whether maxi-
mizing the firm’s value through time is the correct objective. From the point-
of-view of realism, it is generally accepted that this is what a firm should do,
so the problem we solve is consistent with conventional thinking. However,
the literature such as Dierker and Grodal (1986) (see also Dierker and Grodal
(1999)), clearly casts some doubt on this approach. Hart (1985) suggests that
profit maximization may be inappropriate in a monopolistic world because
“the owners of the firm are interested not in monetary profit per se but in
what that profit can buy”. Although Hart discusses maximization of owner’s
utility as an alternative objective, he points out that the question of how to
aggregate is not resolved when there are multiple owners. Similarly, Bonanno
(1990) (in his survey) remarks, “unfortunately we are still far from a satisfac-
tory theory of general equilibrium with imperfect competition”. To this day,
the price normalization and the objective of the firm under imperfect com-
petition in general equilibrium remain open issues. Based on this, we have
simply adopted the most commonly employed price normalization, as well
as the most conventionally considered objective function as the best starting
point, and we focus on the implications.

14
Such a basket is a natural and realistic numeraire to adopt, as argued, for
example, by Pavlova and Rigobon (2003) (for more discussion of prices indexes, see
Schultze (2003)).
22 Suleyman Basak and Anna Pavlova

4 Connection to the Coase Conjecture

As we discussed in Section 3, the time-inconsistency of the firm’s production

plan in our monopolistic pre-commitment equilibrium is similar to the time-
inconsistency arising in the context of a durable good monopoly (e.g., Tirole
(1988, Chapter 1) and references therein). The long-lived firm’s stock in our
model is similar to a durable good in that its value is durable over many
periods. The durability of goods implies that a monopolistic firm in a sense
competes with itself in pricing goods that it produces at other times and must
take this into account in its production behavior. Similarly, in our context,
“current owners” of the firm can be thought of as competing with “future
owners” in maximizing the value of the firm’s stock; they must account for
possible revisions to today’s intertemporal production plan by the future own-
ers. This effective competition between past and future sales in the durable
good problem tends to weaken the monopoly power and may ultimately en-
force competitive behavior. This observation is the Coase (1972) conjecture:
as the time interval between successive decision points shorten, the price of the
durable good charged by the monopolist converges to the competitive price
(marginal cost), and the monopoly profits are driven to zero.
To examine the extension of Coase’s intuition to the case of a long-lived
monopolistic firm manipulating the price of its stock, we explore the limiting
case of our economy as the decision-making interval shrinks.15 We partition
the time horizon [0, T ] in our economy into n small intervals of length ∆, so
that t = 0, ∆, 2∆, . . . , n∆ = T . All flow variables in the economy (ℓ(t), c(t),
f (t) and π(t)) are now interpreted as “flows over the interval t to t + ∆”. We
then take the limit as n → ∞ (∆ → 0). Proposition 5 reports the resulting
optimality conditions and equilibrium characterization for the competitive,
monopolistic pre-commitment and monopolistic time-consistent economies,
assuming appropriate regularity conditions are satisfied.

Proposition 5. In the continuous-time limit, as n → ∞, ∀t:

(i) The competitive firm’s optimal labor demand ℓD satisfies

fℓ (ℓD (t), ε(t)) − w(t) = 0. (34)

Consequently, the competitive equilibrium characterization is given by the

conti-nuous-time analogs of equations (11)–(14).

15
Our goal here is to merely explore what happens to equilibrium quantities as
the firm’s decisions become more frequent, and to not present a continuous-time
extension of our model. Similar discrete-time models with T = ∞ have been argued
to exhibit multiple equilibria as the decision-making interval shrinks, according to
the Folk Theorem (see also Ausubel and Deneckere (1989)), which motivated us to
keep the horizon T finite.
 Monopoly Power and the Firm’s Valuation 23

(ii) The pre-commitment monopolistic firm’s optimal labor demand ℓD satis-

fies
fℓ (ℓD (t), ε(t)) − w(t) = A(t) fℓ (t) π(t) > 0. (35)
Consequently, the monopolistic pre-commitment equilibrium characteriza-
tion is given by the continuous-time analogs of equations (18)–(21).

(iii) Assume further that ε(t) ∈ [k, K], 0 < k < K < ∞. Then the
continuous-time limit exists for the monopolistic time-consistent equilib-
rium. The time-consistent monopolistic firm’s optimal labor demand ℓD
satisfies

π(ℓD (t), ε(t)) = f (ℓD (t), ε(t)) − w(t)ℓD (t) = 0, ∀t. (36)

Consequently, the equilibrium labor for ∀t is given by

v ′ (ℓ̄ − ℓ̂(t))
f (ℓ̂(t), ε(t)) − ℓ̂(t) = 0. (37)
u′ (f (ℓ̂(t), ε(t)))

Furthermore, for ∀t,

ĉ(t) = f (ℓ̂(t), ε(t)), π̂(t) = 0, ˆ = u′ (f (ℓ̂(t), ε(t)),

ξ(t)

v ′ (ℓ̄ − ℓ̂(t))
ŵ(t) = and V̂ (t) = 0.
u′ (f (ℓ̂, ε(t)))
The equilibrium characterization of the competitive and the monopolistic
pre-commitment economies are exactly analogous to the discrete-time charac-
terizations. Our main focus is on the monopolistic time-consistent equilibrium.
There, the firm’s profit and value indeed shrink to zero, as the monopolist’s
decision interval becomes arbitrarily small – in line with the Coase conjec-
ture. In other words, the competing forces between current profits and the
valuation of future profits may diminish shareholder value, fully destroying it
in the continuous-time limit. Note, however, that our monopolist’s profits are
not equated to the competitive ones. This is simply because, in contrast to
the textbook Marshallian framework adopted by Coase where competition im-
plies free entry of firms, competitive behavior in our setting does not yield zero
profits in equilibrium. This distinction is due to our Arrow-Debreu-McKenzie
setting; in particular, to the assumptions that there is a fixed number of firms,
and that the technology of each firm is convex.

5 Conclusion

In this paper, we model a production economy which, along with a standard

representative consumer, includes a large value-maximizing monopolistic firm.
24 Suleyman Basak and Anna Pavlova

The firm manipulates its valuation as well as the price of the good that it pro-
duces. This feature makes its time-0 production plan time-inconsistent. We
address the time consistency problem in two polar ways: first, we assume the
firm can credibly commit to never revoking its time-0 decision (long-term pol-
icy); and second, we assume that the firm takes into account that it will revise
its production plan at each decision point (short-term policy). At a general
equilibrium level, we show that the long-term policy is largely consistent with
the implications of the textbook static monopoly model; as compared to the
competitive economy, the output is decreased and the price of consumption
is increased, yet the profits and the firm’s value can be either increased or
decreased. The short-term policy, however, is at odds with the static model;
the output is increased while the price is decreased. More strikingly, under
the short-term policy, the profits in every period are decreased, and may even
go negative, while the firm’s value can drop below than in the competitive
benchmark. The distinction between the long- and short-term policies be-
comes even sharper in the continuous-time limit of our economy: while the
pre-commitment equilibrium retains its basic discrete-time structure and im-
plications, the time-consistent equilibrium tends to the limit of zero profits
and hence zero firm’s value at all times. Since our primary focus is comparison
between the long- and short-term approaches, and comparison with the text-
book model, we have omitted an analysis of the volatility and risk premium of
the monopolistic firm’s stock price, and of the impact of monopoly power on
the equilibrium interest rate, market price of risk, output and consumption
variabilities. This analysis would be straightforward in the continuous-time
limit of our economy, but less tractable in discrete-time.
Our analysis involving a single monopolistic firm, a single consumption
good, and a representative consumer is, admittedly, simplistic. Our goal in
this paper has been to develop the minimal setting possible capturing the
mechanism through which the firm’s market power may impact valuation
in the economy, and not to produce the most empirically plausible model.
Throughout the paper, we also discuss the robustness of our implications to
alternative modeling approaches. For realism, however, one would need to
extend this work to include multiple imperfectly competitive firms producing
homogeneous, or differentiated, goods, or multiple consumers. The former
(multiple firms) would draw from the results of the oligopoly literature, while
the latter (multiple consumers) would involve aggregation of the consumers’
preferences into a representative agent.

Appendix A. Proofs
Proof of Proposition 1.

Since there is no consumption or production at time 0, we can set ξ(0) = 1/y

w.l.o.g. Upon substitution of (15) into (1), and using the definition (8), we
 Monopoly Power and the Firm’s Valuation 25

obtain an equivalent representation of (16):

T 
 
′
max E u (f (ℓ(t), ε(t)) f (ℓ(t), ε(t)) − w(t)ℓ(t) , (A.1)
ℓ
t=1

where the firm’s objective is now a function of {ℓ(t); t = 1, . . . , T } only. The

first-order conditions for (A.1) are given by

u′′ (f (ℓ(t), ε(t)))fℓ (ℓ(t), ε(t))(f (ℓ(t), ε(t)) − w(t)ℓ(t))

+ u′ (f (ℓ(t), ε(t))) (fℓ (ℓ(t), ε(t)) − w(t)) = 0, ∀t = 1, . . . , T.

Using the definition of A(t) and rearranging above yields (17). The sufficient
condition for concavity in footnote 9 guarantees that the labor solving (17)
is the maximizer for (16). Since the firm has the option to shut down (set
ℓD (t) = 0) during any period [t, t + 1] in this static optimization, π(t) ≥ 0
and u′ (f (ℓ(t), ε(t))π(t) ≥ 0. This together with strict concavity of each term
in (A.1) (by assumption) guarantees that u′ (f (ℓD (t), ε(t))π(t) > 0 at the
optimum, and hence π(t) > 0. π(t) > 0 together with A(t) > 0, fℓ (ℓ, ε(t)) > 0
(by assumptions on preferences and technology), ensures that the expression
on the right-hand side of (17) is strictly positive. ⊓⊔
The following Lemmas A.1 and A.2 are employed in the proofs below.
Lemma A.1 shows that under regularity conditions (satisfied, for example,
by power preferences over consumption u(c) = cγ /γ with γ ∈ (0, 1) and
power production f (ℓ, ε) = εℓν , ν ∈ (0, 1); no additional restriction on v(h)
is required) both equilibrium ℓc and ℓ∗ belong to the interior of [0, ℓ̄]. Lemma
A.2 is employed in the proofs of Propositions 2, 4 and 5.

Lemma A.1.

Under the standard assumptions on preferences and production (see Sections

2.1 and 2.2), there exists a unique solution, ℓc (t) ∈ (0, ℓ̄), to (11). As-
sume further that limℓ→0 u′ (f (ℓ, ε)) f (ℓ, ε)) < ∞, limℓ→0 u′ (f (ℓ, ε)) ℓ < ∞,
limℓ→0 −u′′ (f (ℓ, ε)) f (ℓ, ε)/u′ (f (ℓ, ε)) < 1 and limℓ→0 (u′′ (f (ℓ, ε))) fℓ (ℓ, ε) ℓ +
u′ (f (ℓ, ε)) < ∞ ∀ε. Then there exists a solution, ℓ∗ (t) ∈ (0, ℓ̄), to (18).
Proof of Lemma A.1. Since limℓ→0 u′ (f (ℓ, ε))fℓ (ℓ, ε) = ∞ > limℓ→0 v ′ (ℓ̄ −
ℓ), and limℓ→ℓ̄ v ′ (ℓ̄ − ℓ) = ∞ > limℓ→ℓ̄ u′ (f (ℓ, ε))fℓ (ℓ, ε), and since u′ fℓ
is decreasing in ℓ while v ′ is increasing in ℓ, there exists a unique so-
lution, ℓc ∈ (0, ℓ̄), to (11). Since limℓ→0 u′ (f (ℓ, ε))fℓ (ℓ, ε) − v ′ (ℓ̄ − ℓ) −
A(t)fℓ (ℓ, ε)(u′ (f (ℓ, ε)f (ℓ, ε) − v ′ (ℓ̄ − ℓ)ℓ) = ∞ and limℓ→ℓ̄ u′ (f (ℓ, ε))fℓ (ℓ, ε) −
v ′ (ℓ̄ − ℓ) − A(t)fℓ (ℓ, ε)(u′ (f (ℓ, ε))f (ℓ, ε) − v ′ (ℓ̄ − ℓ)ℓ) = −∞, continuity of
u′ (f (ℓ, ε))fℓ (ℓ, ε) − v ′ (ℓ̄ − ℓ) − A(t)fℓ (ℓ, ε)(u′ (f (ℓ, ε))f (ℓ, ε) − v ′ (ℓ̄ − ℓ)ℓ) on
(0, ℓ̄) together with the boundary behavior at ℓ → 0 and ℓ → ℓ̄ ensures exis-
tence of a solution, ℓ∗ ∈ (0, ℓ̄), to (18). ⊓ ⊔
26 Suleyman Basak and Anna Pavlova

Lemma A.2.
v ′ (ℓ̄−ℓ)
fℓ (ℓ, ε) − u′ (f (ℓ,ε)) is decreasing in ℓ for all ε.

Proof of Lemma A.2. Assumptions on preferences and production and

straightforward differentiation deliver the result. ⊓
⊔

Proof of Proposition 2.

Since the right-hand side of (18) is strictly greater than zero due to Proposition
1, and the right-hand side of (11) is zero, it follows from Lemma A.2 that
ℓ∗ (t) < ℓc (t). The remaining inequalities in (22)–(23) are then straightforward
to derive: they are due to f (ℓ, ε) and w = v ′ (ℓ̄ − ℓ)/u′ (f (ℓ, ε)) being increasing
in ℓ, ∀ε, the good market clearing, and ξ = u′ (f (ℓ, ε)) being decreasing in
ℓ, ∀ε. Equation (6) is automatically satisfied in equilibrium due to clearing in
the good and labor markets, hence y is indeterminate and we can normalize
y = 1. Together with our earlier normalization ξ(0) = 1/y, this yields ξ(0) = 1
in both the competitive and monopolistic pre-commitment equilibria. For w =
w∗ , the function ξπ(ℓ, ε, w) ≡ u′ (f (ℓ, ε))(f (ℓ, ε) − wℓ) achieves its maximum
at ℓ∗ . ξ π is strictly decreasing in w, hence for any w > w∗ , u′ (f (ℓ, ε))(f (ℓ, ε)−
wℓ) < u′ (f (ℓ∗ , ε))(f (ℓ∗ , ε) − w∗ ℓ∗ ), for all ℓ. This together with wc (t) > w∗ (t)
implies that for each term in the expression for V (0) (1), we have ξ ∗ (t)π ∗ (t) >
ξ c (t)π c (t), ∀t. Hence V ∗ (0) > V c (0). Example 1 provides evidence for the last
assertion of the Proposition. ⊓ ⊔

Proof of Proposition 3.

Substituting (15) into (1) and solving (24) at time t = T , we obtain the
optimality condition fℓ (T ) − w(T ) = 0, identical to that of the competitive
firm. Consequently, V (T ) = f (T ) − w(T )ℓD (T ) > 0 and Vex (T ) = 0. At time
t = T − 1, the first-order condition for (24) is
fℓ (ℓ(T − 1), ε(T − 1)) − w(T − 1)
u′′ (f (ℓ(T − 1), ε(T − 1))) fℓ (ℓ(T − 1), ε(T − 1))
−
u′ (f (ℓ(T − 1), ε(T − 1)))2
′
× E[u (fℓ (ℓ(T ), ε(T ))) {f (T ) − w(T )ℓD (T )}|FT −1 ] = 0 .
Using the definition of Vex and rearranging above yields (25) at time T − 1.
Continuing the backward induction, we obtain (25) for all t = 1, . . . T − 1.
The sufficient condition for concavity in footnote 11 guarantees that the labor
solving (25) is the maximizer for (24). V (t), obtained on each step of the
backward induction, is strictly positive because the firm’s objective in (24)
in strictly concave and V (t) ≥ 0 (V (t) < 0 is ruled out since the firm can
shut down at any time and thus increase its value to zero). Consequently,
Vex (t) = E[ ξ(t+1)
ξ(t) V (t + 1) |Ft ] is strictly positive. This together with A(t) > 0
and fℓ (ℓ, ε(t)) > 0 implies that the right-hand side of (25) is strictly negative.
⊓
⊔
 Monopoly Power and the Firm’s Valuation 27

Lemma A.3.

Under the standard assumptions on preferences and production (see Sections

′′ ′′′
2.1, 2.2), and the regularity condition −2 uu′ (c)
(c)
+ uu′′ (c)
(c)
< − ffℓℓ2 (ℓ,ε)
(ℓ,ε)
, c =
ℓ

f (ℓ, ε); ∀ε, ℓ, (footnote 11), there exists a unique solution, ℓ̂(t) ∈ (0, ℓ̄), to
(26).
Proof of Lemma A.3. Since on each step of the backward induction,
limℓ→0 u′ (f (ℓ, ε))fℓ (ℓ, ε) − v ′ (ℓ̄ − ℓ) + A(t)u′ (f (ℓ, ε))fℓ (ℓ, ε)Vex = ∞ and
limℓ→ℓ̄ u′ (f (ℓ, ε))fℓ (ℓ, ε)−v ′ (ℓ̄−ℓ)+A(t)u′ (f (ℓ, ε))fℓ (ℓ, ε)Vex = −∞ and since
u′ (f (ℓ, ε))fℓ (ℓ, ε)−v ′ (ℓ̄−ℓ)+A(t)u′ (f (ℓ, ε))fℓ (ℓ, ε)Vex is decreasing in ℓ, there
exists a unique solution, ℓ̂ ∈ (0, ℓ̄), to (26). ⊓ ⊔

Proof of Proposition 4.

Since the right-hand side of (26) is strictly less than zero due to Proposition 3,
and the right-hand side of (11) is zero, it follows from Lemma A.2 that ℓ̂(t) >
ℓc (t). Consequently, f (ℓ, ε) and w = v ′ (ℓ̄ − ℓ)/u′ (f (ℓ, ε)) being increasing in
ℓ, ∀ε, and the good market clearing yield the comparisons on f , c, w. The
comparison on ξ follows from ξ = u′ (f (ℓ, ε)) being decreasing in ℓ, ∀ε.
To prove the remaining statements, define the equilibrium profit function
v ′ (ℓ̄−ℓ(t))
Π(ℓ(t), ε(t)) ≡ f (ℓ(t), ε(t)) − u′ (f (ℓ(t),ε(t))) ℓ(t). The first-order condition for
maximization of Π(ℓ(t), ε(t)) with respect to ℓ is satisfied by ℓ̃(t) such that

v ′ (ℓ̄ − ℓ̃(t))
fℓ (ℓ̃(t), ε(t)) −
u′ (f (ℓ̃(t), ε(t)))
v ′′ (ℓ̄ − ℓ̃(t)) A(t) fℓ (ℓ̃(t), ε(t)) ′
=− + v (ℓ̄ − ℓ̃(t))
′
u (f (ℓ̃(t), ε(t))) u′ (f (ℓ̃(t), ε(t)))
> 0. (A.2)

Due to Lemma A.2, ℓ̃(t) < ℓc (t), ∀t, ε(t). It is straightforward to verify that
′
for any ℓ such that fℓ − uv′ (f(ℓ̄−ℓ)
(ℓ,ε)) ≤ 0, ∀ε, Πℓ < 0. This together with (26)
implies that Πℓ (ℓc (t)) < 0 and Πℓ (ℓ̂(t)) < 0 ∀t, ε(t). Continuous function
Πℓ (· ; ε) does not change sign on [ℓc (t), ℓ̂(t)], because if it did, there had to be
a point ℓ̆(t) on [ℓc (t), ℓ̂(t)] satisfying Πℓ (ℓ̆(t)) = 0, which is not possible for we
have shown that any such point has to lie to the left of ℓc (t). Consequently,
Πℓ (· ; ε) is monotonically decreasing on [ℓc (t), ℓ̂(t)], and hence π̂(t) < π c (t), ∀t.
It then follows that since ξ(t) ˆ < ξ c (t), ξ(t)π̂(t)
ˆ < ξ c (t)π c (t), and hence from
ˆ V̂ (t) < ξ c (t)V c (t). This together with the argument behind the
(1) that ξ(t)
normalization adapted from the proof of Proposition 2, yields V̂ (0) < V c (0).
Since at time T the equilibrium conditions (11) and (26) of the competitive
and the monopolistic time-consistent economies coincide, ℓ̂(T ) = ℓc (T ), yield-
ing equalization of the remaining equilibrium quantities at time T . Finally, an
28 Suleyman Basak and Anna Pavlova

example can be constructed to verify that V̂ (t) can be lower or higher than
V c (t). Example 2 of Section 3.2 demonstrates this under a modified set of
assumptions on v(h). ⊓ ⊔

Proof of Proposition 5.

Consider an arbitrary partition t = 0, ∆, 2∆, . . . , n∆ = T of [0, T ]. The

consumer-worker’s problem is now given by

T 
 
max E u(c(t)) + v(ℓ̄ − ℓ(t)) ∆
c, ℓ
t=1

T


T 
   
subject to E ξ(t) c(t) − w(t) ℓ(t) ∆ ≤ E ξ(s)π(s)∆ ,
t=1 t=1

where the flow of utility is defined over a rate of consumption and leisure. The
above yields the first order conditions

u′ (c(t))∆ = y ξ(t) ∆ , (A.3)

v ′ (ℓ̄ − ℓ(t))∆ = y ξ(t) w(t) ∆ .

For any ∆, the first-order conditions are equivalent to (4)–(5), consequently,

the consumer demand facing the firm is the same as in the discrete case.
The value of the firm is given by

T −∆ 
 ξ(s) 
V (t) = E (f (ℓ(s), ε(s)) − w(s)ℓ(s))∆ Ft . (A.4)

s=t
ξ(t)

The competitive firm is choosing ℓD to maximize V (0) in (A.4) taking ξ as

given; accordingly, its labor demand ℓD satisfies for all t = ∆, . . . , T − ∆

fℓ (t) ∆ − w(t) ∆ = 0.

The pre-committed monopolist is choosing ℓD and ξ so as to maximize V (0)

in (A.4) subject to ξ(t) = u′ (f (ℓ(t), ε(t)))/y (as implied by (A.3)). His labor
demand ℓD satisfies

fℓ (t) ∆ − w(t) ∆ = A(t)fℓ (t)π(t)∆.

Since the equations above are independent of ∆, equations (34) and (35)
obtain; in addition, the continuous-time limits of the discrete-time competitive
and monopolistic pre-commitment equilibria are now given by the continuous-
time analogs of (11)–(14) and (18)–(21), respectively, for all t ∈ [0, T ].
Similarly, the problem of the time-consistent monopolist is given by (24)
with V (t) specified in (A.4). The backward induction solution yields the fol-
lowing for the firm’s labor demand ℓD
 Monopoly Power and the Firm’s Valuation 29

(fℓ (t) − w(t)) ∆ = −A(t) fℓ (t) Vex (t) ≤ 0 , t = ∆, . . . , T − ∆ . (A.5)

ℓ(t) is bounded from above by ℓ̄, and, anticipating equilibrium, since the right-
hand of (A.5) is nonpositive, it is bounded from below by ℓc (t) > 0 due to
Lemma A.2. ε(t) is bounded by assumption, hence f (ℓ, ε), fℓ (ℓ, ε), u′ (f (ℓ, ε))
and A(t) are bounded. As we take the limit as ∆ → 0 in (A.5), given the
boundedness, its left-hand side tends to zero, hence, so must the right-hand
side. Given the boundedness of all quantities except Vex (t) on the right-hand
side of (A.5), we must have Vex (t) → 0. Consequently, π(t) → 0; in addition,
given our boundedness argument, V (t) → 0, and the firm’s labor demand has
to be such that ℓD (t) yields zero profit π(t) = f (ℓD (t), ε(t)) − w(t)ℓD (t) = 0,
as is stated in (36). In the resulting equilibrium, π̂(t) = 0 as well, hence (37).
(37) yields the equilibrium labor, which in turn determines the remaining
equilibrium quantities. ⊓ ⊔

Appendix B. Alternative Specifications of the Numeraire

Suppose that instead of the consumption good, we specified a basket of com-
modities consisting of α units of the consumption good and (1 − α) units of
labor as the numeraire, α ∈ (0, 1). In other words, the price of this basket is
normalized to 1:

1 = αp(t) + (1 − α)w(t), ∀t = 1, . . . , T, (A.6)

where p(t), w(t) are prices of the consumption good and labor, respectively,
in units of the numeraire.
Under this numeraire, the price-taking consumer-worker’s problem is given
by

T 

max E u(c(t)) + v(ℓ̄ − ℓ(t))
c, ℓ
t=1
T T



   
subject to E ξ(t) p(t)c(t) − w(t) ℓ(t) ≤E ξ(t)π(t) ,
t=1 t=1

where ξ(t) in the state price density process in units of the numeraire basket.
The first-order conditions of this problem are

u′ (c(t)) = y ξ(t)p(t) ,
v ′ (ℓ̄ − ℓ(t)) = y ξ(t) w(t) ,

leading to
v ′ (ℓ̄ − ℓ(t))
p(t) = w(t) . (A.7)
u′ (c(t))
The time-t value of the firm, in units of the numeraire, is given by
30 Suleyman Basak and Anna Pavlova
T


 ξ(s) 
V (t) = E π(s) Ft , (A.8)

s=t
ξ(t)

where π(s) = p(s)f (ℓD (s), ε(s)) − w(s)ℓD (s).

The competitive firm maximizes the time-0 value of the firm, taking all
the prices as given. The optimal labor demand, for all t = 1, . . . , T , of this
firm is given by
1 − (1 − α)w(t)
fℓ (t) − w(t) = 0 , (A.9)
α
where we substituted the consumption good price from (A.6).
A monopolistic firm maximizes the quantity in (A.8) taking into account
the consumer-worker’s demand for its product

u′ (f (ℓ(t), ε(t))) = yξ(t)p(t). (A.10)

The pre-committed monopolist’s problem of maximizing the time-0 value of

the firm subject to (A.10), with p(t) from (A.6), yields the optimal labor
demand, for all t = 1, . . . , T , solving

1 − (1 − α)w(t)
fℓ (t) − w(t) = A(t) fℓ (t)π(t) > 0 . (A.11)
α
By backward induction, we find the time-consistent monopolist’s labor de-
mand, for all t = 1, . . . , T − 1, to be the solution to

1 − (1 − α)w(t)
fℓ (t) − w(t) = −A(t) fℓ (t)Vex (t) < 0 , (A.12)
α
where

Vex (t) ≡
u′ (s) 1 − (1 − α)w(t)
 
1 − (1 − α)w(t)



T
E f (s) − w(s) ℓ(s) Ft

s=t+1
u′ (t) α 1 − (1 − α)w(s)
∀ t = 1, . . . , T − 1 .

At time T , the time-consistent monopoly optimal labor demand ℓD satisfies

1−(1−α)
α fℓ (T ) − w(T ) = 0. The only difference between expressions (A.9),
(A.11), and (A.12) and the corresponding ones in the paper, (9), (17), and
(25), respectively, is the presence of the additional term 1−(1−α)w(t)
α multi-
plying the marginal product of labor – this additional term is the price the
consumption good. Since the right-hand sides of equations (A.11)–(A.12) are
given by the same expressions as in the paper, all our partial equilibrium
implications remain valid.
We now turn to equilibrium comparisons. First, from (A.7), (A.6) and the
good market clearing, c(t) = f (t), the equilibrium consumption good price
and wage satisfy:
 Monopoly Power and the Firm’s Valuation 31

u′ (f (t))
p(t) = > 0,
αu′ (f (t))
+ (1 − α)v ′ (ℓ̄ − ℓ(t))
v ′ (ℓ̄ − ℓ(t))
w(t) = > 0. (A.13)
αu (f (t)) + (1 − α)v ′ (ℓ̄ − ℓ(t))
′

To obtain the equation determining labor ℓc in the competitiveequilib- 

α 1
rium, we multiply (A.9) through by the positive quantity 1−(1−α)w(t) = p(t)
and then substitute in the equilibrium expression for the wage (A.13). Then,
the equilibrium labor ℓc in the competitive economy solves

v ′ (ℓ̄ − ℓc (t))
fℓ (ℓc (t), ε(t)) − = 0. (A.14)
u′ (f (ℓc (t), ε(t)))
α
Analogously, multiplying (A.11) through by 1−(1−α) and using (A.13), we
obtain the expression for the equilibrium labor ℓ∗ in the monopolistic pre-
commitment economy

v ′ (ℓ̄ − ℓ∗ (t))
fℓ (ℓ∗ (t),ε(t)) −
u′ (f (ℓ∗ (t), ε(t)))
v ′ (ℓ̄ − ℓ∗ (t))


= A(t)fℓ (ℓ∗ (t), ε(t)) f (ℓ∗ (t), ε(t)) − ℓ∗

(t) . (A.15)
u′ (f (ℓ∗ (t), ε(t)))

Finally, the equilibrium labor in the monopolistic time-consistent economy ℓ̂

is given by the solution to

v ′ (ℓ̄ − ℓ̂(t))
fℓ (ℓ̂(t), ε(t)) − = −A(t) fℓ (ℓ̂(t), ε(t))
u′ (f (ℓ̂(t), ε(t)))
T


 u′ (f (ℓ̂(s), ε(s)))  v ′ (ℓ̄ − ℓ̂(s)) 
×E f (ℓ̂(s), ε(s)) − ℓ̂(s) Ft .

′ u′ (f (ℓ̂(s), ε(s)))
s=t+1 u (f (ℓ̂(t), ε(t)))
(A.16)

Comparison of (A.14), (A.15), and (A.16) to the corresponding equations

determining the equilibrium labor in the paper (A.14), (A.15), and (A.16),
respectively, reveals that equilibrium labor is invariant to the choice of the
numeraire.
Once we have shown that the equilibrium labor is unchanged, it is straight-
forward to verify that the remaining results listed in Propositions 2 and 4 are
robust as well, apart from the comparison of the state prices ξ(·). However,
the more meaningful quantity to consider instead of ξ(·) in this case is the
price of the consumption good relative to the numeraire basket. The prices
of consumption in the monopolistic pre-commitment, competitive and time-
consistent economies satisfy, for t = 1, . . . , T − 1,

p∗ (t) > pc (t) > p̂(t) ,

32 Suleyman Basak and Anna Pavlova

and p∗ (T ) > pc (T ) = p̂(T ). Similarly to the textbook monopolist, the pre-

committed monopolist in our model raises the price of the good it produces,
while the time-consistent monopolist does the reverse.

Appendix C. The Case of a Monopolistic-Monopsonistic

Firm
Let us suppose we allow the firm to also have power over the wage rate, i.e.,
to solve for all t = 0, . . . T :
 
max V (t) subject to ξ(s) = u′ f (ℓD (s), ε(s)) /y ,
ℓD (s), ξ(s), w(s); s≥t

v ′ (ℓ̄ − ℓD (s))
w(s) = ,
u′ (f (ℓD (s), ε(s)))
∀ s = t, . . . , T.
The pre-committed monopolist-monopsonist (hereafter “monopsonist”) only
solves this problem at t = 0, while the time-consistent solves it backwards for
t = 0, . . . , T . The first-order conditions for the pre-committed monopsonist
are
fℓ (t) − w(t) = A(t)fℓ (t)π(t) + (Aℓ (t) + A(t)fℓ (t))ℓ(t)v ′ (t) > 0 , (A.17)
and for the time-consistent monopsonist are
fℓ (t) − w(t) = −A(t)fℓ (t)Vex (t) + (Aℓ (t) + A(t)fℓ (t))ℓ(t)v ′ (t) , (A.18)
where
v ′′ (t)
Aℓ (t) ≡ − > 0.
v ′ (t)
All other things (A, fℓ , π) being equal, (A.17) suggests that the wage effect
acts in the same direction as the price effect, implying that the pre-committed
monopsonist will decrease his labor (and output) even more than the monopo-
list, and in turn the value of the firm will increase more. For the time-consistent
case, however, (A.18) suggests that the wage effect counteracts the price effect,
implying again a lower labor input (and output) than the monopolist. If the
price effect dominates, comparative static comparisons with the competitive
case will be as in Proposition 4; if the wage effect dominates, they will be as
in Proposition 2. The intuition for this extra term is quite clear; it is in the
firm’s interest to reduce wages, and to do so it will reduce its labor demand.

References
1. Allen F, Gale D (2000) Comparing financial systems. The MIT Press, Cam-
bridge London
 Monopoly Power and the Firm’s Valuation 33

2. Ausubel LM, Deneckere RJ (1989) Reputation in bargaining and durable goods

monopoly. Econometrica 57:511-531
3. Basak S (1997) Consumption choice and asset pricing with a non-price-taking
agent. Economic Theory 10:437–462
4. Blanchard OJ, Fischer S (1989) Lectures on macroeconomics. The MIT Press,
Cambridge London
5. Bonanno G (1990) General equilibrium theory with imperfect competition.
Journal of Economic Surveys 4:297–328
6. Coase RH (1972) Durability and monopoly. Journal of Law and Economics
15:143–149
7. Cox JC, Ingersoll JE, Ross SA (1985) An intertemporal general equilibrium
model of asset prices. Econometrica 53:363–384
8. DeMarzo PM, Bizer DS (1993) Sequential trade and the Coase conjecture: a
general model of durable goods monopoly with applications to finance. Kellogg
School of Management working paper
9. DeMarzo PM, Urosevic B (2001) Optimal trading by a “large shareholder”.
Working paper, Stanford University
10. Dierker H, Grodal B (1986) Non-existence of Cournot-Walras equilibrium in
a general equilibrium model with two oligopolists. In: Hindenbrand W, Mas-
Colell A (eds) Contributions to mathematical economics. North Holland, Am-
sterdam New York Oxford
11. Dierker E, Grodal B (1999) The price normalization problem in imperfect com-
petition and the objective of the firm. Economic Theory 13:257–284
12. Gabszewicz JJ, Vial J-P (1972) Oligopoly “a la Cournot” in a general equilib-
rium analysis. Journal of Economic Theory 4:381–400
13. Grinblatt MS, Ross SA (1985) Market power in a securities market with en-
dogenous information. The Quarterly Journal of Economics 100:143–167
14. Hart OD (1985) Imperfect competition in general equilibrium: an overview of
recent work. In:Arrow K, Honkapohja S (eds) Frontiers of economics. Blackwell,
Oxford
15. Kihlstrom RE (1998) Monopoly power in dynamic securities markets. Working
paper, University of Pennsylvania
16. Lindenberg EB (1979) Capital market equilibrium with price affecting institu-
tional investors. In: Elton EJ, Gruber MJ (eds) Portfolio theory 25 years later.
North Holland, Amsterdam New York Oxford
17. Lucas RE (1978) Asset prices in an exchange economy. Econometrica 46:1429–
1445
18. Mas-Colell A (1982) The Cournot foundations of Walrasian equilibrium theory:
an exposition of recent theory. In: Hindenbrand W (ed) Advances in economic
theory. Cambridge University Press, Cambridge
19. Mas-Colell A, Whinston MD, Green JR (1995) Microeconomic theory. Oxford
University Press, New York Oxford
20. Palley TI (1997) Managerial turnover and the theory of short-termism. Journal
of Economic Behavior & Organization 32:547–357
21. Pavlova A, Rigobon R (2003) Asset prices and exchange rates. NBER Working
Paper No. W9834
22. Schultze CL (2003) The consumer price index: conceptual issues and practical
suggestions. Journal of Economic Perspectives 17:3–22
23. Sundaresan S (2000) Continuous-time methods in finance: a review and an
assessment. The Journal of Finance 55:1569–1622
34 Suleyman Basak and Anna Pavlova

24. Tirole J (1988) The theory of industrial organization. The MIT Press, Cam-
bridge London
Revealed Intertemporal Ine!ciency

Christian Bidard

University of Paris X-Nanterre

Summary. In a stationary state, ine!ciency is established if some marginal per-

turbation leads to a positive surplus for all commodities. We consider the case when,
though the original perturbation itself results in positive and negative changes, an
adequately chosen combination of such perturbations is conclusive. A characteriza-
tion in terms of the positive roots of a polynomial is obtained.

1 Introduction
Let there be a dierentiable model describing a system in a stationary state. In
order to test the e!ciency of that state, we introduce one or several marginal
perturbations at some date, or on some consecutive dates, and contemplate
the eects of this basic operation in terms of surplus. The paper examines
when the operation allows us to conclude to ine!ciency. Section 1 presents
the result in the one-dimension case. Section 2 wonders if, in the presence of
several commodities, ine!ciency for each separate commodity is su!cient to
conclude to global ine!ciency. A general version of the theorem used in both
cases is given in the mathematical appendix.

2 The One-Commodity Case

Considering the impact of marginal perturbations is the classical tool for
studying ine!ciency (Cass, 1972). In this section it is assumed that the basic
operation aects one commodity. The implicit reason is not that there is one
commodity only in the model, but that the marginal perturbation has been
designed in order to have no influence on the surplus of the other commodi-
ties. In the next section, we retain another interpretation: there are positive
or negative eects on the other commodities, but these eects are ignored, as
if the supply of these commodities were illimited.
36 Christian Bidard

Starting from a stationary state, let the eects of the original perturbation
be described by a variation %d = % (d0 > ===> dW ) of the surpluses between dates w
to w + W . Ine!ciency is established if the operation leads to an increase of the
surplus at every date, i.e. if the vector d is positive. The same if the vector
is negative: it su!ces to reverse the operation (linearity of marginal changes
is used here). The interesting question occurs when, for instance, and up to a
positive factor %> the intertemporal changes in the surpluses are represented
by the vector d = (1> 2> 5) at the successive dates w, w+1 and w+2. Though
the operation does not seem conclusive at first sight, let us consider the more
complex transform which consists in repeating the same basic operation one
period later and doubling its activity level (|0 = 1> |1 = 2). The overall result,
represented in the last row of the table below, is semipositive: intertemporal
ine!ciency is revealed.

Table 1. An example of revealed ine!ciency

dates w w+1 w+2 w+3
|0 = 1 1 -2 5
|1 = 2 2 -4 10
overall surplus 1 0 1 10

Note that, since the changes %(1, -2, 5) are linear approximations of the
exact values, the zero component in the overall surplus is only a value close
to zero and might be a small negative scalar. The choice |1 = 2=1 would be
more convincing, or the whole operation might be repeated with a lag, thus
generating the positive surpluses (1, 0, 1, 10, 0) + (0, 1, 0, 1, 10). In general,
the construction of a complex transform that generates a semipositive surplus
is not obvious. However, what we need is not an expression of that complex
transform itself, but a simple criterion that indicates the possibility or the
impossibility to generate a semipositive surplus.
Let the basic operation, which remains unspecified and results in a se-
quence of marginal changes in the surpluses, be characterized by the activity
level |0 = 1= During the next periods, let the same operation be performed at
levels |1 > ===> |Q which may be positive or negative since they reflect changes
in activity levels. We assume that the perturbations and the eects take place
within a finite time. At date w + n, i.e. n periods after the initial perturbation,
the change in the surplus amounts to

vn = |0 dn + |1 dn1 + === + |n d0 (1)

where the coe!cients which are not defined (viz., dn for n A W and |n for
n A Q ) are set equal to zero. Let us associate the formal polynomial D([) =
PW P
Q
dw [ w to the basic operation, the formal polynomial \ ([) = |l [ l
w=0 l=0
 Revealed Intertemporal Ine!ciency 37

to
P the wsequence of activity levels and, finally, the formal polynomial V([) =
vw [ to the sequence of surpluses. The algebraic relationship between these
magnitudes is written V([) = D([)\ ([)= Therefore, ine!ciency is revealed
if V([) has semipositive coe!cients (after multiplication by 1+ [ + ===+[ P ,
it might as well be required that its coe!cients are positive). Let the tech-
nology be linear, but consider finite instead of infinitesimal changes. The dif-
ference is that the scalars |w in formula (1) represent activity levels instead
of changes in activity levels and, therefore, are semipositive (nonnegativity
restrictions would also matter in the marginal case if the initial activity level
is zero). Hence the following definition (to ease the reading, formal polyno-
mials and formal series with semipositivity or positivity restrictions on the
coe!cients are denoted by letters S , T or U):

Definition 1. A marginal change described by the formal polynomial %D([)

reveals ine!ciency in a finite time if and only if there exists a formal polyno-
mial \ ([) such that the formal equality

D([)\ ([) = S ([) (2)

holds, where S is semipositive. A non marginal change reveals ine!ciency in

a finite time if and only if there exist semipositive formal polynomials S and
T such that
D([)T([) = S ([) (3)

It is immediately seen from equality (2) that a necessary condition is that

the real polynomial D({) admits no positive root. Theorem 1 states that
the condition is also su!cient. Curiously enough, condition (3) is not more
restrictive than (2):

Theorem 1. A marginal or non marginal change leading to the sequence of

surpluses (d0 > ===> dW ) reveals ine!ciency in a finite time if and only if the
PW
polynomial D({) = dw {w has no positive root.
w=0

Proof. Let d0 A 0 and apply Theorem 4 of the mathematical appendix to the

case when the matrices Dw have dimension one and are the coe!cients of the
polynomial D({). The alternative between conditions (l) and (lll) of Theorem
4 means that either D({) admits a positive root or D({) is the ratio between
two polynomials with semipositive coe!cients, as in (3).

Up to now we have proceeded to a finite number of successive changes. A

justification of the hypothesis is that, if there were infinitely many nonzero
values of |w , |w might tend to infinity and the marginal approximation would
hold no longer. However, the objection is avoided if we restrict the attention
to uniformly bounded sequences {|}. For instance, the perturbation leading
to the surpluses %(d0 > d1 = 1) with d0 > 1 cannot reveal ine!ciency in a
finite time, but it does so in an infinite time, by repeating the operation for
38 Christian Bidard

all dates at the same activity levels. The example suggests that, in infinite
horizon, what matters is the position of the real roots of D({) with regard to
1. The following notations are adopted: bars refer to formal series, as opposed
to polynomials, and X means that the coe!cients are uniformly bounded.

Definition 2. A marginal change reveals ine!ciency in a finite or infinite

time if and only if the formal equality

D([)X ([) = S ([) (4)

holds, where the formal series X has uniformly bounded coe!cients and the
formal series S is semipositive.

Theorem 2. A marginal change reveals ine!ciency if and only if its asso-

ciated polynomial D({) has no root in ]0> 1[ and the root { = 1, if any, is
simple.

Proof. If D({) has no root in ]0> 1[ and, possibly, a simple root at { = 1> D([)
is written ¡ ¢
D ([) = E ([) l 1  1 l [ [1  [] (5)

¡where E({) has no positive ¢ root, the l s are the roots of D greater than 1
let 1 ?  ? 1  · · ·  n and the presence of the factor [1  [] is optional.
According to Definition 1 and Theorem 1, an equality

E ([) T1 ([) = S1 ([) (6)

holds, where
P S1 ([) and T1 ([) are ¡semipositive
¢ polynomials. The formal
inverse sw [ w of the polynomial l 1  1
l [ is positive and written
¡ ¢1 P
l 1  1
l [ = l ( w w
l [ )
w
P
 l ( w w 1
1 [ ) = l (1  1 [)
1
= (1  1
1 [)
n
w

Therefore
(w + n  1)! w
sw   = yw
w! (n  1)! 1
Since yw+1 /yw = 1
1 (w + n) (w + 1) ? 
1
for w great enough, there exists
some positive scalar p such that sw  yw ? pw for any w. As a consequence
¡ ¢1
the series l 1  1
l [ or, if the case occurs,
¡ ¢1 1
X 1 ([) = l 1  1
l [ [1  [] (7)

has positive and uniformly bounded coe!cients. By multiplying equalities (5),

(6) and (7) side by side, equality

D ([) [T1 ([) X 1 ([)] = S1 ([)

 Revealed Intertemporal Ine!ciency 39

is obtained and condition (4) is met. Hence, ine!ciency is revealed.

Conversely, let equality (4) hold. Assume first that D admits a root {0
in ]0> 1[. The real series X ({) converges at { = {0 and a contradiction is
obtained between equality S ({0 ) = D({0 )X ({0 ) = 0 and the semipositivity of
S = Assume now that D admits no root {0 in ]0> 1[ but that 1 is a root of order
n (n  2) = Then
¡ ¢
D ([) = (1  [)n E ([) l 1  1
l [

where E({) has no positive root and l A 1. There exist positive polynomials
S2 and T2 such that S2 = ET2 = Replace D ([) by the above expression in
(4) and multiply both members of (4) by the positive series
¡ ¢1
U([) = (1  [)2n T2 ([) l 1  1
l [

Equality
(1  [)2 X 2 ([) = S 2 ([)
is obtained, where X 2 = S2 X has uniformly bounded and the coe!cients of
S 2 = S U are positive. (In other terms, a comparison with (4) shows that the
2
original problem has been reduced to the specific case D([) = (1  [) =) Let
V ([) = (1  [) X 2 ([). The inequality (1  [) V ([) A 0 implies 0 ? v0 ?
1
1 ? v2 ? · · · = Therefore the wth coe!cient of X 2 ([) = (1  [)
vP V ([) =
( w [ w )V ([) is greater than wv0 , and a contradiction is obtained with the
uniform boundedness hypothesis.

In the case of finite but bounded changes, instead of marginal changes, the
series X in Definition 2 must moreover be semipositive. An inspection of the
above proof shows that the characterization given in Theorem 2 is unchanged.

3 Several Commodities
We now consider the case when the initial perturbation aects several com-
modities, for instance two commodities D and E during several periods. The
eects are described by the simultaneous changes %(d0 > d1 > ===> dW ) for com-
modity D and %(e0 > e1 > ===> eW ) for E. The question of detecting global ine!-
ciency is that of giving the vectoral version of Theorems 1 and 2 and, therefore,
the expected tool is a theorem on vectors, as the one stated in the appendix.
We explore an alternative path and consider the question component by com-
ponent.
An initial condition is required. Assume first that d0 and e0 are both
nonzero, i.e. the first real eects of the changes occur at the same date. If
d0 and e0 have opposite signs, this will also be the case for any complex
transform derived from the basic operation. Then the surpluses in both com-
modities cannot be simultaneously positive and it is not possible to conclude
40 Christian Bidard

to ine!ciency. In the more complex case when the initial eects on D and E
are not simultaneous, say d0 A 0 but e0 = e1 = 0 6= e2 > the initial condition
is transformed as follows: the first activity levels (|0 = 1> |1 > |2 ) must be such
that

|0 d0 A 0 |0 d1 + |1 d0 > 0 |0 d2 + |1 d1 + |2 d0 > 0 |2 e2 > 0

These inequalities require minimum levels for |1 and |2 , but the sign of the
minimum level of |2 is undetermined.
To avoid these discussions, we return to the simplest case and assume
simultaneous initial changes: d0 A 0 and e0 A 0. Apart from that initial
condition, let us consider the surpluses (d0 > ===> dW ) and (e0 > ===> eW ) separately
and assume that, when each good is considered in isolation, some marginal
perturbation reveals ine!ciency in a finite time (the result for infinite time
would be similar). Can we conclude that the change reveals ine!ciency for
both commodities considered simultaneously? Normally not: the activity lev-
els revealing ine!ciency for D dier from those revealing ine!ciency for E,
whereas the same activity levels should apply to both commodities.
In formal terms, the ine!ciency relative to D is written

< \1 ([) D([)\1 ([) = V1 ([) with V1 A 0 (8)

and the one relative to E is

< \2 ([) E([)\2 ([) = V2 ([) with V2 A 0 (9)

Global ine!ciency, with D and E considered as joint products, is written

< \ ([) D([)\ ([) A 0 and E([)\ ([) A 0 (10)

There is no obvious way to derive (10) from (8) and (9). A curiosum is that
the implication does hold:

Theorem 3. If d0 e0 A 0, the revelation of ine!ciency for commodities D and

E separately su!ces to reveal ine!ciency for the system as a whole.

Proof. If the ine!ciency of D is revealed, the equivalence between conditions

(2) and (3) shows that it is possible to write down an equality of the type (8)
holds where, moreover, \1 is semipositive (if d0 A 0). The same for commodity
E (if e0 A 0). Then condition (10) is met by \ ([) = \1 ([)\2 ([).

4 Conclusion
A marginal perturbation in a stationary system may result in positive and
negative variations of the surplus over several periods. It reveals ine!ciency
if an adequately chosen sequence of such perturbations results in positive
 Revealed Intertemporal Ine!ciency 41

variations only. Though the adequate combination may be di!cult to find,

its existence has been characterized in terms of the roots of a polynomial
associated with the perturbation. In the presence of a multidimensional per-
turbation, an initial condition and ine!ciency for each commodity separately
su!ce to establish global ine!ciency.

5 Mathematical Appendix
Theorem 4 is a slightly adapted version of a result established by Bidard
(1999) to study the theory of fixed capital. A sequence of nonnegative vectors
is called ‘essential’ when W consecutive vectors are not equal to zero. The
hypothesis
@ |0 A 0  DW |0 > 0 (11)
implies that any nonzero sequence of nonnegative vectors satisfying property
(ll) below is essential. Vectors with semipositivity or positivity restrictions
are denoted by letters s> t> u or y.

Theorem 4. Let D0 > D1 > ===> DW be square matrices with D0 semipositive and
indecomposable and DW satisfying (11). The following properties (i) and (ii)
are equivalent. Moreover, either (i) or (ii) holds, or property (iii) holds:
(i) There exists a positive scalar  and a semipositive vector y such that
inequality (12) holds:
à W !
X w
< A0 <yA0   Dw y > 0 (12)
w=0

(ii) There exists an essential sequence of nonnegative vectors yw such that

the formal product
à W !
X X ³X ´
uw [ w =  Dw [ w yw [ w (13)
w=0

has nonnegative elements for any w > W .

(iii) There exist semipositive row-vectors (et0 > ===> teQ ) such that the formal
product ÃQ !Ã W !
Q+W
X X X
m l w
s̃m [ = tel [ Dw [ = (14)
m=0 l=0 w=0

is positive.

Proof. If property (l) holds, then property (ll) holds for the choice yw = w y.
Conversely, assume that property (ll) holds. Consider the set K made of the
qW
‘heads’ {(y0 > y1 > ===> yW )}  U+ of all vector sequences satisfying condition
(13), with uw  0 for w  W . K is a convex cone. Let N be the normalized heads,
42 Christian Bidard

i.e. the intersection of K with the unit simplex. Note the following uniform
boundedness (X E) property: Because of the assumptions on D0 , inequality
(D0 yw + D1 yw1 + === + DW ywW ) > 0 requires that kyw k is not too great with
regard to kyw1 k > ===> kywW k ; therefore, by induction on w, kyw k is uniformly
bounded for all sequences whose head belongs to N. A consequence of (X E)
is that N is closed.
Consider the translation correspondence  defined on K by
 (y0 > ===> yW ) = {yW +1 ; (y1 > ===> yW > yW +1 ) 5 K} = It has compact convex val-
ues. Let the correspondence  : N $ N be defined by  (y0 > ===> yW ) =
{(y1 > ===> yW > (y0 > ===> yW ))}, where  is a normalization factor ( is well-
defined because (y0 > ===> yW ) 5 N excludes y1 = === = yW = 0, thanks to
assumption (11)).  meets the assumptions of the Kakutani theorem and,
therefore, admits a fixed point

< (y0 > ===> yW ) 5 K < A 0 y = y0 = y1 = === = w yW (15)

P P
By definition of N, the W th term of the formal sequence  ( Dw [ w ) ( yw [ w )
is semipositive:
(DW y0 + DW 1 y1 + === + D0 yW ) > 0= (16)
Property (l) results from (15) and (16), therefore the equivalence of (l) and
(ll) is proved.
Assume now that properties (ll) and (lll) hold simultaneously, with yp A
0. Equalities (13) and (14) lead to a contradiction
P Pconcerning
P the sign of the
coe!cient of [ w+Q+p in the product ( tel [l ) ( Dw [ w ) ( yw [ w ) =
Finally, assume that property³ (ll) does not hold. ´ Suppose that, for any
(Q) (Q )
Q , there exists a finite sequence y0 > ===> yw > === > 0, with ky0 k = 1, such
that the vectors uw defined by (13) are semipositive for any w in the interval
[W> W + Q ]. By the (X E) property, every kyw k is bounded independently of Q .
When Q tends to infinity, let us extract successive converging subsequences
and construct an infinite sequence (y0 > ===> y w > ===) as a Cantor diagonal: y 0 =
(Q) (Q )
lim y0 > y 1 = lim y1 , etc. This limit sequence satisfies property (ll). This
Q 5Q0
being excluded by hypothesis, we conclude that there exists some Q such
that the system of linear inequalities uW > 0> ===> uW +Q > 0 has no semipositive
solution. In matricial terms
5 6
DW · · · D0 0
9 DW D0 :
9 :
@y A 0 Py =  9 . . :y > 0
7 .. .. 8
0 DW D0

According to the Theorem of the Alternative, there exists a semipositive row-

vector te such that teP ?? 0. Therefore, property (lll) holds.
 Revealed Intertemporal Ine!ciency 43

References
1. Bidard C. (1999), Fixed Capital and Internal Rate of Return, Journal of Math-
ematical Economics, 31, 523-41.
2. Cass, D. (1972), On Capital Overaccumulation in the Aggregative Neoclassical
Model of Economic Growth: A Complete Characterization, Journal of Economic
Theory, 4, 200-23.
Volatility and Job Creation in the Knowledge
Economy

Graciela Chichilnisky and Olga Gorbachev

Columbia University, New York, NY

Summary. Sectors with increasing returns to scale have been shown to amplify
business cycles exhibiting more volatility than others [13]. Our hypothesis is that
this volatility could be a cause of the "jobless recovery" suggesting policies for em-
ployment generation. To test this hypothesis we introduce a general equilibrium
model with involuntary unemployment. The economy has two sectors: one with in-
creasing returns that are external to the firm and endogenously determined — the
knowledge sector — and the other with constant returns to scale. We define a mea-
sure of employment volatility, a ‘labor beta’ that is a relative of the ‘beta’ used in
finance. A ‘resolving’ equation is derived from which it is proved that increasing re-
turn sectors exhibit more employment volatility then other sectors. The theoretical
results are validated on US macro economic data of employment by industry (2-3
digits SIC codes) of the 1947-2001 period, showing that the highest ‘labor betas’
are in the service sectors with increasing returns to scale. Policy conclusions are
provided to solve the puzzle of the ‘jobless recovery’, where small firms in the ser-
vices industry play a key role. We conclude with policy recommendations on how to
create jobs in the knowledge economy.

1 Introduction
In a recent article [13] we showed that increasing returns to scale sectors am-
plify the business cycle: they grow faster than others during expansions, and
contract faster during downturns. In this sense they exhibit more ‘volatil-
ity’ than other sectors1 . In this paper, we extend the earlier model to analyze
1
The eect of the business cycle on IRS sectors has not been analyzed in the
literature. Real Business Cycle models include IRS in order to generate cyclical
productivity (see for example S. Basu and J. Fernald [3] and [4] for a review of the
literature). In international trade, IRS is seen as determining the pattern and the
factor content of trade (Krugman [23], [24], [25], [26], Panagariya [27], Antweiler
et al. [1]). In growth literature, learning by doing leads to endogenous growth
in the economy (see Arrow[2], Romer[30], and Rivera-Batiz and Romer[29]). In
theoretical macrodynamic general equilibrium models, IRS may lead to unstable
46 Graciela Chichilnisky and Olga Gorbachev

volatility of employment. Our hypothesis is that this volatility could be a cause

of the “jobless recovery”2 suggesting policies for employment generation.
Chichilnisky-Gorbachev (2004) measured volatility using a new concept, a
‘real beta’, which is a statistical relative of the ‘beta’ used in financial markets,
and measures real macro data on output rather than in stock prices. We used
a general equilibrium model with two sectors, one with constant returns to
scale (CRS) and the other with external increasing returns to scale (IRS),
similar to that in Chichilnisky [9], [10] and [11]. In both sectors the firms are
competitive, but the ‘knowledge sector’3 has increasing returns to scale that
are external to the firm and are endogenously determined. The externalities
arise from “learning by doing,” analogous to Arrow (1962), and from the
free transfer of skills from one firm to the other through job turnover or
freely shared skills or R&D. Increasing returns are endogenous because they
depend on the output in the knowledge sector in equilibrium4 . We showed
that IRS sectors exhibit more volatility than other sectors, an outcome we
tested empirically using US data5 ; as these sectors become a larger share
of the economy their volatility increases economic uncertainty. These results
are related to the ‘virtuous and vicious’ cycles in economies with increasing
returns to scale that were developed earlier in Chichilnisky and Heal [14] and
in Heal [20].
The model in Chichilnisky-Gorachev (2004) had full employment in equi-
librium. Extending this model, we introduce here a general equilibrium model
with involuntary unemployment and two sectors, one with external IRS and
systems and may generate “vicious” and “virtuous” cycles in the economy (see
Chichilnisky and Heal [14], and Heal [20], [19]).
2
This is the second jobless recovery recorded in the US, the first was in 1990-1991.
3
To identify ”knowledge sectors” we used benchmarks from the World Knowledge
Competitiveness Index introduced by Robert Huggins Associates, London, UK.
The World Knowledge Competitiveness Index 2004 is an integrated and overall
benchmark of the knowledge capacity, computed utilizing 19 knowledge econ-
omy benchmarks, including employment levels in the knowledge economy, patent
registrations, R&D investment by the private and public sector, education expen-
diture, information and communication technology infrastructure, and access to
private equity (www.hugginsassociates.com).
4
Chipman (1970)[16] provides an excellent general equilibrium model with external
economies of scale. The main dierence with our model is in that his has only one
factor of production and economies of scale are exogenously given, whereas here
(and in Chichilnisky 1993, 1994, and 1998) there are two factors and the extent
of IRS is endogenously determined in one sector of the economy— the other has
constant returns.
5
Our hypothesis is di!cult to prove in complete generality, because the data of
the last ten years is complex: industry classifications have changed and, as is
well know, when using nominal data the increasing returns sectors (whose prices
drop when production expands) are underrepresented in GDP. This observation is
related to comments by William Baumol on productive and unproductive sectors,
and to the November 2002 publication of Survey of Current Business, U.S. [35].
 Volatility and Job Creation in the Knowledge Economy 47

the other with CRS. The external IRS sectors or knowledge sectors are often
characterized by having many small competitive firms. According to the Small
Business Administration O!ce of Advocacy, small firms, defined as having 500
employees or less, represent 50 percent of all employment but generate about
60-80 percent of net jobs in the US [40]. Therefore, while IRS sectors are more
volatile, small firms within IRS sectors create more jobs than they destroy and
more jobs than larger firms, as pointed out by Chichilnisky (2004).
To achieve a meaningful representation of involuntary unemployment in
a general equilibrium context, in the model presented here we postulate that
labor supply is ‘produced’ as an increasing function of several variables rather
than a constant endowment of time (24 hours) possessed by each individual
from birth. The amount of labor supplied is constrained by the amount of
physical energy, health, free time, skills, eort, and education a person pos-
sesses, which in turn relate to wages. An implication of this is that in a period
of structural change, as labor skills become outdated, involuntary unemploy-
ment increases, an observation that has been made by the Federal Reserve
Bank of New York in 2003 [18]. The formulation of our model of ‘produced
labor’ supply has the same mathematical structure as the well-known Shapiro
and Stiglitz e!ciency wage model [32]6 ; therefore, we do not provide a sepa-
rate derivation of the labor market here, but use the Shapiro-Stiglitz model
for this purpose as presented in the Appendix.
The article proceeds as follows: first we provide a general equilibrium model
with involuntary unemployment and solve it analytically finding all prices,
the level of unemployment, and production levels in equilibrium by means
of a single ‘resolving’ equation; second we define the ‘labor beta’ measure
of employment volatility for the dierent sectors of the economy, and prove
formally that employment in external IRS sectors is more volatile than in other
sectors; third, we provide details of the data used and validate the results on
the observed ‘labor betas’ during the 1948-2001 periods; fourth, we discuss the
jobless recovery in the context of this model. We show that small firms and the
service sector play a crucial role and conclude with policy recommendations
on how to create jobs in the knowledge economy.

2 General Equilibrium Model with External Economies

of Scale and Involuntary Unemployment
Internal and External Increasing Returns to Scale

A firm or an industry has increasing returns to scale (IRS) when unit costs fall
with increases in production7 . Economies of scale are internal to the firm when
6
The produced labor equation that we use here, see section 2.03, is equivalent to
the Non-shirking condition of Shapiro-Stiglitz, developed formally in the context
of our model in the Appendix.
7
Sometimes they are defined by ‘average’ unit costs that decrease with production.
48 Graciela Chichilnisky and Olga Gorbachev

a firm becomes more productive as its own size increases, i.e. more e!cient,
in utilizing its resources as is typical to firms with large fixed costs such as
aerospace, airlines, and oil refineries8 . In contrast, increasing returns to scale
are external to the firm when the increased productivity comes about as a
result of decreasing unit costs at the level of the industry as a whole. In the
latter case, each firm could have constant unit costs as its production increases,
and behave competitively9 . Yet as the industry as a whole expands, positive
externalities among the firms are created leading to increased productivity for
all firms in the industry10 . The free movement of skilled workers from one firm
to another can have this eect, as a firm may benefit at no cost from training a
worker received in another firm11 . Equally, a firm can benefit from unspecific
research and development innovations developed in other firms, which are
accessible to it at little or no cost. These positive ‘knowledge spillovers’ often
originate from innovations generated during the course of production. As the
new knowledge spreads to all the firms in the industry, total productivity in
the industry increases and unit costs fall12 .

General Equilibrium with External Increasing Returns to Scale

The economy produces and trades two goods E and L. E is a traditional con-
stant returns to scale (CRS) industry, whereas L is produced under external
increasing returns to scale (IRS). Both goods are produced using two inputs,
labor O and capital N, and the firms in each industry are perfectly compet-
itive. They minimize their costs given the market prices. Firms production
functions are given as:
8
This type of increasing returns can lead to monopolistic competition due to high
entry costs.
9
External IRS is consistent with small competitive firms; internal IRS depends
instead on large size of firms which is typically inconsistent with competitive
behavior.
10
For example, in the period between 1990 and 2000, the expansion in output in
the computer hardware industry led to yearly doubling of the computing power
available per dollar, leading to an exponential increase in CPUs per dollar (a stan-
dardized measure of processing power) and to the corresponding rapid increase
in demand and consumption of CPUs across the entire economy.
11
Workers in the knowledge sectors move between firms more than others, on av-
erage every two years or less.
12
Any industry that depends on knowledge or skilled labor could benefit from such
knowledge spillovers and external economies of scale. In the growth literature
related phenomena are known as ‘learning by doing’, a concept introduced by
Arrow [2] and developed further by Romer [30] in a one sector growth model.
‘Learning by doing’ often refers to increasing returns that are internal to the firm.
We focus instead on increasing returns that are external to the firm, endogenously
determined and internal to one industry within a general equilbirum model with
two goods and two factors, and in which a second sector has constant returns to
scale.
 Volatility and Job Creation in the Knowledge Economy 49

E v = O 1
1 N1 and L v = O2 N21 (1)

where >  5 (0> 1). O1 > N1 are inputs in the E sector, and O2 > N2 are inputs
in the L sector. The total amount of labor and capital in the economy are
Q and N v respectively. The parameter  in the production function for L is
taken as a constant by each firm within this industry. However, at the industry
level,  is endogenously determined and increases with the output of L> i.e.
 = (L v ). For example,  = L  > 0 ?  ? 1. In this case, when externalities are
 1
taken into account the production function for this sector is L v = O21 N21 >
although at the firm’s level the technology that determines the firm’s behavior
is L v = O2 N21 = Observe that  A 1 leads to negative marginal products,
while  ? 0 leads to decreasing returns. Therefore, when the sector L has
increasing returns,  satisfies 0 ?  ? 1> which we now assume13 . Notice that
returns to scale in the sector L are endogenous because the parameter  = L 
is unknown until the equilibrium value of L is determined.
Prices for L and E are sL and sE > respectively, and L is a numeraire good,
or sL = 1, then
\ v = L v + sE E v (2)
By assumption each firm has external IRS so it forecasts a scale parameter
 as exogenously given, and solves the cost minimization problem with first
order conditions as:

z = sE (O1 )1 (N1 )1 and u = sE (1  )(O1 ) (N1 ) in sector E (3)

z = (O2 )1 (N2 )1 and u = (1  )(O2 ) (N2 ) in sector L
We assume that both capital and labor markets are perfectly competitive,
thus the factor prices paid in E sector must equal that of L sector. The ’s
will be determined by equilibrium output and the ’s that the firms projected
will be the ones obtained in general equilibrium14 .

‘Produced’ Labor and Involuntary Unemployment in General

Equilibrium

Our next task is to achieve a representation of involuntary unemployment in

a general equilibrium context. To this end we postulate a ‘produced’ labor
supply curve that is an increasing function of wages (see Figure 2 below).
It is not the purpose of this article to provide an explicit derivation of the
‘produced labor’ supply curve; instead, we use as a model for this ‘produced
13
From equation (19) the marginal product of labor is PS O =

1
13
((s E
C
) 1 )( CsE
)( CsE
CO
)= When  A 1 marginal product is negative. When
 is less than or equal to 0, the L sector exhibits decreasing or constant returns
to scale, and 0 ?  ? 1 is IRS.
14
This notion is similar to that of rational expectations assumptions for technology
and production.
50 Graciela Chichilnisky and Olga Gorbachev

labor’ supply the non-shirking condition that arises from the e!ciency wage
theory of Shapiro and Stiglitz [32] (see Figure 1), the connection between the
two is developed below and in the Appendix, see Figure 1.
The ‘produced’ labor supply function is taken as exogenously given by
workers (hence it is ‘involuntary’). It represents not their willingness to work
but rather the ability to do a certain job at a certain wage given for example
the current skill or education level. In the Appendix, we consider a specific
and familiar representation of this ‘produced labor’ function which delivers
involuntary unemployment as in Shapiro-Stiglitz’ e!ciency wages article [32].
In that paper, the non-shirking condition (NSC) as illustrated in Figure 1
and derived in the Appendix15 acts as the equivalent to our ‘produced’ labor
supply function. Shapiro-Stiglitz observe that in the environment character-
ized by imperfect information (in their case firms are unable to tell the actual
eort level of their workers), firms are forced to pay a higher than market
clearing wage in order to keep workers from shirking, hence the non-shirking
condition and the resulting involuntary unemployment.

Wages Aggregate Aggregate

Labor Demand NSC

Involuntary
Unemployment

w* E
F’(L)

L* N Employment

Fig. 1. Labor Market Equilibrium in General Equilibrium Model with Involuntary

Unemployment Arising from Asymmetric Information from Shapiro and Stiglitz
(1984), see Appendix
15
This figure is reproduced from Figure 2 of Shapiro-Stiglitz paper [32].
 Volatility and Job Creation in the Knowledge Economy 51

Potential
s Labor Force, N
Wages `Produced’ Labor Supply, L

Involuntary
Unemployment

w*

d
Labor Demanded by Firms, L

L* N Labor

Fig. 2. Labor Market Equilibrium in General Equilibrium Model with Involuntary

Unemployment Arising from ‘Produced Labor’ (analogous to Shapiro and Stiglitz
model of 1984, see Figure 1)

The ‘produced labor’ supply function increases with wages because workers
are able to supply more labor at higher wages since acquisition of skills re-
quires resources. For example, in the e!ciency wage models, there is minimal
threshold that workers must reach in order to be considered productive labor
units (due for example to minimal income required to sustain persons’ mini-
mal health). As already mentioned, we do not derive explicitly this function,
taking instead Shapiro-Stiglitz formulation as a foundation.
The concept of involuntary unemployment is illustrated in Figure 216 .
The vertical line represents total potential labor force in the economy or the
total number of individuals, Q= Potential labor force, Q , is dierent from
the ‘produced’ labor supply, which is a positively sloped curve, Ov > because
as mentioned above it is an exogenously give function that depends on the
current levels of workers’ skills or education. Ov is a function of wages, z> and
prices, sE> or Ov = Ov (z> sE ). Labor demand, Og > is a decreasing function of
wage and is derived from firms’ cost minimization problem or after solving for
O in equation (3).
In equilibrium, due to the shape of labor supply curve, there will exist in-
voluntary unemployment, represented by the dierence in labor employed O
16
Notice that Figure 2 is a special case of Figure 1.
52 Graciela Chichilnisky and Olga Gorbachev

and potential labor supply, Q . Unemployment, (Q  O ), is involuntary since

there are workers who are willing to work at a lower wage but because they
cannot credibly promise to do the job since they do not have the required-
skills. Thus, they will not be hired. According to Shapiro and Stiglitz’ model,
unemployment is involuntary because workers who are willing to work at a
lower than equilibrium wage cannot make a credible promise to exert eort
or not to shirk.

Resolving the General Equilibrium Model

This section draws on the results of Chichilnisky [9] and Chichilnisky-Gorba-

chev [13] to solve our general equilibrium model with involuntary unemploy-
ment. To solve this model, there are four prices to be determined: the goods
prices sE and sL , and the two factor prices, wages and return on capital,
z and u . The quantities to be determined in an equilibrium are: the use of
   
factors in each sector N1> N2> O1> O2 > the outputs of the two goods E v and L v ,

the parameter  determining the external economies of scale in L, and the
demand for goods E and L, E g and L g . Thus the model has 13 endogenous
variables to be computed, sE , sL ,z > u > N1>
   
N2> O1> O2 > E v > L v >   > E g and
g
L .
We solve the model by finding an explicit function of a single variable,
sE , (called a ‘resolving’ equation) and 5 exogenously given parameters of the
economy: > > > Q and N v . To obtain the resolving equation we write the
market clearing condition in the L market, demand equals supply, and find a
way to express it as a function of only one variable: sE . Solving this equation
gives the equilibrium value of sE >from which all other endogenous variables
listed above can be found17 . Since the model has constant returns to scale at
the level of the firms18 , we derive the equilibrium relations between supplies
and prices from the marginal conditions, clearing of labor market and full
employment of capital.
For simplicity19 , demand functions for E given as a function of initial
endowments and prices:

E g = E g (Q> N v > sE ) (4)

In equilibrium all markets clear, or:


E v = E g + [E (E market clears) (5)
17
Notice that we have not defined supply behavior outside of an equilibrium; in
particular, there is no information for carrying out stability analysis.
18
As mentioned in the text, individual firms in both sectors of the economy exhibit
CRS. Firms that belong to external IRS sector don’t know that positive external-
ities are generated by their production or hiring decisions. Thus, firms in external
IRS sectors, take  of (1) equation as given.
19
The results of the model do not depend on the demand specification. But in the
Appendix, we solve for a specific demand function.
 Volatility and Job Creation in the Knowledge Economy 53

L v = L g + [L (L market clears)


where [E and [L are equilibrium levels of net exports in E and L sectors
respectively. We assume [L = 0 and [E 
= 0 but the results are true for any
 
given [L and [E .

sE E v + L v = z O + u N  (zero profits)

N  = N v = N1 + N2 (capital market clears)
O = Ov = Og = O1 + O2 (‘involuntary’ labor market clears)
Notice however that labor market clears in the sense that ‘produced’ labor
supply is equal to labor demand. But since ‘produced’ labor supply is given by
Ov = Ov (z> sE ) and is dierent from the potential labor supply, Q , the labor
market is characterized by the existence of involuntary unemployment, since
Q A O , as shown in Figure 1. Workers without a job are willing to work at a
wage below z but will not be hired because below the clearing wage, the Ov
produced by these workers is below firm’s requirement since workers are not
as educated or not as healthy to do the needed job.
We can characterize ‘involuntary’ labor market equilibrium in a more ex-
O1 O2
plicit way. Denoting o1 = N 1
and o2 = N2
> we can rewrite marginal conditions
(3) as:
z = (o1 )1 sE and u = (1  )(o1 ) sE
z = (o2 )1 and u = (1  )(o2 )
Indicating logarithms with the symbol “e” the four equations above can be
rewritten as:

e = (  1)oe1 + 
z e + sf e = oe1 + (1g
E and u  ) + sf
E (6)

z e and ue =  oe2 + (1g

e+
e = (  1)oe2 +   ) + 
e
so that (  1)oe1  (  1)oe2 = ee  sf
E + e
g g
and oe1   oe2 = (1  )  (1  )  sf
E + 
e =
Solving for oe1 and oe2 we obtain:

e  sf
( E  e)()  (1  )[(1g
e+  )  sf g
E  (1  ) + 
e]
oe1 =
[  ]

(  1)[(1g
 )  sf g
E  (1  ) + 
e  sf
e]  ( E 
e+
e)
oe2 =
[  ]
or
sf
E sf
E
oe1 = + D and oe2 = +E (7)
(  ) (  )
where D and E such that:
54 Graciela Chichilnisky and Olga Gorbachev

( e )()  (1  )[(1g
e  )  (1g
 )]  
e
D= (8)
[  ]
(  1)[(1g
 )  (1g e
 )]  [ e]  
e
E=
[  ]
D and E are constants since 
e is taken as a constant by each individual
firm. Also observe that D A 0 and E ? 0 if  ? = Therefore,
1 1
o1 = hD sE and o2 = hE sE (9)

Using (9), marginal conditions can be written as:

1 
z = hD(1) sE and u = (1  )hD sE (10)

Substituting (10) into Ov = Ov (z> sE ), we can rewrite Ov as a function of

sE alone, or:

Ov = Ov (sE ) (11)
‘Involuntary’ labor market clears, or:

Ov (sE ) = Og (sE ) = O20 (12)


Since o2 = NOVO
N1 or O1 = O
1  V
 o2 (N  N1 )= At same time, o1 = O1
N1 > so
that O  o2 (N V  N1 ) = o1 N1 .
The quantity of N and O demanded in the E sector are:

O  o2 N V o1
N1 = and O1 = (O  o2 N V ) (13)
(o1  o2 ) (o1  o2 )
From (9), (12) and (13) N and O are functions of a single variable sE :
1
hD O hD hE N v sE
O1 = D E
 = O1 (sE ) (14)
(h  h ) (hD  hE )
1
O s  hE N v
N1 = D E E  D = N1 (sE ) (15)
(h  h ) (h  hE )
Equations (14) and (15) hold for any level of = In particular, taking  = 1,
we denote production of E and L as  (sE ) and (sE ) respectively. Therefore,
from (1), (14) and (15) we obtain the equilibrium level of output as a function
of equilibrium price sE :
20
Notice that this condition is dierent from the one used in Chichilnisky-Gorbachev
paper [13]. Previous paper had full employment of labor and capital and therefore
no involuntary unemployment. Here due to the ‘produced labor’ function (12)
there is involuntary unemployment.
 Volatility and Job Creation in the Knowledge Economy 55
1 1
hD O hD hE N v sE  O sE

hE N v 1


E v = [  ] [ D  D ]
(hD E
h ) D
(h  h ) E (h  h ) (h  hE )
E

=  (sE )
(16)
1 1
hD O hD hE N v sE  v O sE

hE N v 1


L v = [O   ] [N   ]
(hD  hE ) (hD  hE ) (hD  hE ) (hD  hE )
= (sE )

For industry L, (16), does not express output as an explicit function of equilib-
rium prices alone as we wished, because  = (L), and L = L(> sE ). In order
to obtain output as explicit functions of equilibrium prices we must therefore
find out the equilibrium value of the scale parameter   . This is an additional
“fixed point” problem, since  depends on L> while L depends on = We solve
this as follows.
The industry L has increasing returns which are external to the firms in
this industry, and the parameter  increases with the level of output of L. We
postulated that

 = L  , where 0 ?  ? 1= (17)
At an equilibrium, equations (16) and (17) must be simultaneously satisfied,
i.e.  = [(sE )] =   (sE ) or,  1 = (sE ) =
Thus, 
 = (sE ) (1) (18)
Therefore at an equilibrium from (16) and (18) we obtain a relation be-
tween the outputs of L and sE :
1
L v (sE ) = (sE ) (1) (19)
so that
; 1
? hD O hD hE N v sE 
v 
L (sE ) = [O  D  ]
= (h  hE ) (hD  hE )

1
< 1
1

O 
s h E v
N @
×[N v  D E E  D ]1 (20)
(h  h ) (h  hE ) >

The Resolving Equation F (pE )

We can now give explicitly the ‘resolving’ equation for the model, denoted
I (sE ) below:
g v g 1
I (sE ) = L (sE )  L (sE ) = L (sE )(sE ) (1) = 0
56 Graciela Chichilnisky and Olga Gorbachev

or,
; 1
? hD O hD hE N v sE 
g 
I (sE ) =L (sE ) [O  D  ]
= (h  hE ) (hD  hE )

1
< 1
1


O sE 
h NE v @
× [N v   ]1 =0 (21)
(hD  hE ) (hD  hE ) >

where from (10) and (4), L g (sE ) is a function of sE alone:

1 
L g (sE ) = (hD(1) sE )O + ((1  )hD sE )N v  sE E g (sE ) (22)

Solving the equation I (sE ) = 0> gives an equilibrium value of sE from
which all equilibrium values of other variables (N1 , N2 , O1 , O2 , z , u , E v ,
L v , E g , L g ,   ) can be computed. The model is thus solved.

3 Volatility in the Knowledge Sectors

Our hypothesis is that the level of employment in the increasing returns to
scale sectors (IRS) is more volatile than that of other sectors, in the sense
that employment increases more in IRS sectors than in the rest of the econ-
omy during upturns; during downturns, it contracts more than in the rest
of the economy. To test this hypothesis we introduce a measure of volatility
of employment, similar to that used in our previous paper, and explore its
behavior in a general equilibrium model with involuntary unemployment.

Volatility Index: The ‘Labor Beta’

We define an index to measure volatility that is independent of the scale of

the variables, denoted ‘labor beta’. This is a statistical relative of the financial
markets concept of ‘beta’ that is frequently used to measure volatility of stock
prices:
Fry([> \ )
= (23)
Y du(\ )
Here [ and \ are employment levels rather than stock prices, [ repre-
senting employment in a sector and \ , aggregate employment.  LUV denotes
the ‘labor beta’ associated with the increasing returns sector and  FUV that
of constant returns to scale sector. Our hypothesis can now be stated as:
Hypothesis:  LUV A  FUV =
 Volatility and Job Creation in the Knowledge Economy 57

Volatility: The Eects of Shocks on Employment Across Sectors

To study employment volatility we assume that there are random shocks to

the ‘fundamentals’ of the model (technologies, preferences, demand, initial
endowments of capital and labor). The fundamentals thus vary from period
to period, although there are no intertemporal links between one period or
another. The equilibrium value of sE varies with the fundamentals producing
fluctuations in all equilibrium values of the model, and in particular in equi-
librium levels of labor utilized by each sector: O1 and O2 . Using the general
equilibrium model, we study the attendant variations in O1 and O2 , exploring
the extent to which employment in sector L is systematically more volatile
than that in sector E.
From the ‘resolving’ equation, I (sE ), we know how all the variables of
the model fluctuate with sE . Expressing labor demand for each sector as a
function of output of that sector will allow us to use the main result of the
“Volatility in the Knowledge Economy” article21 to prove that employment
volatility of the IRS sector L is larger than that of CRS sector E. Solving the
firm’s cost minimization problem O1 and O2 can be expressed as functions of
factor prices and total output of their respective sectors, or:
µ ¶ µ ¶
(1  )u (1  )u
O1 = E and O2 = L
z z
From (10), (16), and (19), the above expressions are functions of a single
parameter sE found after solving the ‘resolving’ equation I (sE ) (21), or

   
(1 3 )u(sE ) 1 (1 3 )u(sE )
O1 (sE ) = (sE ) and O2 (sE ) = (sE ) (1)
z(sE ) z(sE )
(24)

Proposition 1. Employment volatility in the increasing returns to scale sec-

tor L is larger than that in the constant returns to scale sector E, i.e. the
‘labor beta’ of external IRS sector,  LUV , is larger than the ‘labor beta’ of the
CRS sector,  FUV , namely  LUV A  FUV , when increasing returns are large
enough, i.e. 0 ?  ? 1 and   1.

Proof. We aim to prove the following inequality:

Fry(O2w > Ow ) Fry(O1w > Ow )

 LUV =  A  FUV = (25)
Y du(Ow ) Y du(Ow )

where Ow = O1w +O2w = Because the denominators of (25) are equal and positive,
we can restate it as:
21
Specifically, in that article we showed that as  < 1,  LUV A  FUV , or that
output of the IRS sector is more volatile than that of the CRS sector.
58 Graciela Chichilnisky and Olga Gorbachev

Fry(O2w > Ow ) A Fry(O1w > Ow ) (26)

Rewriting (26):
W
X W
X
(O2w  O2 )(Ow  O ) A (O1w  O1 )(Ow  O ) (27)
w=1 w=1

1
PW
where O1w > O2w and Ow denote time averages: O1 = W w=1 (O1w )=
Rearranging the terms (26) becomes:
W
X ©£ ¤ ª
(O2w  O2 )  (O1w  O1 ) (Ow  O ) A 0
w=1

Substituting for Ow and O :

W n
X £ ¤h io
(O2w  O2 )  (O1w  O1 ) (O1w + O2w )  (O1 + O2 ) A 0> or
w=1

W
X
{O22w  O21w } + W {(O1 )2  (O2 )2 } A 0
w=1

From (21) and (24) we obtain:

( µ ¶2 µ ¶2 )
PW 2 (1  )u(s E ) 2 (1  )u(s E )
w=1 (sE ) (1)   (sE )
z(sE ) z(sE )
+W {(O1 )2  (O2 )2 } A 0 (28)

For every w, as  $ 1 and w (sE ) A 1> the inequality (28) is satisfied since
2
the first term (w (sE ) (1) ) dominates the equation. It follows therefore, that
 LUV A  FUV .

4 Empirical Issues
In this section we validate empirically our theoretical result that employment
in IRS sectors is more volatile in CRS sectors.

Data Sources and Structure

We use employment data provided by the Survey of Current Business (SCB) 22 ,

prepared by the U.S. Department of Commerce, Bureau of Economic Analysis
(BEA), in Washington, DC, utilizing ‘Full Time and Part Time Employees’
22
Monthly government publications to be found in www.bea.gov.
 Volatility and Job Creation in the Knowledge Economy 59

by industry in thousands for 1948-2001 period (2-3 digit Standard Industry

Codes (SIC))23 . There is a “break” in the time series of this data due to
the SIC reclassification. Thus, 1948-1987 data uses 1972 SIC classification,
whereas 1987-2001 uses 1987 SIC classification (the estimates of 1977-1987
have not been adjusted to 1987 SIC code due to “lack of adequate data”
according to BEA May 2003 publication [34]).

IRS and Traditional Industries

We reviewed the literature and adopted its findings about IRS sectors,24 as
in our 2004 article [13]. We also used a simple correlation between quantities
produced and prices charged on the level of industry to identify IRS sectors
(the sector is considered to exhibit IRS if correlation is negative)25 . Figure 3
lists the IRS sectors identified by the empirical literature, showing in each case
the respective sources. From the list in Figure 3 we separated out industries
that as in the previous paper, we characterize as having internal IRS due to
high fixed costs26 . We were left with 8 industries with external economies of
scale, which we compared with some of the World Knowledge Competitiveness
Index benchmarks (see footnote 2 for details):
1. Credit agencies other than banks (SIC 61)
2. Electronic equipment and instruments (36, 38)
3. Machinery, except electrical (35)
23
Detailed data series are available at the BEA webpage:
http://www.bea.gov/bea/dn2/gpo.htm
24
A study by S. Basu and J. Fernald [3] of US private economy (2-3 SIC) find IRS
in: 1. Metal Mining, 2. Construction, 3. Furniture, 4. Paper, 5. Primary Metals, 6.
Fabricated Metals, 7. Electrical Machinery, 8. Motor Vehicles, 9. Transportation,
10. Communication, 11. Electric Utilities, 12. Wholesale and Retail, 13. Services
(various).
Work by W. Antweiler and D. Trefler [1] examine 27 manufacturing and 7 non-
manufacturing industries (no services) for 71 countries over 1972-1992 period and
identified 11 industries with increasing returns: 1. Petroleum and Coal Products,
2. Pharmaceuticals, 3. Electric and Electronic Machinery, 4. Petroleum Refineries,
5. Iron and Steel Basic Industries, 6. Instruments, 7. Non-Electric Machinery, 8.
Forestry, 9. Livestock, 10. Crude Petroleum and Natural Gas, 11. Coal Mining.
Their general equilibrium model estimates scale for these industries in the rage
of 1.10 to 1.20.
Paul and Siegel [28] find that scale economies are prevalent in US manufactur-
ing. In particular, this study finds evidence of external economies of scale due to
supply-side agglomeration.
25
Specifically, we used chain-type quantity index for GDP by industry and chain-
type price index for GDP by industry from BEA for our correlation computations.
Detailed data files can be downloaded at: http://www.bea.gov/bea/dn2/gpo.htm.
26
It is important to carefully study the structure of each industry before charac-
terizing it as having internal or external IRS. This is a subject of our further
research in this area.
60 Graciela Chichilnisky and Olga Gorbachev

Identification
Agriculture, forestry, and fishing AT
Credit agencies other than banks corr=<0
Coal mining AT
Communications BF
Construction BF
Electronic equipment and instruments BF, AT
Fabricated metal products BF
Furniture and fixtures BF
Machinery, except electrical AT
Metal mining BF
Motor vehicles and equipment BF
Oil and gas extraction AT
Paper and allied products BF
Petroleum and coal products AT
Primary metal industries BF
Retail trade BF
Security and commodity brokers corr<0
Services BF
Telephone and telegraph corr<0
Transportation BF
Wholesale trade BF

Source: U.S. Department of Commerce, Bureau of Economic Analysis.

Detailed annual series can be found at:
http://www.bea.gov/bea/dn2/gpo.htm. y
Antweiler and D. Trefler study of 2000. Correlation coefficient is computed
as correlation between Total nominal GDP by industry with respect to
industry's Price Deflator.

Fig. 3. List of Increasing Returns to Scale Industries Identified by the Literature

Review and the Correlation Coe!cient Specification

4. Retail Trade (52-59)

5. Security and commodity brokers (62)
6. Services (70, 72-73, 75-76, 78-89)
7. Telephone and telegraph (481, 482, 489)
8. Wholesale Trade (50, 51).

Employment in Increasing Returns Sectors Is More Volatile

Total employment’s ‘labor beta’, by definition, is equal to one. Figure 4 pro-
vides ‘labor betas’ for industries with external economies of scale and for some
traditional industries. What is immediately apparent from Figure 4 is that
‘labor betas’ for IRS industries are on average larger than that of traditional
industries, and that the volatility in the service sector drives the results.
 Volatility and Job Creation in the Knowledge Economy 61

In addition, breaking the data into three sub-periods, 1947-1987 and 1988-
2001 (chosen arbitrarily at the break of the series and to be consistent with
our previous paper) provides another interesting view of the data. The ‘labor
betas’ for IRS on average are largest in the later period, or in 1987-2001
period. This phenomenon could be explained by our model: as  $ 1> ‘labor
beta’ of the IRS sector becomes larger.

labor beta % small

1947- 1947- 1988- labor firms
Industry (SIC) 2001 1987 2001 shares 1997
External IRS
Wholesale trade (50, 51) 0.06 0.06 0.03 0.05 65
Services (70, 72-73, 75-76, 78-89) 0.44 0.36 0.53 0.22 55
Retail Trade (52-59) 0.18 0.18 0.15 0.15 51
Machinery, except electrical (35) 0.01 0.02 0.00 0.02 47
Electronic equipment and instruments (36, 38) 0.02 0.03 0.00 0.03 43
Credit agencies other than banks (61) 0.01 0.01 0.01 0.00 35
Security and commodity brokers (62) 0.01 0.01 0.01 0.00 28
Telephone and telegraph (481, 482, 489) 0.00 0.01 0.01 0.01 10
Average IRS 0.26 0.22 0.29 0.48 53
Traditional
Real estate 0.02 0.02 0.01 0.01 88
Miscellaneous manufacturing industries 0.00 0.00 0.00 0.00 67
Lumber and wood products 0.00 0.00 0.00 0.01 64
Furniture and fixtures 0.00 0.00 0.00 0.01 59
Printing and publishing 0.01 0.01 0.00 0.01 51
Leather and leather products 0.00 0.00 0.00 0.00 42
Textile mill products -0.01 -0.01 -0.01 0.01 31
Apparel and other textile products -0.01 0.00 -0.02 0.01 29
Food and kindred products 0.00 0.00 0.00 0.02 28
Paper and allied products 0.00 0.00 0.00 0.01 28
Chemicals and allied products 0.00 0.01 0.00 0.01 21
Average Traditional 0.00 0.00 0.00 0.11 48

Source: U.S. Department of Commerce, Bureau of Economic Analysis. Detailed annual series on “Full Time and Part
Time Employment by Industry” can be found at: www.bea.gov/bea/dn2/gpo.htm; % of small firms within the industry
in 1997 is provided by the U.S. Small Business Administration Office of Advocacy, “US Small Business Indicators
2001,” 2002.
Note: Shares are averages over 1947-2001 periods and are computed as a number of workers employed in an industry
as a share of the total employment. Average IRS and Average Traditional are weighted averages of the series in the
respective columns using the provided employment shares.

Fig. 4. Employment Volatility in IRS Sectors is Larger than in Traditional Sectors

5 The Puzzle of the Jobless Recovery

We saw that employment in sectors with increasing returns to scale is on
the whole more volatile than in other sectors, in the sense of having higher
‘labor betas’. We also saw that within the IRS sectors, service sectors have
the highest volatility, or highest ‘labor betas’. Service sectors employ 22 per-
cent of the U.S. workers (see Figure 4) and are populated by small firms:
55 percent of the firms in the service industry have less than 500 employees.
Small firms generate most new jobs in the US, between 60-80 percent between
1989 and 2001, as illustrated in Figure 5 [40], even though they employ only
62 Graciela Chichilnisky and Olga Gorbachev

about 50 percent of the workers. According to the numerous studies by the

Small Business Administration, 80 percent of industries that generated most
employment (during upturns) and lose most employment (during downturns)
over the last 10 years have been in service sector [41]. The service sector is
key to our study: on the whole it has increasing returns to scale; it generates
most jobs; and it is populated by small firms.

3,300,000

2,900,000
Total Jobs
2,500,000

2,100,000

1,700,000
Small Firms (< 500)

1,300,000

900,000

500,000
Large Firms (500+)
100,000

1989- 1990- 1991- 1992- 1993- 1994- 1995- 1996- 1997- 1998- 1999- 2000- 2001- 2002- 2003-
-300,000 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

-700,000

-1,100,000

-1,500,000

Source: Census Bureau, "County Business Patterns Survey". The total employment data is from January 1989 to March
2004, but employment by size of the firm is from January 1989 until January 2001.

Fig. 5. Net Jobs Generated by Size of Firm

If IRS industries were a large part of the economy, they should, in principle,
generate a large increase in employment during a recovery. The US economy
grew 4.1 percent in real terms in the fourth quarter of 2003 (surpassing its
lowest point of third quarter of 2001) [37]. Yet employment has not yet reached
its pre-recession levels. In order for employment to attain its pre-recession
levels, 1.5 million jobs lost over 2001-2003 periods27 must be replaced [39].
Since the IRS sectors are a significant portion of the economy, they employ
around 50 percent of the workers, then we should have job creation rather
than the observed jobless recovery. How do we resolve this puzzle?
There are at least two possible explanations for the jobless recovery: (i)
due to the increased uncertainty in the economy (geopolitical and economic),
external IRS sectors that are populated by small firms have been particularly
badly hit and thus have been unable to generate new employment; and (ii)
there has been a structural change in the economy that shifted the ‘produced
27
Calculations are done by the authors.
 Volatility and Job Creation in the Knowledge Economy 63

labor’ supply curve to the left, making it more di!cult to generate new jobs.
Both of these situations would lead in our model to a jobless recovery. A
combination of these two explanations is at the core of the jobless recovery
and is discussed below.

Small Firms and the Service Sector: Heightened Uncertainty and

Structural Change

In the current cyclical recovery of 2001-2004, the service sector could have gen-
erated most employment, since it has highest ‘labor betas’ and most new em-
ployment comes from this sector [41]. However, the service sector is populated
by small firms, which finance their operations through equity and retained
earnings rather than debt [42]. Obtaining equity funding or venture capital
has been extraordinarily di!cult in the environment of high uncertainty28
both geopolitical and economic since 2001. Equity funding for small firms, as
illustrated by Figure 6, fell 86 percent over three year period, 2000-200329
[43]. Figure 7 shows that commercial credit, which constitutes 57 percent of
small firms’ debt also fell by 17 percent. Left without funds to finance their
operations, small firms stumbled during this period and therefore generated
fewer jobs. Thus, even though the economy was coming out of a recession, the
continued uncertainty, especially felt by the service sector small firms, reduced
the likelihood of increased job creation generating conditions associated with
a jobless recovery (see Figure 8).
The results of our model support also the Federal Reserve Bank of New
York’s explanation of the jobless recovery [18]. The FRBNY’s explanation for
the jobless recovery is that a structural change30 occurred during the last busi-
ness cycle. This is supported by their recent study in which the authors show
that permanent shifts in the distribution of workers throughout the economy
have contributed to the jobless recovery. FRBNY’s study proposes three expla-
nations for this change: (i) structural decline might be a reaction to a period
of overexpansion, (ii) improved monetary and fiscal policy may have reduced
cyclical swings in employment, and (iii) innovation in firm management may
be promoting a structural shift towards leaner sta!ng.
A structural change would require a change in workers’ skills. Therefore,
even if the recovery created an increased demand for labor as the ‘produced’
labor function (defined in section 2.03) shifted left, this would have lead to
decrease in employment. Figure 9 shows how the new employment level in
equilibrium O could be even less than the previous level O with such a
28
Uncertainty is due to (i) economic and political factors, and (ii) to controversy
over corporate practices after one of the worse collapses in history of the stock
market.
29
Venture capital fell 61% between 2000 and 2001; 45% between 2001 and 2002;
and 37% between 2002 and first half of 2003.
30
Structural adjustments transform a firm or industry by relocating workers or
capital, according to the FRBNY’s definition [18].
64 Graciela Chichilnisky and Olga Gorbachev

120

100

80
billions, $

60

40

20

0
1997 1998 1999 2000 2001 2002 2003

Source: Small Business Administration Office of Advocacy, “SBIC Program Share of Total Venture Capital Financing
to Small Business Reported for Calendar Years 1997-September 2003.”

Fig. 6. Venture Capital Financing for Small Firms

shift even with an expansion in labor demand. Our model therefore supports
the FRBNY’s conclusions.

6 Conclusions and Policies for Job Creation

We showed that employment volatility is higher in the increasing returns to
scale sectors than in the rest of the economy. The highest volatility is observed
in the service sector, an increasing returns to scale sector mostly populated
by small firms with less than 500 employees. It has been shown elsewhere [12]
that smaller firms create most new jobs in the economy. Yet small firms have
been deprived of funding during the recovery because of heightened uncer-
tainty, geopolitical and economic. This combination of circumstances could
be a reason for the jobless recovery.
Indeed, most of small firms’ funding comes from equity financing and ven-
ture capital that declined 86 percent over the 2001-2003 period. Even though
the current US recovery seems to be standard in terms of real output growth,
the dramatically reduced funding undermined the ability of small firms to gen-
erate employment. Thus we conclude that the first explanation, (i) in section
5.011, for the jobless recovery seems justified.
 Volatility and Job Creation in the Knowledge Economy 65

15,000 120,000

14,500 110,000

14,000 100,000

More than $1 million

Less than $1 million

13,500 90,000
Loans to Large Firms

13,000 80,000

12,500 70,000

Loans to Small Firms

12,000 60,000

11,500 50,000

11,000 40,000
1997 1998 1999 2000 2001 2002 2003

Source: Federal Reserve Statistical Release, “E.2 Survey of Terms of Business Lending: 1. Commercial and industrial
loans made by all commercial banks”, 1997-2003. The right hand side axis is for the loans to large firms, i.e. loans that
are greater than $1million; and the left hand side axis is for the loans to small firms, i.e. loans that are less than
$1million.

Fig. 7. Lending by All U.S. Banks by Size of Loan, in thousands of dollars

A second feature of our model that helps explain a jobless recovery is the
impact of structural change on involuntary unemployment. The Federal Re-
serve Bank of New York has documented that structural change has occurred
in this upswing. This could have shifted the ‘produced labor’ supply to the
left making it more di!cult for the economy to create new employment, as in
(ii) section 5.011.
The combination of two circumstances (i) the stumbling of small firms,
and (ii) structural change, is our preferred explanation for the jobless recovery
since 2002.
Several policies can be suggested. Since small firms are viewed as risky,
they have limited access to funds, face higher costs of capital, and have less
access to top talent. Overcoming these obstacles is what will create most new
jobs. Thus, policies should focus on decreasing the risks faced by small firms.
Specific financial policies designed for this purpose were suggested in
Chichilnisky [12]. One is to aggregate the equity of smaller firms into ‘bundles’
that have a lower risk profile due to the ‘law of large numbers’. By securitiz-
ing these bundles they can be sold in financial institutions — such as stock
exchanges — with access to global capital markets. The result is a much larger
pool of funds for smaller firms, and lower cost of capital.
66 Graciela Chichilnisky and Olga Gorbachev

140,000

120,000

Total Goods-Producing

100,000

80,000

60,000

Total Service-Providing
40,000

20,000
1958

1962

1964

1968

1994

1998
1954

1956

1960

1966

1970

1972

1974

1976

1978

1980

1982

1984

1986

1988

1990

1992

1996

2000

2002
Source: U.S. Bureau of Labor Statistics, “Establishment Data Survey: Historical Employment B-1. Employees on
Nonfarm Payrolls by Major Industry Sector, 1954 to date.”

Fig. 8. Employment Evolution from 1954 to 2003 in Goods-Producing vs. Service-

Providing Industries, in thousands

A second proposal is to create credit - enhancement instruments that facil-

itate the smaller firms’ access to capital. A third proposal is to provide access
to better borrowing rates for smaller firms by creating large baskets of equity
across firms which oer less risk due to diversification — much like Freddie
Mae has done successfully in the lower income home mortgage market. In-
stead of home equity, in this case the assets that back the securities would
be the equity in the firms themselves, aggregated throughout many firms in
order to reduce risk.
In addition to financial solutions, providing low cost training to employ-
ees and benefits to ease sectorial displacements will help resolve problems
associated with structural shifts.

Appendix
Workers and the Labor Market (Shapiro and Stiglitz 1984)

Here we provide a specific formulation of the general equilibrium model with

involuntary unemployment following analysis of Shapiro and Stiglitz’ in “Equi-
librium Unemployment as a Worker Discipline Device”[32].
 Volatility and Job Creation in the Knowledge Economy 67

Labor Force, N
Wages `Produced’ Labor Supply
Ls shifts

Involuntary
w** Unemployment

w* Labor Demanded by Firms

Ld shifts

L** = L* N Employment

Fig. 9. Structural changes shift the ‘Produced Labor’ supply curve leading to the
Jobless Recovery

There are Q identical workers in the economy. All workers dislike putting
forth eort, but enjoy consuming goods. Individual’s instantaneous utility
function can be expressed as X (f> h), assuming for simplicity that the utility
function is separable:
X (f> h) = K(f)   (h) (29)
where K(f) is a continuous function of workers’ consumption bundle, such
g2 K
that gK
gf A 0> gf2 ? 0; and  (h) is a continuous function of disutility of eort,
g2 
h, such that g
gh A 0> gh2 ? 0=
Each worker maximizes his utility subject to budget constraint:
sf f = zO + uN
where sf is the index of consumer prices and f is the consumption bundle;
z stands for wages and u for rental on capital received by each worker.
Assume for simplicity that every employed worker supplies one unit of
labor, or O = 1 and that utility from consumption has a Cobb-Douglas form.
Then, worker’s problem can be restated as:
max X (E> L> h) = E  L 1   (h) (30)
v=w= sE E + L  z + uN
where  5 (0> 1)= By solving (30), demand functions for goods E and L,
are given as:
68 Graciela Chichilnisky and Olga Gorbachev

(zO + uN)
Eg = = E g (z> u> sE ) and L g = (1  )(zO + uN) = L g (z> u> sE )
sE
(31)
One can also show that (30) can be rewritten as an indirect utility function

Y (z> u> h) = (z + uN)   (h)

where  = ( sE )1 (1  ) .

Each worker is in one of two states: employed or unemployed. There is a
positive probability e per unit time that a worker will be separated from his job
due to unrelated to his productivity causes (i.e. relocation, bankruptcy, etc.
of his enterprise), which is exogenously given. Workers maximize the expected
present discounted value of utility with a discount rate  A 0.
Z 4
Z =H Y (zw > uw > hw )hw gw (32)
0

The main choice workers make is the selection of the eort or productivity
level. If the worker performs well, or at an expected productivity level, i.e.
he doesn’t shirk, he receives a wage z with probability (1  e). If the worker
shirks, there is a probability t A 0 that he will be caught and fired, and
therefore will enter the unemployment
_
pool. While unemployed, he receives
unemployment compensation z set by the government (and assumed to be
zero from now on).
The utility from work of a shirker for a short interval [0> w] can be stated
as:

YHV = w{(z + uN)   (h)} + (1  w)[(ew + tw)YX + (1  ew)(1  tw)YHV ] (33)

where YX is the expected lifetime utility of an unemployed individual and

YHV is the expected lifetime utility of an employed shirker. YHQ is the expected
lifetime utility of the employed individual-nonshirker, and can be written as:

YHQ = w{(z + uN)   (h)} + (1  w)[ewYX + (1  ew)YHQ ] (34)

Solving (34) and (33) by taking a limit of each of the two equations as
w $ 0, rearranging the terms, taking YX as given, and assuming for simplicity
that  (h) is a discrete function, such that  (h) = 0 if the individual shirks,
and  (h) = h A 0 if he doesn’t, we get:

YHV = (z + uN) + (e + t)[YX  YHV ] (35)

YHQ = (z + uN)  h + (e)[YX  YHQ ]

or

(z + uN) + (e + t)YX

YHV = (36)
+e+t
 Volatility and Job Creation in the Knowledge Economy 69

(z + uN)  h + eYX

YHQ = (37)
+e
The worker will choose to be productive, or to put in h A 0 if and only if
YHQ  YHV or
YX h
z  uN + (t + e + ) (38)
 t
Equation (38) is known as the No-shirking condition (NSC). In order
to determine the wage demanded by the workers, we first have to rewrite
NSC explicitly, taking into account the expected utility of an unemployed
worker,Y x. Thus, analogously to equations (35), Y x can be stated as:
_
YX = (z + uN) + d(YH  YX ) (39)

where d is the job acquisition rate and YH is the expected utility of the
employed worker. In equilibrium YH =YHQ = Now, plugging (37) into (39) and
rearranging the terms, we get:
_
z( + e) + dz dh uN
YX =  + (40)
( + d + e) ( + d + e) 

Substituting (40) into (38), NSC becomes:

_ h h
zz+ + ( + e + d)31 (41)
 t
The implication of NSC in this model is that unless there is a penalty
from being unemployed (or it takes time before a person can get rehired),
there will be no reason for individuals to provide eort32 . Thus, in order for
NSC to be satisfied, there will always be a positive number of unemployed.
The wage demanded by the workers positively depends on the job ac-
quisition rate, d. Following the logic of Shapiro-Stiglitz, d can be related to
the more fundamental parameters of the model. In steady-state, the flow of
workers into the unemployment pool, eO and the flow of workers out of the
unemployment pool, d(Q  O)> (where O is the aggregate employment and Q
is the total labor force in equilibrium), must be equal:

31
Notice that this NSC equation slightly diers from that of Shapiro-Stiglitz. The
wage demanded by the workers as in Shapiro-Stiglitz positively depends on the
unemployment benefits, eort that the worker puts in, probability of loosing the
job e, probability of not getting caught shirking (1 3 t), and the job acquisition
C
rate, d. But z depends positively on sE > since CsE
= (13)1+ ( sE )3 (3 s2 ) ? 0
E
and Cz
C
= 3 h2 (1 + 1t ( + e + d)) ? 0> thus, Cs
Cz
E
= Cz
C
C
W CsE
A 0=
32 V
Since NSC can be also stated as t(YH 3YX ) D h, if an individual could immidiately
obtain employment after being fired, or YHV = YX , or d = +", NSC could never
be satisfied.
70 Graciela Chichilnisky and Olga Gorbachev

d(Q  O) = eO (42)
Substituting (42) into (41):
_ h h eO
zz+ + ( +  + e) (43)
 t Q  O
Thus, wage demanded is a negative function of the number of unemployed,
and as the number of employed (O) increases, wage demanded increases more
than proportionately, because it is aected by two simultaneous factors (i)
an increase in the number of employed, and (ii) a decrease in the number of
unemployed.
Rewriting the above expression we get an explicit labor supply equation:
ht + h + he  zt
Ov  Q (44)
ht + h  zt
Or implicitly

Ov = Ov (z> sE > h)
v v v A
such that CO 33 CO 34 CO
Cz A 0 > CsE ? 0 > and Ch ? 0 depending on the parameters
of the model35 . Figure 1 illustrates this relationship for a specific eort level
h (that could be set to 1 for simplicity).
In equilibrium all markets clear, or:

E v = E g + [E (E market clears) (45)

L v = L g + [L (L market clears)


where [E and [L are equilibrium levels of net exports in E and L sectors
respectively. We assume [L = 0 and [E 
= 0 but the results are true for any
 
given [L and [E .

sE E v + L v = z O + u N  (zero profits)

N  = N v = N1 + N2 (capital market clears)

O = Ov = Og = O1 + O2 (labor market clears)
Notice however that labor market clears in the sense that labor supply is equal
to labor demand. But since labor supply is given by the NSC (43), the labor
market is characterized by the existence of involuntary unemployment, due to
33 C(Ov ) e
Since Cz
= tQh (ht+h3zt) 2 A 0.
COv
34
Since C
= ztQh e
(ht+h3zt)2 A 0 and C
CsE
= (1 3 d)1+d ( sdE )3d (3 sd2 ) ? 0>
E
v v
CO CO C
then = W ? 0=
CsE v C CsE
35 C(O ) ht+h+he3zt Q Q Q
Since Ch = Q (ht+h3zt)h = ht+h3zt
t + ht+h3zt
 + ht+h3zt
e 3
Q
(ht+h3zt)h
zt.
 Volatility and Job Creation in the Knowledge Economy 71

Q A O , as demonstrated by Figure 2. Workers without a job are willing to

work at a wage below z but cannot make a credible promise not to shirk at
such a wage36 , and thus cannot be hired by the firms.
We can characterize labor market equilibrium in a more explicit way. Since
the wage paid by each firm must satisfy the NSC (43), wages paid by firms
given by (10) can be rewritten as:
h h eO 1
+ ( 
+  + e) = hD(1) sE
 t Q  O
Solving the above expression and taking for simplicity h = 1 when workers
exert eort, gives us the total labor used in the economy in equilibrium O as
a function of a single unknown parameter sE :
½ 1
¾
D(1)  1
h sE   t    e
O = Q ½ 1
¾ = O (sE ) (46)
D(1)  1
h sE   t  

The rest of the solution of this general equilibrium model follows the main
text, writing explicitly the ‘resolving’ equation for the model:
; 1
? hD O hD hE N v sE 
g 
I (sE ) = L (sE ) [O  D  ]
= (h  hE ) (hD  hE )

1
< 1
1


O sE  E
h N v @
× [N v   ]1 = 0 (47)
(hD  hE ) (hD  hE ) >

where from (10) and (12), L g (sE ) is a function of sE alone:

½ 1 
¾
g D(1)   D  
L (sE ) =(1  ) (h sE O (sE ) + (1  )h sE N ) (48)

Solving the equation I (sE ) = 0> gives an equilibrium value of sE from
which all equilibrium values of other variables (N1 , N2 , O1 , O2 , z , u , E v ,
L v , E g , L g ,   ) can be computed. The model is thus solved.

References
1. Werner Antweiler; Daniel Trefler, “Increasing Returns and All That: A View
From Trade”. NBER Working Papers #7941, Oct. 2000.
36
As pointed out by Shapiro-Stiglitz, the unemployment in this model is dierent
from that characterized by the search unemployment. Here, all workers and all
firms are identical. There is perfect information about job availability but firms
are assumed unable to perfectly and costlessly observe the work eort applied by
their employees.
72 Graciela Chichilnisky and Olga Gorbachev

2. Kenneth Arrow, “The Economic Implications of Learning by Doing”. Review of

Economic Studies, 1962, Vol. 29, No. 3, pp. 155-173.
3. Susanto Basu; John G. Fernald, “Aggregate Productivity and the Productivity
of Aggregates”. NBER Working Papers #5382, Dec. 1995.
4. Susanto Basu; John G. Fernald, “Returns to Scale in U.S. Production: Estimates
and Implications”. The Journal of Political Economy, Apr., 1997, Vol. 105, No.
2., pp. 249-283.
5. Elise S. Brezis; Paul R. Krugman; Daniel Tsiddon, “Leapfrogging in Interna-
tional Competition: A Theory of Cycles in National Technological Leadership”.
The American Economic Review, Dec., 1993, Vol. 83, No. 5., pp. 1211-1219.
6. Craig Burnside, “Production Function Regressions, Returns to Scale, and Ex-
ternalities”. Journal of Monetary Economics, 1996, Vol. 37, pp. 177-201.
7. David Cass and Herakles Polemarchakis, “Convexity and Sunspots: A Remark”,
Journal of Economic Theory, Dec. 1990, Vol. 52, pp. 433-439.
8. Census Bureau, "County Business Patterns Survey", provided by Virginia Uni-
versity at http://fisher.lib.virginia.edu/collections/stats/cbp/
9. Graciela Chichilnisky, “Traditional Comparative Advantages vs. Economies of
Scale: NAFTA and GATT”. Rivista Di Politica Economica, Apr. 1993, pp. 161-
197.
10. Graciela Chichilnisky, ”Traditional Comparative Advantage vs. Increasing Re-
turns to Scale: NAFTA and the GATT,” International Problems of Economic
Interdependence, New York, 1994 (eds. Baldassarri, M., Di Matteo, M., and
Mundell R.), St. Martin’s Press Inc., pp. 161-197.
11. Graciela Chichilnisky, “Trade Regimes and GATT: Resource Intensive vs.
Knowledge Intensive growth” Chapter 10 in Levi-Livermore,A. (ed.) Handbook
on the Globalization of the World Economy, Chapter 10, pp. 226-249, Edward
Elgar, Cheltenham UK. Northhampton, Mausa, 1998, pp. 147-181.
12. Graciela Chichilnisky, "The Job Creating Machine.", Columbia University
memo.
13. Graciela Chichilnisky and Olga Gorbachev, “Volatility in the Knowledge Econ-
omy”, Economic Theory, Sept. 2004 (forthcoming).
14. Graciela Chichilnisky and Georey Heal, The Evolving International Economy.
Cambridge University Press, 1986.
15. Graciela Chichilnisky, A. Beltratti, and Georey M. Heal, “Sustainable Use of
Renewable Resources,” (with A. Beltratti and G.M. Heal), Chapter 2.1, Sustain-
ability: Dynamics and Uncertainty. (G. Chichilnisky, G.M. Heal, and A.Vercelli,
eds), Kluwer Academic Publishers, The Netherlands, 1998, pp. 49-76.
16. John S. Chipman, “External Economies of Scale and Competitive Equilibrium”,
The Quarterly Journal of Economics, Aug. 1970, Vol. 84, No. 3, pp. 347-385.
17. Federal Reserve Board of Governors, "Statistical Release. E.2 Survey of Terms
of Business Lending: 1. Commercial and industrial loans made by all commercial
banks”, 1997-2003.
18. Erica L Groshen and Simon Potter, "Has Structural Change Contributed to a
Jobless Recovery?" Current Issues in Economics and Finance, Vol. 9, No. 8,
Aug. 2003.
19. Georey M. Heal, ed., The Economics of Increasing Returns. Edward Elgar,
1999.
20. Georey M. Heal, “Macrodynamics and Returns to Scale”. The Economic Jour-
nal, Mar. 1986, Vol. 96, Is. 381, pp. 191-198.
 Volatility and Job Creation in the Knowledge Economy 73

21. Dale W. Jorgenson, “Productivity and Postwar U.S. Economic Growth”. The
Journal of Economic Perspectives, Vol. 2, Iss.4, (Autumn, 1988), pp. 23-41.
22. Dale W. Jorgenson, Mun S. Ho, and Kevin J. Stiroh, “Projecting Productivity
Growth: Lessons from the U.S. Growth Resurgence”. Economic Review, 2002,
Iss.Q3, pp. 1-13.
23. Paul Krugman, “Scale Economies, Product Dierentiation, and the Pattern of
Trade”. The American Economic Review, Dec., 1980, Vol. 70, No. 5., pp. 950-
959.
24. Paul Krugman, “Intraindustry Specialization and the Gains from Trade”. The
Journal of Political Economy, Oct., 1981, Vol. 89, No. 5., pp. 959-973.
25. Paul Krugman, “New Theories of Trade Among Industrial Countries (in Inter-
national Trade)”. The American Economic Review, May, 1983, Vol. 73, No. 2,
Papers and Proceedings of the Ninety-Fifth Annual Meeting of the American
Economic Association. pp. 343-347.
26. Paul Krugman, “Increasing Returns and Economic Geography”. The Journal of
Political Economy, Jun. 1991, Vol. 99, No. 3, pp. 483-499.
27. Arvind Panagariya, “Variable Returns to Scale in Production and Patterns of
Specialization”. The American Economic Review, Mar. 1981, Vol. 71, Is. 1, pp.
221-230.
28. Catherine J. Morrison Paul and Donald S. Siegel, “Scale Economies and In-
dustry Agglomeration Externalities: A Dynamic Cost Function Approach”, The
American Economic Review, Vol. 89, No.1, (Mar. 1999), pp. 272-290.
29. Luis A. Rivera-Batiz, Paul Romer, “Economic Integration and Endogenous
Growth”. The Quarterly Journal of Economics, May 1991, Vol. 106, Is. 2, pp.
531-555.
30. Paul Romer, “Increasing Returns and Long Run Growth”. Journal of Political
Economy, 1986, Vol. 94, No. 5, pp. 1002-1037.
31. Stephanie Schmitt-Grohé, “Endogenous Business Cycles and the Dynamics of
Output, Hours, and Consumption”. The American Economic Review, Dec. 2000,
Vol. 90, No. 5, pp. 1136-1159.
32. Carl Shapiro, Joseph E. Stiglitz, “Equilibrium Unemployment as a Worker Disci-
pline Device”, The American Economic Review 74, Is.3, June 1984, pp. 433-444.
33. Jerome Stein, The Economics of Future Markets, 1987, New York: Basil Black-
well.
34. U.S. Department of Commerce, Bureau of Economic Analysis: Survey of Current
Business, May 2003.
35. U.S. Department of Commerce, Bureau of Economic Analysis: Survey of Current
Business, Nov. 2002.
36. U.S. Department of Commerce, Bureau of Economic Analysis: Survey of Current
Business, “Improved Estimates of Gross Product by Industry for 1947-1998”,
June 2000.
37. U.S. Department of Commerce, Bureau of Economic Analysis, "News Release:
Gross Domestic Product and Corporate Profits", March 25, 2004.
38. U.S. Department of Commerce, Bureau of Economic Analysis: Survey of Current
Business, May 1997.
39. U.S. Department of Labor, Bureau of Labor Statistics, http://data.bls.gov/
servlet/SurveyOutputServlet.
40. U.S. Small Business Administration O!ce of Advocacy "Small Business by the
Numbers", www.sba.gov/advo/ December 2003.
74 Graciela Chichilnisky and Olga Gorbachev

41. U.S. Small Business Administration O!ce of Advocacy, “US Small Business
Indicators,” 1995-2002.
42. U.S. Small Business Administration O!ce of Advocacy, "Financing Patterns of
Small Firms: Findings from the 1998 Survey of Small Business Finance.", Sept.
2003.
43. Venture One, "Annual financing survey by Venture One of San Francisco," web
at www.ventureone.com.
Existence of Sunspot Equilibria and
Uniqueness of Spot Market Equilibria:
The Case of Intrinsically Complete Markets *

Thorsten Hens 1, János Mayer 2 and Beate Pilgrim 3

1
IEW, Department of Economics, University of Zurich, Switzerland
[email protected]
2
IOR, Department of Economics, University of Zurich, Switzerland
3
Reuters AG, Frankfurt, Germany

Summary. We consider economies with additively separable utility functions and

give conditions for the two-agents case under which the existence of sunspot equilib-
ria is equivalent to the occurrence of the transfer paradox. This equivalence enables
us to show that sunspots cannot matter if the initial economy has a unique spot
market equilibrium and there are only two commodities or if the economy has a
unique equilibrium for all distributions of endowments induced by asset trade. For
more than two agents the equivalence breaks and we give an example for sunspot
equilibria even though the economy has a unique equilibrium for all distributions of
endowments induced by asset trade.

1 Introduction

The purpose of this paper is to clarify the relation between the existence of
sunspot equilibria and the uniqueness of spot market equilibria for economies
with intrinsically complete asset markets. In these economies Pareto-efficient
allocations can be attained as competitive equilibria even without asset trade.
[5] introduce their famous paper “Do sunspots matter?” with the ques-
tion: “What is the best strategy for playing the stock market? Should one
concentrate on fundamentals or should one instead focus on the psychology of
the market ”. The sunspot literature, originating from [4] contributes to this
question by emphasizing that even if all market participants are completely
⋆
We would like to thank Anke Gerber, Piero Gottardi, Andreu Mas-Colell and
Mike Jerison for very fruitful discussions. Last but not least we are grateful to
Klaus Schenk-Hoppé for his support in using MATLAB
and to Andreas Tupak
for his help with the manuscript. All remaining errors are ours.
76 Hens et al.

rational (in the sense that they are maximizing expected utility functions and
have rational expectations), still market outcomes can depend on the psy-
chology of the market. The latter is modelled in the sunspot literature as
a publicly observed exogenous random event, nicknamed “sunspot”. Hence,
even without referring to any kind of bounded rationality the best way of
playing the stock market will not only be based on fundamentals! The sim-
plest case in which this reasoning can be shown to be correct arises when the
economic fundamentals allow for multiple equilibria. The sunspot is then used
as a coordination device. In this case sunspot equilibria are similar to corre-
lated equilibria studied in the game theory literature. The point of this paper
is to analyze whether this is all sunspot equilibria are about, i.e. whether
sunspots matter even when spot market equilibria are unique. Throughout
the paper we assume that agents are strictly risk averse. Hence if utilities
differ across sunspot states then sunspots matter because sunspot equilibria
are Pareto-inefficient (cf. [5]).
We show that for the case of two agents, with utility functions being
concave transformations of additively separable functions, the existence of
sunspot equilibria is equivalent to the occurrence of the transfer paradox. As
an application of this equivalence we demonstrate that the transfer paradox
can occur even if the economic fundamentals can be represented by a repre-
sentative consumer, which in particular implies the uniqueness of equilibria in
the initial economy. We also show that the occurrence of sunspot equilibria is
indeed subject to the same critique as the occurrence of the transfer paradox.
In a model with two commodities, sunspot equilibria can only occur if the
initial equilibrium (the equilibrium without asset trade) is not unique. More-
over, if as in the case of Cobb-Douglas economies, uniqueness of equilibria is
guaranteed for (almost) all distributions of endowments then sunspot equilib-
ria cannot occur at all. Then we show, again using the equivalence between
the two paradoxes, that nevertheless the occurrence of sunspot equilibria does
not need to be based on the exogenous selection among multiple equilibria.
We construct a simple example in which the equilibrium of any sunspot state
is not an equilibrium of any other sunspot state. This example is based on
the idea that financial markets may specify incomplete insurance against the
uncertainty that they induce. That is to say, in this example asset payoffs are
sunspot-dependent and incomplete.
Constructing explicit numerical examples in general equilibrium models is
usually done for a class of economies with a simple enough structure so that
excess demand functions remain manageable. Computable general equilibrium
models (cf. [35]), examples for the occurrence of the transfer paradox ([27, 6,
7, 34, 14], etc), and examples for multiple equilibria ([24, 25, 20, 21]) therefore
use the class of CES-utility functions. The class of utility functions assumed
in this paper (monotonic transformations of additively separable functions)
includes CES-functions. Keeping this assumption we broaden our analysis by
looking into the case of more than two agents (countries). We find that the
occurrence of the transfer paradox is then no longer sufficient for the existence
 Sunspots and Uniqueness 77

of sunspot equilibria. Moreover, we give an example with Cobb-Douglas utility

functions and sunspot-dependent assets in which sunspots matter even though
there are unique equilibria for all distributions of endowments the economy
can arrive at using asset trade.
To get the intuition for our results, note that a sunspot equilibrium con-
sists of a set of spot market equilibria (one for each sunspot state) where
the endowments of each spot market economy are obtained by redistributions
resulting from the equilibrium asset market allocation. Hence an unfavorable
redistribution of endowments as it occurs exogenously in the transfer para-
dox can be derived from asset trade only if the resulting income transfers can
be made compatible with the asset allocation the agents choose. In the asset
market equilibrium the sum across states (adjusted by the common proba-
bility beliefs) of the products of marginal utilities and wealth transfers has
to be equal to zero for all agents. Since marginal utilities are non-negative,
this requires first of all that all agents find some state with positive income
transfer and some with negative income transfer arising from asset trade.
Moreover, supposing that marginal utilities are inversely ordered than utility
levels, this requires that for some state in which an agent receives a negative
income transfer he has a higher utility than for some state with a non-negative
income transfer. This is achieved by the transfer paradox. In a sense the trans-
fer paradox then “crosses the order” of marginal utility levels across sunspot
states. Note that otherwise all states with negative income transfers would be
weighted higher than those with positive income transfer and the probabil-
ity adjusted sum of marginal utilities times income transfers cannot equalize
zero. Hence, provided marginal utilities are inversely related to utility levels
the transfer paradox is necessary for sunspots to matter. Moreover, by ap-
propriate choice of the agents’ risk aversions we can also show the converse,
i.e. that a slightly stronger version of the transfer paradox is indeed sufficient
for sunspots to matter. The role of assuming concave transformations of ad-
ditive separable utilities in this reasoning is to guarantee that indeed higher
marginal utilities are implied by lower utility levels. This relation is neither
true without concavity of the transformation, without additive separability or
with more than two agents. In particular, with more than two agents, the ex-
istence of sunspot equilibria is no longer tied to the occurrence of the transfer
paradox. Of course, a rigorous argument for this intuition will be given once
we have made precise the setup of the model considered.
The question concerning the relation of the existence of sunspot equilibria
and the uniqueness of spot market equilibria has also been addressed by [29]
and [16], for example. One of our results shows that the construction suggested
by [29] can indeed be used to show that, using asset trade, an economy can
arrive at an endowment distribution for which there are multiple equilibria
even though for the initial distribution of endowments equilibria are unique.
Note that in [29] it is not shown that “trading towards multiplicity” can be
done by starting at an initial distribution of endowments with a unique spot
market equilibrium because it is not shown that the equilibrium of the initial
78 Hens et al.

economy is unique. Also the Cobb-Douglas example in which sunspot equilib-

ria exist even though spot market equilibria are unique (for all distributions
one can arrive at using asset trade) clarifies a confusion that has recently come
up in [19] and [3]. [19] has claimed that such an example is possible, but [3]
have shown that for the particular parameter values [19] has chosen this is
not true! Note that the sunspot equilibrium [3] suggest to consider instead of
[19]’s example, sunspots do not matter because agents are not risk averse. Our
results in this paper show that the problem in [19] has nothing to do with the
parameter values chosen. It arises because [19] considered an economy with
only two agents. As we show here, with three agents the construction in [19]
is actually possible.
Finally, note that the relation of sunspot equilibria and uniqueness of
spot market equilibria has also been studied in economies in which a Pareto-
efficient allocation cannot be obtained without asset trade. These types of
economies originate in the famous “leading example” of [4]. With first period
consumption, asset trade may occur because of intertemporal substitution
and as a by-product this may introduce extrinsic uncertainty as [4] has first
pointed out.
[16] study an intrinsically incomplete economy a la [4], however with real
instead of nominal assets. In our paper assets also have real payoffs but the
result of [16], who show the existence of sunspot equilibria for a strong unique-
ness assumptions on the underlying economy, does not apply to our setting
because it relies on first period consumption. Also for the same reason the tech-
nique developed by [16] is not applicable here because they control the agents‘
utility gradients both by perturbing the utility functions and by changing the
level of first-period consumption.
In the next section we outline the model. Then we give the definitions
of the transfer paradox and of sunspot equilibria. Thereafter we prove our
result establishing the equivalence of the transfer paradox and the occurrence
of sunspot equilibria. Section 4 applies this result to derive some new insights
both for sunspot equilibria and also for the transfer paradox. Also in Section 4
we also show some implications for the existence of sunspot equilibria when
spot market equilibria are unique. Section 5 concludes.

2 Model
We first outline the sunspot model. The transfer paradox will then be embed-
ded into this model by a new interpretation of the sunspot states.
There are two periods. In the second period, one of s = 1, . . . , S states
of the world occurs. In the first period assets are traded. Consumption only
takes place in the second period. This assumption is important here because
otherwise the sunspot model cannot be linked to the atemporal transfer para-
dox model. There are i = 1, . . . , I agents and l = 1, . . . , L commodities in
 Sunspots and Uniqueness 79

each state. States are called sunspot states because the agents’ characteris-
tics within the states, i.e. the agents’ endowments ω i ∈ X i and their utility
functions ui : X i → IR, do not depend on them. X i is a closed convex subset
of IRL
+ which denotes agent i’s consumption set. In the sunspot literature the
agents’ characteristics [(ui , ω i )i=1,...,I ] are called the economic fundamentals.
Throughout this paper we make the
Assumption 1 (Additive Separability)All agents’ von Neumann-Mor-
genstern utility functions ui are monotonic
transformations of additively sep-
L
arable functions, i.e. ui (xi1 , . . . , xiL ) = f i ( l=1 gli (xil )) for all xi ∈ X i , where
the functions f i and gli , l = 1, . . . , L are assumed to be twice continuously dif-
ferentiable, strictly increasing and the gli are concave. Moreover, we assume
that for every agent i at least L − 1 of the functions gli are strictly concave
and that for all commodities l there is some i for which gli is strictly concave.
Finally, ui , is assumed to be concave.
Note that the assumptions on the functions gli guarantee strict quasi-concavity
of the function ui . (Strict-) Concavity of ui however also depends on the
monotonic transformation f i . The class of utility functions covered by as-
sumption 1 includes all utility functions that are commonly used in ap-
plied general equilibrium theory. In particular, the case of CES utilities,

L  i 1−ρi i ρi 1/ρi
ui (xi ) = l=1 (αl ) (xl ) , defined for all i = 1, . . . , I on X i =
{x ∈ IRL i i i i L
++ |u (x) ≥ u (ω )}, for some ω ∈ IR++ and some 0 < αl < 1,
i
i
l = 1, . . . , L and ρ < 1, is covered by this assumption. Also across states
agents are assumed to have additive separable utility functions:
Assumption 2 (Expected Utility)For all agents, i = 1, . . . , I, the ex-
pected utility functions, defining preferences over state contingent consumption
xi (s) ∈ IRL , s = 1, . . . , S are given by
S

Eui (xi (1), . . . , xi (S)) = π(s)hi (ui (xi (s))) ∀xi ∈ (X i )S ,
s=1

where the hi are twice continuously differentiable, strictly increasing and con-
cave functions. Moreover, Eui is assumed to be strictly concave.
An important subclass
of economies arises if the von Neumann-Morgenstern
utility functions hi (f i ( l gli (xil ))) are concave transformations of additively
separable functions, i.e. if assumption 1 holds and the composite functions
hi ◦ f i are concave.
Assumption 3 (E.U. with Concave Additive Separability)Assump-
tion 1 and 2 hold and the composite functions hi ◦ f i are concave.
To include CES-functions under assumption 3, the convex transformation
i
f i (y) = y 1/ρ has to be transformed by a sufficiently concave function hi
so that hi ◦ f i is concave. However, to satisfy assumptions 1 and 2 one could
80 Hens et al.

choose a strictly concave function that makes the expected utility concave
without requiring concavity of hi ◦ f i . Unfortunately, this subtle difference
will be important for our paper, as we will give examples with CES-utilities,
satisfying assumptions 1 and 2 but not 3, in which sunspot equilibria occur.
In the first period agents can trade j = 1, . . . , J, real assets with payoffs
Aj (s) ∈ IRL if state s occurs. We denote asset prices by q ∈ IRJ . Agent i’s
portfolio of assets is denoted by θi ∈ IRJ . Note that all asset payoffs are real,
i.e. in terms of commodities. Moreover, we allow for sunspot depended asset
payoffs. There is an impressive strand of the sunspot literature originating
from [4] in which asset payoffs are nominal. In this literature asset payoffs
measured in real terms differ across sunspot states if and only if sunspots
matter. The same is effectively also the case in our setting: Supposing spot
market equilibria are unique the equilibrium transfers across states measured
in real terms depend on sunspots if and only if sunspots matter.
All equilibria we consider in this setting are special cases of competitive
equilibria, which are defined in
Definition 1 (Competitive Equilibrium). A competitive equilibrium is an
⋆ ⋆ ⋆ ⋆
allocation (xi , θi ), i = 1, .., I, and a price system (p, q ) such that
1. For all agents i = 1, .., I:
⋆ ⋆ ⋆
(xi , θ i ) ∈ argmaxxi ∈X i ,θi ∈IRJ Ss=1 π(s)hi (ui (xi (s))) s.t. q · θi ≤ 0 and


⋆ ⋆ ⋆
p(s) · xi (s) ≤ p(s) · ω i + p(s) · A(s)θi for all s = 1, . . . , S.

I ⋆ i
I
2. i=1 x (s) = i=1 ω i for all s = 1, . . . , S.
⋆
3. Ii=1 θ i = 0.

Remark 1. To simplify the exposition when analyzing competitive equilibrium

allocations we restrict attention to interior allocations, i.e. to allocations xi
in the interior of X i , i = 1, . . . , I. A sufficient assumption guaranteeing the
interiority of allocations is to impose that the functions f i and gli satisfy the
Inada condition according to which the marginal utility tends to infinity at
the boundary of the consumption set X i ⊂ IRL +.

Note that a competitive equilibrium consists of S spot market equilibria (one

for each spot market economy with endowments ω̂ i (s) = ω i +A(s)θi ) together
with an asset market equilibrium by which the ex-post endowments of the spot
markets are generated. It will be convenient to introduce the spot-market
economy of the economic fundamentals as a point of reference. To abbreviate
notations we therefore let this economy be the spot market economy in the
spot s = 0. Finally, note that when showing the existence of sunspot equilibria
we allow to choose the characteristics not fixed by the economic fundamentals,
the sunspot extension, appropriately. The sunspot extension consist of the
probabilities of the sunspot states π, the asset structure A and the risk aversion
functions hi . The sunspot equilibria are robust with respect to perturbations
 Sunspots and Uniqueness 81

of these characteristics however sunspot equilibria will not exist for all possible
choices of the sunspot extension.
This finishes the description of the model.

3 Sunspot Equilibria and the Transfer Paradox

In the sunspot literature agents transfer commodity bundles across sunspot
states by trading assets. In the international trade literature one thinks of
transfers of commodities arising from donations. Each sunspot state will later
on be associated with different such donations. The transfer paradox is said
to occur if some agent donates some of his resources to some other agent and
the recipients utility decreases. In the case of two agents by Pareto-efficiency
within spot markets the donor’s utility increases. In this statement the utility
comparison is done across the competitive equilibria of the economy before and
after the donation. In the standard case of the transfer paradox, the transfer
was considered to be a transfer of a non-negative amount of commodities
([27]). In order to make the equivalence to the sunspot model more obvious
we consider a slightly more general definition of the transfer paradox which
only requires that the donated commodities have non-negative value in the
competitive equilibrium after the transfer. As [14] have already shown this
generalization is innocuous.
In the following definition we consider alternative possible transfers ∆ω(z)
that we index by some scenarios z. When relating the transfer paradox to
sunspot equilibria these scenarios will be associated with different states of
the world, s = 1, . . . , S. Taking care of potentially multiple equilibria the
transfer paradox is then defined as in1
Definition 2 (Transfer Paradox). Given an economy with fundamentals
[(ui , ω i )i=1,...,I ] the transfer paradox occurs if and only if there exists some

I
transfer of endowments ∆ω(z) ∈ IRLI , with i=1 ∆ω i (z) = 0 such that for
⋆ ⋆
the economy [(ui , ω i + ∆ω i (z))i=1,...,I ] there exists an equilibrium (x(z), p(z))
⋆ ⋆ ⋆
with p(z) · ∆ω 1 (z) ≥ 0 so that u1 (x1 (z)) < u1 (x1 (0)) for some equilibrium
⋆ ⋆
(x(0), p(0)) of the economic fundamentals, [(ui , ω i )i=1,...,I ], in the reference
scenario without transfers, s = 0.
Note that under certain conditions and if the economic fundamentals have at
least two equilibria then even without any transfers the transfer paradox oc-
curs. Our definition covers this case because then ∆ω = 0 is already sufficient
⋆ ⋆
to obtain u1 (x1 (z)) < u1 (x1 (0)) for the two equilibria s = 0, z. Of course if
the resulting equilibria are regular then in this case one can also find some
1
To save on notation we define the transfer paradox with respect to the value of
the transfers and changes in utility of agent 1.
82 Hens et al.

transfers of endowments that have positive value and yet the recipients utility
decreases. Making the transfer paradox a bit more paradoxical.
We will show that the occurrence of the transfer paradox is a necessary
condition for sunspots to matter. To show a converse of this claim we consider
the following slightly stronger notion of the transfer paradox.
Definition 3 (Strong Transfer Paradox). Given an economy with funda-
mentals [(ui , ω i )i=1,...,I ] the strong transfer paradox occurs if and only if there

I
exist some transfers of endowments, ∆ω(z) ∈ IRLI , with i
i=1 ∆ω (z) = 0
I
and ∆ω(s̃) ∈ IRLI , with i


i=1 ∆ω (s̃) = 0 such that for the economies
i i i
[(u , ω + ∆ω (s))i=1,...,I ], s = z, s̃
⋆ ⋆ ⋆ ⋆ ⋆
1. there are some equilibria (x(z), p(z)), (x(s̃), p(s̃)) with p(z)·∆ω 1 (z) ≥ 0
⋆
and p(s̃) · ∆ω 1 (s̃) ≤ 0 and
⋆ ⋆ ⋆
2. it holds that u1 (x1 (z)) < u1 (x1 (s̃)) < u1 (x1 (0)) for some equilibrium
⋆ ⋆
(x(0), p(0)) of the economic fundamentals [(ui , ω i )i=1,...,I ], in the refer-
ence scenario without transfers s = 0.
Note, that if the economic fundamentals have at least three equilibria then by
the same reason as given for the transfer paradox the strong transfer paradox
occurs. Hence the existence of at least (three) two equilibria is sufficient for
the (strong) transfer paradox. Of course, in regular economies we know that
if there are at least two equilibria then there also are at least three equilib-
ria (cf. [9]). This observation indicates that in regular economies the transfer
paradox and the strong transfer paradox are actually equivalent. Indeed this
it true as the next proposition shows. Recall that in regular economies equi-
libria are well determined, i.e. in a neighborhood of regular equilibria (being
defined by full rank of the Jacobian of market excess demand) there exists
a smooth mapping from the exogenous parameters of the economy to the
endogenous equilibrium values (cf. [8]). In the following argument regular-
ity needs only be required for the spot market equilibria of the economic
fundamentals. This property holds generically in the set of agents’ initial en-
dowments IRLI ++ (cf. [8]).

Proposition 1. Suppose all spot market equilibria of the economic funda-

mentals [(ui , ω i )i=1,...,I ] are regular. Then the transfer paradox and the strong
transfer paradox are equivalent.
Proof
The strong transfer paradox implies the transfer paradox. To establish the
converse suppose that the transfer paradox

Iholds. I.e. there exists some trans-
fer of endowments ∆ω(z) ∈ IRLI , with i=1 ∆ω i
(z) = 0, such that for the
⋆ ⋆
i i i
economy [(u , ω + ∆ω (z))i=1,...,i ] there exists an equilibrium (x(z), p(z))
⋆ ⋆ ⋆
with p(z) · ∆ω 1 (z) ≥ 0 so that u1 (x1 (z)) < u1 (x1 (0)) for some equilibrium
⋆ ⋆
(xi (0), p(0)) of the economic fundamentals, s = 0.
 Sunspots and Uniqueness 83

We need to show that there also exists some ∆ω(s̃) ∈ IRLI , with
⋆
1 ⋆1 1 ⋆1

I i 1
p
i=1 ∆ω (s̃) = 0 such that (s̃) · ∆ω (s̃) ≤ 0 and u (x (z)) < u (x (s̃)) <
⋆
u1 (x1 (0)).
This is of course the intuitive case in which a negatively valued transfer
leads to a loss in utility. However, we need to ensure that this is the outcome in
the spot market equilibrium after the transfer and that the utility loss is not
too severe as compared to the loss in the transfer paradox case. This is ensured
by the regularity of the equilibrium of the economic fundamentals from which
we construct the transfer appropriately: Consider the utility gradient of agent
⋆
1, ∇u1 (x1 (0)) at the equilibrium of the economic fundamentals. Choose the
⋆
transfers (∆ω 1 (s̃)), such that ∇u1 (x1 (0))(∆ω 1 (s̃)) < 0. By the first order
condition of utility maximization in the reference situation s = 0 we get
⋆
that this wealth transfer evaluated at the pre-transfer prices is negative, p(0)·
⋆
(∆ω(s̃)) < 0. Since ∇u1 (x1 (0))(∆ω(s̃)) < 0, by proposition 31.2 (ii) in [28]
⋆ ⋆
we can find some 1 ≥ α > 0 such that u1 (x1 (0) + α(∆ω(s̃))) < u1 (x1 (0)).
Moreover, by the regularity of the economy, α > 0 can be chosen small enough
so that also the utility at the induced equilibrium is smaller than in the
⋆ ⋆
reference situation without transfers, u1 (x1 (s̃)) < u1 (x1 (0)). This is because in
⋆
regular economies the induced change in the equilibrium allocation x1 (s̃) can
⋆ ⋆
be held small so that |u1 (x1 (s̃))−u1 (x1 (0)+α∆ω(s̃))| is also small. Moreover,
by the same continuity argument this can be done such that ∆ω 1 (s̃) evaluated
⋆
at prices after the transfer is non-positive, i.e. p(s̃)·∆ω 1 (s̃) ≤ 0. Finally, all
this can be done without decreasing the utility level too much, so that for
⋆ ⋆ ⋆
agent 1 we get the inequality u1 (x1 (z)) < u1 (x1 (s̃)) < u1 (x1 (0)).

The strong transfer paradox ensures the order crossing property mentioned
in the introduction. To see this note that it is always possible to find transfers
of resources, say ∆ω(ŝ), such that the transfer to agent 1 has negative value in
⋆
the resulting equilibrium, i.e. p(ŝ)·∆ω 1 (ŝ) ≤ 0, and agent 1 gets a level of utility
that is smaller than any of the utility levels considered in the definition of the
⋆ ⋆ ⋆ ⋆
strong transfer paradox, i.e. u1 (x1 (ŝ)) < u1 (x1 (z)) < u1 (x1 (s̃)) < u1 (x1 (0)).5
By this observation we get three transfers, two with negative value and one
with positive value so that the utility decreases for all transfers. As we will see,
by assumption 1, in the case of two agents, we then get that the order of the
marginal utilities does not coincide with the order or the reverse order of the
5
Note that these losses in utility as compared to the equilibrium of the economic
fundamentals do not conflict with the fact that trade is voluntary because it may
well be that the utility of agent 1 derived from his initial endowments is even smaller
than the expected utility obtained in the spot market equilibria. Also the agent is
assumed to be a price taker, i.e. he cannot enforce the equilibrium of the economic
fundamentals.
84 Hens et al.

transfer values, i.e. the order crossing property is also obtained for marginal
utilities.
Note that the transfer paradox concerns the ordering of income trans-
fers and utility levels. In the first order condition for asset demand however
marginal utilities and not utility levels themselves play a role. Hence we need
to know how the levels of marginal utility are related to the utility levels.
Keeping prices fixed across different states, by concavity of the utility func-
tion, marginal utilities are inversely related to utility levels. This feature oc-
curs for example if the agents have identical and homothetic preferences. In
this case however neither sunspot equilibria matter nor the transfer paradox
occurs. In general, changes in relative prices induced by redistributions of in-
come are decisive to determine both the level of utility and of marginal utility.
It is these changes from which the transfer paradox and also the existence of
sunspot equilibria are derived. Nevertheless, with two agents, whose utilities
are concave transformations of additively separable functions, we show that
marginal utilities are negatively associated to the level of utilities. To make
these ideas precise, we first define the agents’ indirect utility function and
their marginal utility of income within each state without considering the
monotonic transformations hi :
Let
 L 

v i (s) = v i (p(s), bi (s)) = max
i i
fi gli (xil (s)) s.t. p(s)·xi (s) ≤ bi (s)
x ∈X
l=1

be the indirect utility of agents i in state s. Since the functions gli , l = 1, . . . , L

are concave and since at least L − 1 of them are strictly concave there is a
unique point xi at which the utility attains its maximum, given that for all
commodities the prices pl (s), l = 1, . . . , L and the income bi (s) are positive.
In our model the income bi (s) will be given by p(s) · (ω i + A(s) · θi ). I.e., the
values of the transfers are given by ri (s) = p(s)·A(s) · θi . In the analysis of
the sunspot model the agents’ marginal utility of income will be important
λi (s) = ∂v hi (v i (s)) ∂b v i (p(s), bi (s)) .
Hence the marginal utilities that determine the asset allocation are given by
the marginal utilities within each state, ∂b v i (p(s), bi (s)), multiplied by the first
derivative of the agents’ concave transformations hi determining the agents’
risk aversion.
The association between levels of utilities and of marginal utilities is an
important link between the transfer paradox and sunspot equilibria which we
therefore need to define properly:
Definition 4 (Inverse Association of Utilities and Marginal Utili-
ties). We say that for some agent i the levels of marginal utility are inversely
associated to the levels of utility if
v i (1) ≥ v i (2) ≥ . . . ≥ v i (S)
 Sunspots and Uniqueness 85

implies
λi (1) ≤ λi (2) ≤ . . . ≤ λi (S) .
Moreover, if v i (s̃) < v i (z) for some s̃, z ∈ {1, . . . , S} then the corresponding
inequality in the marginal utilities of income should also be strict.
This definition puts us in the position to state the equivalence of the
occurrence of the transfer paradox and the existence of sunspot equilibria.
Theorem 1 (Equivalence Sunspot Equilibria and Transfer Paradox).
Suppose assumption 2 holds and all agents’ level of marginal utility are in-
versely associated to their level of utility. Then
1. the transfer paradox is a necessary condition for sunspots to matter and
2. if there are only two agents then the strong transfer paradox is a sufficient
condition for sunspots to matter.
Proof
1. To link the transfer paradox to the sunspot economy consider
⋆ ⋆
ri (s) := p(s)·A(s) θi ,

i.e. the transfer of income to agent i as generated by asset trade in some

competitive equilibrium.
A necessary condition for optimal portfolio choice is
S

λi (s) π(s) ri (s) = 0 , i = 1, . . . , I ,
s=1

which we call the first-order conditions for asset demand.6

Now suppose that the transfer paradox does not hold. Then negative
transfers ri (s) < 0 are associated with lower utility levels than positive
transfers ri (s) > 0 are. If moreover marginal utilities are inversely associ-
ated to utilities then negative transfers are associated with higher marginal
utility levels than positive transfers are. Hence the first order condition for
asset demand requires that the probability weighted sum of the absolute
values of negative transfers is smaller than the probability weigthed sum
of positive transfers:
 
π(s) ri (s) > − π(s) ri (s) , for all i = 1, .., I . (1)
s:r i (s)>0 s:r i (s)<0

This however conflicts with asset market clearing, which implies that in-
come transfers must be balanced:
6
λi (s) π(s) p(s) A(s) = γ i q together with q · θi =


This condition follows from s
0.
86 Hens et al.
I

ri (s) = 0 , for all s = 1, . . . , S . (2)
i=1

To see this, multiply equation (2) by π(s) and sum those equations over
states to obtain:
S  I
π(s) ri (s) = 0. (3)
s=1 i=1

Rearranging terms this implies

⎛ ⎞
 I  
⎝ π(s) ri (s) − π(s) ri (s)⎠ = 0 , (4)
i=1 s:r i (s)>0 s:r i (s)<0

which contradicts equation (1).

2. Suppose the strong transfer paradox occurs, then there exist transfers
indexed by s̃, z such that

r1 (z) ≥ 0 , r1 (s̃) ≤ 0 and for some equilibria v 1 (z) < v 1 (s̃) < v 1 (0)

where v 1 (0) refers to agent 1’s utility in an equilibrium of the spot economy
s = 0. Given the utility functions u1 , u2 and given the total endowments
ω 1 + ω 2 consider the set of Pareto-efficient allocations as being parame-
terized by the income transfers r.
Now we have to distinguish three cases:
Case 1: If r1 (z) > 0
then we know that b1 (z) > 0 and therefore there exists r1 (ŝ) < 0 suffi-
ciently small such that for the induced b1 (ŝ) = (b1 (z) + r1 (ŝ)) ≥ 0 we
get v 1 (ŝ) < v 1 (z) for some equilibrium in ŝ. By thisobservation and the
strong transfer paradox we have the order crossing property:

r1 (ŝ) ≤ 0 , r1 (s̃) ≤ 0 , r1 (z) > 0

while
v 1 (ŝ) < v 1 (z) < v 1 (s̃)
so that by the negative association of marginal utilities to the level of
utilities
λ1 (ŝ) > λ1 (z) > λ1 (s̃) .
To construct the sunspot equilibrium consider an economy with the three
states s = ŝ, s̃, z. In this case the first-order conditions for asset demand
become:

λi (ŝ) π(ŝ) |ri (ŝ)| + λi (s̃) π(s̃) |ri (s̃)| = λi (z) π(z) |ri (z)| , i = 1, 2 .

Now choose π(z) < 1 sufficiently large (and accordingly π(ŝ) > 0 and
π(s̃) > 0 sufficiently small) such that
 Sunspots and Uniqueness 87

π(ŝ) |ri (ŝ)| + π(s̃) |ri (s̃)| < π(z) |ri (z)| .

Note that ∂hi is any continuous, positive and decreasing function. Recall
that, λ1 (s̃) < λ1 (z) and that v 1 (ŝ) is the smallest utility level in the
three states. Hence we can choose h1 such that λ1 (ŝ) is sufficiently large
to solve the first order condition for i = 1. Analogously it follows that
λ2 (ŝ) < λ2 (z) and we can choose h2 such that λ2 (s̃) is sufficiently large
to solve the first order condition for i = 2.
To complete the proof we follow the analogous steps as in [29]. Choose
A ∈ IR3L×2 such that

r1 (s) = p(s)·(A1 (s) − A2 (s)) for s = s̃, ŝ, z . (5)

Finally, note that

 
λ1 (s) π(s) p(s)·A1 (s) = λ1 (s) π(s) p(s)·A2 (s)
s s

so that we can choose q1 = q2 . Accordingly we choose θ1 = (1, −1),

θ2 = (−1, 1) so that q · θi = 0, i = 1, 2 and θ1 + θ2 = 0. Since we have
chosen an economy with two assets, the
first-order conditions for asset
trade are equivalent to the conditions s λi (s) π(s) p(s) A(s) = γ i q.
Case 2: If r1 (z) = 0 and r1 (s̃) = 0
then by the strong transfer paradox, even without trading any asset, there
is a competitive equilibrium in which sunspots matter.
Case 3: Finally, the case r1 (z) = 0 and r1 (s̃) < 0
is already covered by the reasoning of the first case if one changes the
point of view from agent 1 to agent 2.

Remark 2. Note that in the theorem above part 1 has been shown for the
most general statement without evoking any particular assumption on the
asset structure A ∈ IRSL×J . part 2 however is a stronger claim the more the
set of asset structures can be restricted. The choice of the asset structure
matters in equation (5) of the proof. One way of restricting A is to only
consider numeraire assets so that all assets pay off in the same commodity.
Allowing for sunspot dependent assets this is a possible choice in the solution
of equation (5). If assets are not allowed to depend on the sunspot states then
one can still find an asset structure solving equation (5), provided the three
price vectors p(s), s = s̃, ŝ, z are linearly independent. The latter then requires
to have at least 3 commodities.

To complete this section we first show that under assumption 3 in the case
of two agents the order of the marginal utilities of income is inverse to the
order of their (indirect) utilities. Hence not only in the trivial case of identical
and homothetic preferences we get this property but we also get it for all
88 Hens et al.

numerical examples with two agents that have so far been considered in the
sunspot and in the transfer paradox literature.
Lemma 1. Suppose there are only two agents. Without loss of generality as-
sume that in a competitive equilibrium

v 1 (1) ≤ v 1 (2) ≤ . . . ≤ v 1 (S).

Then under assumption 3 it follows that

λ1 (1) ≥ λ1 (2) ≥ . . . ≥ λ1 (S)

and that
λ2 (1) ≤ λ2 (2) ≤ . . . ≤ λ2 (S) .
Moreover, if v 1 (s̃) < v 1 (z) for some s̃, z ∈ {1, . . . , S} then the corresponding
inequality in the marginal utilities of income is also strict.
Proof
Assume that

v 1 (s̃) ≤ v 1 (z) ( resp. that v 1 (s̃) < v 1 (z) ) for some s̃, z ∈ {1, . . . , S} .

Then, by monotonicity of the utility function, for some commodity, say k ∈

{l, . . . , L} we must have that

x1k (s̃) ≤ x1k (z) ( resp. that x1k (s̃) < x1k (z) ) .

Moreover, Pareto-efficiency within spot markets implies that for all states
s = 1, . . . , S the marginal rates of substitution are equal across agents, i.e.
1
∂gm (x1m (s)) 2
∂gm (x2m (s))
=
∂gl1 (x1l (s)) ∂gl2 (x2l (s))

for any pair of commodities (l, m). Note that x2m (s) = ωm 1 2
+ ωm − x1m (s),
i
s = 1, . . . , S. Hence if the functions gl are concave and if for some agent the
function gli is strictly concave then it follows that

x1l (s̃) ≤ x1l (z) ( resp. that x1l (s̃) < x1l (z) ) for all l = 1, . . . , L .

Without loss of generality assume that l = n is the numeraire in all states

s = 1, . . . , S, where n is chosen such that gn1 is strictly concave. Hence we have
shown that

v 1 (1) ≤ v 1 (2) ≤ . . . ≤ v 1 (S) ( with v 1 (s̃) < v 1 (z) for some s̃, z )

implies for the numeraire that

x1n (1) ≤ x1n (2) ≤ . . . ≤ x1n (S) ( with x1n (s̃) < x1n (z) for some s̃, z ) .
 Sunspots and Uniqueness 89

From the first order condition to the maximization problem


max gl1 (x1l (s)) s.t. p(s)·x1 (s) ≤ b1 (s)
xi ∈X i
l

we get that λ1 (p(s), b1 (s)) = ∂y (h1◦f 1 )(y) ∂gn1 (x1n (s)) for all s =

1, . . . , S. Since
h1 ◦f 1 , gn1 are (strictly) concave and since x1n (s) and y 1 (s) = l gl1 (x1l (s)) are
increasing (resp. strictly increasing) in s we get that
λ1 (1) ≥ λ1 (2) ≥ . . . ≥ λ1 (S) ( resp. that λ1 (z) > λ1 (s̃) ) .
The claim for i = 2 follows analogously from the inverse inequalities
x2l (1) ≥ x2l (2) ≥ . . . ≥ x2l (S) for l = 1, . . . , L ,
and from
v 2 (1) ≥ v 2 (2) ≥ . . . ≥ v 2 (S) ,
the latter inequalities being implied by Pareto-efficiency within spot markets.

Before passing to the next section we want to point out that the assump-
tion of additive separability is indeed tight. The inverse association between
the levels of marginal utilities and that of utilities, as shown in lemma 1 does
not necessarily hold without additive separability. As the following example
shows without additive separability one can find that lower utilities are as-
sociated with lower marginal utilities. The endowments in this example are
supposed to be the ex-post endowments. Hence they are allowed to depend
on the sunspot states since the asset payoffs may depend on them.
Remark 3. Consider a two-agent economy with two commodities. The utility
functions are7 :

u1 (x1 ) = x11 x12 + x12 and u2 (x2 ) = x21 x22 + x21 .

Note that neither of the two utility functions is additively separable but both
are strictly monotonically increasing and strictly concave on IR2++ and both
satisfy the Inada-conditions. Moreover, note that both utility functions are
homogenous of degree one implying that both goods are normal. In situation
s = 1 the ex-post endowments are
ω̂11 (1) = 1 , ω̂21 (1) = 5 and ω̂12 (1) = 4 , ω̂22 (1) = 2 .
There is a unique equilibrium8 with prices p(1) = (1, 0.7125). The equilibrium
budgets are:
7
The transformations hi ◦f i are assumed to be the identity, which is a strictly
increasing concave function.
8
All values have been rounded to 4 decimal digits. The exact values can be
found at the page http://www.iew.unizh.ch/home/hens/sunspot. Uniqueness can
be seen from the graph of the excess demand also shown on the webpage.
90 Hens et al.

b1 (1) = 4.5623 and b2 (1) = 5.4249 .

The resulting allocation is:

x11 (1) = 0.5380 , x12 (1) = 5.6485 and x21 (1) = 4.4620 , x22 (1) = 1.3515 .

The utility levels are:

u1 (1) = 7.3917 and u2 (1) = 6.9177 .

i


Marginal utilities within state 1, ∂b v i (1) = vbi (1)
(1)
, are:
i=1,2

∂b v 1 (1) = 1.6202 and ∂b v 2 (1) = 1.2752 .

Now consider a second situation s = 2 with the same total endowments but
with a distribution of ex-post endowments as:

ω̂11 (2) = 5 , ω̂21 (2) = 5 and ω̂12 (2) = 0 , ω̂22 (2) = 2 .

Again, there is a unique equilibrium, now with prices p(2) = (1 , 1.5113). The
equilibrium budgets are:

b1 (2) = 12.5563 and b2 (2) = 3.0225 .

The resulting allocation is:

x11 (2) = 2.3164 , x12 (2) = 6.7758 and x21 (2) = 2.6836 , x22 (2) = 0.2242 .

The utility levels are:

u1 (2) = 10.7375 and u2 (2) = 3.4594 .

Marginal utilities within state 2 are:

∂b v 1 (2) = 0.8552 and ∂b v 2 (2) = 1.1445 .

Note that the second agent’s utility and his marginal utility has decreased is
passing from situation 1 to situation 2. Finally, note that we could also have
chosen two strictly concave functions hi such that the same ordering still holds
for the marginal utilities λi (s) = ∂v hi (v i ) ∂b v i (s).
The next example shows that for more than two agents the strong transfer
paradox is no longer sufficient for the existence of sunspot equilibria. The
simple reason is that for agent 1 the strong transfer paradox may occur while
the two other agents will not find income transfers of opposite sign.
Remark 4. The example to construct the strong transfer paradox is the famous
three country example from [6].

There are three agents and two goods. The utility functions are:
 Sunspots and Uniqueness 91

u1 (x11 , x12 ) = min(x11 , 4.0 x12 ) ,

u2 (x21 , x22 ) = min(x21 , x22 ) ,
u3 (x31 , x32 ) = min(2.8 x31 , x32 ) .

Note that this example uses Leontief preferences. Hence strictly spoken our
assumption 1 is not satisfied. However, these preferences can be attained as a
limit case of CES-utility functions. That is to say, perturbing the preferences
slightly within the CES-class will establish an example satisfying assump-
tion 1 and if we like to transform it by a sufficiently concave function hi
also assumption 3 can be satisfied, as we mentioned above. Moreover, note
that both utility functions, ui are homogenous of degree one implying that
both goods are normal. Consider the situations s = 0, s̃, z, ŝ as required by the
strong transfer paradox.
Let the matrix of endowments (for both goods per agent and state), with
rows corresponding to states s = 0, ŝ, z, s̃ and with columns corresponding to
agents, be: ⎛ ⎞
(1 , 1.00) (2 , 1.00) (1 , 3.00)
⎜ (1 , 0.10) (2 , 2.40) (1 , 2.50) ⎟
ω=⎜ ⎝ (1 , 1.10) (2 , 1.00) (1 , 2.90) ⎠ ,
⎟

(1 , 0.80) (2 , 1.25) (1 , 2.95)

In all situations there is a unique equilibrium9 . The equilibrium price vectors
are:

p(0) = (5.9084 , 1) ,
p(s̃) = (4.4892 , 1) ,
p(z) = (9.6382 , 1) ,
p(ŝ) = (0.9438 , 1) .

Evaluated at these equilibrium prices the transfers as compared to situation

s = 0 are:
r1 (s̃) = −0.2 , r2 (s̃) = 0.25 and r3 (s̃) = −0.5 ,
r1 (z) = 0.1 , r2 (z) = 0.00 and r3 (z) = −0.1 ,
r1 (ŝ) = −0.9 , r2 (ŝ) = 1.40 and r3 (ŝ) = −0.5 .

The equilibrium budgets are:

b1 (0) = 6.9084 , b2 (0) = 12.8169 and b3 (0) = 8.9084 ,

b1 (s̃) = 5.2892 , b2 (s̃) = 10.2284 and b3 (s̃) = 7.4392 ,
b1 (z) = 10.7382 , b2 (z) = 20.2765 and b3 (z) = 12.5382 ,
b1 (ŝ) = 1.0438 , b2 (ŝ) = 4.2876 and b3 (ŝ) = 3.4438 .
9
All values have been rounded to 4 decimal digits. The exact values can be
found at the page http://www.iew.unizh.ch/home/hens/sunspot. Uniqueness can
be seen from the graph of the excess demand function which is displayed at the
webpage together with the computations.
92 Hens et al.

The resulting allocations are:

⎛ ⎞
(1.1218 , 0.2804) (1.8552 , 1.8552) (1.0230 , 2.8643)
⎜ (1.1161 , 0.2790) (1.8634 , 1.8634) (1.0206 , 2.8576) ⎟
x=⎜ ⎝ (1.0860 , 0.2715) (1.9060 ,
⎟,
1.9060) (1.0080 , 2.8225) ⎠
(0.8744 , 0.2186) (2.2058 , 2.2058) (0.9120 , 2.5756)

The corresponding utility levels are:

u1 (0) = 1.1218 , u2 (0) = 1.8552 and u3 (0) = 2.8643 ,

u1 (s̃)= 1.1161 , u2 (s̃) = 1.8634 and u3 (s̃) = 2.8576 ,
u1 (z) = 1.0860 , u2 (z) = 1.9060 and u3 (z) = 2.8225 ,
u1 (ŝ)= 0.8744 , u2 (ŝ) = 2.2058 and u3 (ŝ) = 2.5756 .
i


Marginal utilities ∂b v i (s) = vbi (s)
(s)
are:
i=1,2,3.

∂b v 1 (0) = 0.1624 , ∂b v 2 (0) = 0.1448 and ∂b v 3 (0) = 0.3215 ,

∂b v 1 (s̃) = 0.2110 , ∂b v 2 (s̃) = 0.1822 and ∂b v 3 (s̃) = 0.3841 ,
∂b v 1 (z) = 0.1011 , ∂b v 2 (z) = 0.0940 and ∂b v 3 (z) = 0.2251 ,
∂b v 1 (ŝ) = 0.8377 , ∂b v 2 (ŝ) = 0.5145 and ∂b v 3 (ŝ) = 0.7479 .

Note that the second agent’s transfers are never negative while that of the
third agent are never positive. Hence these transfers cannot be sustained by
asset trade. Finally, note that as compared to situation s = 0 the third agent’s
utility and his marginal utility has decreased in passing to situation s = z.

4 Sunspot Equilibria and Uniqueness of Spot Market

Equilibria
Having established the link between the transfer paradox and the existence of
sunspot equilibria we now derive some new results on the existence of sunspot
equilibria when spot market equilibria are unique on the one hand and also
on the possibility of the transfer paradox on the other hand. Applying part 1
of our theorem and [36] we can rule out sunspot equilibria if marginal util-
ities are inversely related to utilities and spot market equilibria are unique
at all non-negative distributions of endowments. Once again, applying part 1
of our theorem and [2] we can rule out sunspot equilibria in the case of two
commodities and two agents if the economic fundamentals do have a unique
equilibrium. On the other hand, applying the construction of [29], which is
done for the case of two agents, part 2 of our result shows that the trans-
fer paradox can occur even if for the economic fundamentals there exists a
representative consumer. We continue, using once more part 2, to show that
sunspot equilibria need not be derived from multiple equilibria of the spot
market economy that is obtained by asset trade leading to the same endow-
ment distribution in all states. While [29]’s construction uses that there are
 Sunspots and Uniqueness 93

multiple equilibria for the induced distribution of endowments, this example

however exploits the fact that there are multiple equilibria for the initial dis-
tribution of endowments. Therefore, we give one more example in which spot
market equilibria are unique for all distributions of endowments that can be
arrived at by asset trade and still sunspots matter. This final example then
settles the question whether sunspot equilibria need to be based on some sort
of multiplicity of spot market equilibria.
The following terminology is quite useful. A randomization equilibrium is
a competitive equilibrium in which for some ex-post endowments the equi-
librium allocation in every state s is a spot market allocation for the same
economy. If, for example, the economic fundamentals allow for multiple equi-
libria then there is a randomization equilibrium. [29] has shown that there can
also be randomization equilibria if there are multiple equilibria for some dis-
tribution of endowments that is attainable via asset trade. In both cases the
equilibrium allocation of such a competitive equilibrium is a randomization
among the set of equilibria of some underlying economy. In randomization
equilibria sunspots are a device to coordinate agents’ expectations. This case
of sunspot equilibria has found many applications. In the international trade
literature, for example, currency crises are modelled by randomization sunspot
equilibria. See, for example, the seminal papers by [31] and [32] and also the
interesting empirical papers on this issue by [22] and [23]).
The question that arises from these observations is whether sunspot equi-
libria could be identified with randomization equilibria. This would then make
them very similar to publicly correlated equilibria known in the game theoretic
literature ([1])10 . Hence, the results of this literature would then be applicable
to sunspot equilibria.
It is obvious that in our setting with intrinsically complete markets
sunspot equilibria necessarily are randomization equilibria if assets are sun-
independent assets, if A(s) = A(1), s = 1, . . . , S. It is, however, not obvious at
all whether with a general asset structure there can also be sunspot equilibria
which are different from randomization equilibria. To clarify this point some
more definitions are needed.
Definition 5 (Attainable Endowment Distributions). Given the econo-
mic fundamentals [(ui , ω i )i=1,...,I ] and given the asset structure A the endow-
ment distributions ω̂ i (s), s = 1, . . . , S, i = 1, . . . , I is attainable if there exists
⋆
some competitive equilibrium with asset allocation (θi ), i = 1, . . . , I, such that
⋆
ω̂ i (s) = ω i + A(s)θi , s = 1, . . . , S, i = 1, . . . , I.
Based on the attainability concept we now define the uniqueness concept
suggested in [29]. This condition has later been called no potential multiplicity
by [16].

10
See [12] for relating sunspot equilibria to correlated equilibria in an overlapping
generations model.
94 Hens et al.

Definition 6 (Strong Uniqueness). The economy with the fundamentals

[(ui , ω i )i=1,...,I ] satisfies the strong uniqueness property for some asset struc-
ture A, if the spot market equilibria are unique for every attainable endowment
distribution.

Remark 5. With intrinsically incomplete markets sunspots are known to mat-

ter even if the economic fundamentals satisfy the strong uniqueness property
(cf. [4] and the literature that has emerged from it, [17] and [16]).

As said above, when assets are sun-independent then in our setting with in-
trinsically complete markets the strong uniqueness property rules out sunspot
equilibria. The following corollary to our theorem shows that for general asset
structures the strong uniqueness property restricted to non-negative alloca-
tions is still sufficient to rule out sunspot equilibria when marginal utilities
are inversely related to levels of utility.
Corollary 1. Under assumptions 1 and 2 and if marginal utilities are in-
versely related to levels of utility then sunspots do not matter if there are
unique spot market equilibria for all non-negative distributions of endowments
in the Edgeworth box.
Proof
Suppose sunspots did matter, then from our main result we know that the
transfer paradox needs to occur. However, as [36] has shown, this requires to
be able to trade to some non-negative distribution of endowments for which
there are multiple equilibria, which is a violation of the strong uniqueness
property.

Hence applying our lemma 1, in particular in the case of two agents,

sunspots do not matter if there are unique spot market equilibria for all dis-
tributions of endowments in the Edgeworth box.
However, asking for the strong uniqueness property may be asking too
much since not many economic fundamentals will satisfy this property. Hence
the question arises whether sunspot equilibria can exist when there are multi-
ple equilibria for some distribution of endowments the economy can arrive at
using asset trade. As [29] has shown this is definitely the case. However, [29]
did not show that this is still the case if for the initial distribution of endow-
ments there is a unique equilibrium. Note if there were also multiple equilibria
at that point then sunspots would matter even without evoking asset trade.
The question then is whether ”trading from uniqueness to multiplicity” is
possible. As the following corollary shows, this is not possible if there are two
agents and two commodities.
Corollary 2. Under assumption 3, in the case of two commodities and two
consumers sunspots do not matter if the economic fundamentals have a unique
equilibrium.
 Sunspots and Uniqueness 95

Proof
Suppose sunspots did matter then from our main result we know that the
transfer paradox needs to occur. However, as for example [2]11 has shown, in
the case of two commodities this requires to have multiple equilibria for the
initial distributions of endowments.

From corollary 2 we can see that in the case of two commodities and
sun-independent assets it is not possible to ”trade from uniqueness to multi-
plicity”. This is because with sun-independent assets sunspots can only matter
at distributions of endowments for which there are multiple equilibria.
[19] has claimed that for an economy with two agents and two commodi-
ties in which utility functions are concave transformations of Cobb-Douglas
utility functions sunspots matter. The corollary 2 shows that this claim is in-
correct. Moreover the mistake in [19] cannot be cured by changing the values
of the parameters for the same example12 because that example falls into the
broad class of economies which are covered by this note. Indeed for Cobb-
Douglas economies the equilibrium at the initial distribution of endowments
is unique and the strong uniqueness requirement is satisfied for almost all
asset structures A.
Making further assumptions on the agents’ risk aversion functions hi , [33]
shows that the construction of [29] does not work if there are only two com-
modities. Indeed, [33] shows for an economy with an arbitrary number of
consumers with additive separable utilities ui and non-decreasing relative risk
aversion, with two commodities and with asset payoffs that are independent
of the sunspot states, non-trivial sunspot equilibria do not exist.
The next example shows that with more than two commodities [29]’s con-
struction can be followed to demonstrate that it is well possible to trade from
uniqueness to multiplicity. Hence sunspot equilibria do occur even if for the
initial distribution of endowments there is a unique equilibrium. By our the-
orem then also the transfer paradox needs to occur because the economy we
construct has two consumers with CES-utility functions. That the transfer
paradox arises even if the initial equilibrium is unique should be no surprise
since this is well known in that literature. In the example we give we do how-
ever have some more structure. For the initial distribution of endowments the
excess demand function of the economic fundamentals can be generated by a
single representative consumer. Hence our example shows that ”the represen-
tative consumer is not immune against sunspots” and that even under this
strong structural assumption the transfer paradox is possible.

11
See also the solution to exercise 15.B.10C from [30] that is given in [18].
12
This possibility is left open by the observation of [3] who demonstrate that for
the specific parameter values chosen in [19] sunspots do not matter!
96 Hens et al.

Corollary 3. There exist economic fundamentals, satisfying assumptions 1

and 2, with two agents for which there is a representative consumer and yet
sunspot equilibria and the transfer paradox occur.
Proof
The example we give to prove this claim follows the construction of [29] making
sure that it can be done even under assumption 1 and 2 and more importantly,
that it can be done even for economic fundamentals with a representative
consumer:
There are four commodities and two CES-consumers with utility functions:
4
1/ρi
 i i
ui (xi ) = (αli )1−ρ (xil )ρ
l=1

The utility parameters are13 : ρ1 = −250.00000, ρ2 = −250.00000 and

α1 = (0.24986 , 0.24986 , 0.25041 , 0.24986) ,

α2 = (0.25261 , 0.25224 , 0.24758 , 0.24758) .

Note that utility functions satisfy assumption 1 and that they are linearly
homogenous. Moreover, agents utilities are close to the Leontief case which
is convenient because then the Slutzky-subsitution effects contributing to the
uniqueness of equilibria are small. Initial endowments are:

ω 1 = (0.01024 , 0.01021 , 0.01006 , 0.0101) ,

ω 2 = (0.01011 , 0.01009 , 0.00994 , 0.01) .

Note that the second agent’s initial endowments multiplied with 1.00971 gives
the first agent’s endowments, i.e. endowments are collinear and hence, by [11]’s
theorem, for this economy there exists a representative consumer.
Suppose there are three sunspot states, s = 1, 2, 3. We claim that trade in
(sun-independent) assets can be used to arrive at the following distribution
of endowments for which there are three spot market equilibria:

ω̂ 1 = (0.01414 , −0.01804 , 0.01259 , 0.0197) ,

ω̂ 2 = (0.00621 , 0.03834 , 0.00741 , 0.00037) .

The resulting spot market equilibrium prices are:

p(1) = (1 , 1.5801 , 5.05786 , 0.88298) ,

p(2) = (1 , 1.37985 , 0.88617 , 0.35661) ,
p(3) = (1 , 1.6344 , 11.79925 , 3.67601) .

Note that the rank of the price matrix is three since the determinant of all
prices excluding the numeraire is −11.11 . The equilibrium budgets are:
13
All values have been rounded to 5 decimal digits. The exact values can be found
at the page http://www.iew.unizh.ch/home/hens/sunspot.
 Sunspots and Uniqueness 97

b1 (1) = 0.06669 and b2 (1) = 0.10461 ,

b1 (2) = 0.00742 and b2 (2) = 0.06582 ,
b1 (3) = 0.20558 and b2 (3) = 0.15770 .
The resulting allocations are:
x1 (1) = (0.00785 , 0.00783 , 0.00781 , 0.00785) ,
x2 (1) = (0.01250 , 0.01246 , 0.01218 , 0.01226) ,

x1 (2) = (0.00205 , 0.00204 , 0.00205 , 0.00206) ,

x2 (2) = (0.01830 , 0.01825 , 0.01795 , 0.01801) ,

x1 (3) = (0.01142 , 0.01140 , 0.01134 , 0.01136) ,

x2 (3) = (0.00893 , 0.00890 , 0.00866 , 0.00871) .
As was to be expected in a case close to the Leontief-preferences, for both
agents and in all states the demand of any good l as a fraction of the sum of
all demands stay close to the agents utility parameters αli even though prices
vary substantially.
The corresponding utility levels are:
u1 (1) = 0.03132 and u2 (1) = 0.04935 ,
u1 (2) = 0.00819 and u2 (2) = 0.07249 ,
u1 (3) = 0.04544 and u2 (3) = 0.03513 .
i


Marginal utilities ∂b v i (s) = ubi (s)
(s)
are:
i=1,2,3.

∂b v 1 (1) = 0.46960 and ∂b v 2 (1) = 0.47178 ,

∂b v 1 (2) = 1.10458 and ∂b v 2 (2) = 1.10130 ,
∂b v 1 (3) = 0.22104 and ∂b v 2 (3) = 0.22277 .
Note that agent 1’s marginal utilities are ordered inversely to his utilities while
agent 2’s marginal utilities are ordered as his utility levels. This is possible
here because, as mentioned above, the assumption 3 is not satisfied for CES-
utilities. Evaluating the change in endowments at the new equilibrium prices,
the resulting income transfers from agent 1’s point of view are:
r1 (1) = 0.00201 , r1 (2) = 0.02195 , r1 (3) = −0.05989 .
Next we argue that for these values we can solve the first order condition for
asset demand:

h′i (v i (s)) ∂b v i (s) π(s) ri (s) = 0 , i = 1, 2 .
s

To do so we can still choose the vector of probabilities π and also the degree
of the agents’ risk aversion given by three values of the function h′i which are
98 Hens et al.

positive and decreasing with v i . However, the functions hi are not sufficiently
concave to make the composition hi ◦f i concave.
The resulting values are π = (0.8 , 0.05 , 0.15) for the probabilities and

h′1 = (1.98877 , 2.00032 , 1.97707) and h′2 = (1.01074 , 1.00064 , 1.02074)

for the risk aversions. Note that these values are smaller the higher the utility
levels are.
Since the three spot price vectors are linearly independent, we can now
find an asset matrix A ∈ IR3×4×2 such that

r(s) = p(s)·(A1 (s) − A2 (s)) for s = 1, 2, 3 . (6)

The vector A1 − A2 we have chosen is:

A1 − A2 = (−0.00597 , 0.02618 , −0.00457 , −0.01165) .

Finally, note that

 
λ1 (s) π(s) p(s)·A1 (s) = λ1 (s) π(s) p(s)·A2 (s)
s s

so that we can choose q1 = q2 . Accordingly we choose θ1 = (1, −1), θ2 =

(−1, 1) so that q · θi = 0, i = 1, 2 and θ1 + θ2 = 0. Since we have chosen
an economy with two assets,
the first-order conditions for asset trade are
equivalent to the conditions s λi (s) π(s) p(s)·A(s) = γ i q.
Comparing the vector of income transfers and the resulting utility levels we
find that agents’ utility decrease when they receive positive income transfer,
i.e. the transfer paradox occurs even though the order crossing property does
not.

u1 (no-asset-trade) = 0.03224 and u2 (no-asset-trade) = 0.03242 .

Last but not least, both agents profit from trade, i.e. their reservation utilities
are smaller than their utilities when participating in the markets14 :

u1 (no-trade) = 0.03218 and u2 (no-trade) = 0.04831

Eu1 = 0.03228 and Eu2 = 0.04838 .

The next corollary shows that as in the case of intrinsically incomplete

markets also with intrinsically complete markets sunspots can still matter
even if they do not serve as a coordination device among multiple equilibria.
14
In the computation of the expected utility we have ignored the functions hi
because they are sufficiently close to the case of risk neutrality. Indeed in this ex-
ample, the first-order-conditions for asset demand could also be solved without the
functions hi .
 Sunspots and Uniqueness 99

Corollary 4. With sun-dependent assets, under assumption 3, even for the

case of two commodities, there are sunspot equilibria which are not random-
ization equilibria.
Proof
The example we give to prove this corollary is adapted from the Example
15.B.2 in [30]15 . There are two commodities and two agents with endow-
ments [(ω11 , ω21 ), (ω12 , ω22 )] = [(2, r), (r, 2)]. Consumption sets are X i = {x ∈
IRL i i i
++ | u (x) ≥ u (ω )} and utility functions are given by

1 1
u1 (x1 ) = x11 − (x12 )−8 and u2 (x2 ) = − (x21 )−8 + x22 .
8 8
8 1
Aggregate endowments are ω = (2 + r, 2 + r) where r = 2 9 − 2 9 ≈ 0.77 .
Figure 1 shows the Edgeworth Box of this economy.16 The convex curve is

2.5 0
s̃

z
1.5

ŝ

0.5

0
0 0.5 1 1.5 2 2.5

Fig. 1. Edgeworth-Box

the set of Pareto-efficient allocations that lie in the interior of the Edgeworth
1
Box. It is given by the function x12 = 2+r−x 1 . The competitive equilibrium
1
allocations of our example will be constructed out of these interior allocations.
In figure 1 we have also drawn some budget lines indexed by s = ŝ, z, s̃, 0 ,
15
See [18] for the solution to the original example.
16
The figures 1 and 2 have been generated with MATLAB
. The scripts can be
downloaded from the page http://www.iew.unizh.ch/home/hens.
100 Hens et al.

supporting four different Pareto-efficient allocations which are equilibrium

allocations in the spot markets once appropriate spot market endowments
have been chosen. The sunspot equilibrium we construct exploits the fact that
in this example there are three equilibria for the distribution of endowments
[(ω11 , ω21 ), (ω12 , ω22 )] = [(2, r), (r, 2)]. Taking these endowments as the reference
point for the economy s = 0, we consider the transfer of endowments as
visualized in figure 2. From the three equilibria at [(2, r), (r, 2)] we have chosen
the one with the highest first agent utility to be the equilibrium allocation
for the reference situation s = 0. The asset structure A consists of numeraire

0.776

0.774
z
0.772
o
0.77 s̃
0.768

0.766 ŝ

0.764

0.762

0.76

0.758

0.756

1.99 1.992 1.994 1.996 1.998 2 2.002 2.004 2.006 2.008 2.01

Fig. 2. Zoom of the rectangle in 1

assets denominated in the second commodity. The vertical line in figure 2

indicates the possible direction of endowment redistributions. With reference
to s = 0, in the situation z the first agent has received a transfer of the
second commodity but his utility decreases. In the situations ŝ, s̃ the first
agent has donated some of the second commodity and with reference to s = 0
his utility decreases. While in ŝ it falls to the lowest of the four values, in s̃
it obtains a value between the utility in s = 0 and s = z. Hence for these
transfers the strong transfer paradox occurs and by application of our main
result there exists a sunspot equilibrium with spot market endowments given
by the intersection of the budget lines ŝ, z, and s̃ with the vertical line through
the point (2, 0), while the selected equilibrium in reference economy has the
budget line 0.
 Sunspots and Uniqueness 101

Although this sunspot equilibrium lies in a neighborhood of a randomiza-

tion equilibrium it is itself not a randomization equilibrium because all spot
market endowments differ.

This example exploits the multiplicity of equilibria of the economic funda-

mentals in the sense that in the neighborhood of the endowment distribution
leading to multiple equilibria the budget lines have various slopes that are not
ordered as the utility levels resulting in the ex post spot market equilibria (see
figures 1 and 2). This property could however also occur with a unique equi-
librium for the economic fundamentals. Imagine for example that the area in
the rectangle of figure 1 would lie outside the Edgeworth-Box. Then keeping
the line along which the transfers are defined inside the Edgeworth-Box the
same construction could be done. Unfortunately, we could however give no
simple utility functions as in Example 15.B.2 in [30] which would lead to this
feature.
Our final example shows that sunspot equilibria do exists even if the strong
uniqueness property holds. This example follows the construction of [19]. It
uses Cobb-Douglas utility functions and sun-dependent assets. As compared
to [19] the dimension of the example has been increased as we now use three
agents, three commodities, three assets and four states.
Proposition 2. With sun-dependent assets, and under assumptions 1 and 2,
the strong uniqueness assumption does not rule out the existence of sunspot
equilibria.
Proof
There are three commodities and three Cobb-Douglas-consumers with utility
functions:
3

ui (xi ) = exp αli ln(xil )
l=1

The utility parameters are17 :

α1 = (0.4072 , 0.5112 , 0.0816) ,

α2 = (0.0269 , 0.9301 , 0.0430) ,
α3 = (0.0088 , 0.0019 , 0.9892) .

Note that utility functions satisfy assumption 1 and that they are linearly
homogenous. Initial endowments are:

ω 1 = (0.1093 , 0.9045 , 0.1008) ,

ω 2 = (0.0715 , 0.0672 , 0.3781) ,
ω 3 = (0.8192 , 0.0283 , 0.5211) .
17
All values have been rounded to 4 decimal digits. The exact values can be found
at the page http://www.iew.unizh.ch/home/hens/sunspot.
102 Hens et al.

Suppose there are four sunspot states, s = 1, 2, 3, 4, which are equally likely
to occur. We claim that trade in numeraire but (sun-dependent) assets can
be used to arrive at the following distribution of endowments:
Because asset trade only redistributes the first good, to specify the asset
structure A ∈ IR3×4×3 , it is sufficient to give the pay offs in the numeraire,
good 3: ⎛ ⎞
410.0959 726.8543 384.2747
⎜ −19.8893 −35.3459 −18.6341 ⎟
A3 = ⎜⎝ −386.5250 −685.5971 −362.1722 ⎠ .
⎟

0.1150 0.8172 0.0893

Suppose equilibrium asset prices are q = (1.0108 , 1.7917 , 0.9472) and agent’s
asset allocation are:
⎛ ⎞
0.9911 −0.1125 −0.8787
θ = ⎝ −0.0544 0.0777 −0.0233 ⎠ .
−0.9549 −0.0269 0.9818

In this matrix rows correspond to assets and columns correspond to agents.

Then the resulting allocation of good three is:
⎛ ⎞
0.1013 0.3767 0.5220
⎜ 0.1032 0.3708 0.5260 ⎟
⎝ 0.1137 0.3385 0.5477 ⎠ ,
A3 θ = ⎜ ⎟

0.0851 0.4262 0.4887

where rows correspond to states and columns correspond to agents.

The induced spot market equilibria have the following price vectors, allo-
cations, values of utilities and of marginal utilities:
⎛ ⎞
0.4358 0.9623 1
⎜ 0.4324 0.9518 1 ⎟
p=⎜ ⎝ 0.4141 0.8954 1 ⎠ .
⎟

0.4638 1.0489 1

We give the allocations for each agent in a matrix xi with rows corresponding
to states and columns corresponding to goods:
⎛ ⎞
0.9525 0.5415 0.0832
⎜ 0.9525 0.5432 0.0825 ⎟
x1 = ⎜⎝ 0.9527 0.5531 0.0791 ⎠ ,
⎟

0.9522 0.5285 0.0885

⎛ ⎞
0.0291 0.4567 0.0204
⎜ 0.0289 0.4550 0.0200 ⎟
x2 = ⎜⎝ 0.0278 0.4449 0.0184 ⎠ ,
⎟

0.0307 0.4698 0.0228

 Sunspots and Uniqueness 103
⎛ ⎞
0.0184 0.0018 0.8965
3
⎜ 0.0185 0.0018 0.8974 ⎟
x =⎜
⎝ 0.0194
⎟.
0.0020 0.9025 ⎠
0.0171 0.0017 0.8887
The corresponding levels of indirect utility are:
⎛ ⎞
0.7176 0.4248 0.8651
⎜ 0.7183 0.4229 0.8661 ⎟
V =⎜ ⎝ 0.7225
⎟.
0.4122 0.8714 ⎠
0.7123 0.4388 0.8570

Finally, the levels of marginal utility within states are:

⎛ ⎞
0.1121 0.1538 0.2478
⎜ 0.1729 0.1537 0.2478 ⎟
∂b v = ⎜
⎝ 0.1075
⎟.
0.1527 0.2476 ⎠
0.1177 0.1531 0.2480

The resulting income transfers are:

⎛ ⎞
0.0004 −0.0013 0.0009
⎜ 0.0024 −0.0073 0.0049 ⎟
A3 θ = ⎜
⎝ 0.0129 −0.0396
⎟.
0.0266 ⎠
−0.0157 0.0481 −0.0324

Note that the transfer paradox does not occur because higher transfers are
always associated with higher levels of utility. However, this raises no problem
here to the existence of sunspot equilibria because higher utility levels are not
always associated with lower marginal utilities!
To complete the argument we need to solve the first order conditions for
asset demand 
h′i (v i (s)) ∂b v i (s) ri (s) = 0 , i = 1, 2 ,
s

by appropriate choice of the degree of the agents’ risk aversion:

⎛ ⎞
0.6414 0.6519 0.9985
′
⎜ 0.9980 0.6583 0.9985 ⎟
h =⎜ ⎝ 0.6513
⎟.
0.6930 0.9983 ⎠
0.6282 0.5979 0.9986

Note that agents are risk averse since smaller values of h′i (s) are associated
with higher values of utility. However, the degree of risk aversion changes
in a non-trivial way with the level of utility. Those changes in risk aversion
are know to be very important for asset demand. See the related papers by
[20, 21, 15] and also the monograph by [10] for an extensive study.


104 Hens et al.

5 Conclusion

Throughout the paper we restricted attention to economies with additively

separable utility functions since those functions are commonly used when ap-
plying general equilibrium models. We showed that for the case of two agents
(countries) the existence of sunspot equilibria is equivalent to the occurrence
of the transfer paradox. This equivalence enabled us to show that sunspots
cannot matter if the economy has a unique equilibrium for all distributions
of endowments induced by asset trade or if the initial economy has a unique
spot market equilibrium and there are only two commodities. Then we gave
an explicit example to show that with more than two commodities it is ac-
tually possible to trade from an endowment distribution leading to a unique
equilibrium towards one for which there are multiple equilibria. Moreover, we
gave two examples showing that sunspot equilibria need not result from mul-
tiplicity of spot market equilibria. In one of those examples, however, there
are multiple equilibria at the initial distribution of endowments. In the other
example there are sunspot equilibria even though the economy has a unique
equilibrium for all distributions of endowments induced by asset trade. This
latter example is possible because with more than two agents the existence of
sunspot equilibria is no longer tied to the occurrence of the transfer paradox.
We hope that this paper has clarified the relation between the existence of
sunspot equilibria and the uniqueness of spot market equilibria for economies
with intrinsically complete asset markets. Moreover, further exploring the con-
nection between sunspot equilibria and the transfer paradox, future research
might also show interesting new results in related settings. For example one
could try to get new insights for the sunspot literature by exploring the re-
sults on the transfer paradox in the overlapping generations model ([13]).
And similarly there might be new results in storage analyzing economies with
transaction costs respectively tariffs ([26]).

References
1. Aumann R (1974) Subjectivity and Correlation in Randomized Strategies. Jour-
nal of Mathematical Economics 1(1):67–96
2. Balasko Y (1978) The Transfer Paradox and the Theory of Regular Economies.
International Economic Review 19(3):687–694
3. Barnett RC, Fisher EO’N (2002) Comment on: ”Do Sunpots Matter When
Spot Market Equilibria Are Unique?” Econometrica 70:393–396
4. Cass D (1989) Sunspots and Incomplete Financial Markets: The Leading Ex-
ample. In: Feiwell G (ed) The Economics of Imperfect Competition and Em-
ployment: Joan Robinson and Beyond. MacMillan; London
5. Cass D, Shell K (1983) Do sunsposts matter? Journal of Political Economy
91:193–227
6. Chichilniski G (1980) Basic Goods, the Effect of Aid and the International
Economic Order. Journal of Development Economics 7(4):505–519
 Sunspots and Uniqueness 105

7. Chichilniski G (1983) The Transfer Problem with Three Goods once again:
Characterization, Uniqueness and Stability. Journal of Development Economics
13:237–248
8. Debreu G (1970) Economies with a Finite Set of Equilibria. Econometrica
38:603–615
9. Dierker E (1972) Two Remarks on the Number of Equilibria of an Economy.
Econometrica 40:951–953
10. Eeckhoudt L, Gollier C (1996) Risk: Evaluation, Management and Sharing.
Harvester Wheatsheaf; New York, London, Toronto
11. Eisenberg B (1961) Aggregation of Utility Functions. Management Science
7:337–350
12. Forges F, Peck J (1995) Correlated Equilibria and Sunspot Equilibria. Economic
Theory 5(1):33–50
13. Galor O, Polemarchakis HM (1987) Intertemporal Equilibrium and The Trans-
fer Paradox. Review of Economic Studies 54(1):147–156
14. Geanakoplos J, Heal G (1983) A Geometric Explanation of the Transfer Para-
dox in a Stable Economy. Journal of Development Economics 13:223–236
15. Gollier C, Pratt JW (1995) Risk Vulnerability and the Tempering Effect of
Background Risk. Econometrica 64:1109–1123
16. Gottardi P, Kajii A (1999) The Structure of Sunspot Equilibria: The Role of
Multiplicity. Review of Economic Studies 66:713–732
17. Guesnerie R, Laffont JJ (1988) Notes on Sunspot Equilibria in Finite
Economies. In: Volume en l’Honneur d’Edmont Malinvaud,Economica-EHESS
(English Version) MIT Press; Cambridge MA
18. Hara C, Segal I, Tadelis S (1997) Solutions Manual for Microeconomic Theory
by Mas-Colell, Whinston and Green. Oxford University Press
19. Hens T (2000) Do Sunpots Matter When Spot Market Equilibria Are Unique?
Econometrica 68:435–441
20. Hens T, Schmedders K, Voß B (1999) On Multiplicity of Competitive Equilibria
when Financial Markets Are Incomplete. In: Herings PJJ, van der Laan G,
Talman AJJ (eds) The Theory of Markets. North-Holland; Amsterdam
21. Hens T, Laitenberger J, Löffler A (2002) Two Remarks on the Uniqueness of
Equilibria in the CAPM. Journal of Mathematical Economics 37(2):123–133
22. Jeanne O (1997) Are currency crises self-fulfilling? A test. Journal of Interna-
tional Economics 43:263–286
23. Jeanne O, Masson P (2000) Currency crises, sunspots and Markov-switching
regimes. Journal of International Economics 50:327–350
24. Kehoe TJ (1985) Multiplicity of Equilibria and Comparative Statics. The Quar-
terly Journal of Economics 100:119–147
25. Kehoe TJ (1991) Computation and Multiplicity of Equilibria. In: Hilden-
brand W, Sonnenschein HF (eds) Handbook of Mathematical Economics
Vol. 4 : Chapter 38. North-Holland; Amsterdam
26. Lahiri S, Raimondos P (1995) Welfare Effects of Aid under Quantitative Trade
Restrictions. Journal of International Economics 39(3-4):297–315
27. Leontief W (1936) Note on the Pure Theory of Transfer. Explorations in Eco-
nomics, Taussig Festschrift, Vol. 84–92 New York
28. Magill M, Quinzii M (1996) Theory of Incomplete Markets. MIT Press; Cam-
bridge MA
106 Hens et al.

29. Mas-Colell A (1992) Three Observations on Sunspots and Asset Redundan-

dancy. In: Dasgupta P, Gale D, Hart O, Maskin F (eds) Economic Analysis of
Markets and Games: Essays in Honor of Frank Hahn. MIT Press; Cambridge
MA
30. Mas-Colell A, Whinston MD, Green JR (1995) Microeconomic Theory. Oxford
University Press
31. Obstfeld M (1994) The Logic of Currency crises. Cahier Economiques et Mon-
etaires (Banque de France) 43:189–213
32. Obstfeld M (1996) Models of Currency Crises with Self-Fulfilling Features. Eu-
ropean Economic Review 40:1037–1047
33. Pilgrim B (2000) A Brief Note on Mas-Colell’s Observations on Sunspots. Dis-
cussion Paper No. 450, University of Bielefeld, Germany
34. Rao M (1992) On the advantageous reallocation paradox. Social Choice and
Welfare 9:131–139
35. Shoven JB, Whalley J (1995) Applying General Equilibrium. Cambridge Uni-
versity Press
36. Trannoy A (1986) Paradoxe Global des Transfers et Multiplicité des Equilibres:
Deux Résultats. Annales D’Economie et de Statistique 4:53–61
Survival, Uncertainty, and Equilibrium Theory:
An Exposition

Mukul Majumdar1 and Nigar Hashimzade2

1
Department of Economics, Cornell University, Ithaca, New York, USA
[email protected]
2
School of Business and Economics, University of Exeter, Exeter, UK
[email protected]

Summary. The last twenty years have witnessed a significant growth of the lit-
erature on the “survival problem” , primarily in the context of the causes and
remedies of famines. Once a subject essentially of empirical development economics,
economic survival became an issue of analytical economics and, most recently, of
general equilibrium theory. We review several issues of the survival problem in the
general equilibrium framework. First, we consider a Walrasian economy with intrin-
sic uncertainty affecting the endowments and compute the asymptotic probability
of survival under different assumptions of dependence among the agents. Next, we
show how the presence of extrinsic uncertainty in a dynamic economy may lead to
a ruin in a sunspot equilibrium. Finally, we analyze the link between survival and
specialization in production.

Key words: Survival, General Equilibrium, Sunspot Equilibrium.

1 Introduction: A Historical Note

In his assessment of the important developments of general equilibrium the-
ory in early fifties, T.C. Koopmans ([20], p.59) observed that “most authors
have ignored the analytical difficulty of formulating a model that ensures the
possibility of survival, blithely admitting any nonnegative rates of consump-
tion as sustainable. Arrow and Debreu face this issue and find it to be a
complicated one...” Koopmans gave a verbal description/interpretation of the
Arrow-Debreu assumptions (see Section 4 of [2]) that guaranteed the existence
of equilibrium, noting that “they assume that each consumer can, if necessary,
survive on the basis of the resources he holds and the direct use of his own la-
bor, without engaging in exchange, and still have something to spare of some
type of labor which is sure to meet with a positive price in any equilibrium. If,
contrary to the authors’ indications, their model were given a stationary state
interpretation, it would be found best suited for describing a society of self-
sufficient farmers who do a little trading on the side. In modern society few
108 Mukul Majumdar and Nigar Hashimzade

of us can indeed survive without engaging in exchange...” Koopmans felt that

“there is considerable challenge to further research on the survival problem”
and did touch upon some directions and interpretations somewhat informally,
and did also recognize the “inadequacies of any model unable to recognize the
element of uncertainty in individual survival.”
We note that in his celebrated article Nikaido ([26], p.136) assumed that in
his exchange economy the initial endowment vector of each agent was strictly
positive. However, in his subsequent note [27], he recognized that this was a
“rather strict assumption” in his method of proof (relying upon what came
to be known as the Debreu–Gale–Nikaido lemma and generated a substan-
tial literature) of the existence of Walrasian equilibrium. Nikaido proceeded
to present a crisp and beautiful extension of the existence theorem to the
case where the initial endowment vector was positive (“non-zero, non-negative
bundle”) and the sum of the initial endowment vectors (over all agents) was
strictly positive. Concerned with the application of the model to international
trade theory , he felt (quite justifiably) that this weaker assumption was “rea-
sonable enough.”
For brevity, we touch upon just one more landmark in equilibrium the-
ory. In Debreu’s Theory of Value [9], the word “survival” did not figure at
all, but some remarks relevant for the interpretations and applications of the
Walrasian model are worth recalling. First, he provided a treatment of “uncer-
tainty” in which “the contract for the transfer of a commodity ... specifies,in
addition to its physical properties, its location and its date, an event on the
occurrence of which the transfer is conditional. This .. definition of a com-
modity allows one to obtain a theory of uncertainty free from any probability
concept and formally identical with the theory of certainty...”. This treatment
of “uncertainty” postulating the existence of a complete set of markets for all
“commodities” originated in Arrow’s remarkable paper [1]. With such a def-
inition of “commodities” it is clear that the technology or production set Yj
of producer j and the consumption set Xi of consumer i, are “in general,
contained in a coordinate subspace of Rl with a relatively small number of
dimensions” ([9] p.38 and p.51). Secondly, Debreu (p.50) asserted that the
consumption set Xi of all the possible consumption plans reflected “a priori
constraints (for example, of a physiological nature)” ( a “concrete” example
appears on p.51 with diagrams illustrating consumption sets):
“... the decision for an individual to have during the next year as sole
input one pound of rice and as output one thousand hours of some
type of labor could not be carried out”.
Another example (p.52) concerns consumption at two dates, with a minimal
level of consumption at date 1 which allows the consumer to survive until
date 2. In this example, by choosing xi the consumer “implicitly” chooses
“his life span”. Finally, in his existence theorem on pp.83–84 Debreu assumed
(assumption (c) on p.84) that “there is some x0i ∈ Xi such that x0i ≪ ωi ”. In
his Notes following the text of Chapter 5 (on pp.88–89), he indicated (refer-
 Survival, Uncertainty, and Equilibrium Theory: An Exposition 109

ring to the contributions of Mckenzie and David Gale) how the assumption (c)
could be replaced with other weaker( but, in our view, much less transparent)
assumptions involving the interior of the asymptotic cone of the aggregate pro-
duction possibility set. These assumptions play an essential role in establishing
the continuity of the budget set correspondence which is a step towards get-
ting the upper semicontinuity of the excess demand correspondence to which
the Debreu–Gale–Nikaido lemma is used, and also in ensuring that the wealth
of each agent is positive in equilibrium. Newman [25] has commented on the
proper interpretation of the consumption possibility set and the difficulties in
handling the survival problem.
It is fair to say that survival was not a serious theme in the subsequent
progress of Walrasian equilibrium analysis. However, the importance of this
problem was stressed in several prominent publications by Amartya Sen, pri-
marily in the context of his study of famines. Sen [32] quoted Koopmans’
remark on the Arrow–Debreu model (that we noted above) and felt that “the
problem that is thus eliminated by assumption in these general equilibrium
models is precisely the one central to a theory of survival and famines.” He
commented further that “ ‘the survival problem’ for general equilibrium mod-
els calls for a solution not in terms of a clever assumption that eliminates it
irrespective of realism, but for a reflection of the real guarantees that actually
prevent starvation deaths in advanced capitalist economies.” The entitlement
approach developed by Sen provided the foundation for modern theoretical
analysis of famines. However, it is difficult to capture the sweep of this ap-
proach by using formal models, particularly when uncertainty has to be ex-
plicitly modeled (see [8] for a deterministic analysis). Sen emphasized that for
a better understanding of the problems of survival, one must recognize: (i)
for a consumer to survive, his wealth at the equilibrium price system must
allow him to obtain the basic necessities, and (ii) “starvation can develop for
a group of people as its endowment vector collapses, and there are indeed
many accounts of such endowment declines on the part of sections of poor
rural population in developing countries... but starvation can also develop
with unchanged asset ownership through movement of exchange entitlement
mapping” ([33], pp.47–48).
Among the themes in the subsequent literature on famines, we note the
following:
(1) famines often occur without substantial decline in aggregate food avail-
ability:
“· · · starvation is a matter of some people not having enough food
to eat, and not a matter of there being not enough food to eat.
While the latter can be a cause of the former, it is clearly one of
many possible influences.” [32]
An example is the Bangladesh famine of 1974, where the availability of
food per head, including food production and net imports, in 1974 was
higher than in any other year during 1971–76 (see [12]). Thus, a partial
110 Mukul Majumdar and Nigar Hashimzade

equilibrium analysis focusing on the food market may be unable to capture

the complexity of events leading to an entitlement failure, and may render
misleading policy prescriptions.
(2) the impact of famines may differ for distinct groups of population, in par-
ticular, different occupational groups. For example, during the Ethiopian
famine of 1972–74 agricultural community suffered the most, and within
this group nomadic herdsmen were hit the hardest. The famine itself was
initiated by droughts, which resulted in reduced food supply. However,the
herdsmen “were affected not merely by the drought but also by the growth
of commercial agriculture, displacing some of these communities from
their traditional dry-weather grazing land, thereby vastly heightening the
impact of the drought”[33]. Other occupational categories that were in
the destitution groups included farm servants, rural laborers, craftsmen,
women in service occupations. One can see that the most suffering groups
were the ones who did not have command over food production, and whose
own production or labor, being less essential for survival, was no longer in
demand when food supply dropped.
“The characteristics of exchange relations between the pastoral and
agricultural economies contributed to the starvation of the herds-
men by making price movements reinforce – rather than counteract
– the decline in the livestock quantities. The pastoralist, hit by the
drought, was decimated by the market mechanism.” ([33], p.112)
Another example is the Bangladesh famine of 1974, during which the fam-
ilies of rural laborers suffered the most, even without decline in the aggre-
gate food availability. The reason in this case was the loss of employment
as a result of floods: floods prevented planting of the rice, which “would
reduce the food output later, but its impact on employment was imme-
diate and vicious” [12]. To understand this aspect of famines one has to
either study the economy at a disaggregated level with specialized occupa-
tions/endowments or allow for some commodities to play essential role in
survival;
(3) expectations about future food prices can play a significant role in market
behavior and result in food deprivation of certain groups of population.
According to the study by Ravallion [28], a sudden increase in rice prices
in Bangladesh during the 1974 famine could be explained by “excessive
hoarding” by stockholders, who overestimated the damage to the future
crop from floods, and, subsequently, have over-optimistic price expecta-
tions. To study the role of expectations one needs to turn to dynamic
general equilibrium framework.
A formal approach to the analysis of these themes seems to call for general
equilibrium models with specific structures (or, in Lindbeck’s words, “con-
crete” Walrasian systems) and with an explicit recognition of uncertainty. In
this paper we review some attempts in this direction.
 Survival, Uncertainty, and Equilibrium Theory: An Exposition 111

First, using the earlier works of Hildenbrand ([18], [19]) and Bhattacharya
and Majumdar [3] an attempt was made in [4] to define the probability of
survival in the presence of intrinsic uncertainty affecting the endowments. It
is reviewed in Section 2 with an improvement of one of the principal results
on the asymptotic behavior of the equilibrium price as the size of the econ-
omy increases. It is shown (see (16) that “ruin” may occur either as a result
of a collapse of endowments or as a result of an unfavorable movement in
the price system. For the case of a weak correlation in endowments and the
case of dependence in the form of dependency neighborhoods, the asymptotic
results are similar for the ones for the case of independent agents. The is-
sue of modeling a group of agents exposed to a common shock leads to the
study of exchangeable random variables, and the problem of characterizing
the probability of ruin in a large economy becomes more subtle.
Nest, we turn to the possible role of extrinsic uncertainty (a topic to which
David Cass has made notable contributions) and give an example of ruin
in a sunspot equilibrium. Using a model with overlapping generations and
constant endowments we show that there can exist multiple equilibria with
inter-generational trade, in some of which all agents survive, and in others
old agents are ruined, with all fundamentals being the same. These multiple
equilibria are supported by self-fulfilling beliefs, and the agents co-ordinate
their beliefs using “sunspots” as a co-ordinating device. Next we turn to a
model which links the survival problem to specialization. The main result, in
line with [10], is that in an economy with specialization in production a group
of agents, whose produce is less essential for survival, is more vulnerable to
starvation.

2 The Probability of Survival: Intrinsic Uncertainty in

the Cobb–Douglas–Sen Economy
In this and the next section we introduce a survival problem in the Walrasian
framework. For simplicity of exposition, we assume that the preferences of
the agents can be represented by a Cobb–Douglas utility function. This as-
sumption enables us to compute the Walrasian equilibrium explicitly (see (17):
such a computation can be extended to more than two goods). A more general
treatment, using Hildenbrand’s path-breaking work, of some of the issues can
be found in [4], although the “language” becomes unavoidably technical. Our
exposition focuses on the central economic issues with a rather minimal set
of techniques from probability theory, and, hopefully, has pedagogical value.
In what follows, R++ is the set of positive real numbers, x = (xk ) ∈ Rl
is non-negative (written x ≥ 0) if xk ≥ 0 for all k, and x is strictly positive
(written x ≫ 0) if x ∈ R++. Define x > 0 as positive (non-negative and
non-zero).
Consider, first, a deterministic Walrasian exchange economy with two
goods. Assume that an agent i has an initial endowment ei = (ei1 , ei2 ) ≫ 0,
112 Mukul Majumdar and Nigar Hashimzade

and a Cobb–Douglas utility function

u(xi1 , xi2 ) = xγi1 x1−γ

i2 (1)

where 0 < γ < 1 and the pair (xi1 , xi2 ) denotes the quantities of goods 1 and
2 consumed by agent i. Thus an agent i is described by a pair αi = (γ, ei ).
Let p > 0 be the price of the first good. In a Walrasian model with two
goods, we can normalize prices so that (p, 1−p) is the vector of prices accepted
by all the agents. The typical agent solves the following maximization problem
(P):
maximize u(xi1 , xi2 ) (2)
subject to the “budget constraint” defined as

pxi1 + (1 − p)xi2 = wi (p)

where the income or wealth wi of the i-th agent is defined as the value of its
endowment computed at (p, 1 − p):

wi (p) = pei1 + (1 − p)ei2 . (3)

Solving the problem (P) one obtains the excess demand for the first good as:

ζi1 (p, 1 − p) = [(1 − p)/p]γei2 − (1 − γ)ei1 (4)

One can verify that

pζi1 (p, 1 − p) + (1 − p)ζi2 (p, 1 − p) = 0 (5)

The total excess demand for the first good at the prices (p, 1 − p) in a
Walrasian exchange economy with n agents is given by:
n

ζ1 (p, 1 − p) = ζi1 (p, 1 − p) (6)
i=1

In view of (5) it also follows that

pζ1 (p, 1 − p) + (1 − p)ζ2 (p, 1 − p) = 0 (7)

The “market clearing” Walrasian equilibrium price system is defined by

ζ1 (p∗ , 1 − p∗ ) = ζ2 (p∗ , 1 − p∗ ) = 0 (8)

and direct computation gives us the equilibrium price p∗n (we emphasize the
dependence of equilibrium price on the number of agents by writing p∗n ) as:
 n   n n

  
∗
pn = Xi / Xi + Yi (9)
i=1 i=1 i=1
 Survival, Uncertainty, and Equilibrium Theory: An Exposition 113

where
Xi = γei2 , Yi = (1 − γ)ei1 . (10)
To be sure, one can verify directly that demand equals supply in the market
for the second good when the excess demand for the first good is zero.
Finally, let us stress that a Walrasian economy is “informationally decen-
tralized” in the sense that agent i has no information about (ej ) for i
= j.
Thus it is not possible for agent i to compute the equilibrium price p∗n .
One of the suggestions made by Sen ([33], Appendix A) to deal with sur-
vival explicitly is now recalled using our notation. Let Fi be a (nonempty)
2
closed subset of R++ . We interpret Fi as the set of all combinations of the
two goods that enable the i-th agent to survive. Now, given a price system
(p, 1 − p), one can define a function mi (p) as

mi (p) = min {pxi1 + (1 − p)xi2 } (11)

(xi1 ,xi2 )∈Fi

Thus, mi (p) is readily interpreted as the minimum expenditure needed for

survival at prices (p, 1 − p).
2
Example: Let (ai1 , ai2 ) ≫ 0 be a fixed element of R++ . Let
2
Fi = {(xi1 , xi2 ) ∈ R++ : xi1 ≥ ai1 , xi2 ≥ ai2 } (12)

Here mi (p) = pai1 + (1 − p)ai2 .

In our approach we do not deal with the set Fi explicitly. Instead, let us
suppose that, in addition to its utility function and endowment vector, each
agent i is characterized by a continuous function mi (p) : [0, 1] → R++ , and
say that for an agent to survive at prices (p, 1 − p), its wealth wi (p) (see (3)
must exceed mi (p). Hence, the i-th agent fails to survive (or, is ruined ) at
the Walrasian equilibrium (p∗n , 1 − p∗n ) if

[p∗n ei1 + (1 − p∗n )ei2 ] ≤ mi (p∗n ) (13)

or, using the definition (3)

wi (p∗n ) ≤ mi (p∗n ) (14)

From (13) and (14) one can see that an agent may face ruin due to (a) a possi-
ble endowment failure or (b) the equilibrium price system adversely affecting
its wealth relative to the minimum expenditure. This issue is linked to the lit-
erature on the “price” and “welfare” effects of a change in the endowment on
a deterministic Walrasian equilibrium (see the review of the transfer problem
by Majumdar and Mitra [22]).
Consider now a case of intrinsic uncertainty: suppose that the endowments
ei of the agents (i = 1, 2, · · · n) are random variables. In other words, each
ei is a (measurable) mapping from a probability space(Ω, F, P ) into the non-
negative orthant of R2 . One interprets Ω as the set of all possible states of
114 Mukul Majumdar and Nigar Hashimzade

environment, and ei (ω) is the endowment of agent i in the particular state ω.

The distribution of ei (·) is denoted by µi [formally each µi is a probability
measure on the Borel σ field of R2 , its support being a nonempty subset of the
strictly positive orthant of R2 ]. From the expression (9), the “market clearing”
equilibrium price p∗n (ω) is random, i.e., depends on ω:
 n   n n

  
∗
pn (ω) = γei2 (ω) / γei2 (ω) + (1 − γ) ei1 (ω) (15)
i=1 i=1 i=1

The wealth wi (p∗n (ω)) of agent i at p∗n (ω) is simply p∗n (ω)ei1 (ω) + [1 −
p∗n (ω)]ei2 (ω). The event
Rin = {ω ∈ Ω : wi (p∗n (ω)) ≤ mi (p∗n (ω))} (16)
is the set of all states of the environment in which agent i does not survive.
Again, from the definition of the event Rin it is clear that an agent may be ru-
ined due to a meager endowment vector in a particular state of environment.
In what follows, we shall refer to this situation as a “direct” effect of endow-
ment uncertainty or as an “individual” risk of ruin. But it is also possible
for ruin to occur through an unfavorable movement of the equilibrium prices
(terms of trade) even when there is no change (or perhaps an increase!) in the
endowment vector. A Walrasian equilibrium price system reflects the entire
pattern of endowments that emerges in a particular state of the environment.
Given the role of the price system in determining the wealth of an agent and
the minimum expenditure needed for survival, this possibility of ruin through
adverse terms of trade can be viewed as an “indirect” (“terms of trade”) effect
of endowment uncertainty.
Our first task is to characterize P (Rin ) when n is large (so that the as-
sumption that an individual agent accepts market prices as given is realistic).
While this task is certainly made easier by the structure of the model that
allows us to compute p∗n (ω) explicitly (15), the convergence arguments are
still somewhat technical, in particular when we attempt to dispense with the
assumption of stochastic independence.
To begin with let us make the following assumptions:
A1. {Xi }, {Yi } are uniformly bounded3 : there exists M > 0 such that 0 ≤
Xi < M , 0 ≤ Yi < M for i = 1, 2, · · ·.
{Xi } are uncorrelated,
A2.   {Yi } are uncorrelated.
 
 
A3. (1/n) EXi converges to some π1 > 0, (1/n) EYi converges to
i i
some π2 > 0 as n tends to infinity.
In the

special case when the distributions of ei are the same for all i (so that
1/n i EXi = π1 , where π1 is the common expectation of all Xi ; similarly
for π2 ), A3 is satisfied.
3
Recall that Xi and Yi are non-negative by non-negativity assumption on endow-
ments.
 Survival, Uncertainty, and Equilibrium Theory: An Exposition 115

Under A1–A3, if the number n of agents increases to infinity, as a con-

sequence of the strong law of large numbers we have the following limiting
property of equilibrium prices p∗n :
Proposition 1. Under A1–A3, as n tends to infinity, p∗n (ω) converges with
probability 1 (almost surely) to the constant

p0 = π1 / [π1 + π2 ] (17)

For the proof, we apply Corollary 6.2 in Bhattacharya and Waymire [5]
n n
1 1
(p.649) to the sequences { (Xi − E(Xi ))} and { (Yi − E(Yi ))} and
n i=1 n i=1
conclude that each of these sequences converges to zero with probability 1.
n n
1 1
Therefore, { Xi } converges with probability 1 to π1 , and { Yi } con-
n i=1 n i=1
verges with probability 1 to π2 (see, for example, Rohatgi [31], p. 252, Theorem
13). Next, we consider g(x, y) ≡ x/(x + y), a continuous function of a vector
a.s.
(x, y) on R++ . By definition of almost sure convergence, (Xn , Yn ) → (X, Y )
a.s.
implies g(Xn , Yn ) → g(X, Y ) (see, for example, Ferguson [14], p.9). Therefore,
a.s.
p∗n → p0 . Q.E.D.
Roughly, one interprets (17) as follows: for large values of n, the equilib-
rium price will not vary much from one state of the environment to another,
and will be insensitive to the exact value of n, the number of agents.
In [17] pn (ω) was shown to converge only in probability to p0 under weaker
assumptions. In our context, the boundedness assumption A1 seems quite
innocuous. Since we do not assume stochastic independence, the proof relies
on a relatively recent version of the Strong Law of Large Numbers due to
Etemadi [13] (see [5] for further discussion).
For the constant p0 defined by (17), we have the following characterization
of the probability of ruin in a large Walrasian economy:

Proposition 2. If p0 ei1 (ω) + (1 − p0 )ei2 (ω) has a continuous distribution

function,

lim P (Rin ) = P {ω : p0 ei1 (ω) + (1 − p0 )ei2 (ω) ≤ mi (p0 )}

 
(18)
n→∞

Remark : The probability on the right side of (18) does not depend on n, and
is determined by µi , a characteristic of agent i, and p0 .

2.1 Dependence

Case studies of famines often indicate that a famine is typically confined to a

particular geographic region or affects people belonging to the same occupa-
tion group (see, for example, [11]). To account for this property one needs to
116 Mukul Majumdar and Nigar Hashimzade

introduce stochastic dependence among the agents in the model. Of course,

the most difficult question is how to model the dependence among agents.
Also, dependence among random variables complicates asymptotic theory,
and, to obtain analytic results, one has to assume a particular structure of
the form of dependence. In this section we consider few examples of stochastic
dependence among agents in which the limiting results can be derived explic-
itly. Results parallel to Proposition 1 were obtained in [4] and [16]. Stated
informally, three interesting models were tractable: a model involving “weak”
correlation among agents ([4], Proposition 1.3); a model with appropriate re-
strictions on the size of the dependency neighborhood (a concept introduced
by Stein [35] and studied by Hashimzade [16] in the present context), and a
model with exchangeable (conditionally independent) agents. The last model
is particularly important in recognizing that the terms of trade effect may
remain significant even in a large economy. We stress the importance of this
point with a precise statement of the basic result proved in [4].

2.2 Exchangeability (Exposure to a Common Shock)

To capture the probability of exposure to a common shock to endowments in
a simple manner, let us say that µ and ν are two possible probability laws of
{ei (·)}i≥1 . Think of Nature conducting an experiment with two outcomes “H”
and “T” with probabilities (θ, 1 − θ), 0 < θ < 1. Conditionally, given that “H”
shows up, the sequence {ei (·)}i≥1 is independent and identically distributed
with common distribution µ. On the other hand, conditionally given that “T”
shows up, the sequence {ei (·)}i≥1 is independent and identically distributed
with common distribution ν. Let π1µ and π1ν be the expected values of X1
under µ and ν respectively. Similarly, let π2µ and π2ν be the expected values of
Y1 under µ and ν. It follows that pn (·) converges to p0 (·) almost surely, where
p0 (·) = π1µ /[π1µ +π2µ ] = p0µ with probability θ and p0 (·) = π1ν /[π1ν +π2ν ] =
p0ν with probability 1 − θ. We now have a precise characterization of the
probabilities of ruin as n tends to infinity. To state it, write
2
J = {(u1 , u2 ) ∈ R+ : p0µ u1 + (1 − p0µ )u2 ≤ mi (p0µ )};

ri (µ) = µ(du1 , du2 ). (19)
J

Similarly, define ri (ν) obtained on replacing µ by ν in (19).

Proposition 4. Assume that p0 ei1 (ω) + (1 − p0 )ei2 (ω) had a continuous dis-
tribution function under each distribution µ and ν of ei = (ei1 , ei2 ).
(a) Then, as the number of agents n goes to infinity, the probability of ruin of
the i-th agent converges to ri (µ), with probability θ, when “H” occurs and
to ri (ν), with probability 1 − θ when “T” occurs.
(b) The overall, or unconditional, probability of ruin converges to
θri (µ) + (1 − θ)ri (ν).
 Survival, Uncertainty, and Equilibrium Theory: An Exposition 117

Here, the precise limit distribution is slightly more complicated, but the
important distinction from the case of independence (or, “near independence”)
is that the limit depends not just on the individual uncertainties captured by
the distributions µ and ν of an agent’s endowments, but also on θ that retains
an influence on the distribution of prices even with large n.

3 Survival and Extrinsic Uncertainty: An Example With

Overlapping Generations

We now turn to extrinsic uncertainty: when the uncertainty affects the beliefs
of the agents (for example, the agents believe that market prices depend on
some “sunspots”) but the fundamentals are the same in all states. Clearly,
with respect to the probability of survival, the extrinsic uncertainty has no
direct effect, because it does not affect the endowments. However, it may have
an indirect effect: self-fulfilling beliefs of the agents regarding market prices
affect their wealth, and some agents may be ruined in one state of environment
and survive in some other state, even though the fundamentals of the economy
are the same in all states. To study the indirect, or the adverse term-of-trade
effect of extrinsic uncertainty on survival we need a dynamic economy.
Consider a discrete time, infinite horizon OLG economy with constant
population. We use Gale’s terminology [15] wherever appropriate. For exposi-
tory simplicity, and without loss of generality we assume that at the beginning
of every time period t = 1, 2, · · · there are two agents: one “young” born in
t, and one “old” born in t − 1. In period t = 1 there is one old agent of
generation 0. There is one (perishable) consumption good in every period.
The agent born in t (generation t) receives an endowment vector et = (eyt , eot )
and consumes a vector ct = (cyt , cot ). We consider the Samuelson case 4 and
assume, without loss of generality, et = (1, 0). We assume that the preferences
of the agent of generation t can be represented by expected utility function
Ut (·) = E [U t (ct )] with Bernoulli utility U t (ct ), continuously differentiable
and almost everywhere twice continuously differentiable, strictly concave and
2
strictly monotone in D, compact, convex subset of R++ . The old agent of
generation 0 is endowed with one unit of fiat money, the only nominal asset
in the economy. In every period the market for the perishable consumption
good is open and accessible to all agents. Denote the nominal price of the
consumption good at time t by pt . Define a price system to be a sequence of
positive numbers, p = {pt }∞ t=0 , a consumption program to be a sequence of
pairs of positive numbers c = {ct }∞ t=0 , a feasible program to be a consumption
program that satisfies cyt + cot−1 ≤ eyt + eot−1 = 1. The agent of generation t
4
If a population grows geometrically at the rate γ, so that γ t agents is born in pe-
riod t, and there is only one good in each period, the Samuelson case corresponds
to marginal rate of intertemporal substitution of consumption under autarky,
U1 (ey , eo )/U2 (ey , eo ), being less than γ. In our case γ = 1.
118 Mukul Majumdar and Nigar Hashimzade

maximizes his lifetime expected utility in the beginning of period t. In period

1, the young agent gives its saving (sy1 ) of the consumption good, to the old
agent in exchange for one unit of money (the exchange rate is determined by
p1 ). Thus, p1 s1 = 1. This unit of money is carried into period 2 (the old age
of agent born in period 1) and is exchanged (at the rate determined by p2 ) for
the consumption food saved by the young agent born in period 2 (sy2 ). The
process is repeated.

3.1 Perfect Foresight Equilibrium

If there is no uncertainty, with perfect foresight the price-taking young agent’s

optimization problem is the following:

max U (cyt , cot )

subject to

cyt = 1 − syt
cot = pt syt /pt+1

(0 ≤ syt ≤ 1, t = 1, 2, · · ·).
Here, syt ≡ eyt − cyt is savings of the young agent (this is the Samuelson
case, in Gale’s definitions [15]). A perfect foresight competitive equilibrium is
defined as a feasible program and a price system such that
(i) the consumption program c̄ = {c̄t } solves optimization problem of each
agent given p̄ = {p̄t } : (c̄yt , c̄ot ) ∈ D, c̄yt = 1 − st and c̄ot = p̄t st /p̄t+1 with


y y p̄t
st = arg max U (1 − st ) , st
0≤sy t ≤1
p̄t+1

and
(ii) the market for consumption good clears in every period:

c̄yt + c̄ot−1 = 1 (demand = supply for the consumption good)

p̄t st = 1 (demand = supply for money)

for t = 1, 2, · · ·.
By strict concavity of the utility function U (cyt , cot ), the young agent’s
optimization problem has a unique solution. Hence, we can express st as a
single-valued function of pt /pt+1 , i.e. we write st = st (pt /pt+1 ). This function
(called savings function) generates an offer curve in the space of net trades,
as price ratios vary. In the perfect foresight equilibrium

st (pt /pt+1 ) = 1/pt . (20)

 Survival, Uncertainty, and Equilibrium Theory: An Exposition 119

The stationary perfect foresight monetary equilibrium is a sequence of

constant prices p and constant consumption programs (1 − s̄, s̄), where s̄ =
s(1).5

3.2 Sunspot Equilibrium

Now consider an extrinsic uncertainty in this economy. There is no uncer-

tainty in fundamentals, such as endowments and preferences, but the agents
believe that market prices depend on realization of an extrinsic random vari-
able (sunspot). We assume that there is one-to-one mapping from the sunspot
variable to price of the consumption good. Because the agents cannot observe
future sunspots, they maximize expected utility over all possible future real-
ization of the states of nature. We examine the situation with two states of
nature, σ ∈ {α, β}, that follow a first-order Markov process with stationary
transition probabilities,  αα αβ
π π
Π= (21)
π βα π ββ
′
where π σσ > 0 is the probability of being in state σ ′ in the next period given
that current state is σ, and π σα + π σβ = 1. A young agent born in t observes
price pσt and solves the following optimization problem:
 
max π σα U (cy,σ
t , co,α
t ) + π σβ
U (c y,σ o,β
t , ct )

subject to

cy,σ
t = 1 − sσt
′
co,σ
′
t = pσt sσt /pσt+1
′
(0 ≤ sσt ≤ 1, sσt ≥ 0, σ, σ ′ ∈ {α, β}).
We restrict our attention to stationary equilibria, in which prices depend
on the current realization of the state of nature σ, and do not depend on the
calendar time nor the history of σ. A stationary sunspot equilibrium, SSE, is
a pair of feasible programs and nominal prices, such that for every σ ∈ {α, β}
(i) the consumption programs solve the agents’ optimization problem:
′
sσ (pσ /pσ ) =
arg max [π U ((1 − sσ ) , sσ pσ /pα ) +
σα
0≤sσ ≤1

+π σβ U (1 − sσ ) , sσ pσ /pβ
 
(22)
and

5
Given our assumptions on preferences and endowments, the stationary perfect
foresight monetary equilibrium exists and is optimal (see, for example, [21], Chap.
8).
120 Mukul Majumdar and Nigar Hashimzade

(ii) markets clear in every period, in every state.

cy,σ + co,σ = 1
pσ sσ = 1

It is easy to see that a stationary sunspot equilibrium exists when the

equation
pα α pα


β
β p
s − s =0 (23)
pβ pβ pα
has positive solutions for pα /pβ other than 1 . Solution pα /pβ = 1 corresponds
to the equilibrium in which uncertainty does not matter. It can be shown that,
if sunspot equilibria exist in this economy, there are at least two of them,
with pα /pβ > 1 and pα /pβ < 1 (see, for example, [7], [34]). This means that
in the sunspot equilibrium consumption of old agents is above the certainty
equilibrium consumption of olds in one state of nature and below in the other.
Suppose, we introduce an exogenous minimal subsistence level of consumption
(independent of σ ∈ {α, β}). It may be the case that in one of the states of
nature consumption of old agents falls short of minimal subsistence level:
old agents are ruined. Note that the endowments are not affected by the
uncertainty, and, therefore, there is no direct effect of uncertainty on ruin.
The event of ruin is caused purely by an indirect, or term-of-trade effect: the
equilibrium price system is such that the wealth of old agents does not allow
them to survive. The following numerical example illustrates this possibility
for the case of quadratic utility.

3.3 Ruin in Equilibrium

Let the preferences of the agents be represented by expected utility function

with

U (c) = u(cy , co ) − v(co )

√ 1 1
u(cy , co ) = 2a cy co + q cy + r co − b(cy )2 − d(co )2
2 2
o 2
θ o
(A − c ) , 0 < c ≤ A
v(co ) = 2
0, co > A

where a, b, c, q, r, θ, A are positive constants such that the utility function is

increasing and jointly concave in its arguments in D. v(·) is the disutility of
consuming less than A, the minimal subsistence level.6 As above, agents in
6
It may seem odd that the disutility from starvation is finite, but this can be
justified by the willingness of the agents to take a risk. Consider the following. In
the continuous time, if the consumption of an old agent is above A, he lives to the
end of the second period. If his consumption is below A, perhaps, he does not die
immediately. Albeit low, the amount consumed allows him to live some time in
 Survival, Uncertainty, and Equilibrium Theory: An Exposition 121

each generation receive identical positive endowments e = 1 of consumption

good when young and zero endowments when old; the initial olds are endowed
with one unit of money.

Benchmark case: perfect foresight

For the above preferences, savings function st (pt /pt+1 ) is implicitly defined
by 
a ρt st /(1 − st )) + q − b (1 − st )
ρt =  , (24)
a (1 − st )/(ρt st ) + r − d ρt st − v ′ (ρt st )
where ρt ≡ pt /pt+1 . The offer curve is described by

 
 
y x ′
(1 − x) a +q −bx −y a + r − d y − v (y) = 0 (25)
x y
In the stationary (deterministic) perfect foresight monetary equilibrium
consumption plan of an agent is (x, 1 − x), where x solves
  
x 1−x
a − + x (b + d) + v ′ (1 − x) + q − r − b = 0 (26)
1−x x

Stationary sunspot equilibria

Two states of nature, α and β evolve according to a stationary first-order

Markov process. The states of nature do not affect the endowments. Agents
can trade their real and nominal assets. In a stationary sunspot equilibrium
with trade sα , sβ solve the following system of equations:
 
sα sβ
π αα a + (1 − π αα
) a + q − b (1 − sα )
1 − sα 1 − sα
 
1 − sα
= π αα a + r − d sα − v ′ (sα ) (27)
sα
  
αα 1 − sα β ′ β sβ
+ (1 − π ) a β
+ r − d s − v (s )
s sα

the second period, and his lifespan in the second period is the longer, the closer
is his consumption to A. In the discrete time this translates into probability of
survival in the second period as a function of consumption. Thus, the old agent
survives with probability 1 if co ≥ A and with probability less than 1 if co < A.
Suppose, the objective of the agent is to maximize the probability of survival
(or maximize his expected lifespan). Then it can be presented equivalently as
the objective to minimize the disutility from consumption at the level below A.
Clearly, this disutility can be finite, at least in the vicinity of A, if the agent is
willing to take a risk. The authors are indebted to David Easley for this argument.
122 Mukul Majumdar and Nigar Hashimzade

and  
ββ sβ ββ sα
+ q − b 1 − sβ
   
π a β
+ 1−π a β
1−s 1−s
 
ββ 1 − sβ β ′ β
=π a + r − d s − v (s ) (28)
sβ
  
 ββ
 1 − sβ α ′ α sα
+ 1−π a α
+ r − d s − v (s )
s sβ

It is easy to see that one solution is sα = sβ = 1 − x, where x solves the

equation for the perfect foresight above. This solution does not depend on the
transition probabilities, prices and consumption are not affected by the un-
certainty: sunspots do not matter in this equilibrium. However, there may be
more solutions. For example, for a = 2, b = 0.5, d = 7, q = 0.02, r = 0.6 θ =
0.05 A = 0.3 and π αα = π ββ = 0.15 there are three stationary monetary
equilibria in the economy: one coinciding with the perfect foresight equilib-
rium and two sunspot equilibria. Prices and consumption programs for these
equilibria are given in the following table.

Multiple Equilibria Under Extrinsic Uncertainty.In every entry, the first

number is consumption of the young, the second is consumption of the old, and the
third is the nominal price of consumption good.

State PFE 1st SSE 2nd SSE

α (0.6670; 0.3330; 3.00) (0.5973; 0.4027; 2.48) (0.7518; 0.2482; 4.03)

β (0.6670; 0.3330; 3.00) (0.7518; 0.2482; 4.03) (0.5973; 0.4027; 2.48)

The consumption programs in sunspot equilibria are Pareto inferior to

the program in the perfect foresight equilibrium. Furthermore, in two sunspot
equilibria old agents survive in one state of nature and fail to survive in an-
other with the same amount of resources, because equilibrium price is too
high. (We intentionally considered the case where agents survive in the cer-
tainty equilibrium to demonstrate that survival is always feasible. Also, in this
model young agents always survive, – otherwise, the overlapping generations
structure collapses.)
 Survival, Uncertainty, and Equilibrium Theory: An Exposition 123

4 Survival and Specialization

The entitlement approach in the study of famines suggests that one has to ex-
plicitly take into account: (1) some goods or services produced in the economy
are less essential for survival than others, and (2) some groups of agents are
involved in production of these “less essential” goods and services and supply
them to the market in exchange for “more essential” ones, and, therefore, are
more vulnerable to starvation. The first observation can be formalized as the
existence of asymmetry in preferences, and the second one – as the existence
of specialization in output (or endowments) among agents.
A model of a static economy with asymmetric preferences and complete
specialization was introduced by Desai in [10]. The idea is the following. In
order to survive an agent needs to consume an essential good (or goods) at or
above some minimum level. Only after attaining the minimum level of con-
sumption of the essential good the agent can derive utility from consumption
of other, non-essential goods. If some agents in the economy are not initially
endowed with the essential good, they have to purchase the essential good
in the market. We modify Desai’s model to incorporate the probability of
survival when the consumption of the essential good falls below the mini-
mum subsistence level7 . There are two goods, essential (“food”, labeled x)
and non-essential (“non-food”, labeled y) in consumption, and two agents, a
food producer (agent 1) and a non-food producer (agent 2). The food pro-
ducer is endowed with f > 0 units of good x, and the non-food producer with
e > 0 units of good y. The agents have identical preferences, and each needs
a minimum quantity of food, fi∗ , to survive. The probability P of survival of
agent i is the function of i’s consumption of food:
⎧
⎨ 0, xi ≤ x̄i
P [i survives] = v(xi ), x̄ < xi ≤ fi∗
1, xi > fi∗
⎩

where v(x) is continuous and strictly increasing 8 . Without loss of generality

we further assume x̄i = 0 and fi∗ = f ∗ for i = 1, 2. Hence, if xi ∈ (0, f ∗ ), i
starves, and is ruined with probability 1 − v(xi ) because of starvation. Given
his budget constraint, an agent i, first, maximizes his probability of survival
(or minimizes probability of ruin), and, second, if he survives with probability
one, maximizes his utility of consumption above the survival level:

ui = ui ((x1 − fi∗ ) , x2 )
7
This formulation eliminates the “degeneracy” of the equilibrium with starvation,
in which an agent enjoys “minus infinity” utility, and, furthermore, prefers one
“minus infinity” level of utility to another “minus infinity”, see [10], p.434.
8
More generally, probability function is non-decreasing and right-continuous. We
use stronger assumptions to ensure uniqueness.
124 Mukul Majumdar and Nigar Hashimzade

(the survival level of the non-essential good y is zero), where ui (·) is strictly
concave, strictly increasing and twice continuously differentiable. We can for-
mally combine these two consecutive objectives of an agent into one objective
function:
xi ≤ f ∗

v (xi ) ,
Ui (xi , yi ) =
v (f ∗ ) + u ((xi − f ∗ ) , yi ) , xi > f ∗
We assume
αi
u(·) = (xi − f ∗ ) yiβi , (29)
∗
xi > f , yi > 0, αi , βi ∈ (0, 1), αi + βi ≤ 1. The functional form of v(xi ) in a
static model is irrelevant, as long as it is strictly increasing in xi . Let p be the
price of food in terms of non-food consumption good. The market equilibrium
in this economy is the set of consumption vectors {(x1 , y1 ) , (x2 , y2 )} and price
p such that
(i) given p, agent i maximizes his objective function Ui subject to his budget
constraint:
p xi + yi ≤ p fi + ei
i = 1, 2, (f1 , e1 ) = (f, 0), and (f2 , e2 ) = (0, e).
(ii) markets clear:

x1 + x2 = f,
y1 + y2 = e.

Because of the asymmetric preferences and complete specialization the

non-food producer is more vulnerable to starvation 9 . Consider different cases.
Case 1. Absolute scarcity.
Suppose, the harvest is so low that it cannot feed the food producer himself:
f ≤ f ∗ . The food producer consumes all his endowment and survives with
positive probability. There is no trade, and the non-food producer is ruined
with probability 1.
Case 2. Aggregate scarcity.
If the endowment of the food producer exceeds his minimum subsistence level,
he sells some of good x to the non-food producer in exchange for good y.
However, if the total amount of food is not enough to feed both agents, f ∗ <
f < 2f ∗ , the non-food producer starves and is ruined with positive probability.
Case 3. Aggregate availability.
Suppose now, that the total amount of food is enough to feed both agents,
f > 2f ∗ . However, as the analysis below shows, this condition is necessary,
but not sufficient to allow the non-food producer to survive with probability
one in the equilibrium. Below we derive the necessary and sufficient condition
of survival of both agents with probability one. We show that this condition
does not involve endowment and preferences of the non-food producer.
9
We assume free disposal for good y.
 Survival, Uncertainty, and Equilibrium Theory: An Exposition 125

Now we proceed to the formal analysis. Case 1 is irrelevant for our purpose,
because there is no trade in that case. In two other cases consumption of agent
1 (food producer) is


∗ α1 ∗ β1 ∗
(x1 , y1 ) = f + (f − f ) , p (f − f ) .
α 1 + β1 α 1 + β1

At price p the wealth, in terms of food, of agent 2 (non-food producer) is e/p.

If this wealth is below f ∗ , he sells all his endowment in good y for good x,
and his consumption is then


e
(x2 , y2 ) = ,0
p
From the market clearance condition the equilibrium price is
α 1 + β1 e
p∗ = . (30)
β1 f − f ∗
 
β1
Hence, agent 2 starves and is ruined with probability 1 − v α1 +β1 (f − f ∗ )
(notice, that this probability depends only on the endowment and preferences
of agent 1). This happens when e/p∗ < f ∗ < f , or, using (30),


α1
f∗ < f < f∗ 2 + . (31)
β1
Clearly, it can happen that the non-food producer starves even when the
amount of food is more than enough to feed both agents, i.e.,


∗ ∗ α1
2f < f < f 2 + .
β1
This is an example of exchange entitlement failure.
If the wealth of agent 2 is above f ∗ , he survives with probability one, and
his consumption is


 
∗ α2 e ∗ β2 ∗
(x2 , y2 ) = f + −f , (e − pf ) .
α 2 + β2 p α 2 + β2
Equilibrium price in this case is
α2 / (α2 + β2 ) e
p∗ =  . (32)
β1 / (α1 + β1 ) f − f ∗ 1 + β2 (α1 +β1 )
β1 (α2 +β2 )

Hence, the non-food producer survives with probability 1 when e ≥ p∗ f ∗ , or,

using (32),

∗ α1
f ≥f 2+ . (33)
β1
126 Mukul Majumdar and Nigar Hashimzade

This is the necessary and sufficient condition for survival of both types. One
can see that, holding the aggregate amount of food fixed, agent 2 is more
likely to starve, the more agent 1 prefers his own good over the other good.
Neither non-food producer’s tastes nor his endowment affect his probability
of survival.
An important policy implication is that under condition (31) increase in
the endowment of the non-food producer will not improve his food purchasing
power. Hence, the famine remedy in this case can be (i) redistribution or
(ii) direct food support. An interesting question in this regard is whether
universal or targeted food support is more efficient. Drèze and Sen provide
an extensive discussion of this issue in [12], Chapter 7. In the context of
this model, the universal support means giving equal amount of food to both
agents, and targeted support means giving the total amount of food aid to
the most vulnerable agent, i.e. to the non-food producer. If the objective
of the relief agency is to maximize the probability of survival of all agents,
then in situation (31), because type 1 survives with probability 1 given his
endowment, the relief agency can do either of the following:
(a) Give food to type 2 only. Type 2 will survive with probability 1 if the
β1
amount of food aid is, at least, fa ≡ f ∗ − (f − f ∗ );
α 1 + β1
(b) Give equal amounts of food to both types. If each type receives 1/2fb ,
such that 1/2fb < f ∗ , then type 2 will survive with probability 1 if the
value (in terms of food) of
his endowment  is at least f ∗ − 1/2fb. Simple
f
calculations render fb = 2 f ∗ − .
2 + α1 /β1
It is straightforward to show that fa < fb , i.e. targeted support is more
efficient than universal support (requires less resources, holding cost of dis-
tribution equal), if and only if f ∗ < f < f ∗ (1 + α1 /2β1 ). In other words,
the model suggests that at relatively high aggregate amount of food in the
economy the food aid from outside should be distributed equally among food-
and non-food producers. At relatively low aggregate amount of food in the
economy the food aid from outside should be directed to the non-food pro-
ducers.

5 Concluding Remarks
Joan Robinson ([30], p.189) wrote that “the hidden hand will always do its
work, but it may work by strangulation.” It has of course been an achievement
of high order to spell out the conditions under which the price system can play
an effective role in coordinating decentralized decisions in order to generate a
Pareto efficient allocation of resources. From all indications it appears, sadly
enough, that the strangulation by the invisible hand will haunt millions, es-
pecially in the century of globalization. General equilibrium analysis is very
 Survival, Uncertainty, and Equilibrium Theory: An Exposition 127

much relevant in understanding the full implications of economic policy to

improve the probability of survival. But help will not come from models in
which uncertainty has no essential role to play or in which the consumers have
the luxury of “choosing their life spans” explicitly or implicitly. Much remains
to be done in developing models that can throw light on the survival issues.

References
1. K. J. Arrow, Le Rôle des Valeurs Boursières pour la Répartition la Meilleure des
Risques, Econométrie (1953), pp.41–48, Paris, Centre National de la Recherche
Scientifique. (English translation: The Role of Securities in the Optimal Alloca-
tion of Risk-bearing, Revew of Economic Studies 31 (1964), pp.91–96.)
2. K. J. Arrow and G. Debreu, Existence of an Equilibrium for a Competitive
Economy, Econometrica 27 (1954), pp.265–290.
3. R. N. Bhattacharya and M. Majumdar, Random Exchange Economies, Journal
of Economic Theory 6 (1973), pp. 37–67.
4. R. N. Bhattacharya and M. Majumdar, On Characterizing the Probability of
Survival in a Large Competitive Economy, Review of Economic Design 6 (2001),
pp. 133–153.
5. Rabi N. Bhattacharya and Edward C. Waymire, Stochastic Processes with Ap-
plications. Wiley Series in Probability and Mathematical Statistics, 1990.
6. D. Cass and K. Shell, Do Sunspots Matter? Journal of Political Economy 91
(1983), pp. 193–227.
7. S. Chattopadhyay and T.J. Muench, Sunspots and Cycles Reconsidered, Eco-
nomic Letters 63 (1999), pp. 67–75.
8. Jeffrey L. Coles and Peter Hammond, Walrasian Equilibrium Without Survival:
Existence, Efficiency, and Remedial Policy, in “Choice, Welfare, and Develop-
ment: a Festschrift in Honour of Amartya K.Sen” (K. Basu, P. Pattanaik and
K. Suzumura, Eds.), pp. 32–64. Oxford: Clarendon Press, 1995.
9. G. Debreu, Theory of Value: An Axiomatic Analysis of Economic Equilibrium.
New Haven and London: Yale University Press, 1959.
10. Meghnad Desai, Rice and Fish: Asymmetric Preferences and Entitlement Fail-
ures in Food Growing Economies with Non-Food Producers, European Journal
of Political Economy, 5 (1989), pp. 429–440.
11. Jean Drèze (ed.), The Economics of Famine. Cheltenham, UK, Northampton,
MA: An Elgar Reference Collection, 1999.
12. Jean Drèze and Amartya Sen, Hunger and Public Action. Oxford: Clarendon
Press, 1989.
13. N. Etemadi, On the Laws of Large Numbers for Nonnegative Random Variables,
Journal of Multivariate Analysis, 13, pp.187–193.
14. Thomas S. Ferguson, A Course in Large Sample Theory. Chapman & Hall/CRC,
Texts in Statistical Science Series, 2002.
15. D. Gale, Pure Exchange Equilibrium of Dynamic Economic Models, Journal of
Economic Theory 6 (1973), pp. 12–36.
16. N. Hashimzade, Probability of Survival in a Random Exchange Economy with
Dependent Agents, Economic Theory 21 (2003), 907–912.
128 Mukul Majumdar and Nigar Hashimzade

17. N. Hashimzade and M. Majumdar, Survival Under Uncertainty in an Exchange

Economy. In: Athreya, K. et al. (Eds.), “Probability, Statistics and their Ap-
plications: Papers in Honor of Rabi Bhattacharya.” Institute of Mathematical
Statistics Lecture Notes - Monograph Series, v. 41 (2003), pp. 187–208.
18. W. Hildenbrand, On Economies with Many Agents, Journal of Economic Theory
2 (1970), pp.161–188.
19. W. Hildenbrand, Random Preferences and Equilibrium Analysis, Journal of
Economic Theory 3 (1971), pp.414–429.
20. T. C. Koopmans, Three Essays on the State of Economic Science. New York:
McGraw-Hill, 1957.
21. L. Ljungqvist and T. Sargent, Recursive Macroeconomic Theory. Cambridge,
Massachusetts; London, England: The MIT Press (2000).
22. M. Majumdar and T. Mitra, Some Results on the Transfer Problem in an Ex-
change Economy. In: Dutta, B. et al. (Eds.), “Theoretical Issues in Development
Economics.” New Delhi: Oxford University Press (1983), pp. 221–244.
23. M. Majumdar and R. Radner, Linear Models of Economic Survival Under Pro-
duction Uncertainty, Economic Theory 1 (1991), pp.13–30.
24. M. Majumdar and V. Rotar, Equilibrium Prices in a Random Exchange Econ-
omy with Dependent Agents, Economic Theory 15 (2000), pp.531–550.
25. P. Newman, Consumption Sets. In: The New Palgrave: General Equilibrium.
Eatwell, J., Milgate, M., Newman, P. (Eds.) Macmillan, London (1989), pp.108–
111.
26. H. Nikaido, On the Classical Multilateral Exchange Problem, Metroeconomica
8 (1956), pp.135–145.
27. H. Nikaido, A Supplementary Note to “On the Classical Multilateral Exchange
Problem”, Metroeconomica 9 (1957), pp.209–210.
28. M. Ravallion, Markets and Famines. Oxford: Clarendon Press, 1987.
29. Y. Rinott and V. Rotar, A Multivariate CLT for Local Dependence with
n−1/2 log n Rate, and Applications to Multivariate Graph Related Statistics,
Journal of Multivariate Analysis 56 (1996), pp. 333–350.
30. J. Robinson, The Pure Theory of International Trade (1946). In: J. Robinson,
“Collected Economics Papers,” I. Oxford: Basil Blackwell (1966).
31. V.K. Rohatgi, An Introduction to Probability Theory and Mathematical Sta-
tistics. Wiley Series in Probability and Mathematical Statistics (1976).
32. A. Sen, Ingredients of Famine Analysis: Availability and Entitlements, Quarterly
Journal of Economics 96 (1981), pp. 433–464.
33. A. Sen, Poverty and Famines: An Essay on Entitlement and Deprivation. Oxford:
Oxford University Press (1981).
34. S. Spear, Sufficient Conditions for the Existence of Sunspot Equilibria, Journal
of Economic Theory 34 (1984), pp. 360–370.
35. C. Stein, Approximate Computation of Expectations. Harvard, CA: IMS (1986).
On Price-Taking Behavior in Asymmetric
Information Economies⋆

Richard McLean1 , James Peck2 , and Andrew Postlewaite3

1
Rutgers University, New Brunswick, NJ 08901, USA
2
The Ohio State University, Columbus, OH 43210, USA
3
University of Pennsylvania, Philadelphia, PA 19104, USA

Summary. It is understood that rational expectations equilibria may not be in-

centive compatible: agents with private information may be able to affect prices
through the information conveyed by their market behavior. We present a simple
general equilibrium model to illustrate the connection between the notion of infor-
mational size presented in McLean and Postlewaite (2002) and the incentive proper-
ties of market equilibria. Specifically, we show that fully revealing market equilibria
are not incentive compatible for an economy with few privately informed producers
because of the producers’ informational size, but that replicating the economy de-
creases agents’ informational size. For sufficiently large economies, there exists an
incentive compatible fully revealing market equilibrium.

1 Introduction
In many markets of interest, agents are asymmetrically informed. Sellers of
stock or automobiles often possess information that potential buyers do not
have. In the presence of informational asymmetries, prices may reveal infor-
mation to some agents. A particularly low price for shares in a company may
signal to an uninformed agent that better informed agents are not buying
the stock, or may be selling the stock. The notion of rational expectations
equilibrium is one generally accepted extension of Walrasian equilibrium to
economies with asymmetrically informed agents. As in the case of symmetric
information, agents are assumed to maximize expected utility in a rational
expectations equilibrium. In the rational expectations model, however, agents
maximize expected utility not with respect to an exogenously given probabil-
ity distribution. Instead, agents maximize expected utility with respect to an
updated probability distribution that combines their initial information with
the additional information conveyed by the prices.
⋆
Postlewaite gratefully acknowledges support from the National Science Founda-
tion. We want to thank David Easley for helpful conversations.
130 Richard McLean, James Peck, and Andrew Postlewaite

Informational asymmetries can have serious consequences for the perfor-

mance of an economy. While Walrasian equilibria are Pareto efficient under
quite general conditions when agents are symmetrically informed, market out-
comes in the simplest of economies can be inefficient when the agents are
asymmetrically informed, as shown clearly in Akerlof’s Lemon’s paper (Ak-
erlof (1970)). In at least one case, however, asymmetric information does not
result in inefficiency. This is the case in which the rational expectations equi-
librium is fully revealing, that is, when the price reflects all of the agents’
information. If the price conveys all the information that agents have, then
each agent’s decision problem is equivalent to the problem that would be
solved if all the information were publicly available. In this case, the welfare
theorems assure that efficient outcomes are obtained.
Fully revealing rational expectations equilibria are ex-post efficient, but
open the question of the reasonableness of the price-taking assumption. As
in the case of Walrasian analysis of symmetric information economies, ra-
tional expectations equilibria assume that agents ignore the effect of their
market behavior on prices. In economies with symmetrically informed agents,
this assumption is sometimes justified by a heuristic argument that in large
economies agents will not be able to affect the price. There is also a formal
foundation for this argument that relies on the explicit modeling of agents’
strategic possibilities in a general equilibrium setting and provides conditions
under which the Nash equilibria of the strategic market game are approxi-
mately Walrasian.4 Roughly, it can be shown that, in plausible strategic mar-
ket games, agents will have little effect on the price of a good when they control
a small portion of the good. This provides some justification that price-taking
behavior is a plausible assumption for agents in large economies.
The situation with asymmetric information is more complicated since
agents can affect the price not only through the quantity of a good that they
trade, but also through the information their trades reveal. This second chan-
nel through which agents can affect prices means that it is not enough that
the quantity of a good that an agent controls is small relative to the aggregate
quantity of that good. An agent with a small amount of a particular good may
affect the price of that good because of the information he possesses. It is well
understood that there may be a conflict between the information contained
in rational expectations equilibrium prices and an agent’s incentive to reveal,
directly or indirectly, his information.5
This conflict should not in itself be surprising, since the incentive not to
take prices as given exists even when agents are symmetrically informed. The
most that one would hope for is that the effect of an agent’s behavior on prices,
via the information that his market behavior reveals, will be negligible in large
economies. Palfrey and Srivastava (1986) considered a stochastic replication
4
See, e.g., Mas Colell, Dubey and Shubik (1980) or Postlewaite and Schmeidler
(1978, 1981).
5
See, e.g., Blume and Easley (1983).
 On Price-Taking Behavior in Asymmetric Information Economies 131

procedure for an economy in which the incentive compatibility problems asso-

ciated with rational expectations equilibria asymptotically vanish. However,
their stochastic replication procedure has the property that, with probability
one, each agent’s private information is duplicated as the number of agents
increases. In a large economy, a single agent’s information is redundant in the
presence of the information of all other agents.
We are interested in situations in which a single agent’s information is
not redundant. The prototypical large economy that we envision is one in
which preferences and technology depend on the state of the world, which
is not directly known. Each agent has some information (a signal) regarding
the relative likelihoods of states. When agents’ signals are conditionally inde-
pendent (given the true state), the signal of a single agent can still provide
additional information about the true state, even in the presence of many
agents. However, the incremental value of that signal vanishes as the num-
ber of agents becomes large. In this world, agents become “informationally
small” as economies grows, but they never become “informationally irrele-
vant.” There is a large literature analyzing competitive models that ignores
the asymmetric information that must surely be present in any real-world
problem. The usefulness of analyses that ignore such asymmetric information
hinges on the belief that the incentive problems brought on by asymmetrically
informed agents become negligible in large economies.
We present and analyze a simple general equilibrium example with asym-
metrically informed agents similar to that described above. In the example,
asymmetrically informed agents make production decisions based on their pri-
vate information. Markets then open in which the produced goods are traded.
When the number of producers is small, the fully revealing market equilibria
are not incentive compatible; an agent’s market behavior can reveal private in-
formation, and the revealed information can affect prices in ways detrimental
to that agent. Consequently, when agents take into account the informational
impact of their market behavior, the outcome may be different from the com-
petitive outcome. However, when the economy is replicated in a natural way,
agents become informationally small, where the notion of informational size
is essentially that introduced in McLean and Postlewaite (2002). As a conse-
quence of their asymptotically vanishing informational size, agents will have
no incentive to manipulate prices in large economies.
We discuss within the example several of the issues that arise in modelling
general equilibrium economies with asymmetric information, including the
completeness of markets and multiple equilibria.

2 Example
The economy. There are two states of nature that are equally likely, θ1 and
θ2 and two periods. There are two kinds of agents, producers and consumers,
132 Richard McLean, James Peck, and Andrew Postlewaite

and three commodities: type 1 widgets, type 2 widgets, and money (denoted
m).
Producers. There are n producers, each of whom can make exactly one
widget using his own labor and chooses which type to produce in the first pe-
riod. The producers’ choice of widgets is simultaneous. Producers have iden-
tical state independent payoff functions defined simply as their final holdings
of money. That is, they value neither widgets nor their own labor input.
Consumers. Each consumer is endowed with 20 units of money but no
widgets. Consumers have the same utility function u(·) that depends on the
state, the number x1 of type 1 widgets consumed, the number x2 of type 2
widgets consumed and the final holding of money as given in the table below:

u(x1 , x2 , m, θ1 ) = m + 25x1
u(x1 , x2 , m, θ2 ) = m + 10x2

Note that, in state θi ,only type i widgets yield positive utility.

Information. Prior to production, each producer receives a noisy signal
of the state. The conditional distributions of the signal a producer receives in
each state are given in the table below:

state θ1 θ2
signal
s1 .8 .2
s2 .2 .8

Producers’ signals are conditionally independent.

Markets. There are no markets open in the first period. In the second pe-
riod competitive markets open in which the widgets that have been produced
can be exchanged for money. Since producers incur no opportunity cost in
making widgets, each of the n̄ producers makes a widget. We denote by n the
number of widgets of type 1 produced in period 1 (hence, there are n̄ − n type
2 widgets produced). If producers choose to make different types of widgets
when they have observed s1 than they make when they have observed s2 ,
the mix of widgets on the market in period 2 will convey information about
the state θ. If consumers rationally take this information into account, the
competitive price in period 2 will depend on the mix of widgets offered for
exchange.
A strategy for a producer is a mapping σ : {s1 , s2 } → {1, 2} specifying
which type of widget to produce as a function of the observed signal. We
consider symmetric equilibria in which all producers employ the same (pure)
strategy. If σ(s1 ) = σ(s2 ), then no information will be conveyed by the mix
of widgets on the market. However, if σ(s1 )
= σ(s2 ), the number of widgets
of type 1 will reveal the number of producers who received each of the two
signals.
 On Price-Taking Behavior in Asymmetric Information Economies 133

Consumers form expectations given the (common) strategy of the produc-

ers and the number of widgets of each type that are offered on the market in
the second period. When σ(s1 )
= σ(s2 ), Bayes rule uniquely determines the
posterior distribution on Θ, but if σ(s1 ) = σ(s2 ) there are producer choices
lying off the equilibrium path where one or more producers produce the wid-
get which was not the strategic choice for either of the signals. We denote by
µ(·|n) = (µ(θ1 |n), µ(θ2 |n)), a consumer’s beliefs when he observes n widgets
of type 1 on the market. We assume that µ(θ1 |n) is the Bayesian posterior
probability on θ1 when the number of widgets of type 1, n, is consistent with
producers’ (common) strategy σ, and unrestricted otherwise.6 No restriction
is placed on consumers’ beliefs when n is not consistent with σ.
Let Jn̄ = {1, .., n̄}. A price is a function Jn → R2+ where p(n) = (pn1 , pn2 )
is the pair of prices for widgets 1 and 2 respectively when n widgets of type 1
are produced in period 1. (The price of m is normalized to 1.) Given a price
function p(·) , consumers maximize expected utility (with respect to their
beliefs µ(·|n)). A market equilibrium is a price p(·) and optimizing behavioral
rules for producers and consumers for which markets clear.7

Definition: Given beliefs, n → µ(·|n), a market equilibrium (ME) is a price

function p(·) and a common producer strategy σ(·), for which
i. The symmetric strategy profile (σ(·), ..., σ(·)) is a Bayes equilibrium and
ii. For each n, consumer demand for widgets at price p(n) is equal to
number the of widgets produced.
When consumers’ have large initial endowments of money, they will want
to purchase a large number of widgets of a given type if the expected utility
of that type of widget is greater than the price. Similarly, they will purchase
none if the expected utility is below the price. Consequently, the only possible
market clearing prices for sufficiently large m are those for which the market
price of each widget is equal to the expected value of that widget. The expected
value of widget 1 when there are n widget 1’s on the market is (given producer
strategy σ) 25µ(θ1 |n), which then must be the price of widget 1 when n widget
1’s are offered on the market. Analogously, the expected value and the price
of widget 2 when n widget 1’s are offered is 10µ(θ2 |n).
We are interested in the existence (or nonexistence) of a fully revealing
market equilibrium (FRME), that is a market equilibrium for which the equi-
librium price reveals the private information that agents (producers) have.
When σ(s1 )
= σ2 (s2 ), the price reflects the number of widgets of type 1 on
the market, which is the same as the number of producers who have observed
6
It will be consistent with σ if either (i) σ(s1 ) = σ(s2 ); or (ii) σ(s1 ) = σ(s2 ) = 1
and n = n̄; or (iii) σ(s1 ) = σ(s2 ) = 2 and n = 0.
7
We use the term market equilibrium rather than rational expectations equilibrium
because producers’ choices may not maximize expected profit at the given prices
since their decisions must be taken prior to time at which the market opens. We
discuss this further in the last section.
134 Richard McLean, James Peck, and Andrew Postlewaite

signal s1 . Hence, a ME is fully revealing if the common producer strategy is

separating. We will show that fully revealing ME cannot exist when the num-
ber of producers is too small, but they will exist if the number of producers
is sufficiently large.

The case of a single producer

In the presence of a single producer, the problem reduces to the existence

of a separating equilibrium in a simple sender-receiver game. In a separating
equilibrium, the producer chooses one type of widget when he observes signal
s1 and the other type when he observes signal s2 .
Suppose that σ(s1 ) = 1 and σ(s2 ) = 2. Note first that, given this strategy,
the consumers’ beliefs are µ(θ1 |1) = .8 if he sees widget 1 and µ(θ1 |0) = .2 if he
sees widget 2. The equilibrium prices must therefore be p11 = 25µ(θ1 |1) = 20,
p12 = 10µ(θ2 |1) = 2 and p01 = 25µ(θ1 |0) = 5, p02 = 10µ(θ2 |0) = 8. We claim
that this proposed strategy is not an equilibrium. To see this, suppose the
producer receives signal s2 . If the producer produces widget 2, his payoff will
be p02 = 8, while his payoff from deviating and producing widget 1 is p11 = 20.
Thus, there cannot be a separating equilibrium in which the producer chooses
widget 2 when he receives signal s2 .
The same type of argument demonstrates that there cannot be a sepa-
rating equilibrium in which the producer chooses widget 2 when he receives
signal 1. Thus there cannot be a fully revealing equilibrium when there is a
single producer. We note that welfare is maximized when the producer chooses
widget 1 after receiving signal 1 and widget 2 after signal 2.

The case of two producers

Suppose there are two producers. We will show that as in the case of
a single producer, there can be no separating equilibrium. Let σ(·) be the
common strategy and suppose that σ(s1 )
= σ(s2 ). In particular, suppose that
σ(s1 ) = 1 and σ(s2 ) = 2. Finally, suppose that producer 1 receives signal
s2 . If both producers are following this strategy the consumers’ beliefs about
state 1 following 0, 1 or 2 widget 1’s being offered on the market are given in
the table below.

Widget Production Consumers’ beliefs Expected value Expected value

of widget 1 of widget 2
2 widget 1′ s µ(θ1 |2) = .94 23.5 .5
1 widget 1 µ(θ1 |1) = .5 12.5 5
0 widget 1′ s µ(θ1 |0) = .06 1.5 9.4
As before, the market clearing price of widgets must be the expected value
of the widget. If a producer chooses to produce a type 2 widget, then the most
that he will get for this widget is 9.4. If he produces a type 1 widget, then his
 On Price-Taking Behavior in Asymmetric Information Economies 135

payoff is at least 12.5. Consequently it cannot be an equilibrium for producers

to produce widget 1 following signal 1 and widget 2 following signal 2.
As in the case of a single producer, the calculations for the strategy which
prescribes producing widget 1 following signal 2 and widget 2 following signal
1 are the same as in case 1. Consequently, this producer strategy will not be
an equilibrium either.
In summary, when there are 2 producers there is no symmetric separating
equilibrium. It is easy to see why: a consumer values widget 1 more highly in
state θ1 than he values widget 2 in θ2 . Producer 1 affects consumers’ beliefs
through his choice of widget. If consumers expect producers to choose widget 2
after seeing widget 2, a producer’s payoff will be higher if he instead produces
widget 1.

The case of many producers

When there are many producers, there will exist separating equilibria in
which each producer chooses widget i after receiving signal si , i = 1, 2.8
Suppose producers receive conditionally independent, noisy signals of the state
that are accurate with probability .8 (that is, P (si |θi ) = .8). If n is large and
if all producers follow the separating strategy proposed above, then, with
high probability, approximately 80% of the widgets offered for sale will be
of type 1 when the state is θ1 (so that n ≈ .8n) and approximately 80%
of the widgets offered for sale will be of type 2 when the state is θ2 (so
that n ≈ .2n ). This observation is simply an application of the law of large
numbers. If n is large and if approximately 80% of the widgets are of type
1 (i.e., if n ≈ .8n ), a simple calculation verifies that the consumers’ beliefs
will ascribe probability close to 1 to state θ1 (i.e., (µ(θ1 |n), µ(θ2 |n)) ≈ (1, 0)).
If n is large and if approximately 80% of the widgets are of type 2 (i.e.,
if n ≈ .2n ), the same calculation verifies that the consumers’ beliefs will
ascribe probability close to 1 to state θ2 (i.e., (µ(θ1 |n), µ(θ2 |n)) ≈ (0, 1)). If the
true but unobserved state is θ1 , then, with high probability, the price vector
(pn1 , pn2 ) ≈ (p1.8n , p2.8n ) ≈ (25, 0) and if the true but unobserved state is θ2 ,
then, with high probability, the price vector (pn1 , pn2 ) ≈ (p1.2n , p2.2n ) ≈ (0, 10).
When n is large, the production decision of a single producer has only a small
effect on the ratio nn . If n is large, it follows that, with probability close to 1,
any single producer who changes production from one type of widget to the
other will have only a small effect on consumers’ beliefs, and hence on the
prices of the widgets.
Suppose that n is large and that producers are employimg the separating
strategy. Consider a producer who receives signal s1 . He believes that the
price of a type 1 widget will be close to 25 with probability P (θ1 |s1 ) = .8
and close to 0 with probability P (θ2 |s1 ) = .2, yielding an expected price of 20
8
In addition, there will exist equilibria which are pooling, and hence, are not fully
revealing. This is discussed in the last section.
136 Richard McLean, James Peck, and Andrew Postlewaite

for type 1 widgets. On the other hand, he believes that the price of a type 2
widget will be close to 0 with probability P (θ1 |s1 ) = .8 and close to 10 with
probability P (θ2 |s1 ) = .2, yielding an expected price of 2 for type 2 widgets.
Consequently, a producer who observes signal s1 will produce a type 1 widget.
A similar calculation will be made by a producer who receives signal s2 . He
believes that the price of a type 1 widget will be close to 25 with probability
P (θ1 |s2 ) = .2 and close to 0 with probability P (θ2 |s1 ) = .8, yielding an
expected price of 5 for type 1 widgets. On the other hand, he believes that
the price of a type 2 widget will be close to 0 with probability P (θ1 |s2 ) = .2
and close to 10 with probability P (θ2 |s2 ) = .8, yielding an expected price of
8 for type 2 widgets. Consequently, a producer who observes signal s2 will
produce a type 2 widget..
In summary, there will exist a fully revealing market equilibrium when
the number of agents is sufficiently large. It will be an equilibrium for each
producer to produce the widget that maximizes expected value given his own
information alone if other producers are doing the same. Any deviation from
this will have a vanishingly small effect on price as the number of producers
becomes large, making such deviations unprofitable.

3 Modelling the Consumer

The concept, market equilibrium, models producers as strategic, specifying
precisely what actions are available to them, but does not do the same for
consumers. In this section, we model the second stage game as a Shapley-
Shubik market game (Shubik (1973), Shapley (1974)), in which producers put
goods and consumers put money on widget-1 and widget-2 trading posts,
with the prices determined to clear the markets. We will consider limits, as
the number of consumers approaches infinity, of symmetric perfect Bayesian
equilibria in which all producers offer their widgets for sale. These equilibria
will define beliefs, producer strategies, and prices that constitute a market
equilibrium. We view this section as justifying the simpler model of section 2.
Wherever possible, we maintain the notation of section 2. The timing of
the game is now as follows. At stage 1, producers observe their signal, either s1
or s2 , and decide which widget to produce. We restrict attention to equilibria
in which all producers adopt the same production strategy, σ, and supply
their widget to the appropriate trading post. At stage 2, consumers observe
the number of widgets of each type that were produced, where n denotes the
number of type 1 widgets and n − n is the number of type 2 widgets. After
observing n, consumers decide how much money to bid for type 1 widgets and
type 2 widgets, at their respective trading posts.
Let h be the number of consumers, and let j index a particular consumer.
For j = 1, ..., h, let bj1 (n) denote the amount of money that consumer j bids on
the widget-1 trading post when the number of type-1 widgets produced is n,
 On Price-Taking Behavior in Asymmetric Information Economies 137

and let bj2 (n) denote the amount of money that consumer j bids on the widget-
2 trading post when the number of type-1 widgets produced is n. A strategy
for consumer j is a mapping, ψ j : Jn → R2+ , such that bj1 (n) + bj2 (n) ≤ 20
holds for all n.
The market clears according to the following allocation rule.

nbj (n)
xj1 (n) =
h 1 j ′ , (1)
j ′ =1 b1 (n)

(n − n)bj2 (n)
xj2 (n) =
h j′
, and (2)
j ′ =1 b2 (n)

mj (n) = 20 − bj1 (n) − bj2 (n),

where xj1 (n) denotes consumer j’s purchases of widget 1 when the number of
type-1 widgets produced is n, xj2 (n) denotes consumer j’s purchases of widget
2 when the number of type-1 widgets produced is n, and mj (n) denotes con-
sumer j’s money consumption when the number of type-1 widgets produced
is n. The money received by a firm selling a particular widget is the price of
that widget. These prices are given by

h j′
j ′ =1 b1 (n)
pn1 = and
n

h j′
j ′ =1 b2 (n)
pn2 = .
(n − n)

The allocation rule guarantees that all trade on a market takes place at
the same price, which is the total amount of money bid divided by the total
amount of widgets supplied. From (3.1) and (3.2), we see that the percentage
of the widgets up for sale that consumer j purchases is equal to the percentage
of the money that consumer j bids. If numerator and denominator are both
zero in (3.1) or (3.2), then consumers do not receive any widgets. Therefore,
we adopt the convention that 00 = 0 in (3.1) and (3.2). However, prices are a
different story. If, say, there are no type-1 widgets produced and no money is
bid for type-1 widgets, then the price of type-1 widgets is indeterminate. The
resolution of this indeterminacy is irrelevant for the characterization of perfect
Bayesian equilibrium, but could affect the comparison to market equilibrium.
We will comment on this later.
We restrict attention to symmetric perfect Bayesian equilibria, in which all
widgets produced are supplied to the market. To find an equilibrium, we find
consumer j’s best response to the common strategy played by all other con-
sumers, (b1 (n), b2 (n)), after n type-1 widgets are produced. We then impose
the condition that consumer j’s best response is in fact (b1 (n), b2 (n)). Given
beliefs, µ, the optimization problem for consumer j is to choose (bj1 (n), bj2 (n))
to solve
138 Richard McLean, James Peck, and Andrew Postlewaite

25nbj1 (n)
max[20 − bj1 (n) − bj2 (n)] + µ(θ1 | n)
bj1 (n) + (h − 1)b1 (n)
10(n − n)bj2 (n)
+µ(θ2 | n) .
bj2 (n) + (h − 1)b2 (n)

Computing the first order conditions, imposing (bj1 (n), bj2 (n)) = (b1 (n), b2 (n)),
and simplifying, we have

µ(θ1 | n)25n(h − 1)
b1 (n) =
h2
µ(θ2 | n)10(n − n)(h − 1)
b2 (n) = .
h2
Notice that the above equilibrium bids are uniquely determined, as long as
we impose symmetry. Plugging the above bids into the formula for prices, we
have
h−1
pn1 = ( )25µ(θ1 | n) and (3)
h
h−1
pn2 = ( )10µ(θ2 | n). (4)
h
The prices in (3.3) and (3.4) are uniquely determined from the ratio of
bids and offers, except for pn1 when n = 0 holds, and pn2 when n = n holds. In
these cases, bids and offers are zero, but either n or (n − n) appear in both
the numerator and denominator, and cancel each other. Thus, we will define
the prices associated with a symmetric PBE by (3.3) and (3.4).9

Proposition. Consider a sequence (σ h , ψ h , µh )h where (σ h , ψ h , µh ) is a sym-

metric PBE for the game with h consumers and consider the associated se-
quence of prices, (pn,h n,h h′ h′ h′
1 , p2 )h . If (σ , ψ , µ )h′ is a convergent subsequence,
′ ′
then (limh′ →∞ (pn,h , pn,h
′
1 2 ), limh′ →∞ σ h ) is a market equilibrium for beliefs
′
limh′ →∞ µh .

Proof. From the definition of PBE, and because there are only four possible
producer strategies, σ h , there exists h, σ, and µ such that h′ > h implies:
(1) σ h = σ, and
(2) µh (θ1 | n) = µ(θ1 | n) and µh (θ2 | n) = µ(θ2 | n) for all n occurring
with positive probability, given σ. Thus, µ is consistent with σ, according to
the criterion required for a PBE. This also implies that µ is consistent with
σ, according to the criterion required for a market equilibrium.
9
The prices given by (3.3) and (3.4), for the case of markets with zero supply and
demand, would arise if we placed ε offers of widgets on each market, and let ε
approach zero. See the discussion of virtual prices in, say, Dubey and Shubik
(1978).
 On Price-Taking Behavior in Asymmetric Information Economies 139

>From (3.3) and (3.4), we see that, for h′ > h, the incentives for producers
to deviate are exactly the same in the PBE as they are in a market equilibrium.
Sequential rationality of σ in the PBE implies {σ} satisfies part (i) of the
definition of a market equilibrium, given beliefs µ. The limiting price function,
p(n) = (25µ(θ1 | n), 10µ(θ2 | n)), satisfies part (ii) of the definition of a market
equilibrium.

4 Discussion
Incomplete markets

In the example, market equilibrium cannot be fully revealing when the

number of producers is small, but is fully revealing when the number of pro-
ducers is sufficiently large. Because producers are informationally small in
large economies, they cannot gain by attempting to manipulate prices. How-
ever, even for large economies, a fully revealing market equilibrium is not
a rational expectations equilibrium. In a rational expectations equilibrium,
producers can observe the prices of widget-1 and widget-2, infer the state of
nature, and produce the widget corresponding to the correct state. In the
example, the market equilibrium is fully revealing, but only after output has
been produced. A producer will produce the wrong widget with probability .2,
so a market equilibrium is not ex post efficient. Also, if we were to change the
parameter for observing the correct signal from .8 to .6, then there is no fully
revealing market equilibrium. Producers are informationally small, so there is
no incentive to manipulate market prices, but producers receiving signal s2
are better off gambling that their signal is wrong and producing widget 1.
Markets are incomplete in the example: there is no forward market in which
producers can sell their planned output before producing it. As mentioned in
the introduction, the structure of markets is crucial for the example. Suppose
instead that the only market available operated in the first period, in which
producers could offer widgets of either type for delivery in the second period.
Whatever prices prevail in this forward market, all producers will wish to sell
the same widget – the widget with the higher price. Thus, it is impossible that
producers with different signals will behave differently. But when all producers
behave identically regardless of their information, the price cannot reflect their
information.10 On the other hand, suppose that a securities market operated
in the first period, on which producers could trade money, contingent on
whether the number of widgets produced was greater than, or less than, n2 .
Now production could depend on the prices of securities, so that full revelation
would lead to efficient production decisions. In future work, we will explore
10
This is similar to the phenomenon in Grossman and Stiglitz (1980).
140 Richard McLean, James Peck, and Andrew Postlewaite

the conjecture that, in this more complete market structure, fully revealing
market equilibria exist, and correspond to rational expectations equilibria.
One might imagine a non-tatonement process that reveals producers’ infor-
mation (for example, a bargaining process between buyers and sellers), with
trade taking place only after revelation has taken place. Assuming such an
unmodelled process is unsatisfactory, however. The point of the present exer-
cise is to understand when agents’ private information will be revealed when
those agents are behaving strategically with respect to the revelation. Any
interesting analysis addressing this issue must model the process by which
agents’ information is reflected in prices. In other words, it is necessary to
specify exactly what actions agents can take and the mapping of their actions
into prices and outcomes.11
Specifying that producers choose which widgets to produce, with prices
and outcomes arising from competitive behavior subsequent to the choices,
provides a precise and plausible mechanism by which informed agents’ infor-
mation is incorporated into prices. One can, of course, think of alternative
mechanisms that link agents’ actions and resulting outcomes, but the intu-
ition in the example is quite general. Whatever the mechanism linking actions
and prices, if strategic behavior is modelled by Bayes equilibria, the revela-
tion principle applies. An agent’s incentive to misreport his information will
be limited by the degree to which his report affects the expected price. Said
differently, those agents whose information is likely to have a trivial effect on
price have little to gain from misreporting that information. For many nat-
ural mechanisms, when the gains from altering behavior to affect the price are
small, equilibrium actions will be close to actions that are optimal ignoring
the effect on price.

Multiple equilibria

We have demonstrated the existence of a fully revealing incentive compat-

ible ME when the number of producers was sufficiently large. This does not
mean that all incentive compatible ME’s are fully revealing. The nonrevealing
ME in which sellers produce widget 1 regardless of their signal, at a price of
(12.5, 10) remains an incentive compatible ME. This will be a perfect Bayes
equilibrium if consumers’ beliefs following the disequilibrium choice of widget
2 by a producer were that that producer had seen signal s1 with probability
.5. Even if consumers beliefs were such that they believed that a producer
who made this disequilibrium choice had seen signal s2 , this would have a
negligible effect on the subsequent price when there are many producers. As
a result, the return to a producer who chose to produce widget 2 would be
lower than producing widget 1.
Our point is not that a large number of agents necessarily leads to infor-
mation revelation but only that a large number (and the consequent infor-
11
See Dubey, Geanakoplos and Shubik (1987) for an early argument along these
lines and the general treatment of the question in Jackson and Peck (1999).
 On Price-Taking Behavior in Asymmetric Information Economies 141

mational smallness) makes the return to manipulation of prices through the

information revealed vanish asymptotically.

Interim vs. ex post incentive compatibility

The revealing ME is incentive compatible because at the time the seller

makes his decision about which widget to produce, a change will have a small
effect on the price with high probability. This is because the law of large num-
bers implies that for “most” realizations of the sellers’ signals, the posterior
on Θ given the signals puts probability close to 1 on the true state, and any
single deviation in the sellers’ choice of widgets will have a small effect on the
posterior. For some realizations, however, a single seller’s change in the widget
produced can have a nonnegligible effect on the posterior. Suppose that there
are 1001 sellers, and the vector of signals s is such that 500 sellers receive s1
and 501 sellers receive s2 . P (θ1 |s) = .4 in this case. Consider, however, the
vector of signals s′ in which one s2 is changed to an s1 ; P (θ1 |s′ ) = .6. In
other words, a single seller’s change in the choice of the widget to produce
causes a nontrivial change in the posterior distribution on Θ, given the change
in inference resulting from the production change. The nontrivial change, of
course, translates into a nontrivial change in the market price.
Regardless of the number of replicas, a single seller’s actions will have a
nontrivial effect on market prices for some realizations of the other agents’
signals. However, when the number of sellers is large, the probability that the
other sellers’ signals are such that any given seller will have a nontrivial effect
on the price is small. Since the potential gains from any change in price are
bounded, the expected price change resulting from a change in production will
be small when there are many sellers.
The presence of many other sellers makes a given seller informationally
small. Given the other sellers’ information, the given seller’s signal provides
little additional information, and the posterior distribution on Θ is not likely
to be very sensitive to his information, and hence not likely to be sensitive to
his market behavior.
142 Richard McLean, James Peck, and Andrew Postlewaite

References
1. Akerlof, G. (1970). “The Market for Lemons: Quality Uncertainty and the Mar-
ket Mechanism.” Quarterly Journal of Economics. 89: 488-500.
2. Blume, L. and D. Easley (1983). “On the Game-Theoretic Foundations of Mar-
ket Equilibrium with Asymmetric Information”
3. Dubey, P., J. Geanakoplos, and M. Shubik (1987). ”The Revelation of Informa-
tion in Strategic Market Games: A Critique of Rational Expectations Equilib-
ria,” Journal of Mathematical Economics, 16, pp 105-137.
4. Dubey, P., A. Mas Colell, and M. Shubik (1980). “Efficiency Properties of Strate-
gic Market Games: An Axiomatic Approach” Journal of Economic Theory 22,
363-376.
5. Dubey, P. and M. Shubik (1978). “A Theory of Money and Financial Institu-
tions. 28. The Noncooperative Equilibria of a Closed Trading Economy with
Market Supply and Bidding Strategies,” Journal of Economic Theory 17, 1-20.
6. Grossman, S. and J. Stiglitz (1980). “On the Impossibility of Informationally
Efficient Markets,” American Economic Review, Vol. 70, 393-408.
7. Jackson, M. and J. Peck (1999). “Asymmetric Information in a Competitive
Market Game: Reexamining the Implications of Rational Expectations,” Eco-
nomic Theory, Vol. 13, No. 3, pp 603-628.
8. McLean, R. and A. Postlewaite (2002). “Informational Size and Incentive Com-
patibility,” Econometrica, 70, 2421-2454.
9. Palfrey, T. and S. Srivastiva (1986). “Private Information in Large Economies,”
Journal of Economic Theory, 39, 34-58.
10. Postlewaite, A. and D. Schmeidler (1978). “Approximate Efficiency of Non-
Walrasian Nash Equilibria,” (with D. Schmeidler), Econometrica, 46, No. 1, pp.
127-137.
11. Postlewaite, A. and D. Schmeidler (1981). “Approximate Walrasian Equilibria
and Nearby Economies,” International Economic Review, Vol. 22, No. 1, pp.
105-111.
12. Shapley, L. S. (1974). “Noncooperative General Exchange,” Rand Corporation
#P-5286.
13. Shubik, M. (1973). “A Theory of Money and Financial Institutions: Commodity
Money, Oligopoly Credit and Bankruptcy in a General Equilibrium Model,”
Western Economic Journal, 11, 24-38.
Ambiguity aversion and the absence of indexed
debt⋆

Sujoy Mukerji1 and Jean-Marc Tallon2

1
Department of Economics, University of Oxford, Oxford OX1 3UQ and
University College, Oxford OX1 4BH, United Kingdom.
[email protected]
2
CNRS– EUREQua, Université Paris I, 106-112 bld de l’Hôpital, 75647 Paris
Cedex 13, [email protected]

Summary. Following the seminal works of Schmeidler (1989), Gilboa and Schmei-
dler (1989), roughly put, an agent’s subjective beliefs are said to be ambiguous if the
beliefs may not be represented by a unique probability distribution, in the standard
Bayesian fashion, but instead by a set of probabilities. An ambiguity averse decision
maker evaluates an act by the minimum expected value that may be associated with
it.
In spite of wide and long-standing support among economists for indexation of
loan contracts there has been relatively little use of indexation, except in situations
of extremely high inflation. The object of this paper is to provide a (theoretical)
explanation for this puzzling phenomenon based on the hypothesis that economic
agents are ambiguity averse. The present paper considers a general equilibrium model
based on (Magill and Quinzii, 1997), with ambiguity averse agents, where both
nominal and indexed bond contracts are available for trade and all relevant prices
are determined endogenously. We obtain conditions which prompt an endogenous
cessation of trade in indexed bonds: i.e., conditions under which there is no trade in
indexed bonds in any equilibrium; only nominal bonds are traded. We argue that the
obtained conditions mirror the known stylized facts about trade in indexed financial
contracts.

1 Introduction

In spite of wide and long-standing support among economists for indexation

of loan contracts there has been relatively little use of indexation, except

⋆
We thank seminar members at Birkbeck, Oxford, Paris I, Southampton and Tel
Aviv, the audience at the ’00 European Workshop on General Equilibrium Theory,
and especially, E.Dekel, I. Gilboa, D. Schmeidler and A. Pauzner for helpful com-
ments. The first author acknowledges financial assistance from an Economic and
Social Research Council of U.K. Research Fellowship (# R000 27 1065).
144 Sujoy Mukerji and Jean-Marc Tallon

in situations of extremely high inflation. Indeed, except in cases where in-

flationary circumstances forced them to do so, few governments, and fewer
private borrowers, have issued indexed bonds. People seem to have a prefer-
ence for specifying their obligations and opportunities in nominal units. As
(Shiller, 1997) remarks:
That the public should generally want to denominate contracts in
currency units–despite all the evidence that it is not wise to do so and
despite the obvious examples from nominal contracts of redistributions
caused by unexpected inflation–should be regarded as one of the great
economic puzzles of all time.
The object of this paper is to provide a (theoretical) explanation for this
puzzling phenomenon based on the hypothesis that economic agents are am-
biguity averse. The analysis throws up testable hypotheses and insights on
policy that are distinctive, compared to what a more standard analysis based
on the assumption that decision makers are (subjective) expected utility max-
imizers would suggest.
Suppose an agent’s subjective knowledge about the likelihood of contin-
gent events is consistent with more than one probability distribution. And
further that, what the agent knows does not inform him of a (second order)
probability distribution over the set of ‘possible’ (first order) probabilities.
Roughly put, we say then that the agent’s beliefs about contingent events
are characterized by ambiguity. If ambiguous, the agent’s beliefs are captured
not by a unique probability distribution in the standard Bayesian fashion but
instead by a set of probabilities. Thus not only is the outcome of an act un-
certain but also the expected payoff of the action, since the payoff may be
measured with respect to more than one probability. An ambiguity averse
decision maker evaluates an act by the minimum expected value that may be
associated with it: the decision rule is to compute all possible expected values
for each action and then choose the act which has the best minimum expected
outcome. This notion of ambiguity aversion, an intuition about behavior un-
der subjective uncertainly famously noted in (Ellsberg, 1961) and earlier by
(Knight, 1921), inspires the formal model of Choquet expected utility (CEU)
preferences introduced in (Schmeidler, 1989). The present paper considers a
competitive general equilibrium model of goods, bonds and money markets
populated by agents with CEU preferences3 , where both nominal and indexed
bonds are available for trade and prices of all goods and bonds are determined
3
Recent literature has debated the merits of the CEU framework as a model of
ambiguity aversion. For instance, (Epstein, 1999) contends that CEU preferences
associated with convex capacities (see section 2, below) do not always conform with
a “natural” notion of ambiguity averse behavior. On the other hand, (Ghirardato
and Marinacci, 2002) argue that ambiguity aversion is demonstrated in the CEU
model by a broad class of capacities which includes convex capacities.
Further on it will be seen that a property of portfolio inertia typical to the be-
havior of CEU agents is what crucially underpins the whole argument in the paper.
 Ambiguity aversion and the absence of indexed debt 145

endogenously. We obtain conditions which prompt an endogenous cessation

of trade in indexed bonds among private agents: i.e., conditions under which
there is no trade in indexed bonds in any equilibrium and only nominal bonds
are traded. It is worth clarifying, at this point, that while we “explain” the
veritable absence of indexed debt by showing that no trade in indexed bonds
is the unique equilibrium outcome under certain conditions, the analysis does
not imply that this is an efficient outcome. Indeed, as will be evident from the
analysis, making appropriate changes to the way price indices are constructed
would lead, if the theory presented in this paper is empirically valid, to more
widespread indexation of debt and an accompanying Pareto improvement.
An important point of inspiration for the analysis was to note that indexing
does not eliminate all (price) risk—rather it substitutes one risk for another—
an observation, we believe, originally due to (Magill and Quinzii, 1997). An
indexed bond, whose payoffs by definition are denoted in units of a reference
bundle of goods and services, will be secure against the aggregate price level
risk arising from changes in the money supply—the monetary risk, but un-
avoidably picks up the real risks arising from fluctuations in the relative prices
of the goods in the reference bundle. A nominal contract on the other hand
implies susceptibility to monetary risk but less so to real risk. The basic intu-
ition is straightforward. Being paid in terms of an index essentially amounts
to being paid units of the reference bundle of goods. Typically, the reference
bundle contains items that are not part of a given individual’s consumption
basket. Hence, effectively the individual is left to exchange goods in the refer-
ence bundle not in his consumption basket with goods he actually consumes.
Thus, a change in the price of goods not in his basket will affect the “worth”
of his remuneration in terms of the goods he does consume. Since the presence
of both types of risk is typical, standard portfolio analysis will advise that the
optimal portfolio should contain both nominal and indexed contracts (or to
put it somewhat differently, partial indexation). Given this one would expect,
and a result in this paper confirms, trade in indexed bonds will always be
observed in a market consisting of SEU (subjective expected utility) agents
so long as there were some inflation, however small. Under ambiguity aversion
the market outcome, though, may be dramatically different.
More specifically suppose, with respect to any two agents wishing to trade
in indexed bonds, the following is true:
1. the indexation bundle contains at least one good which is not consumed
by either of the agents;
2. the agents’ beliefs about the change in the price of good(s) not consumed
by either of them, relative to the average price level, is ambiguous;
3. agents are ambiguity averse.

Mukerji and Tallon (2003a) shows that this inertia property may be derived from
the primitive notion of ambiguity presented in Epstein and Zhang (2001) without
relying on a parametric preference model such as the CEU.
146 Sujoy Mukerji and Jean-Marc Tallon

The main result of the paper shows that, if agents believe general inflation
will not exceed a given bound and if ambiguity of beliefs about the relative
price movements is sufficiently high, then agents will have zero holdings of the
indexed bond in any equilibrium.
The result appears to fit well with what is commonly observed, both, in
terms of the plausibility of the hypotheses it rests on and in terms of its con-
sistency with regularities widely associated with trade in indexed debt. First,
consider the plausibility of the assumptions. Even at the best of times and
even in the most developed nations, information (say, formal forecasts) about
relative price movements are very hard to come by. Pick any two agents in
the economy; it is inevitable that the consumer price index will include goods
and services that are not part of the consumption basket of either agent. For
instance, it will inevitably include housing in regions that the agents have no
interest in. It is a plausible assumption that agents will have, at best, very
sparse informal knowledge about possible (relative) price movements of goods
and services that they never consume. Thus, if agents are typically ambigu-
ity averse when confronted with vague information, as much experimental
evidence suggests (see (Camerer, 1995)), then it would seem compelling (a
priori) to argue that they would behave in an ambiguity averse manner when
acting on beliefs about relative price movements of goods that never figure in
their consumption plans4 .
Next, consider some “stylized facts” about trade in indexed debt. As has
been already mentioned, barring certain exceptions trade in indexed bonds,
especially private bonds, is negligible. The exceptions are, one, situations of
extremely high and variable inflation, and two, situations (like in U.K., Israel)
where even though inflation is currently moderate, many wage payments are
statutorily index linked. A second exception occurs in economies which have
tamed inflation in the recent past but have experienced bouts of high inflation
in the more distant past (many S. American countries and also, Israel). It will
be explained that the logic of the main result does not only imply that trade
in indexed debt will be observed during episodes of relatively high inflation;
such an episode actually serves to ensure that trade in indexed debt would
endure for many periods beyond the original trigger even if agents do not
expect further bouts of high inflation. It is also observed that individuals in
countries with high and variable inflation commonly denominate their debt
(or even transactions like rental contracts for housing, as in Israel) in U.S.
Dollars even though they may rarely choose to tie payments to an index like
the CPI. This practice, at the least, demonstrates that agents understand
the vulnerability of nominal contracts denominated in terms of the (relatively
4
It is interesting to note in this context, as reported in (Shiller, 1997), when
asked for reasons for not opting for indexation many agents say that they are in-
hibited by their doubts that the government inflation numbers were valid for their
individual circumstances (pp 183, 188-190, 208). (Apparently, the concern was that
the official price index referred to a basket of goods that was possibly different from
the individual’s.)
 Ambiguity aversion and the absence of indexed debt 147

inflationary) domestic currency and do act on their understanding. They are

not as “naive” or “confused” as may be supposed in the first instance. If
anything, this observation deepens the puzzle: if they do understand the point
of indexation why do they refuse to use CPI and use the Dollar instead? The
results presented in the paper provide an explanation of this practice. Finally,
even though ours is an analysis of the private debt market in that we do not
model government behavior per se, we will argue the phenomenon actually
modelled here also provides an explanation for the lack of trade in indexed
public debt.
Finally, we relate our paper to some of the literature that has applied
CEU preferences to financial asset choice and financial market equilibrium.
Although a fuller discussion of the precise relationship between this contri-
bution and existing work is best deferred till the model has been spelled out,
some preliminary remarks might help the reader to grasp the intuition behind
our result. CEU preferences were identified by (Dow and Werlang, 1992) as
a potential source of inertia in portfolio holding: an agent having a riskless
endowment will not want to hold an uncertain asset on a (degenerate) price
interval only if he perceives the asset return as ambiguous and if he is ambigu-
ity averse. As noted in (Mukerji and Tallon, 2001) this inertia result does not
translate to no trade in an equilibrium model unless some extra ingredients
are added. Following a result in (Epstein and Wang, 1994) on sufficient condi-
tions for an equilibrium to be supported by multiple asset prices, we precisely
identified in that paper what these ingredients were: part of the asset return
has to be idiosyncratic (i.e., uncorrelated with the agents’ endowments), and
that part has to be ambiguous. Then, we showed, it is possible to establish
that assets that are of this form may not be traded at any equilibrium. The
present paper may be seen to be following that line of research by specifying
what form these idiosyncrasies might take in a more concrete setting: here,
they are precisely the ”noise” introduced in the return of an indexed asset
via the presence of ”irrelevant” goods (irrelevant in the sense that the agents
trading the asset neither consume nor are endowed with them) in the index-
ation bundle. The way this paper ”operationalizes” the idea is by restricting
the financial market trade to agents who consume a bundle of goods that
does not include some of the goods that are used in the indexation bundle.
The ”irrelevant” goods are consumed by some other ”prop” agents who do
not participate in financial markets; their role in the model is to endogenize
the (stochastic) prices of these goods. Their role is crucial in the sense that
without them we would lose the source of the noise and therefore the no trade
result.
The rest of the paper is organized as follows. The following section provides
an introduction to the formal model of Choquet expected utility. Section 3
works through a leading example with the aim of conveying the essential
intuition of the argument in a partial equilibrium setting. Section 4 contains
the general equilibrium model and the main result. Section 5 concludes the
148 Sujoy Mukerji and Jean-Marc Tallon

paper with a discussion of the related literature as well as the interpretation

and empirical significance of the findings.

2 Ambiguity aversion
2.1 The Ellsberg urn

One classic experiment illustrating how ambiguity aversion may affect behav-
ior, due to Daniel Ellsberg (1961), runs as follows:
There are two urns each containing one hundred balls. Each ball is ei-
ther red or black. The subjects are told of the fact that there are
fifty balls of each color in urn I. But no information is provided
about the proportion of red and black balls in urn II. One ball is
chosen at random from each urn. There are four events, denoted
IR, IB, IIR, IIB, where IR denotes the event that the ball chosen
from urn I is red, etc. On each of the events a bet is offered: $100 if
the events occurs and $0 if it does not.
The modal response is for a subject to prefer every bet from urn I (IR or IB)
to every bet from urn II (IIR or IIB). That is, the typical revealed pref-
erence is IB ≻ IIB and IR ≻ IIR. (The preferences are strict.) Clearly,
the decision maker’s beliefs about the likelihood of the events, as revealed in
the preferences, is not consistent with a unique probabilistic prior. The story
goes: People dislike the ambiguity that comes with choice under uncertainty;
they dislike the possibility that they may have the odds wrong and so make a
wrong choice (ex ante). Hence they go with the gamble where they know the
odds — betting from urn I. A slight restatement of the intuition conveyed
by the observed (modal) choice provides a useful perspective on what is to
follow. Notice, betting on IIR is the same as betting against IIB, and vice
versa, since the events are complementary. But to decide whether to bet on or
against IIR requires information about relative likelihood of the event IIR
and its complement. That is, of course, what ambiguity about IIR precludes.
On the other hand, betting on, say, IB allows the decision maker (DM) to
choose a prospect whose evaluation is unaffected by ambiguity.

2.2 Choquet expected utility

N
Let Ω = {ωi }i=1 be a finite state space, and assume that the DM chooses
among acts with state contingent payoffs, z : Ω → R. In the CEU model
((Schmeidler, 1989)) an ambiguity averse DM’s subjective belief is represented
by a convex non-additive probability function (or a convex capacity), ν such
that, (i) ν(∅) = 0, (ii) ν(Ω) = 1 and, (iii) ν(X ∪Y ) ≥ ν(X)+ν(Y )−ν(X ∩Y ),
for all X, Y ⊂ Ω. Define the core of ν, (notation: ∆ (Ω) is the set of all additive
probability measures on Ω):
 Ambiguity aversion and the absence of indexed debt 149

C (ν) = {π ∈ ∆ (Ω) | π(X) ≥ ν(X), for all X ⊆ Ω.}

Hence, ν(X) = minπ∈C(ν) π(X). Hence, convex capacity may be interpreted

as representing a convex set of (additive) probabilities. The ambiguity 5 of the
belief about an event X is measured by the expression A(X; ν) ≡ 1 − ν(X) −
ν(X c ) = maxπ∈C(ν) π(X) − minπ∈C(ν) π(X).
Like in SEU, a utility function u : R+ → R, u′ (·) ≥ 0, describes DM’s
attitude to risk and wealth. (Gilboa and Schmeidler, 1989) showed, that given
a convex non-additive probability ν, the Choquet expected utility 6 of an act is
simply the minimum of all possible ‘standard’ expected utility values obtained
by measuring the contingent utilities possible from the act with respect to each
of the additive probabilities in the core of ν:

 

CEν u (z) = min u (z (ω)) π (ω) ≡ u (z (ω)) dν
π∈C(ν) Ω
ω∈Ω

The fact that the same additive probability in C (ν) will not in general ‘mini-
mize’ the expectation for two different acts, explains why the Choquet expec-
tations operator is not additive, i.e., given any acts z, w : CEν (z) + CEν (w) ≤
CEν (z + w). The operator is additive, however, if the two acts z and w are
comonotonic, i.e., if (z(ωi ) − z(ωj ))(w(ωi ) − w(ωj )) ≥ 0.
Next, we state the notion of independence of convex non-additive proba-
bilities, proposed by (Gilboa and Schmeidler, 1989), used in this paper. Es-
sentially, the idea is as follows. Start with the set of probabilities in the core
of each capacity, select a probability from each such set and multiply to ob-
tain the corresponding product probability in the usual way and repeat for
all possible selections, thereby obtaining a set of product probabilities. The
lower envelope of the set of product probabilities, obtained in this way, is the
product capacity.

Definition 1. Let ν and µ be two convex non-additive probabilities, defined

 Ων and Ωµ respectively. The independent product of ν
on contingency spaces
and µ , denoted ν µ, is defined as follows
  
ν µ (A) ≡ min {(πν × πµ ) (A) : πν ∈ C (ν) , πµ ∈ C (µ)}

for every A ⊆ Ων × Ωµ .

5
(Fishburn, 1993) provides an axiomatic framework for this definition of am-
biguity and (Mukerji, 1997) demonstrates its equivalence to a more primitive and
epistemic notion of ambiguity (expressed in term’s of the DM’s knowledge of the
state space).
6
The Choquet expectation operator may be directly defined with respect to a
non-additive probability, see (Schmeidler, 1989). Also, for an intuitive introduction
to the CEU model see Section 2 in (Mukerji, 1998).
150 Sujoy Mukerji and Jean-Marc Tallon

It is well-known that it is possible to define more than one notion of inde-

pendence for non-additive beliefs. (Ghirardato, 1997) presents a comprehen-
sive analysis of the various notions. The definition invoked above, suggested
by (Gilboa and Schmeidler, 1989), is arguably the more prominent in the
literature. However, the formal analysis in the present paper, given the prim-
itives of our model, does not hinge on this particular choice of the notion
of independence. An important finding of Ghirardato’s analysis was that the
proposed specific notions of independent product give rise to a unique product
for cases in which marginals have some additional structural properties. The
capacity we use in our model is a product of two two-point capacities (i.e.,
each capacity is defined on a state-space consisting of two states). A two-point
capacity is, of course, a convex capacity and (trivially) a belief function. As
is explicit in Theorems 2 and 3 in (Ghirardato, 1997), if marginals satisfy
the structural properties the marginals we use do, then uniqueness of product
capacity obtains.

2.3 A portfolio inertia

(Dow and Werlang, 1992), identified an important implication of Schmeidler’s

model. They showed, in a model with one risky and one riskless asset, that if
a CEU maximizer has a riskless endowment then there exists a set of asset
prices that support the optimal choice of a riskless portfolio. The intuition
behind this finding may be grasped in the following example. Consider an
asset that pays off 1 in state L and 3 in state H and assume that ν (L) = 0.3
and ν (H) = 0.4. Assuming that the DM has a linear utility function, the
expected payoff of buying an unit of z, the act zb , is given by CEν (zb ) =
0.6 × 1 + 0.4 × 3 = 1.8. On the other hand, the payoff from going short on an
unit of z (the act zs ) is higher at L than at H. Hence, the relevant minimizing
probability when evaluating CEν (zb ) is that probability in C (ν) that puts
most weight on H. Thus, CEν (zs ) = 0.3 × (−1) + 0.7 × (−3) = −2.4. Hence,
if the price of the asset z were to lie in the open interval (1.8, 2.4), then the
investor would strictly prefer a zero position to either going short or buying.
Unlike in the case of unambiguous beliefs there is no single price at which to
switch from buying to selling. Taking a zero position on the risky asset has
the unique advantage that its evaluation is not affected by ambiguity.

3 The single decision maker’s problem: the intuition in a

simplified set up
In this section we consider the problem of a decision maker (DM) who wants
to transfer an amount S, S > 0, from today (Period 0) to tomorrow (Period
1). Goods prices that will obtain tomorrow are uncertain at the moment and,
for the purposes of this section, taken to be exogenously determined. We will
examine, in particular, the DM’s choice of portfolio given that the DM has
 Ambiguity aversion and the absence of indexed debt 151

access to only two kinds of assets, nominal bonds and indexed bonds, whose
prices are known and exogenously given. While the model in this section is
simpler in many details than the one in the next section, it is instructive in
that it will reveal to us how the trade-offs involved, given ambiguity aversion,
are such that the DM will strictly prefer to maintain a zero holding of the
indexed bond over a non-degenerate interval of indexed bond prices. This, as
we will see in the next section, is a key intuition to understanding why no
trade in indexed bonds might emerge as an equilibrium outcome.

3.1 A simple portfolio problem

We assume that there are just two goods in the economy: x and y. The
agent consumes only good x and is endowed in Period 1 with a (non-random)
endowment of that good, x̄. The agent does not consume good y nor is he
endowed with that good. However, the indexed bond pays off a unit of good
x and a unit of good y. The nominal bond pays in units of money.
The money supply in the economy in Period 1 can be either high (M ) or
low (m). When the money supply is low, suppose that prices can be equal
to either (px , pL H H L
y ) or (px , py ), with py > py , i.e., we assume that the price
of good y can be affected by factors that do not affect the price of good x.
When the money supply is high, we assume that prices can be either equal
to (λpx , λpL H
y ) or (λpx , λpy ) where λ = M/m > 1. This is reminiscent of the
quantity theory of money.
The following four states exhaustively describe the price uncertainty faced
by the individual:

State Prices
Return from an indexed bond
H H
1 p , p
x y H
p x + p y
λ × px + pH

2 λp
x , λp y
y
L
3 px , py L
p x + pLy
λ × px + pL

4 λpx , λpy y

In this section we leave the decision problem concerning the Period 0 con-
sumption unspecified and simply assume that the agent wants to save a given
amount S. Let xs denote the agent’s consumption in state s, bi the agent’s
indexed bond holding, q i its price, bn the agent’s nominal bond holding, and
q n its price. The DM’s budget constraints may then be rewritten to obtain:
   
pH n pH i
x1 = x̄ + 1 + pyx bi + pb x = x̄ + 1 + pyx − qnqpx bi + qnSpx
   
pH bn pH qi
x2 = x̄ + 1 + pyx bi + λp x
= x̄ + 1 + y
px − n
q λpx bi + qnSλpx
   
pL n pL i
x3 = x̄ + 1 + pyx bi + pb x = x̄ + 1 + pyx − qnqpx bi + qnSpx
   
pL bn pL qi
x4 = x̄ + 1 + pyx bi + λp x
= x̄ + 1 + y
px − n
q λpx bi + qnSλpx
152 Sujoy Mukerji and Jean-Marc Tallon

The budget constraints reveal how each of the two kinds of bonds provide a
hedge against a particular type of risk while simultaneously making the agent
vulnerable to another type of risk. The agent does not consume y, hence
given that the indexed bond pays a unit each of x and y, on maturity (of the
indexed bond) the agent is effectively left to exchange units of y obtained for
units of x. Therefore, even though payoff from an indexed bond is immune
to monetary shocks (it is independent of λ) it changes with changes in the
price of y, relative to the price of x. On the other hand, while the payoff (to
the agent) of a nominal bond is not affected by shocks to the relative price
of y, it is affected by monetary shocks (i.e., the value of λ). Hence, if bi = 0,
x1 = x3 and x2 = x4 , while, if bn = 0, then x1 = x2 and x3 = x4 . Notice also
that, given our assumptions, if bi > 0, then x1 > x3 and x2 > x4 , while, if
bi < 0, then x1 < x3 and x2 < x4 ; i.e., the agent’s ranking of the states (1,3
and 2,4) according to consumption reverses when switching from a long to a
short position on the indexed bond.
We next explore the consequences of ambiguity of beliefs about relative
price movements on the agent’s decision whether or not hold indexed bonds.
We assume that the agent is risk averse, with a utility index u : R+ → R,
which is increasing, strictly concave and differentiable. Suppose the agent has
precise probabilistic beliefs concerning the money supply7 (and consequently
whether the “price level” is high or low) but has ambiguous beliefs concerning
the realization of the idiosyncratic shock that affects only the price of good
y (a good he is not endowed with, and which he does not consume). More
precisely, we assume that the agent can assess the probability of the event
{1, 3} to be, say, µ and that of event {2, 4} to be 1 − µ. On the other hand,
conditional on a monetary state, the agent has only vague beliefs on whether
the price of good y is high or low, which is represented by the fact that sub-
jective beliefs are described by capacities ν H ≡ ν ({1, 2}) and ν L ≡ ν ({3, 4}),
with ν L + ν H < 1. We then assume that the overall beliefs of the agent are
simply the independent product of µ and ν. The preferences of the agent are
then represented by a utility functional, denoted V (x1 , x2 , x3 , x4 ), obtained by
taking the Choquet integral of u (xs ) with respect to the independent product
belief µ ⊗ ν.
If bi > 0, then V (x1 , x2 , x3 , x4 ) is given by:

µ ν H u(x1 ) + (1 − ν H )u(x3 ) + (1 − µ) ν H u(x2 ) + (1 − ν H )u(x4 )

If bi < 0, then V (x1 , x2 , x3 , x4 ) is given by:

µ (1 − ν L )u(x1 ) + ν L u(x3 ) + (1 − µ) (1 − ν L )u(x2 ) + ν L u(x4 )

Note that if ν H + ν L = 1 then the two expressions above coincide.

7
The actual equilibrium model in the next section allows beliefs about the money
supply to be ambiguous too.
 Ambiguity aversion and the absence of indexed debt 153

3.2 A price interval supporting zero holding of the indexed bond

We now establish, following (Dow and Werlang, 1992) that there is a non-
i
degenerate interval of relative bond prices, qqn , at which the agent optimally
wants to hold a zero position in the indexed bond. Below, we present an infor-
mal, intuitive argument. A more formal argument appears in the Appendix.
Suppose the agent presently holds only nominal bonds and is considering
buying/selling an arbitrarily small unit of indexed bonds. The agent’s present
utility level
 in each ofthe four states may then be represented generically by
U (λ) ≡ u x̄ + qnSλpx , where λ = 1 in states 1 and 3 and λ = 1 + ε, ε > 0, in
′′ ′
states
 2 and 4.  Since u (x) < 0, the marginal utility in each state, U (λ) ≡
′ S
u x̄ + qn λpx , is increasing in λ; the intuition being that a higher inflation
will affect the saver adversely. Now consider the gross increase (decrease) in
welfare at each state if the agent were to buy (sell) an infinitesimal unit of an
indexed bond:  
pH
State 1 : 1+ y
px U ′ (1)
 
pH
State 2 : 1 + y
px U ′ (1 + ε)
 
pL
State 3 : 1 + y
px U ′ (1)
 
pL
State 4 : 1 + y
px U ′ (1 + ε)

Figure 1, depicts these “payoffs” for an ε “small enough”.

Figure 1. Contingent payoffs from an indexed bond

Notice, as arranged in the figure the payoffs are increasing

  from left
pL
′
to right. Since U (λ) is increasing in λ and ε > 0, 1 + px U ′ (1) <
y

     
pL pH pH
1 + pyx U ′ (1 + ε) and 1 + pyx U ′ (1) < 1 + pyx U ′ (1 + ε) . Since u
154 Sujoy Mukerji and Jean-Marc Tallon
 
pL
is continuous and pH y > pL
y , for ε small enough 1 + y
px U ′ (1 + ε) <
 
pH
1 + px U ′ (1) . Now, to simplify matters dramatically, suppose ν H = ν L =
y

0. Hence, if the agent were to go long in the indexed bond the  payoffs in
pL
events where the monetary shock is low and high are 1 + px U ′ (1) and
y

 
pL
1 + pyx U ′ (1 + ε), respectively. Similarly, if the agent were to go short in
the indexed the payoffs inevents where
 bond   the monetary shock is low and
pH
y ′ pH
y ′
high are 1 + px U (1) and 1 + px U (1 + ε), respectively. Hence, the
i
most (in terms of the relative price qqn ) the agent would want to bid for an unit
 
pL
of the indexed bond is 1 + pyx U ′ (1 + ε) . On the other hand the minimum
 
pH
the agent would ask for going short on an indexed bond is 1 + px U ′ (1).
y

Hence, there must be a non-degenerate interval of prices at which the agent

strictly prefers to maintain a zero holding of the indexed bond. Notice, as de-
picted in the 
figure, the
 effect of increasing
 ε would
 be to increase the distance
pL pH
l, since both 1+ y
px U ′ (1 + ε) and 1+ y
px U ′ (1 + ε) will rise relative to
the other payoffs. This implies that, for ε large enough, the portfolio inertia
interval will collapse. Notice, it is not necessary that A = 1 − ν H − ν L = 1 for
the portfolio inertia interval to emerge. A portfolio inertia interval will exist
for A < 1, as long A is high enough. Since the ranking of states (according
to consumption) reverses when switching between a long position and a short
position, the “relevant” probability also switches between evaluating a long
and a short position. It were as if when evaluating a long position the priors
are “concentrated” on states 1 and 2, while they are concentrated on 3 and 4
when evaluating a short position. It is this feature which causes a “kink” in
the utility functional at the zero holding position and leads to the portfolio
inertia interval.

Remark 1. An analogous argument, mutatis mutandis, shows that even for

a lender (i.e., S < 0) one would obtain a portfolio inertia interval. While
reconstructing the argument, essentially, the only adjustment one needs to
make is to bear in mind is that for a lender U ′ (λ) is decreasing in λ and
hence the “ordering” of the states is reversed (compared to the ordering in the
case of a saver).

Remark 2. One could also wonder whether there is a range of prices at which
there is a zero holding of the nominal bond. Indeed, in the set up we have
described so far, if we were to have, in addition, ambiguous beliefs about the
inflation (i.e., µ ({1, 3}) + µ ({2, 4}) < 1), then for a high enough level of
 Ambiguity aversion and the absence of indexed debt 155

this ambiguity, by an argument analogous to the one given above, there will
i
be an interval of relative bond prices, qqn , at which the agent will only have
a zero holding of the nominal bond. However, this is not very compelling as
the result is not robust in an important way. It does not hold any more if the
agent were to have some second period income, preset in nominal terms, that
is not derived from bond holdings.
Imagine for instance that the agent receives a state contingent income
4
stream, {ms }s=1 , preset in nominal terms. This could represent any previ-
ously contracted arrangement like, wage income, pension, social security ben-
efits, etc., that are no more than partially indexed (i.e., they have a nominal
component). In this “richer” model, the second period budget constraints are
as follows:
(px +pH )

q n  bn m1
x1 = x̄ + qi pxy S + 1 − px + pH

y qi px + px
(px +pH )

q n  bn m2
x2 = x̄ + qi pxy S + λ1 − px + pH

y qi px + λpx
(px +pL )

q n  bn m3
x3 = x̄ + qi pxy S + 1 − px + pL

y qi px + px
(px +pL )

q n  bn m4
x4 = x̄ + qi pxy S + λ1 − px + pL

y qi px + λpx

immediately, except in the very specific case wherein m2 , m4 =

Now,
1it follows
λ m , m3 , having a zero holding of nominal bonds does not allow the agent to

be rid of the inflation risk. Indeed, bn = 0 implies x1 = x2 and x3 = x4 if and

2 4 1 3

only if m , m = λ m , m . This, of course, is the case only if all income

is fully indexed or there is no preset nominal income (ms = 0, ∀s) . Hence,
whenever, m2 , m4 = λ m1 , m3 and bn = 0 consumption is strictly ordered

across states 1 and 2 and across states 3 and 4. (For instance, if ms = m > 0
∀s, one has x1 > x2 and x3 > x4 .) This strict ordering is preserved in the ǫ-
neighborhood bn ∈ (−ǫ, ǫ) . Hence, there is no switch in the probability the DM
“applies” when evaluating going short and going long on the nominal bond.
Thus the usual expected utility logic applies and the derivative (of the utility
functional) is continuous at the point of zero holding, implying that there is no
non-degenerate interval of (bond) prices at which the agent holds zero nominal
bond.

Remark 3. We have assumed constant x endowment. This is essentially for

expositional ease and to make the model as simple as possible. If one were
to introduce uncertain endowment of good x, a similar reasoning would hold:
given a value of the endowment, there would still be two other (orthogonal)
sources of uncertainty, namely, the price of good x and the value of the indexed
bundle (the price of good y). While it is true that Dow and Werlang’s result
assumed that the agent’s endowment was riskless, a crucial contribution of
Epstein and Wang (1994) was to show that the result could be generalized to
the case of a non-stochastic endowment so long as there were a part of the
asset payoff that was orthogonal to the endowment. If one were to introduce
156 Sujoy Mukerji and Jean-Marc Tallon

say two values for the level of endowment (say, x̄, x), the actual number of
states would be eight, and for sufficiently high ambiguity aversion, the agent
would not be willing, for an interval of prices, to use the indexed bond to hedge
the risk linked to his x-endowment.

4 No-trade in indexed bonds: a general equilibrium

framework

4.1 The institutional setup

The previous section shows that for both types of agents, those who save and
those who borrow, there exists a range of relative bond prices, correspond-
ing to each agent, at which the agent maintains a zero holding in indexed
bonds. However, this does not immediately translate into a conclusion about
conditions under which no-trade in indexed bonds is the unique equilibrium
outcome. To be able to get to that conclusion several questions remain to be
answered. What would ensure that the bid-ask price intervals of the various
agents “overlap”? Why should equilibrium bond price fall within the zone
of “overlap”? Further, since we know that for a portfolio inertia interval to
emerge the goods prices have to vary across states in particular ways, a related
question is are such state-contingent price variations consistent with competi-
tive equilibrium in money, goods and bond markets? To deal with such issues
we turn next to a two-period monetary general equilibrium model, general in
the sense that all prices are obtained endogenously by (simultaneous) market
clearing in bond, goods and money markets. Since the overall aim is to lay
out the logic of no trade as transparently as possible, we have chosen the sim-
plest model we could. For instance, given the crucial role of the movement of
goods prices in obtaining no-trade, the relationship between such prices and
the parameters of the model has been kept as tractable as possible. Arguably,
the more realistic source of sectoral price movements are shifts in preferences
and/or technological shocks. However, the analysis here is exposited within
a framework of the simplest general equilibrium model known to economists,
an exchange economy without any production, wherein relative-price move-
ments are derived by perturbing endowments. The point, we emphasize, is
transparency, not realism per se.
There are two groups of agents in the model. The first group (whose agents
are indexed by h = 1, . . . , H) are those who trade on financial markets, while
the second group (whose agents are indexed by k = 1, . . . , K) has no access
to any financial markets and therefore all the agents in this group consume
all the revenue from their endowment spot by spot. There are three goods
in this economy, x, y, and z. Agents h consume only goods x and z while
agents k consume only goods y and z. We also assume that agents h have
real endowments only in goods x and z, while agents k have real endowments
only in goods y and z. In addition, agents h may have nominal endowments.
 Ambiguity aversion and the absence of indexed debt 157

Nominal endowments are any precontracted transfers, positive or negative,

between agents that are (at least partly) set in nominal terms. Examples
include, wages and salaries, house rents or even something like alimony or
child support payments. (In the U.S. alimony and child support payments are
almost never indexed, see (Shiller, 1997).)
To see the rationale of “type-casting” agents as above recall, from what
was noted in the introductory section, we want to ensure in the model that
with respect to any two agents wishing to trade in indexed bonds, it is true
that the indexation bundle contains at least one good which is not consumed
by either of the agents. This condition, of course, would not be satisfied if an
h-type agent were to trade bonds with a k-type agent. We have each type of
agent consuming two goods rather than one, unlike in the model in the pre-
vious section, so that there may be market exchange among agents, thereby
obtaining well-defined prices at equilibrium (reflecting the common utility
gradients). Informally put, the focus of the “show” will be the intertemporal
exchange between the h-type agents, with the role of k-type agents being es-
sentially that of a necessary “prop”, enabling the determination of the relative
price of the good not figuring in the consumption baskets of agents trading
bonds.
There are two periods in the model; uncertainty essentially comes into play
in the final period. The endowment of h-type agents is uncontingent, given by
((x̄0h , z̄h0 ), (x̄h , z̄h , mh )), where (x̄0h , z̄h0 ) is the endowment in the initial period,
Period 0, and (x̄h , z̄h , mh ) is the endowment in the final period, Period 1. mh
denotes the nominal endowment, so that mh
0 and since transfers should
balance across households, we have
H

mh = 0.
h=1

Note though, the endowments vary across households; this heterogeneity

is the reason why h-type agents trade intertemporal transfers. The endow-
ment of k-type agents are given by (ȳk0 , z̄k0 ) in the initial period8 . Their final
period endowment in good z is uncontingent and equal to z̄k . We assume,
though, their endowments in good y is contingent: in state t, ȳkt , are such that
K t L
K t H
k=1 ȳk = y for, say, t = 1, . . . , τ and k=1 ȳk = y for t = τ + 1, . . . , T .
Thus, in terms of total endowments, there are two “aggregate” states: one
where the total endowment of good y is low (y L ) and another, where the total
endowment of y is high (y H ). As will be seen, it is this variation in aggregate
endowment which completely determines the variation in the relative price of
y.

8
Note, h-type agents do not have nominal endowments in the initial period. k-
type agents do not have nominal endowments at all. This is just to save on notation;
introducing such endowments would not make the slightest difference to any of our
results.
158 Sujoy Mukerji and Jean-Marc Tallon

There is also (outside) money in the model, whose supply in the Period 0
is fixed at M 0 but may take on two values in the Period 1, m or M , where
M ≡ λm, λ > 1. The role of money is simply to facilitate exchange. Hence, at
each spot, we assume the standard fiction that agents sell to a central authority
all their endowments against currency issued by the central authority and then
buy back from that authority the goods they want to consume (see (Magill
and Quinzii, 1992)). The money obtained from the central authority by agent
h (respectively, k) from the sale of endowments in state s is denoted msh
(respectively, msk ).
Uncertainty in the model is exhaustively represented by the state space
S ≡ {0} ∪ {{1, ..., T } × {m, M }} ,
where, {0} refers to the only state of the world in Period 0, {1, ..., T } indexes
contingencies in Period 1 obtaining due to variation in real endowments of
agents, {m, M } indexes the variation in money supply. Let s ∈ S be an index
for states, s = 0, 1, . . . , S. We denote the (absolute) prices of goods x, y, and
z as psx , psy , and psz , respectively, in state s.
There are two financial assets in the model, traded in Period 0. The first is
a nominal bond, bn , that pays off one unit of money in all states and with its
price denoted q n . The second is an indexed bond, bi , that pays off a bundle of
goods at each state in Period 1. We take this bundle to be state-independent
and comprising of a unit of each good traded in the economy. Hence, the
monetary return to holding a unit of this indexed bond is psx + psy + psz in state
s = 1, . . . , S. Its price is denoted q i .
For the moment, denote
an agent h’s preferences by a functional
Vh (x0h , zh0 ), . . . , (xSh , zhS ) , on which we’ll impose assumptions detailed later
on. His maximization problem is hence :
M axxh ,zh ,bih ,bnh Vh (x0h , zh0 ), . . . , (xSh , zhS )

p0x x̄0h + p0z z̄h0 = m0h

⎧
⎪
p0x x0h + p0z zh0 = m0h − q i bih − q n bnh
⎪
⎨
s.t.
p x̄ + ps z̄h + m̄h = msh
s s
⎩ x h s zs
⎪
px xh + psz zhs = msh + bnh + (psx + psy + psz )bih , s = 1, . . . , S
⎪

Agents k, who have no access to financial markets, have to solve S + 1

separate maximization programs. We assume that their preferences at each
spot take the simple form of a Cobb-Douglas function: (yks )α (zks )1−α for α ∈
(0, 1). Hence, their maximization problem for s ∈ S, is:

(yks )α (zks )1−α

M axyks ,zks 
psy ȳks + psz z̄k = msk
s.t.
psy yks + psz zks = msk
An equilibrium of this model is therefore an allocation (x, y, z, m, bi , bn )
and prices (px , py , pz , q i , q n ) such that, given these prices agents solve their
maximization problems and markets clear.
 Ambiguity aversion and the absence of indexed debt 159

Observe that money as we introduced it is simply a veil and we can rewrite

the budget constraints as follows, for agent h:
 0 0
px xh + p0z zh0 = p0x x̄0h + p0z z̄h0 − q i bih − q n bnh
psx xsh + psz zhs = mh + psx x̄sh + psz z̄h + bnh + (psx + psy + psz )bih s = 1, . . . , S

and, for agent k in state s (s = 0, 1, . . . , S):

psy yks + psz zks = psy ȳks + psz z̄k

One can also use the particular structure of the model to simplify the
market clearing condition for good z. Indeed, adding the budget constraints
H H
in state s of agents h, one gets, at equilibrium, h=1 zhs = h=1 z̄h (under the
s
assumption that px > 0, which is met since preferences are assumed strictly
monotonic). Similarly, for agents k, one obtains, from adding their budget
constraints in state s and using equilibrium condition on the market for good
K K
y (plus the fact that, at an equilibrium, psz > 0), that k=1 zks = k=1 z̄k .
Thus, the market clearing conditions on the market for good z can be split in
two equalities as follows :
H
 H
 K
 K

zhs = z̄h and zks = z̄k s = 0, . . . , S
h=1 h=1 k=1 k=1

Hence, the market for good z can be “divided in two”, agents h exchanging
among themselves, and similarly for agents k. The intuition for this is fairly
obvious once one lifts the “veil of money” and considers the nature of “real”
exchange in the model. The point is, given that the two types of agents share
only one good between their respective consumption baskets, there cannot be
any “real” exchange between these groups on spot markets.
Finally, notice that the market clearing condition on the money market
can be written as
H
 K
 H
 K

psx x̄h + psy ȳks + s
pz ( z̄h + z̄k ) = M s s = 0, . . . , S
h=1 k=1 h=1 k=1

H i
while the market clearing condition on the bond markets are h=1 bh =
H n
h=1 bh = 0.

4.2 Equilibrium prices in goods markets

We can further reduce the model by noticing that only aggregate states “mat-
ter”. Indeed, note that there are two sources of (aggregate) uncertainty in this
model: one is linked to the money supply, the second stems from the random-
ness in the (aggregate) endowment in good y of agents k. As we will be only
interested in the equilibrium allocation of the h agents (and in particular
160 Sujoy Mukerji and Jean-Marc Tallon

whether they hold indexed bonds or not), the only way this last source of
uncertainty is relevant to h agents is through the effect it has on prices. Now,
observe that we can solve for the equilibrium relative price of y with respect
to z, spot by spot. Indeed, agents k demand functions are easily computed
and are equal to:
psy ȳks + psz z̄k psy ȳks + psz z̄k
yks (psy , psz ) = α and zk
s s
(py , ps
z ) = (1 − α)
psy psz
Hence, at equilibrium,
K
psy α k=1 z̄k
=
psz 1−α K s

k=1 ȳk
and therefore, the ratio of the prices psy and psz depends only on the aggregate
(among k agents) endowments of good y and z, and thus, can take on only
two values, whether aggregate endowment in y is high (y H ) or low (y L ). Note
that the price levels do depend on the money supply. To sum up, we need, for
our purposes, concentrate only on four states ω ∈ Ω ≡ {1, 2, 3, 4} defined as
follows9 :
ω = 1 : low money supply (m) & low aggregate y-endowment (y L )
ω = 2 : high money supply (M ) & low aggregate y-endowment (y L )
ω = 3 : low money supply (m) & high aggregate y-endowment (y H )
ω = 4 : high money supply (M ) & high aggregate y-endowment (y H )
We now describe agent’s h preferences, following partly a specification
due to (Magill and Quinzii, 1997). At state 0, h-type agents’ utility function
is written as u(x0h , zh0 ), where u : R+ × R+ → R is strictly increasing, con-
cave, differentiable and homogeneous of degree 1. In state ω = 1, 2, 3, 4, the
spot utility function of agent h is given by fh (u(xω ω
h , zh )), where fh : R → R
is strictly increasing, strictly concave and differentiable, and u is as defined
above. The linear homogeneity of u will facilitate the tractability of the equi-
librium (contingent) price function (essentially, the assumption ensures that
goods prices, in each spot market, are independent of the distribution of wealth
among agents) while the concavity of fh is a simple way of endowing agents
with risk aversion as well as a desire to smooth consumption across periods.
Type-h agents are endowed with (common) beliefs about the money sup-
ply process and the process generating the (aggregate) y-endowments. The
capacity µ = (µm , µM ) denotes their (marginal) belief about the money sup-
ply: µm is the (possibly non-additive) probability the money supply is m and
µM is the probability that it is M ; µm + µM ≤ 1. Their (marginal) beliefs on
the process generating the y-endowments (a good they do not consume and
are not endowed with) are represented10 by ν = (ν L , ν H ), with ν L + ν H ≤ 1.
9
We denote these aggregate states by ω to distinguish them from the underly-
ing states s which encompass some heterogeneity among k-type endowments. We’ll
continue to use the notation ω = 0 to denote the first period.
10
Note that, ν H refers now to the state with high y-endowment, and therefore
low py , while it referred to the high py in the previous section.
 Ambiguity aversion and the absence of indexed debt 161

The overall beliefs on Ω is given by the independent product µ⊗ν. Preferences

of agent h are thus represented by the functional Vh , where,

Vh (x0h , zh0 ), . . . , (x4h , zh4 ) ≡ u(x0h , zh0 ) + CEµ⊗ν fh (u(xω ω

h , zh )) .

Under our assumptions (essentially, the linear homogeneity of u) at an

equilibrium, the maximization problem of each h agent can be decomposed to
separate the financial decision about the allocation of resources across states
from the decision about the consumption mix at each state. Turning first to
the latter, given a stream (Rh0 , Rh1 , . . . , Rh4 ) of income with Rh0 = p0x x̄0h +p0z z̄h0 −
q i bih − q n bnh and Rhω = pω ω ω ω ω ω i n
x x̄h + pz z̄h + (px + py + pz )bh + bh + m̄h , ω = 1, . . . , 4,
the agent solves the problem :

M axxωh ,zhω u(xω ω

h , zh )
s.t. pω ω ω ω ω
x xh + pz zh = Rh

At the optimal choice, one gets for all ω = 0, 1, ..., 4 :

∇1 u(xω ω
h , zh ) pω
x
ω ω = ω
∇2 u(xh , zh ) pz

where ∇i u(xω ω
h , zh ) is the derivative of u with respect to its i
th
component.
By homogeneity of degree one, the gradients are collinear among agents only
if their consumption vectors are collinear as well. Recall, at an equilibrium
agents h only trade among themselves and do not trade with the k agents.
Hence, each agent h’s consumption in state ω is a fraction αhω of total endow-
ment of h-agents with

p0x x̄h + p0z z̄h − q i bih − q n bnh

αh0 = H H
p0x h=1 x̄0h + p0z h=1 z̄h0
pω ω ω ω ω i n
x x̄h + pz z̄h + (px + py + pz )bh + bh + m̄h
αhω = H H
, ω = 1, ..., 4
pω ω

x h=1 x̄h + pz h=1 z̄h

Hence, at an equilibrium,
H H
 
 
(xω ω
h , zh ) = αhω x̄h , z̄h
h=1 h=1

Therefore, agent h’s utility, at an equilibrium, can be rewritten in state ω =

1, ..., 4, u (xω ω ω
h , zh ) = αh u, where u is simply the utility at the endowment
H H
point, i.e., u ≡ u h=1 x̄h , h=1 z̄h . Observe that the same holds in the
 
H H
first period, i.e., u xh , zh = αh0 u0 , with u0 ≡ u
0 0
0 0
x̄
h=1 h , h=1 h .
z̄
Finally, since relative prices of good x and z in state ω are equal to the
gradient of an agent h’s utility function, it is easy to see that
162 Sujoy Mukerji and Jean-Marc Tallon

p1x p2 p3 p4
1
= x2 = x3 = x4 ≡ ζ
pz pz pz pz

given that endowments of goods x and z are constant across states. In fact,
it turns out that, the absolute price of goods x and z do not depend on the
amount of good y available in the economy. In other words, and since there is
no uncertainty on the total endowments of goods x and z, their price depends
only on the money supply. This is the content of Proposition 1, below, which
is proved in the appendix. A direct corollary is that the price of y, conditional
on the monetary state, is completely determined by the aggregate endowment
in y. Proposition 2 (proved in the appendix) which essentially shows that
monetary equilibrium requires the price vector in state 2 (respectively, 4) is
simply a λ-multiple of prices in state 1 (respectively, 3), completes the required
characterization of equilibrium prices.

Proposition 1. At an equilibrium, p1x = p3x , p2x = p4x , p1z = p3z , and p2z = p4z .
p1y
From this proposition, it is easy to show that, at an equilibrium, p3y =
yH p2y p1y p3y yH
yL
= p4y (this follows from the fact that p1z = p3z yL
and p1z = p3z ).

Proposition 2. At an equilibrium, (p1x , p1y , p1z ) = λ1 (p2x , p2y , p2z ) and

(p3x , p3y , p3z ) = λ1 (p4x , p4y , p4z ).

The following table, then, summarizes the equilibrium prices, and the cor-
responding return from an unit of an indexed bond at each state ω, ω ∈ Ω.
The reader will recall the table is identical (but for price of the good z) to the
one presented in Section 3.

State ω Prices
Return from an indexed bond
H
1 px , pyH, pz
p x + pHy + pz

2 λp , λp
x Ly
z, λp λ × p x + pH
y + pz
L
3 px , pyL, pz
p x + py + pz

4 λpx , λpy , λpz λ × px + pL y + pz

4.3 The nature of equilibrium in bond markets

We now turn to the intertemporal maximization problem of agent h and derive

the principal formal conclusions of our analysis. The first result, which is in
the nature of a benchmark, shows that, for generic endowments, if beliefs
are not ambiguous, there will always be trade in indexed bonds. The second
and main result shows that, at equilibrium, if ambiguity of belief about the
y−prices (A (ν) ≡ 1−ν L −ν H ) is large enough and inflation risk (λ) is not too
high, the indexed bond is not traded and only the nominal bond is traded. As
we show, this result holds irrespective of the degree of ambiguity about the
money supply. In what follows, we first explain an intuition of the equilibrium
 Ambiguity aversion and the absence of indexed debt 163

reasoning underlying the results and then state the theorems, with the formal
proofs appearing in the appendix.
We begin by considering the nature of the equilibrium in the indexed bond
market at two values of λ, λ = 1 and λ = 1 + ε, where ε is a positive number
arbitrarily close to 0. Consider, first, the case wherein λ = 1. Without any
inflation risk at all, clearly, all borrowing and lending will be done through
nominal bonds, at equilibrium. Take two h-type agents, h′ and h who save
and borrow, respectively, in the initial period. Their utility
 in the final period,

Sh′
with slight abuse of the notation, may be written as fh′ u(x̄h′ , z̄h′ ; λq n) and
 
Sh
fh u(x̄h , z̄h ; λqn ) , where Sh′ > 0 and Sh < 0 denote the amount saved and
the amount borrowed
 by h′ and  h respectively. Define
 the “marginalutilities”,
′ ′ Sh ′ ′ Sh′
Uh (λ) ≡ fh u(x̄h′ , z̄h′ ; λqn ) and Uh′ (λ) ≡ fh′ u(x̄h′ , z̄h′ ; λq n) . Notice,
Uh′ ′ (λ) ↑ in λ while Uh′ (λ) ↓ in λ, since Sh′ > 0 and Sh < 0; intuitively,
inflation affects the welfare of savers and borrowers differently. Furthermore,
it must be necessarily true at an equilibrium that Uh′ (λ = 1) = Uh′ ′ (λ = 1) .
This is so since if there is no inflation risk, h-agents are effectively trading in a
complete market when trading only in nominal bond, and thus the equilibrium
is Pareto optimal.
First suppose, ceteris paribus, there were no ambiguity, i.e., 1−ν L −ν H = 0
and 1−µm −µM = 0, so that the DM’s behavior were that of an SEU agent. At
λ = 1, the “utility return” from an (infinitesimal) unit of an indexed
bond
at
equilibrium must be Uh′ (λ = 1) × E px + pω ′ ω

y = Uh′ (λ = 1) × E p x + p y ≡
i ′
q (λ = 1; h, h ). Putting it differently,
q i (λ = 1; h, h′ ) is the price at which the agents (h and h′ ) are indifferent
between not trading and trading an infinitesimal amount of indexed bonds.
Similarly, for an arbitrary λ, define q i (λ; h) (respectively, q i (λ; h′ )) as the
minimum (maximum) price h (h′ ) is willing to accept (pay) to trade in the
indexed bond. Next, consider a perturbation of λ to λ = 1 + ε. Recalling the
effect of a change in λ on Uh′ (λ) and Uh′ ′ (λ), it is straightforward to see that

q i (λ = 1 + ε; h′ ) ≡ Uh′ ′ (λ = 1 + ε) × E px + pω

> q i (λ = 1; h, h′ )
> Uh′ (λ = 1) × E px + pω i

y ≥ q (λ = 1 + ε; h) .

Intuitively, since the saver is affected adversely by inflation, relative to the

debtor, the indexed bond is more valuable to the saver in the presence of
inflation, and also, more valuable than it is to the debtor. Hence, inevitably,
with inflation risk creeping up there will be gains from trading in indexed
bonds and indexed bonds will be traded at equilibrium under SEU. This is
depicted in Figure 2, below, in the left-hand-side panel and formally stated in
our first theorem11 .
11
Both theorems refer to properties that hold “generically”. The term is applied
in a way that is now standard in economic theory. Notice, endowments are points in
164 Sujoy Mukerji and Jean-Marc Tallon

Theorem 1. Suppose, µm + µM = 1 and ν L + ν H = 1. Then, for generic first

period aggregate endowments, there is trade in the indexed bond whenever
λ = 1.

Figure 2. Equilibrium in the indexed bond market

Next suppose, agents have CEU preferences and beliefs about the y-prices
are ambiguous, i.e., 1 − ν L − ν H > 0. As before, we first consider the equilib-
rium at λ = 1. As would be evident from our discussion in Section 3, there
would exist a portfolio inertia interval: there will be a bid price correspond-
ing to (perceived) marginal gain from moving (infinitesimally) into a indexed
bond, and an ask price, that is strictly lower, corresponding to the (perceived)
marginal gain from going short on the indexed bond. The bid-ask interval will
correspond to an interval of expected marginal utilities, where the lower end of
the interval is evaluated by applying the probability measure that minimizes
the expectation for an agent going long and the upper end is evaluated by
applying the probability measure that minimizes the expectation for an agent
going short:
Next, consider the equilibrium given the perturbation λ = 1 + ε. Noting
again the effect of the perturbation on Uh′ (λ) and Uh′ ′ (λ), it is straightforward
to see that for h′ , the saver, the entire interval moves up, whereas for h, the
debtor, the entire interval moves down. The extent of movement is greater,
the greater increase in λ. Hence, for λ > 1, but small enough, the intervals
overlap and the bid price of the saver remains (strictly) lower than the ask
price required by the lender:

R6 and the µ-beliefs are simply points on the 2-dimensional simplex. We say that a
property is satisfied for generic endowments (respectively, µ′ s) if, for every endow-
ment (respectively, µ) vector there is an open dense neighborhood of endowment
(respectively, µ) vectors that generate economies that satisfy this property. Thus,
if a property is satisfied for all generic endowments (respectively, µ′ s), small per-
turbations of the endowment (respectively, µ) in any economy can generate a new
economy that satisfies this property robustly, even if the original economy does not.
 Ambiguity aversion and the absence of indexed debt 165

q i (1 + ε; h′ ) ≡ Uh′ ′ (1 + ε) × E px + pω ′ ω

y < Uh (1 + ε) × Ē px + py

≡ q̂ i (1 + ε; h)

This is represented in Figure 2 in the right-hand side panel. It is also evident

in the figure that if the increase in λ were large enough then the intervals
move apart enough not to overlap. Hence, we have, for λ small enough, there
is no trade in indexed bonds at equilibrium. This is formally stated in the
theorem below.
Finally, recalling Remark 2, if there is at least one agent with a positive
nominal endowment (and hence, at least one other agent with a negative nom-
inal endowment) it follows that there is at least a pair of agents who do not
optimally choose a zero-holding of the nominal bond over a non-degenerate
interval of relative bond prices. Hence, for these agents the situation is no dif-
ferent from the SEU case described earlier (and depicted on the left-hand-side
panel of Figure 2). “Typically” such agents would want to trade in nominal
bonds. The discussion is summarized by the two parts of Theorem 2, below.

Theorem 2. Suppose, µm + µM ≤ 1 and ν L + ν H < 1. Then, there exists a

bound δ, δ > 1, such that, if λ < δ, there exists γ, 0 < γ < 1, such that if
A (ν) > γ, then at an equilibrium,
(a) the indexed bond is not traded, i.e., bih = 0 for all h,
(b) for generic µ-beliefs, there is trade in the nominal bond as long as there
exists an h such that m̄h > 0.

Remark 4. Notice, both parts of the theorem hold regardless of the level of
ambiguity w.r.t. the µ-beliefs. Indeed, the formal proof is a lot shorter (and
simpler) if one were to assume that µ-beliefs were unambiguous. We, however,
do not impose this restriction in the analysis so that one may obtain a more
informed idea of the nature of robustness of the result. Of course, nominal
endowments would no longer play a role in the argument if µ-beliefs were
assumed to be unambiguous.

Remark 5. Parameter λ refers to inflationary risk: recall that it is equal to

the ratio M/m, that is, the relative inflation in the “high money supply” state
with respect to the “low money supply state”. Since money is not a store of
value in this model, the only risk related to second period money supply is
indeed that of differential inflation in the various states of the world, which
is precisely what is picked up by λ. Thus, the theorem states that if this risk
is sufficiently small then, ambiguity about relative price movement prevents
trade in indexed assets.

Remark 6. The proof of the theorem, in the Appendix, provides an explicit

characterization of the bound δ in terms of the fundamentals of the model. As
would have been intuitively evident from the discussion preceding the theorem,
the degree to which inflation risk would be “tolerated” before indexed bonds
166 Sujoy Mukerji and Jean-Marc Tallon

are traded depends (positively) on the “variability” of the relative prices. For
2+ H x̄ /y L

instance, the δ is equal to h=1 h H when all agents ( h and k) have
2+ Hh=1 h /y
x̄
Cobb-Douglas (.5, .5) as their respective u (., .) functions.
Remark 7. The logic underlying the result in Theorem 2(b) is actually instruc-
tive, indirectly, as to why ambiguity about the price movements of goods not
in the( h-agents’) consumption basket was the crucial factor in obtaining no-
trade in indexed bonds. Putting it differently, if we allowed the absolute prices
of say, x, to vary in response to supply shocks and assumed agents had am-
biguous beliefs about such price movements, that would not obtain the no-trade
in indexed bonds (without ambiguity about price of y). The reasoning here is
analogous to the one showing that the presence of nominal endowments pre-
cludes no-trade in nominal bonds. Since x is present in the endowment and/or
affects utility directly (of h-agents), maintaining a zero position on the in-
dexed bond would not get rid of the risk due to the variability of the price of
x. Consequently, the ordering of states would not switch at the zero holding
position. This is why we need the “prop” agents in the model: they are the
ones who are the source of volatility of the price of good y, which is the cru-
cial factor underlying the result. If we dropped these prop agents we would be
taking away the y-good and therefore, the risk in an indexed asset orthogonal
to the asset traders’ endowment and consumption. If the asset did not contain
this idiosyncratic risk, there would be trade in the two bonds, much like in a
SEU economy (for further clarification see the discussion related to Figure 1
on page 887 and Example 2 in Mukerji-Tallon (2001)).
Remark 8. In the real world inflation and relative price movements are cor-
related. Recall, though, the model assumes that the processes generating the
money supply and the real shocks (to the aggregate endowment of y) are be-
lieved to be independent. However, what is really crucial is that the process
generating py must have at least one component that is believed to be orthog-
onal to the money supply: i.e., there has to be at least two states of the world
across which the price of y changes even though the money supply stays con-
stant. In other words, for the reasoning to hold, prices cannot be perfectly
correlated with the money supply. Similarly, the logic behind our result does
not require that py and px (or pz ) have to be independent, but that they are
not perfectly correlated.

5 Concluding discussion
The discussion in this section is in two parts. In the first part we relate the
results obtained in this paper to those in the relevant (theoretical) literature.
The second part considers to what extent the analysis here may be thought to
explain the empirics of trade in indexed debt any more than the explanations
already advanced. Implications of the results in terms of policy required to
encourage trade in indexed bonds is also discussed.
 Ambiguity aversion and the absence of indexed debt 167

5.1 Related (theoretical) literature

We begin by linking the result here to the findings in the literature applying
ambiguity aversion to financial markets. Then we discuss the theoretical lit-
erature in the standard Savage paradigm that seeks to explain observed facts
relating to trade in indexed debt.
As has already been noted, (Dow and Werlang, 1992) showed that a zero
position may be held on a price interval if the agent’s endowments were risk-
less. Obviously, an economy where all agents’ endowments were unvarying
across all states the question of asset trading and risk sharing is an uninter-
esting question. (Epstein and Wang, 1994) significantly generalized the (Dow
and Werlang, 1992) result to find that price intervals supporting the zero po-
sition occurred (in equilibrium) if there were some states across which asset
payoffs differ while endowments remain identical; in other words, asset pay-
offs have component of idiosyncratic risk. However, the focus of (Epstein and
Wang, 1994) was the issue of asset pricing. In their model endowments are
Pareto optimal, and consequently, the issue of whether ambiguity aversion
cause assets not to be traded is not examined. (Mukerji and Tallon, 2001),
building on the results in the two papers cited, finds conditions for an econ-
omy wherein the agents’ price intervals overlap in such a manner such that
every equilibrium of the economy involves no trade in an asset, and more
importantly, conditions under which ambiguity aversion demonstrably “wors-
ens” risk sharing and incompleteness of markets. One of the conditions, the
presence of idiosyncratic risk, identified in (Mukerji and Tallon, 2001), is es-
sentially the same as in the result of (Epstein and Wang, 1994) explained
above. As has been suggested, it is possible to see that, for h-type agents,
payoffs of indexed bonds contain an element of idiosyncratic risk derived from
the risk inherent in the relative price of y. This is, essentially, how the finding
in this paper links up with the results in the papers cited above.
The paper closest to ours, within the Savage paradigm, which seeks to
explain the lack of indexed debt is (Magill and Quinzii, 1997). That paper
compares the welfare improvements obtained from introducing within an in-
complete markets setting, in turn, a nominal bond and an indexed bond.
The welfare improvements derive from, essentially, the increase in the span of
available assets (or, in other words, the “lessening” of incompleteness) that
comes about due to the introduction of each type of bond. The more relevant
result is that the welfare gain from introducing the indexed bond may be less
(respectively, more) than that from introducing a nominal bond if the infla-
tion risk was “small” (respectively, large) compared to the relative price risk.
In contrast to the analysis in this paper, (Magill and Quinzii, 1997) does not
actually obtain a equilibrium with no-trade in indexed bonds; indeed, as we
confirm in Theorem 1, Savage rational agents will necessarily trade in indexed
bonds as long as there is some inflation. Also, (Magill and Quinzii, 1997) do
not allow both indexed and nominal bonds to be available for trade simulta-
neously; one or the other is available.
168 Sujoy Mukerji and Jean-Marc Tallon

Another paper, within the Savage paradigm, which studies a similar ques-
tion, but in a rather different framework is (Freeman and Tabellini, 1998).
The framework in that paper is an overlapping generations model. The paper
finds that nominal contracts are optimal in the presence of price level shocks
if all the contracting parties have the same degree of constant relative risk
aversion. With respect to relative price shocks, the optimality of nominal con-
tracts follows from the fact that, in their model, agents cannot insure each
other against this risk as they all are ex ante identical with respect to such
shocks.
The present paper also complements the finding in Mukerji and Tallon
(2004) which shows that ambiguity aversion (with CEU preferences) may
help to explain why we see so little wage indexation. Among other things,
the framework in that paper is one of bilateral contracting, not a general
equilibrium market environment like it is here. Hence, the result there does
not follow from the result here even. However, as we understand it, the same
intuition explains both results, a point that is significant in so far as it shows
that the intuition is robust across seemingly different trading environments.
Finally, the paper adds to the growing literature on the economics applications
of the idea of ambiguity aversion (see Mukerji and Tallon (2003b) for a survey).

5.2 Explaining the empirics of trade in indexed debt

Recall the intuition underlying the main result. Taking a long or a short
position on the indexed bond implies betting on or against the (ambiguous)
event wherein the (relative) price of good y will be high. To decide whether
to bet on or against a particular event one has to reach a fine judgement
about the relative likelihood of the event compared to its complement. Hence,
the attraction of the zero holding position to the ambiguity averse agent.
Moving from the zero position, in either direction, requires a compensating
“ambiguity premium”. Hence, the portfolio inertia intervals for the indexed
bond. At low levels of inflation the bid-ask intervals of the borrower and the
saver overlap and agents only trade in the nominal bond. As inflation rises,
the saver is affected adversely while the borrower is made better off. As a
consequence, the most the saver is willing to pay for the indexed bond goes
up and the minimum the borrower would ask decreases. Hence if inflation
were high enough, agents do trade indexed bonds. We also argued that, so
long as agents held (non-zero) nominal endowments, this reasoning does not
apply quite symmetrically to trade in nominal bonds. For instance, with a
positive holding of the nominal bond, one would be betting on the event
that (average) price level will be low. But, if one had, say, positive nominal
endowments, moving to a zero holding would still imply one is betting in favor
of the event that the price level will be low. Indeed, this would continue to
be true even if one were to move marginally into a negative holding of the
nominal bond. Hence, the portfolio inertia interval no longer occurs at the zero
holding and trade occurs. Of course, by the same token, if some endowments
 Ambiguity aversion and the absence of indexed debt 169

were indexed, the argument for no trade in indexed bonds will be affected
similarly.
Thus, according to the theory presented in this paper, it is the comparative
lack of information about relative, as opposed to average, price movements,
the comparative preponderance of nominal, as opposed to indexed, endow-
ments that explains why trade in indexed bonds is observed only in excep-
tional circumstances but trade in nominal bonds is so widespread. While these
are testable hypotheses that could provide the basis for a specific empirical
investigation, that is a matter for future research. However, a lot that we do
know about trade in indexed bonds is broadly consistent with the theory. Ar-
guably, the theory is very consistent with the fact that typically indexed bonds
are traded almost exclusively under extreme inflationary circumstances. Also,
while trade in indexed bonds is negligible in most non-inflationary economies,
it is more than negligible (though still quite small) in the few such economies
where, in addition, there are some instances of indexed endowments, statu-
tory wage indexation, as is the case, for example, in the U.K. and in Israel
(statutory wage indexation in a limited number of sectors of the economy).
Indeed, the reasoning predicts that one would observe a kind of hysteresis
in the market for indexed bonds. In economies with inflationary past, where
indexed bonds were traded when high inflation reigned, indexed bonds would
continue to be traded even after inflation has been brought down to moder-
ate levels because of the presence of indexed bonds as endowments. Perhaps,
this explains the continued trade in indexed financial instruments observed in
some South American economies (and even Israel) where inflation has lately
been tamed. One may also argue that an analogous reasoning explains why
in countries, like Turkey, where use of dollar is widespread in spot market
transactions, so is the use of dollar-indexed debt.
The theory presented, strictly interpreted, demonstrates a reason why in-
dexed bonds are not exchanged by private individuals. But in the case of
trade in government bonds, at least at the point of issue, one of the parties to
the trade is not a utility maximizing private agent. However, note our theory
would apply just as well to any secondary trade of indexed government bonds
between private individuals. Thus our theory predicts, this secondary mar-
ket typically be a rather thin market. In turn, this carries the implication of
indexed government debt not being a particularly liquid asset. Clearly, given
rational, forward looking individuals, this would, in itself, ensure that demand
for such debt would be weak even at the point of issue.
Finally, we turn briefly to some policy implications. One obviously welfare
increasing move would be to publish (trustworthy) indexes that are more
particular and focused on fewer goods and services than the CPI. Looking back
to the formal model, if there were an index composed purely of the prices of x
and z the resulting allocation would indeed be a Pareto improvement on the
allocation obtained with index composed of prices of x, z and y (the market
would become as good as complete for the h-type agents). Government action
in introducing statutory indexation of some payments, say of wages in some
170 Sujoy Mukerji and Jean-Marc Tallon

sectors, would increase the trade in indexed debt. However, it is far from clear,
given the abstractions in our model, that we may conclude that issuing such
a fiat would at all be welfare enhancing.

6 Appendix
6.1 Slice-comonotonicity
The computation of the Choquet expectation operator using product capaci-
ties is particularly simple for slice comonotonic functions ((Ghirardato, 1997)),
defined below. Let X1 , ..., Xn be n (finite) sets and let Ω = X1 × ... × Xn . Cor-
respondingly, let νi be convex non-additive probabilities defined on algebras
of subsets of Xi , i = 1, ..., n.
Definition 2. Let ϕ : Ω → R. We say that ϕ has comonotonic
xi -sections
if for every (x1 , ..., xi−1 , xi+1 , ..., xn ) , x′1 , ..., x′i−1 , x′i+1 , ..., x′n ∈ X1 × ... ×
X i−1 × Xi+1 × ... × Xn , ϕ (x
1 , ..., xi−1 , ·, xi+1 , ..., xn ) : Xi → R, and
ϕ x′1 , ..., x′i−1 , ·, x′i+1 , ..., x′n : Xi → R are comonotonic functions. ϕ is called
slice-comonotonic if it has comonotonic xi -sections for every i ∈ {1, ..., n}.
The utility function of agent h, fh (u(xω ω
h , zh )), is actually slice comonotonic,
since u is strictly monotone and hence Fact 1, below, which follows from
Proposition 7 and Theorem 1 in (Ghirardato, 1997), applies to the calculation
of Choquet expected utility for agent h.
Fact 1 Suppose that ϕ : Ω → R is slice comonotonic. Then
  
ϕ (x1 , ..., xn ) d (⊗νi ) = ... ϕ (x1 , ..., xn ) dνn ...dν1
Ω X1 Xn

6.2 A formal confirmation of the argument in Section 3

To see the argument put more formally, compute first the right-hand-side
derivative of V (x1 , x2 , x3 , x4 ), in which we replace xs by its expression as a
function of bi , to obtain a function W (bi ) :
dW (bi )
| bi =0+ =
dbi     
H
pH
y qi H
pL
y qi
µ ν 1+ − n + (1 − ν ) 1 + − n ×
px q px px q px
 
′ S
u x̄ + n +
q px
    
H
pH
y qi H
pLy qi
(1 − µ) ν 1+ − n + (1 − ν ) 1 + − n ×
px q λpx px q λpx
 
S
u′ x̄ + n
q λpx
 Ambiguity aversion and the absence of indexed debt 171

The same computation, but for the left-hand-side derivative yields :

dW (bi )
| bi =0− =
dbi     
L
pH
y qi L
pL
y qi
µ (1 − ν ) 1 + − n +ν 1+ − n ×
px q px px q px
 
′ S
u x̄ + n +
q px
    
H i L i
py q p y q
(1 − µ) (1 − ν L ) 1 + − n + νL 1 + − n ×
px q λpx px q λpx
 
S
u′ x̄ + n
q λpx

Hence, if the following conditions hold, then bi = 0 is optimal for the agent:
   
pH pL
i

µ 1 − qnqpx + ν H pyx + (1 − ν H ) pyx u′ x̄ + qnSpx +
   
pH pL
i

(1 − µ) 1 − qnqλpx + ν H pyx + (1 − ν H ) pyx u′ x̄ + qnSλpx

≤0≤
   
pH pL
i

µ 1 − qnqpx + (1 − ν L ) pyx + ν L pyx u′ x̄ + qnSpx +
   
pH pL
i

(1 − µ) 1 − qnqλpx + (1 − ν L ) pyx + ν L pyx u′ x̄ + qnSλpx
172 Sujoy Mukerji and Jean-Marc Tallon

This amounts to:

pH pL
νH y
px + (1 − ν H ) pyx ≤
i i
µ qnqpx u′ (x̄+ qnSpx )+(1−µ) qnqλpx u′ (x̄+ qn S
λpx )
−1
(
µu′ x̄+ qnSpx )+(1−µ)u′ (x̄+ qnSλpx )
pH pL
≤ (1 − ν L ) pyx + ν L pyx

If one looks at the limiting case in which λ = 1, this further reduces to :

qi
ν H pH H L
y + (1 − ν )py + px ≤ ≤ (1 − ν L )pH L L
y + ν py + px
qn
where it is easily seen that there is a range of relative prices for the indexed
bond with respect to the nominal bond such that it is not held in the optimal
portfolio, as long as ν L + ν H < 1. If, on the other hand, the capacity ν
is actually a probability measure ν L + ν H = 1, there is only one relative
price q i /q n such that the agent does not want to hold any indexed bond. By
continuity, the same is true when λ is strictly greater than 1. Observe that the
length of the interval at which the agent does not want to hold any position
in the indexed bond is increasing in the ambiguity of the beliefs, measured by
1 − νL − νH .

6.3 Proofs of results in Section 4

Proof of Proposition 1.
p1 p3
Proof. Since px1 = px3 , there exists β such that (p1x , p1z ) = β1 (p3x , p3z ). We want
z z
to show that β = 1. From the equilibrium condition on the money market and
the definition of the states (which entails that M 1 = M 3 ), we have that :
H
H K

  
1 1 L 1
px x̄h + py y + pz z̄h + z̄k
h=1 h=1 k=1
H H K
 
  
= p3x x̄h + p3y y H + p3z z̄h + z̄k
h=1 h=1 k=1

Replacing p3x and p3z by βp1x and βp1z respectively, in the equation above, and
recalling that, at equilibrium,

p1y p3y y H
=
p1z p3z y L
and hence,
p3y p1y y L
= β
p1z p1z y H
 Ambiguity aversion and the absence of indexed debt 173

one gets :
H
H K

p1x  p1y  
(1 − β) x̄h + 1 (1 − β)y L + (1 − β) z̄h + z̄k = 0
p1z pz
h=1 h=1 k=1

Hence, β = 1. The same reasoning shows that p2x = p4x and p2z = p4z .

Proof of Proposition 2
p1 p2 p1 p2y
Proof. Recall that px1 = ζ = px2 and py1 = p2z . Hence, (p1x , p1y , p1z ) and
z z z
(p2x , p2y , p2z ) are proportional. Furthermore,
H H K
 
  
p1x x̄h + p1y y L + p1z z̄h + z̄k =m
h=1 h=1 k=1

and
H H K
 
  
p2x x̄h + p2y y L + p2z z̄h + z̄k =M
h=1 h=1 k=1

and hence
1 2 2 2
(p1x , p1y , p1z ) =
(p , p , p )
λ x y z
An analogous argument holds for states 3 and 4.

Notation 1 To simplify notation, the remaining proofs apply the terms

αhω (bih , bnh ) and ξhω (bih , bnh ) (that we will write as αhω and ξhω , respectively, for
the sake of simplicity) defined as follows:

pω ω ω ω ω i
x x̄h + pz z̄h + (px + py + pz )bh + bh + m̄h
n
αhω (bih , bnh ) ≡ H H
,
pω ω

x h=1 x̄h + pz h=1 z̄h

fh′ αhω (bih , bnh )u

ω i n
ξh (bh , bh ) = H H u.
pω
x h=1 x̄h + pz
ω
h=1 z̄h

Lemma 1. At an equilibrium, if bih > 0, then αh1 > αh3 and αh2 > αh4 . If
bih < 0, then αh1 < αh3 and αh2 < αh4 .

Proof. By definition of αhω and since p1x = p3x and p1z = p3z , and the endowment
in x and z are non random, one has:

(p1y − p3y )bih

αh1 − αh3 = H H
px 1 1
h=1 x̄h + pz h=1 z̄h

Hence, the sign of αh1 − αh3 is the same as the sign of bih since p1y > p3y (recall
that p1y /p3y = y H /y L > 1).
174 Sujoy Mukerji and Jean-Marc Tallon

Proof of Theorem 1
Proof. (Sketch.) Since the hypothesis of this theorem is that µm + µM = 1
and ν H + ν L = 1, we may write the proof by setting µm ≡ µ, µM ≡ 1 − µ and
ν H = ν, ν L = 1 − ν.
Write down the f.o.c. to agent h’s program:
−q i u0 + µ(1 − ν)ξh1 (bih , bnh )(p1x + p1y + p1z )+
(1 − µ)(1 − ν)ξh2 (bih , bnh )(p2x + p2y + p2z )
+µνξh3 (bih , bnh )(p3x + p3y + p3z )+
(1 − µ)νξh4 (bih , bnh )(p4x + p4y + p4z ) = 0

−q n u0 + µ(1 − ν)ξh1 (bih , bnh ) + (1 − µ)(1 − ν)ξh2 (bih , bnh )

+µνξh3 (bih , bnh ) + (1 − µ)νξh4 (bih , bnh ) = 0

Recall that (p2x , p2y , p2z ) = λ(p1x , p1y , p1z ), (p3x , p3y , p3z ) = λ(p4x , p4y , p4z ), p1x = p3x ,
p1z
= p3z , p2x = p4x , p2z = p4z , and p1y /p3y = y H /y L = p2y /p4y .
Observe that ξh1 (0, bnh ) = ξh3 (0, bnh ) and ξh2 (0, bnh ) = ξh4 (0, bnh ). Hence, at an
equilibrium, if bih = 0 for all h, it must be the case that:
−q i u0 + µ(1 − ν)ξh1 (0, bnh )(p1x + p1y + p1z )+
(1 − µ)(1 − ν)ξh2 (0, bnh )λ(p1x + p1y + p1z )
+µνξh1 (0, bnh )(p1x + p1z + p1y y L /y H )+
(1 − µ)νξh2 (0, bnh )λ(p2x + p2z + p2y y L /y H ) = 0

−q n u0 + µξh1 (0, bnh ) + (1 − µ)ξh2 (0, bnh ) = 0

that is
−q i u0 + [p1x + p1z + p1y (1 − ν + νy L /y H )][µξh1 (0, bnh ) + (1 − µ)λξh2 (0, bnh )] = 0
−q n u0 + µξh1 (0, bnh ) + (1 − µ)ξh2 (0, bnh ) = 0
One can see straight away on these two equations that, unless λ = 1, there
is “little chance” that there exists (q i , q n ) and (bnh )h=1,...,H such that they hold
for all h. The following argument sketches how we might put this formally.
We define a mapping φ from RH+3 to R2H+1 as follows:
φ(bn1 , . . . , bnH , q i , q n , u0 ) =

..
⎡ ⎤
.
q i u0
⎢ ⎥
⎢ µξ 1 (0, bn ) + (1 − µ)λξ 2 (0, bn ) − ⎥
⎢ h h h h 1 1 1 L H
px +pz +py (1−ν+νy /y ) ⎥
µξh1 (0, bnh ) + (1 − µ)λξh2 (0, bnh ) − q n u0
⎢ ⎥
⎢ ⎥
..
⎢ ⎥
⎢ ⎥
⎣ . ⎦
H n
h=1 h b
Observe, the equilibria with no trade in the indexed bond are the zeros of this
function. The argument then runs as follows. Note that the Jacobian of φ, a
 Ambiguity aversion and the absence of indexed debt 175

(2H + 1) × (H + 3) matrix, has full rank. Hence, by a transversality argument,

we may conclude that for generic u0 , the rank of the matrix is full. Now,
this implies that the system does not have a solution, since “there are more
equations than unknowns”. Hence, bih = 0 for all h cannot be an equilibrium
for generic first period aggregate endowments (the latter is sufficient to change
u0 ).

Proof of Theorem 2(a)

Proof. Now, to write down the Choquet integral w.r.t. an agent’s portfolio,
one has to consider all possible cases (drop subscript h), whether bi > 0 or
bi < 0 and whether bn + m̄ > 0 or bn + m̄ < 0. Observe that:

bi > 0 ⇒ α1 > α3 α2 > α4

bi < 0 ⇒ α1 < α3 α2 < α4
bn + m̄ > 0 ⇒ α1 > α2 α3 > α4
bn + m̄ < 0 ⇒ α1 < α2 α3 < α4

For each case, there are two possible orders, but these two orders give the same
(probabilistic) “decision weights”. For instance, if bi > 0 and bn + m̄ > 0, the
two following orders are possible: α1 > α2 > α3 > α4 or α1 > α3 > α2 > α4 .
If one computes the Choquet integral in these two cases, one sees that they
take the same form, i.e. the switch in the “middle position” has no effect.
Decision weights associated to the different cases:

ω=1 ω=2
m
bi > 0, bn +m̄ > 0 ν L µm ν L (1 − µ )
n M
bi > 0, b +m̄ < 0 ν L (1 − µ ) ν L µM
i n H m H m
b < 0, b +m̄ > 0 (1 − ν )µ (1 − ν )(1 − µ )
i n H M H M
b < 0, b +m̄ < 0 (1 − ν )(1 − µ ) (1 − ν )µ
ω=3 ω=4
i n L m L m
b > 0, b +m̄ > 0 (1 − ν )µ (1 − ν )(1 − µ )
bi > 0, bn +m̄ < 0 L M L M
(1 − ν )(1 − µ ) (1 − ν )µ
bi < 0, bn +m̄ > 0 H m
ν µ H m
ν (1 − µ )
bi < 0, bn +m̄ < 0 H M
ν (1 − µ ) H M
ν µ

Suppose first that bih > 0. One gets, from the first order conditions

qi πν L ξh
1
(p1x +p1y +p1z )+(1−π)ν L ξh 2
(p2x +p2y +p2z )
qn = 1 +(1−π)ν L ξ 2 +π(1−ν L )ξ 3 +(1−π)(1−ν L )ξ 4 +
πν L ξh h h h
π(1−ν L )ξh 3
(p3x +p3y +p3z )+(1−π)(1−ν L )ξh 4
(p4x +p4y +p4z )
1 2 3
πν L ξh +(1−π)ν L ξh +π(1−ν L )ξh +(1−π)(1−ν L )ξh 4

where π depends on the agent’s position on the nominal bond market : if

bnh + m̄h > 0, π = µm , if bnh + m̄h < 0, π = (1 − µM ) and, lastly, if bnh + m̄h = 0,
π is some number lying in the interval µm , 1 − µM .
176 Sujoy Mukerji and Jean-Marc Tallon

Observe that (αh1 (bih , bnh ), αh3 (bih , bnh )) and (p1y , p3y ) are positively dependent
(see chapter 3 in (Magill and Quinzii, 1996)). Hence, since fh is concave,

cov αh1 bih , bnh , αh3 bih , bnh , p1y , p3y < 0

Similarly,
αh2 bih , bnh , αh4 bih , bnh , p2y , p4y < 0

cov
Hence, a necessary condition for having bih > 0 at an equilibrium is that:

qi π (ν L ξh
1
+(1−ν L )ξh
3
)(ν L (p1x +p1y +p1z )+(1−ν L )π(p3x +p3y +p3z ))
qn < π (ν L ξh
+
h) ( h h)
1 +(1−ν L )ξ 3 +(1−π) ν L ξ 2 +(1−ν L )ξ 4

(1−π)(ν L ξh 2
+(1−ν L )ξh
4
)(ν L (p2x +p2y +p2z )+(1−ν L )(p4x +p4y +p4z ))
π (ν L ξh h) ( h h)
1 +(1−ν L )ξ 3 +(1−π) ν L ξ 2 +(1−ν L )ξ 4

and therefore
qi
< max ν L (p1x + p1y + p1z ) + (1 − ν L )(p3x + p3y + p3z ),

qn
ν L (p2x + p2y + p2z ) + (1 − ν L )(p4x + p4y + p4z )

Recalling that
yL 1
p3x = p1x , p3y = p ,
yH y
and p3z = p1z
yL 2
p4x = p2x , p4y = p ,
yH y
and p4z = p2z
one gets that a necessary condition for bih > 0 at equilibrium is that :
⎛  L  L
 ⎞
qi p1x + yyH + ν L 1 − yyH p1y + p1z ,
< max ⎝  L  L
 ⎠
qn p2x + yyH + ν L 1 − yyH p2y + p2z

Furthermore, since (p2x , p2y , p2z ) = λ(p1x , p1y , p1z ), a sufficient condition for agent
h to hold a positive position in the indexed bond at equilibrium is:

qi
 L
yL
   
1 y L 1 1
< λ p x + + ν 1 − py + p z
qn yH yH

Observe that this condition is independent of the position of the agent on the
nominal bond market, as the weights π, which depend on the sign of bnh + m̄h ,
do not show up in this condition.
Consider now an agent who, at an equilibrium, holds a negative amount
of the indexed bond, that is bih < 0. One gets, from the first order conditions

qi π(1−ν H )ξh
1
(p1x +p1y +p1z )+(1−π)(1−ν H )ξh 2
(p2x +p2y +p2z )
qn = H 1 H 2 H 3
π(1−ν )ξh +(1−π)(1−ν )ξh +πν ξh +(1−π)ν H ξh 4 +
πν H ξh 3
(p3x +p3y +p3z )+(1−π)ν H ξh4
(p4x +p4y +p4z )
π(1−ν H )ξh 1 +(1−π)(1−ν H )ξ 2 +πν H ξ 3 +(1−π)ν H ξ 4
h h h

where, as above, the value of π depends on the sign of bnh + m̄h for agent
h. Note that given that we look at one particular agent, the π that appears
 Ambiguity aversion and the absence of indexed debt 177

here is the same as the one that appeared in the f.o.c. for agent h if he had a
the
positive position in indexed bond (b
ih > 0).
1 i
Noticing that αh (bh , bhn ), αh3 (bih , bnh ) and p1y , p3y are now negatively de-

pendent, and hence

cov αh1 bih , bnh , αh3 bih , bnh , p1y , p3y > 0

as well as
αh2 bih , bnh , αh4 (bih , bnh ) , p2y , p4y > 0

cov
one gets the following necessary condition for an agent to go short on the
indexed bond:
qi
 L
yL
 
1 y H
> px + + (1 − ν ) 1 − p1y + p1z
qn yH yH

Here also, the exact value of π does not matter given that it does not appear
in the expression.
Therefore, if
 L
yL
   
1 y L 1 1
λ px + +ν 1− H py + p z (1)
yH y
 L
yL
 
y
< p1x + + (1 − ν H
) 1 − p1y + p1z
yH yH

then, at an equilibrium, no agent wishes to trade the indexed bond, and

therefore, bih = 0 for all h. This condition can be expressed as follows, writing
1 − ν H = A (ν) + ν L :
 L  L

p1x + yyH + (A (ν) + ν L ) 1 − yyH p1y + p1z
λ<  L
 L

p1x + yyH + ν L 1 − yyH p1y + p1z

Let
p1x + p1y + p1z
δ= yL 1
p1x + p
yH y
+ p1z

corresponding to A (ν) = 1 and ν L = ν H = 0. Notice that δ > 1. Furthermore,

if λ < δ, there exists γ < 1 such that if A (ν) > γ, then the condition (1)
is met, and, therefore, at an equilibrium, bih = 0 for all h. Notice that δ can
be expressed as a function of the fundamentals of the model, since, say p1x /p1z
and p1y /p1z are known as functions of the aggregate endowments and the utility
function u.
178 Sujoy Mukerji and Jean-Marc Tallon

Proof of Theorem 2(b)

Proof. We show that if (bih , bnh ) = (0, 0) for all h is an equilibrium of the model
when beliefs about the money supply are represented by the capacity µ, then,
perturbing slightly µ, one gets a new equilibrium at which bnh = 0 and bnh′ = 0,
for some h, h′ .Note first that, if m̄h > 0, bnh = 0 implies that α1 > α2 and
α3 > α4 , while if m̄h′ < 0, bnh′ = 0 implies that α1 < α2 and α3 < α4 .
Now, the first order condition with respect to bnh , taken at bnh = 0 is:

q n u0 = πµm ξh1 (bih , 0) + π(1 − µm )ξh2 (bih , 0)

+(1 − π)µm ξh3 (bih , 0) + (1 − π)(1 − µm )ξh4 (bih , 0)

where π is some decision weight related to the capacity ν that we need not
specify at this stage. Indeed, observe that ξh1 (0, 0) = ξh3 (0, 0) and ξh2 (0, 0) =
ξh4 (0, 0). Hence, given that m̄h > 0, a necessary condition for (bih , bnh ) = (0, 0)
to be a solution to h’s maximization program is simply that:

q n u0 = µm ξh1 (0, 0) + (1 − µm )ξh2 (0, 0)

Similarly, a necessary condition for (bih′ , bnh′ ) = (0, 0) to be a solution to h′ ’s

maximization program is that:

q n u0 = (1 − µM )ξh1′ (0, 0) + µM ξh2′ (0, 0)

Therefore, a necessary condition for (bih , bnh ) = (0, 0), h, h′ , at an equilibrium

is that

µm ξh1 (0, 0) + (1 − µm )ξh2 (0, 0) = (1 − µM )ξh1′ (0, 0) + µM ξh2′ (0, 0)

Now, either this condition does not hold, and then (0, 0) is not an equilibrium
of the economy and therefore, since we know there is no trade in the indexed
bond, this means that there is some trade in the nominal bond. Or else, this
condition does hold. However, this essentially means that

(1 − µM )ξh1′ (0, 0) + µM ξh2′ (0, 0) − ξh2 (0, 0)

µm =
ξh1 (0, 0) − ξh2 (0, 0)

Hence, if this were the case, that property would not hold for any other econ-
omy in which we would have changed µm ever so slightly.
 Ambiguity aversion and the absence of indexed debt 179

References
Camerer, C. (1995): “Individual Decision Making,” in Handbook of Experimental
Economics, e1d. by J. K. Kagel, and A. Roth. Princeton.
Dow, J., and S. Werlang (1992): “Uncertainty Aversion, Risk Aversion, and the
Optimal Choice of Portfolio,” Econometrica, 60(1), 197–204.
Ellsberg, D. (1961): “Risk, ambiguity, and the Savage axioms,” Quarterly Journal
of Economics, 75, 643–669.
Epstein, L. (1999): “A Definition of Uncertainty Aversion,” Review of Economic
Studies, 66, 579–608.
Epstein, L., and T. Wang (1994): “Intertemporal Asset Pricing under Knightian
Uncertainty,” Econometrica, 62(3), 283–322.
Epstein, L., and J. Zhang (2001): “Subjective probabilities on subjectively unam-
biguous events,” Econometrica, 69, 265–306.
Fishburn, P. (1993): “The Axioms and Algebra of Ambiguity,” Theory and Decision,
34, 119–137.
Freeman, S., and G. Tabellini (1998): “The Optimality of Nominal Contracts,” Eco-
nomic Theory, 11, 545–562.
Ghirardato, P. (1997): “On Independence for Non-Additive Measures, with a Fubini
Theorem,” Journal of Economic Theory, 73, 261–291.
Ghirardato, P., and M. Marinacci (2002): “Ambiguity Made Precise: A Comparative
Foundation,” Journal of Economic Theory, 102, 251–289.
Gilboa, I., and D. Schmeidler (1989): “Maxmin Expected Utility with a Non-Unique
Prior,” Journal of Mathematical Economics, 18, 141–153.
Knight, F. (1921): Risk, uncertainty and profit. Houghton Miffin.
Magill, M., and M. Quinzii (1992): “Real effects of money in general equilibrium,”
Journal of Mathematical Economics, 21, 301–342.
Magill, M., and M. Quinzii (1996): Theory of Incomplete Markets, vol. 1. MIT Press.
Magill, M., and M. Quinzii (1997): “Which Improves Welfare More: A Nominal or
an Indexed Bond?,” Economic Theory, 10, 1–37.
Mukerji, S. (1997): “Understanding the Nonadditive Probability Decision Model,”
Economic Theory, 9(1), 23–46.
Mukerji, S. (1998): “Ambiguity Aversion and Incompleteness of Contractual Form,”
American Economic Review, 88(5), 1207–1231.
Mukerji, S., and J.-M. Tallon (2001): “Ambiguity Aversion and Incompleteness of
Financial Markets,” Review of Economic Studies, 68(4), 883–904.
Mukerji, S., and J.-M. Tallon (2003a): “Ellsberg’s 2-Color Experitment, Portfolio
Inertia and Ambiguity,” Journal of Mathematical Economics, 39, 299–316.
Mukerji, S., and J.-M. Tallon (2003b): “An Overview of Economic Applica-
tions of David Schmeidler’s Models of Decision Making Under Uncertainty,”
Discussion Paper no. 165, Department of Economics, Oxford University,
http://www.econ.ox.ac.uk/Research/wp/pdf/paper165.pdf.
Mukerji, S., and J.-M. Tallon (2004): “Ambiguity Aversion and the Absence of Wage
Indexation,” Journal of Monetary Economics, 51, 653-670.
Schmeidler, D. (1989): “Subjective Probability and Expected Utility Without Ad-
ditivity,” Econometrica, 57(3), 571–587.
Shiller, R. J. (1997): “Public Resistance to Indexation: A Puzzle,” Brookings Papers
on Economic Activity, 1.
Sunspots, Indeterminacy and Pareto
Ine!ciency in Economies With Incomplete
Markets?

Tito Pietra

Universita’ G.D’Annunzio, Chieti-Pescara, Italy

Summary. We consider two periods economies with both intrinsic and extrinsic
uncertainty. Asset markets are incomplete in the certainty economy. If assets are
nominal, there are enough commodities and the number of agents is greater than two
and smaller than the total number of states of nature “tomorrow” (minus one), then
a sunspot-invariant equilibrium is generically Pareto dominated by some sunspot
equilibria. When assets are real, and there are enough commodities, whenever there
are sunspot equilibria, typically there are also sunspot equilibria Pareto dominating
the sunspot-invariant equilibria under the same restriction on the number of agents
(and stronger restrictions on the number of commodities).

1 Introduction
Cass [3] provided the first example of the existence of a continuum of sunspot
equilibria in economies with incomplete markets. Since then, existence and
structure of sunspot equilibria have been extensively studied in this class of
economies with nominal (see, [18], [4], [14], [16], [20] and [10]) and with real
assets (see, [13] and [9])). Existence and “number” of sunspot equilibria in
financial economies crucially depend upon the nature of asset payos: With
nominal assets, they generically exist in a neighborhood of a sunspot-invariant
equilibrium and their set contains a non-zero dimensional manifold. Its dimen-
sion depends upon the way asset prices are treated: Generically, it is (  L)
if asset prices are fixed, (  1) if they are variable ( is the number of states
of nature “tomorrow”, I the number of assets). When asset payos are real
(and if there are enough commodities), sunspot equilibria exist for an open
set of asset structures and are locally unique.
Our second theme is Pareto dominance. As pointed out in [6], in convex
economies sunspot equilibria are Pareto ine!cient. When asset markets are
B
I wish to thank Paolo Siconolfi for helpful suggestions and comments. I aknowl-
edge the financial support of M.I.U.R. and the kind hospitality of C.C.D.R. in
Summer 2003.
182 Tito Pietra

incomplete at the sunspot-invariant equilibria, they are themselves typically

Pareto ine!cient and can possibly be Pareto dominated by equilibria of the
same economy where extrinsic uncertainty matters. Parametric examples of
this phenomenon in incomplete market economies have been provided in the
literature ([15] gives an example with = 3> two agents and one commodity).
Our purpose is to investigate the (comparative) e!ciency properties of
sunspot and sunspot-invariant equilibria. When asset payos are nominal,
we show that, generically, sunspot-invariant equilibria are Pareto dominated
by some sunspot equilibria, if the number of agents (K) is greater than two
and smaller than (  1) and there are enough commodities. We focus the
analysis on this comparison. However, an obvious corollary is that, typically,
sunspot equilibria are in turn dominated by some other equilibria. A second
corollary is that, when 2 ? K ? (V  1) (V is the number of intrinsic
events “tomorrow”), certainty equilibria are typically Pareto dominated by
some other certainty equilibria.
The notion of genericity adopted is somewhat weak, because we show
that, given a vector of numeraire, there is an open, dense set of economies
(parameterized by utility functions and endowments) such that our result
holds. A stronger result would be to show that, generically, all the equilibria
in a full measure set of sunspot invariant equilibria are Pareto dominated
by some sunspot equilibrium. I am not ready to formulate a conjecture on
the truthfulness of this result, which could possibly be analyzed using the
approach exploited in [17] to study real indeterminacy from a global viewpoint
in economies with nominal assets and incomplete markets.
Given [11] (where real indeterminacy of equilibria is re-interpreted in terms
of real eects of monetary policies), one possible interpretation of our result is
in terms of the existence of Pareto improving randomized monetary policies.
Another possible interpretation is as another robust example of a class of
economies where the value of information can be negative.
We also study the case of economies with real assets, restricting ourselves
to financial structures such that sunspot equilibria exist and with enough
commodities. We show that, for a generic set of economies (parameterized
by utility functions and endowments), there are asset structures such that
sunspot equilibria exist and Pareto dominate at least one of the sunspot-
invariant equilibria.
The structure of the paper is the following: Section 2 describes the model.
Section 3 establishes that, with nominal assets, sunspot-invariant equilibria
are, generically, regular in the sunspot economy and (if they are not Pareto
e!cient) satisfy some further restrictions on the vector of excess demand and
Lagrange multipliers. Then, following the approach outlined in [19] and de-
veloped in [8], [5] and [7], we establish that the extended system of equations
describing the equilibria of the economy and the utility functions of the agents
define a system of independent equations (i.e., the derivative of this transfor-
mation has maximal rank, equal to the number of equations). This immedi-
ately implies that, given a sunspot-invariant equilibrium, there are open sets
 Sunspots, Indeterminacy and Pareto Ine!ciency 183

of Pareto superior sunspot equilibria. While the logic of our argument is the
same as the one of the quoted papers, the technical details are quite dierent.
Those papers establish the full rank property exploiting in a crucial way ar-
bitrary perturbations of the second order derivatives of the utility functions.
This approach works in our framework when (V  1)  K, it is bound to fail
when K A (V  1), i.e., in the more interesting case when “sunspot matters”
in terms of feasibility of Pareto improvements. Therefore, we have to follow
a more cumbersome approach.
The last section studies economies with real assets, building on the previ-
ous results.

2 The model
We consider a two period model with both intrinsic and extrinsic uncertainty.
Spot v = 0 is “today”, v = 1> ===> V denotes one of the V intrinsic events
“tomorrow”. There are also N mutually exclusive extrinsic events, so that, in
the sunspot economy, there are NV = states “tomorrow”. A generic spot
will be denoted either by >  = 0> (11)> ===> (1V)> ===> (N1)> ===> (NV)> or by the
more explicit notation nv, when it is important to emphasize the intrinsic
event characterizing the state of nature.
To simplify notation, and without any loss of generality, we assume that
the probability of each extrinsic event is 1@N=
There are K agents, denoted by k = 1> ===> K. At each spot there are F
commodities, f = 1> ===> F, so that the total number of commodities is J =
(( + 1)F). Agent k’ consumption vector at spot  is {k = ({1 F
k > ===> {k )>
0
while his consumption vector is {k = ({k > ===> {k )= In a similar fashion, ex-
cess demand vectors are }k and }k > while commodity prices are denoted by
s = (s0 > ===> s ) and s = (s1 > ===> sF )= Also, let  (s) be the (( + 1) × J)
dimensional matrix
5 0 6
s
9 :
 (s) = 7 . . . 8.
s

There are L assets. Asset l has payos | l = [| 1l > ===> | l ]> its price is t l =
Let \ be the ( × L) dimensional matrix of asset payos and let
 ¸
t
U(t) =
\

be the price-payos matrix. For each k> ek 5 <L is the portfolio vector=
The payos of asset l can be nominal or can be given, at each state, by the
value of a commodity bundle vl = [vl1 > ===> vlF ]=
When assets are nominal, we need three restrictions on asset payos:
0
i. | nv = | n v > for each v and n> n0;
184 Tito Pietra
£ ¤
ii. The matrix | n1 · · · | nV is in general position;
iii. | 1 = 1> for each  A 0> i.e., asset one is inside money.
Evidently, i guarantees the existence of n-invariant equilibria, iii could be
weakened.
For the case of real assets, we will explicitly construct the n-invariant
collection vl > for each v and l, in Section 4.
Assumptions on consumers are standard: For each k, the endowment vector
n0 v
is hk = (h0k > h11 NV J nv
k > ===> hk ) 5 <++ > satisfying hk = hk , for each v> n and n0.
Given the n-invariance restriction, we will identify the endowment space with
(V+1)F
<++ = Preferences are described by a Von Neumann-Morgenstern, strictly
concave, strictly increasing, F 2 > utility function
P
Yk ({k ) = n xk ({0k > {n1 nV
k > ===> {k )@N

such that the Bernoulli utility index xk ({0k > {1k > ===> {Vk ) satisfies the usual
(V+1)F
boundary restriction: the set {({0k > {1k > ===> {Vk ) 5 <++ | xk ({0k > {1k > ===> {Vk )
0 1 V (V+1)F
 xk ({k > {k > ===> {k )} is closed in <++ > for each ({0 1 V
k > {k > ===> {k ) 5
(V+1)F
<++ =
In the sequel, we will take as given asset payos \ (or ) and probabilities,
Q (V+1)F
= The space of economies is identified with E = k (Uk × <++ ), where
Uk is the set of Bernoulli utility indexes satisfying the conditions above. As
(V+1)F
usual, we endow Uk with the C2 > compact-open topology, <++ with the
Euclidean topology and E with the product topology. H will denote an element
of E.
When dealing with nominal assets, we need an appropriate parameteri-
zation of the set of equilibria. Here, the most convenient is the following:
Fix sF = 1> for each > and (locally) parameterize the set of equilibria with
a vector  = (1>  11 >  12 > ===>  1 > 1), where we identify  with its variable
1
coe!cients, so that  5 <++ = Then, the matrix  (s) is replaced by
5 0 6
s
9 y11 s11 :
9 :
 (s> ) = 9 .. :.
7 . 8
s
1
In the sequel, s 5 P = {s 5 <J ++ |s
F
= 1} and  5 <++ . Also, let
\
J\ = (( + 1)(F  1)) and let s\ 5 <J ++ be the vector of the first (F  1)
commodity prices ©
at each spot. Finally, let’s restrict t to the
ª set of noarbitrage
asset prices, Q = t 5 <L | U(t) = 0, for some  5 < +1 ++ .
Consumer behavior and equilibrium are defined as usual:
1
Definition 1. Given  5 <++ > an equilibrium is a pair (===> ({k > ek )> ====) and
(s> t) 5 P × Q> such that:
i. P P ({k > ek ) 5Parg max Yk ({k ) subject to  (s> )}k = U(t)ek ;
For each h,
ii. k {k = k hk and k ek = 0;
 Sunspots, Indeterminacy and Pareto Ine!ciency 185

iii. An equilibrium is a sunspot equilibrium if, for some h and some s, there
n0 v
are k and k’ such that {nv
k 6= {k = Otherwise, it is a k-invariant equilibrium.

Bear in mind that, to avoid too many repetitions, we use the term sunspot
equilibrium only to refer to equilibria where sunspots matter. Equilibria of the
sunspot economy where sunspots do not matter are called n-invariant equi-
libria. Finally, we use the term certainty equilibria to refer to the equilibria
of the certainty economy, i.e., (with some abuse of language) of the economy
with no extrinsic uncertainty.
As is well known, n-invariant equilibria are related in an obvious way
to certainty equilibria. In particular, ((s0 > ===> sV )> t  ) 5 <(V+1)F+L is a
V1
certainty equilibrium associated with  5 <++ , and with allocation (...,
   V+1
({k > ek )> ===) and Lagrange multipliers k 5 <++ , for each k, if and only if
((s0 > ===> (s1 > ===> sV )> ===)> t  ) 5 <J+L is a n-invariant equilibrium associated
with  = (1>  > ===>  ), allocation (..., (({0 1 V 
k > ===> ({k > ===> {k )> ===> ek )> ===) and
0 1 V +1
Lagrange multipliers (k > ===> (k @N> ===> k @N)> ===) 5 <++ = We will repeat-
edly exploit this fact.
In the sequel, we will study equilibria making reference to the entire sys-
tem of equations implicitly described in Definition 1, after getting rid of the
redundant equations, i.e., to the so called extended system of equations. Let’s
define I (===> ({k > ek > k )> ===> (s> t> )> ===> (xk > hk )> ===) as
5 6
===
9 G{k Yk ({k )  k  (s> ) :
9 :
9
9 U(t)W Wk :
:
I (> > H) = 9  (s> )}k + U(t)ek :
9
:>
9 ==== :
9 P \ :
7 8
Pk }k
k ek

\
where }k denotes agent k’s vector of excess demand for all commodi-
ties but commodity F at each spot, while  = (===> ({k > ek > k )> ===> (s> t)) and
q = (K(J+L + +1)+J\ +L)= I (=) summarizes the first order conditions of
each agent and, after applying appropriately Walras’ law, the non-redundant
market clearing conditions for assets and commodities as a function of the ex-
ogenous parameter > of the endogenous variables  5 <q and of endowments
1
and utility functions. Evidently, I : <q × <++ × E $ <q = Let IH (> ) be
the map above for given H 5 E : Then, the set of equilibria of H is IH1 (0)=

3 Pareto improving sunspot equilibria with nominal

assets
In this section, we consider economies with nominal assets and focus the
analysis on the case of variable asset prices.
186 Tito Pietra

The basic idea of the proof is standard: The first step is to establish that,
for a generic set of economies, n-invariant equilibria are regular and satisfy
some additional restrictions (Lemma 2). Then, we show that these equilibria
are Pareto dominated by some sunspot equilibria (Theorem 3).
The proof of the theorem requires us to be able to perturb independently
the utility functions of two agents at each n-invariant equilibrium. In this
class of economies, this is not necessarily possible, due to real indeterminacy.
1
Therefore, we resort to fix  5 <++ : For such a given , we will establish the
existence of an open and dense set of economies such that their n-invariant
equilibria can be Pareto improved by changing appropriately .
In the proof of our main theorem, we need to restrict ourselves to regular
equilibria where the matrix
£ ¤
] \1 · · ·  ] \N E Z \ =
5 6
=== === ===
7 [===>  } \ > ===]A0 [0 ek ] [===>  s }  > ===]| AA0 8 >
k k k k k
=== === ===
evaluated at the n-invariant equilibrium, has maximal rank K=
1
Lemma 1. Let K ? (V(F  2) + 2) and L ? V. Given  5 <++ > there is an
j j
open and dense subset of E, E , such that, for each H 5 E > at each k-invariant
equilibrium,
udqnG IH (> )|IH (=)=0 = q>
£ ¤
udqn ] \1 · · ·  ] \N E Z \ = K.
Proof. The first result is basically established in [4], [14] and [16]. The second
follows by a routine argument.
As is well known, the regularity part of the Lemma holds independently
of V (and )=
The rank condition obviously implies that the equilibrium is Pareto inef-
ficient. Similar rank conditions are common in the literature on constrained
Pareto ine!ciency. Our is slightly dierent (and stronger than the usual one)
because we do not consider the (F  1) columns of the excess demands for
commodity 0f, f ? F. In terms of economic interpretation, bear in mind
that the matrix is given by the gradients of the indirect utility functions with
respect to s\ >  A 0, t and =
Given Lemma 2, we can restrict the analysis of the welfare eects of
sunspot phenomena to regular sunspot equilibria, satisfying the additional
rank condition above=
Using the standard approach, consider the system of equations
5 6
I (> > H)
9 Y1 ({1 )  Y1 :
(> > H) =9
7
:=
8
===
YK ({K )  YK
 Sunspots, Indeterminacy and Pareto Ine!ciency 187

Suppose that, at the equilibrium  (associated with some ),

udqnG(>) H (> ) = (q + K) =
Then, there exists a vector (g> g) such that
 ¸  ¸
g 0
G(>) H (; ) = >
g 
for each  AA 0= Hence, by the implicit function theorem, we can find
an appropriate perturbation of  such that the associated equilibrium alloca-
tion Pareto dominates the original one. As mentioned above, our argument
is an application of the approach originally introduced in [19] and developed
and applied, in economies with incomplete markets, to issues related to con-
strained Pareto ine!ciency (see, [8] and [7]) and to Pareto improving financial
innovation (see, [5]).

Theorem 1. Let 3  K ? (  1)> L ? V and N ? (F  1). Then, for

each H 5 E S > an open, dense subset of E j > given  = 1> for each k-invariant
equilibrium with allocation {0 there is a sunspot equilibrium with allocation {j
0
such that, for each h, Yk ({”k ) A Yk ({k )=

Evidently, N ? (F  1) and K ? (  1) imply that K ? (V(F  2) + 2),

so that Lemma 2 applies. We need at least three agents for technical reasons,
related to the details of the proof of Lemma 7. It is an open issue if a similar
result could be established for K = 2. As an immediate consequence of the
theorem, we obtain

Corollary 1. If 3  K ? (V  1) and L ? V, then, for each H 5 E S > an open,

dense subset of E j > given  = 1> for each certainty equilibrium with allocation
{0 there is a certainty equilibrium with allocation {” such that, for each k,
0
Yk ({”k ) A Yk ({k )=

Remark 1. Both Theorem 3 and Corollary 4 take as a starting point a regular,

n-invariant equilibrium. We can reinterpret the model as follows: Let h denote
a type of agents and, therefore, assume that there is a continuum of agents of K
dierent types. Then, our result holds true for economies with a continuum of
agents, provided that the characteristics of agents are, a.e., not “too dierent”
from the ones specified in the original K types (see, [5]).

Remark 2. Dealing with certainty equilibria, we do not face additional re-

strictions (but symmetry, negative semidefiniteness and continuity) on the
perturbation of the Hessian matrix of the utility functions. Therefore, the
result in Corollary 4 could also be established directly, by a straightforward
application of the usual technique of proof (in fact, with K  2). However,
given the additional restrictions across extrinsic events that we must satisfy
in sunspot economies, this technique cannot be applied in the more general
framework of Theorem 3.
188 Tito Pietra

The proof of Theorem 3 is based on three steps: First, we observe that

udqnG(>) H (> ) = (q + K) if and only if
" #
G(s\ >t)  G 
udqn £ ¤£ ¤ = (J\ + L + K)>
] \ -E Z \

where (=) denotes the aggregate excess demand for all commodities but
commodity F at each spot, while, now, ] \ is the (K × J) dimensional matrix
with typical coe!cient (k }kf ) > each . Next, given H = (x> h) with an
equilibrium (s> t)> ({> e) and with Lagrange multipliers , we pick an economy
H̄ = (x> h̄) such that the initial equilibrium is still an equilibrium of this
modified economy and study udqnG(>) H̄ (> ). We show that, generically
in the space of the utility functions, udqnG(>) H̄ (> ) = (q + K). In fact,
to restore its full rank, it su!ces to perturb the vector of marginal utilities of
two agents. Then, a standard argument shows that (given the new collection
of utility functions and modulo a perturbation of the original endowment h in
the n-invariant direction (h  h̄)) udqnG(>) H (> ) = (q + K).
We make precise our argument splitting it into three Lemma.
Let’s start displaying G(>) H (> )>
5 6
..
. ··· ··· ···
9 :
9 · · · G({k >ek >k ) I RFk · · · G(s>t>) I RFk :
9 :
9 .. .. .. .. :
9 . . . . :
9  \ ¸ :
9 LJ 0 0 :
9··· ··· 0 :=
9 0 LL 0 :
9 :
9 . . . . :
9 .. .. .. .. :
9 :
9··· G{k Yk ··· 0 :
7 8
.. .. .. ..
. . . .
In the matrix above,5 6
G{2k Y1 0  (s> )W
G({k >ek >k ) I RFk = 7 0 0 U(t)W 8 >
 (s> y) U(t) 0
and 5 6
\
k () 0 (s> k )
9 0 0k LL¸ 0 :
G(s>t>) I RFk = 9 7  :>
8
\ e k \
 (}k > ) Zk
0
where
5 \0 6
}k
9 .. :
\ 9 . :
 (}k > ) = 99 :
1 \ 1 :
7  }k 8
\
}k
 Sunspots, Indeterminacy and Pareto Ine!ciency 189
¡ ¢
is a ( + 1) × J\ dimensional matrix,
5 6
0 0
1
9 s }k 1 :
9 :
\ 9 . .. :
Zk = 9 :
9 :
7 ( 1) 8
s( 1) }k
0 0
is a (( + 1) × (  1)) dimensional matrix, (s> k ) is a (J + L)×
(  1) dimensional matrix with typical column  ( 6= 0> NV) given by
\ ¡ ¢
(0> ===> k s > 0> ===)W = Finally, k () is the (J + L) × (J\ + L) dimensional
matrix with all F rows identically zero. Restricted to the other rows, the
matrix is diagonal with coe!cients k   =
Given that the block diagonal matrices G({k >ek >k ) I RFk are readily es-
tablished to have full rank, G({>e>) I RF 1 exists. By premultiplying G(>) 
by
 ¸
G({>e>) I RF 1 0
>
0 L
applying the obvious operations on the rows of the matrix so obtained
and, finally, premultiplying by
 ¸
G({>e>) I RF 0
>
0 L
we conclude that
5 6
G({>e>) I RF G(s\ >t) I RF G I RF
udqnG(>)  = udqn 7 0 £ G(s\\ >t)  ¤ G \
8=
0 ] E Z
The next step is to compute the rank of the bottom right matrix for an
appropriately selected economy H̄. Notice that to keep in mind the entire
extended system of equations, and, hence, the matrix G(>) (> > H)> is es-
sential
£ to keep track ¤ of the exact eects that the operations on the columns
of ] \ E Z \ have on the columns of G(s\ >t>) (=) via their eects on
the columns of G(s\ >t>) I RFk (=)=
Pick  = 1 and an associated n-invariant equilibrium (s > t  ) with allo-
cation (===> {k > ===) = We want to construct a new economy H̄ with the same
equilibrium prices and allocation, but with (n-invariant) individual endow-
ments selected so that they allow us to drastically simplify the computations.
Bear in mind that prices and allocations are, by assumption, n-invariant, so
that we may sometime omit the index n, to emphasize this symmetry.
Let v̂ = plq {v|v(F  2)  (K  2)} = The economy H̄ is defined as follows:
For each k, x̄k = xk = Endowments are, instead, changed: For agent 1, set
h̄vf vf
1 = {1 for all the commodities, but commodity f = 1 at each state v> v A 0,
190 Tito Pietra
¡ 0F ¢ ¡ v1 ¢
and commodity F at v = 0. Set h̄0F 1 = {1 + t 1 > h̄v1
1 = {1  1@s
v1
=
Evidently, given h̄1 > ({1 > (1> 0> ===)) is agent 1’s optimal choice at prices (s > t  ).
For agent k> K A k A 1> define in the obvious way a one-to-one map
i : {2> ===> K  1} $ {12> ===> 1 (F  1) > 22> ===> 2 (F  1) > ===> v̂(F  1)} >
i (k) = vf= Then, set h̄vf k = {k
vf
for each vf but i (k) = vf> and com-
vf vf
modity¡ F at the same ¢ spot. Set h̄k = ({k  1) > for vf = i (k) and
vF vF
h̄k = {k + s i (k)
= Evidently, given h̄k > ({k > 0) is agent k’s optimal choice
at prices (s > t  ). His excess demand is nil for all commodities, but commod-
i (k)
ity vf = i (k)¡P and commodity P F ¢at the same spot and }k = 1= For agent
 
K> set h̄K = k k{  k?K kh̄ =
It is straightforward to check that, given  = 1, (s > t  ) with allocation
({ ) is an equilibrium of the economy H̄, supported by the portfolios ek = 0,
for k 6= 1> K> and e1 = eK = (1> 0> ===)=
In words, agent 1 buys one unit of inside money and, at each spot, spends
the revenue to buy commodity 1. At each v  v̂> one unit of each commodity
is traded with agent K by agent h such that vf = i (k), who finances the
purchase selling commodity F at the same spot. Also, each agent (but 1 and
K) trades just once.
(V+1)F
Observe that not necessarily h̄ 5 <++ = However, given that, at  = 1
 
and in a neighborhood of (s > t )> for each k and v, sv h̄vk A 0> this is irrelevant.
The result driving the entire proof is summarized in the next Lemma.
The proof £ is in Appendix. ¤ The basic idea is the following: Given H̄, the
matrix ] \ E Z \ has a very simple structure. In particular, using
£ ¤
columns operations, it can be transformed into a matrix ] \ 0 0 where
] \0 = 0> while each block ] \n contains only one nonzero term, in the column
corresponding to commodity (nvn). Of course, these columns operations also
aect the matrix G(s\ >t>) (=)> transforming it into a matrix G(s\ >t>)   (=)=

Let s be the vector of prices of the commodities snvn > each n. The key step of
the proof is to show that, given x1 and modulo a perturbation of xK and x2 >

the (J\ + L) columns of G(s\ >t>)   (=) referred to (s\ > t> )> where the vector

s\ does not includes commodity nvn, each n, are linearly independent. When
this is true, relabelling columns, we obtain
 ¸  ¸
G \  G  G(s\ >t>y)   Gs   ¡ ¢
udqn £ (s\ >t) ¤ \ = udqn \ = J\ + L + K =
] E Z 0 ]
This sketch of the proof should help to understand why we need a restric-
tion on the number of extrinsic events : We need to replace column Gsnvn  
with column G nv   > each nv 6= NV. This works as long as N ? (F  1)= The
economy H̄  constructed in the Lemma depends upon the particular equilib-
rium we start with. However, for given > IH1 (0) is a finite set (generically).
Hence, iterating the procedure, we can show that the same result holds for
each equilibrium, associate with the given > of an open and dense set of
economies.
 Sunspots, Indeterminacy and Pareto Ine!ciency 191

Lemma 2. Let 3  K ? (  1)> L ? and N ? (F  1). Given > H̄ and

any open set Y (H̄), there is H̄  = (x > h̄) 5 Y (H̄) with the same equilibrium
(s> t) and ({> e) and such that udqnG(>) H̄  = (q + K) =
Now, let’s go back to the actual economy H. Here, we exploit an argument
which has been introduced by [1] and subsequently used, in dealing with
sunspot economies, by [4] and [10].
Lemma 3. Let 3  K ? (  1)> L ? and N ? (F  1). Given > H
and any open set Y (H), there is an open, dense subset Y 0(H)  Y (H), such
0
that, for each H 5 Y 0(H)> udqnG(>) H 0 |I (=)=0 = (q + K) =

Proof. Start with the economy H and construct H̄  = (x > h̄)= By Lemma
7, udqnG(>) H̄  = (q + K) = In fact, in Appendix, we establish the result
dropping the redundant columns of G H̄  , so that we obtain a square ma-
\ \ \
trix G(>) H̄  = Given that udqnG(>) H̄  = (q + K) > ghwG(>) H̄  6= 0= Let
\
h(w) = (wh̄ + (1  w)h) and H̄  (w) = (x > h(w))= Evidently, ghwG(>) H̄  (w) is a
\
polynomial in the variable w. Given that ghwG(>) H̄  (1) 6= 0> the polynomial
\
is non trivial and, therefore, ghwG(>) H̄  (w) = 0 has a finite number of solu-
\
tions, {w1 > ===> wW }. Hence, if ghwG(>) H̄  (0) = 0> it is su!cient to perturb the
initial endowment in the (n-invariant) direction (h̄h) to obtain an (arbitrarily
close) economy H0 = (x > h0) with ghwG(>) \ |H 0 6= 0= Hence, Y 0(H) is dense.
Given that the initial equilibrium is regular in the economy H, openness of
Y 0(H) follows immediately, for Y (H) su!ciently small.
We are finally ready to establish Theorem 3.
Proof. Density of E S follows directly from Lemma 2, 7 and 8. Given the
boundary conditions, a standard argument establishes that E S is open as
well.

4 Pareto improving sunspot equilibria with real assets

As is well known, while sunspot equilibria may exist in economies with real
assets, they do not exist for arbitrarily given payos structures. We will extend
the previous result showing that, in economies with incomplete markets (at
the n-invariant equilibrium) and such that sunspot equilibria exist, there are,
generically, asset payos structures such that the original equilibrium is Pareto
dominated. To guarantee the existence of sunspot equilibria, we will follow the
approach originally proposed in [13] and developed in [9].
Our construction rests heavily on the previous results. Pick an economy H,
nominal asset structure \ and  0 such that the associated equilibrium (more
properly, one of the equilibria) (s0> t0) with allocation ({0> e0) is Pareto dom-
inated by a sunspot equilibrium associated with ”> (s”> t”), with allocation
192 Tito Pietra

({”> e”)= We will show that, typically, there is a n-invariant structure of real
assets  such that (s0> t0), ({0> e0) and (s”> t”)> ({”> e”) are both equilibria of
the new economy.
Notice that, by Corollary 4, if K ? (V  1) and assets are nominal,
certainty equilibria are typically Pareto dominated by some other certainty
equilibrium, so that the possibility of Pareto improvements within the set of
equilibria is not necessarily related to sunspots phenomena. This is in general
not true with real assets. It is easy to check that our construction works also
for economies with a unique certainty equilibrium.
The proof of the main result exploits an additional property of sunspot
equilibria. Define the ((N + 1) × F) dimensional matrix
5 0 0 6
 v sv
9  1v s1v” :
”

(v) = 9 7
:
8
===
Nv” Nv”
 s
where the first row is given by state s certainty equilibrium prices at  0 ,
while the last N rows are given by equilibrium prices at the states (nv) at  00 .

Lemma 4. Let 3  K ? (  1)> L ? V and N ? (F  2). Given the

nominal asset structure \ , let (s0> t0) be the k-invariant equilibrium associated
with  0 and (s”> t”) be a sunspot equilibrium associated with ”> satisfying
y nd v ” 6= y ne v ”for each s and each pair kd , ke . Then, for an open and dense
set E   E, (v) has maximal rank (N + 1), for each s.

Proof. By Lemma 2, there is no loss of generality in assuming that both

n-invariant and sunspot equilibria are regular and su!ciently close (in the
economy with nominal assets). Hence, it su!ces to show the density part of
the thesis, i.e., that for each open ball Y (H), there is H  5 Y (H) such that
udqn(v) = (N + 1)> for each v, at the equilibrium (s > t  )= By a standard
argument, modulo a perturbation of the utility function in the certainty econ-
0 0
omy, there is an k, say k = 1, such that, at each > s }1 6= 0= By regularity
” ”
of the sunspot equilibrium and continuity, we can assume that s }1 6= 0,
at each = Given that y nd v” 6= yne v” > each v and each pair nd > ne , this im-
0
plies {n1d v” 6= {n1e v” , each v and each pair nd > ne > and xv1 6= {nv” 1 > each n.
Hence, there exists a collection of N open balls E1 ({n” 1 ), each one centered
on {n” 0” n1” nV”
1 = ({1 > {1 > ===> {1 ), n = 1, ===> N, and a collection of closed balls
% n” (V+1)F (V+1)F
E1 ({1 )  <++ such that E1 ({n” % n”
1 )  E1 ({1 )  <++ , each n, and
n ” 0
% nd ” % % n”
E1 ({1 ) _ E1 ({1 ) = >> each kd , ne and {1 5
e
@ E1 ({1 )> each n. Also, let
n ({1 > E1% ({n”
1 ))> n = 1> ===> N> be a collection of bump functions taking the
value 1 at {1 5 E1 ({n” 1 )> the value 0 at {1 5 @ E1% ({n”
1 ) (see, [12])= Assume
that udqn(v) ? (N + 1), for some v. Consider an arbitrarily small per-
turbation of its last N rows (keeping snv1 = 1> each n), so that the new
matrix  0 (v) has full rank. Repeat the operation for each v A 0. Let s
 Sunspots, Indeterminacy and Pareto Ine!ciency 193
P
be the price vector so obtained andPlet  (s > t  > ”) = kA1 }k (s > t  > ”)
and }1 = }1 (s > t  > ”)= Given that k }k (s”> t”> ”) = 0> by continuity, || 
(s > t  > ”)  }1 (s > t  > ”)|| can be made arbitrarily close to 0 by choosing a
price vector s su!ciently close to s” and still preserving the full rank of  0 (v)>
each v. Consider now the P economy H  with hk = hk > each k, xk = xk > k A 1,
 n % n”
and x1 ({1 ) = x1 ({1 )+ n  P ({1 > E1 ({1 ))[G{1 x1 ({1 )|}1 G{1 x1 ({1 )| ]{1 =
I claim that, ( > e1 =  kA1 ek (s > t  > ”)) is agent 1’s optimal choice
given (s > t  ), ” and utility function x1 = Evidently, ( > e1 ) satisfies
agent 1’s sequence P of budget constraints. Moreover, given that, with util-
ity x1 > at }1 > n (n)G{1 x1 ({1 )|}1 = W 
1  (s > ”)> for some vector  such

W 
P  n
P n
that
P 1 U(t ) = 0> n (n)G{1 x1 ({1 )| = n (n)G
P {1 x1 ({1 )| +
n (n)[G x ({
{1 1 1 }1)|   G x ({ )|
{1 1 1   ] = n (n)[G {1 x1 ({1 )|}1
W   
= 1  (s > ”)= Hence, ( > e1 ) solves the FOC of agent 1’s optimiza-
tion problem at (s > t  ) and ”= By construction, the allocation (}1 =
 > }2 (s > t  > ”), ..., }K (s > t  > ”)) satisfies the market clearing condi-
tions. Given that \ is a full rank matrix, this means that the market
clearing conditions for assets are satisfied as well. Hence, (s > t  ) is a
sunspot equilibrium of the economy H  > given ”= Moreover, given that
0
agent 1’s utility function is not changed at {1 > the n-invariant equilibrium
is not aected. As noticed above, for a su!ciently small perturbation of
the equilibrium prices s”, H  can be made arbitrarily close to H, hence
H  5 Y (H)=

Finally,

Theorem 2. Let 3  K ? (  1), L ? V and N ? (F  2). Then, there

exists an open, dense set of economies E s  E such that, for each H 5 E S >
there is an asset structure  such that there are sunspot equilibria Pareto
superior to a n-invariant equilibrium.

Proof. Given Lemma 9, it su!ces to establish the density part. Given H,

any open set Y (H) and any asset structure \ , with L ? V, pick H  such
that the regular n-invariant equilibrium associated with  is Pareto dom-
inated by a sunspot equilibrium associated with ”= Given the previous
Lemma, we may assume that, at the sunspot equilibrium s”, udqn(v)=
(N + 1), each v. This implies that, for each v and each l, there is a solu-
tion l (v) to the system of equations (v)l (v) = [| lv > | lv > ===> | lv ]= Then,
let  define the structure of real asset of the economy. Evidently, (s0> t0) is
a n-invariant equilibrium, while (s”> t”) is a sunspot equilibrium of the new
economy.

Remark 3. In the economy with real assets, neither k-invariant nor

sunspot equilibria are necessarily regular (because G(s>t) (=) in the economy
with real assets is obviously dierent from the same derivative in the economy
with nominal assets). Given [9], our result could be strenghtened by consid-
ering equilibria which are regular in the economy with real assets, too. This
194 Tito Pietra

would allow to strenghten Theorem 10, establishing openness of the set of

asset structures allowing for Pareto improvements. The argument is straight-
forward and, therefore, omitted.

Appendix: Proof of Lemma 7

Throughout the proof, we will keep fixed  = 1 and an associated equilibrium
(s> t) and perturb the economy H̄ so that (s> t) 5 IH̄1 0 (0)= The technique of
proof requires us to be able to perturb (in n-invariant directions) the gradients
of two agents. It is well known that this can be obtained via perturbations of
the utility functions (see, [13], [14] and [16]). The following result is reported
here for completeness.

Claim. Let L ? V= Fix   > H̄, any open set Y (H̄) and a k-invariant equilibrium
(s > t  )> ({ > e ) with associate Lagrange multipliers  = Then, for each  5
< +1 ,  small enough and such that U(t) W = 0, there is H̄  5 Y (H̄) such that
(s > t  ), ({ > e ) is an equilibrium of H̄  with associated Lagrange multipliers
k = (k + ) =

The proof of Lemma 7 is based on two steps:

1. We exploit the n-invariance of the allocation to get rid of all but K
linearly independent columns of the submatrix [] \ E Z \ ]. Given x1> this
may require a perturbation of agent K’s utility function.
2. We show that, given x1 > xK and modulo an arbitrarily small pertur-
bation of x2 , the (J\ + L) dimensional, square matrix G(s\ >t>) > defined in
the text above, has full rank.
The Lemma follows immediately.

Step 1

Bear in mind that N ? (F  1) and K ? (  1).

Without loss of generality, but for notational simplicity, fix  nv = 1 for
each v A v̂ and drop the corresponding columns of Z \ and G =
Remember that v̂ = plq {v|v(F  2)  (K  2)}) and, without any loss
of generality, assume v̂(F  2) = (K  2)=
Also, for notational convenience, let ck = N(11 NV 0
k > ===> k )@k = Hence,
f 1 V 0
if ck = (ck > ===> ck ) is the vector of period 1, normalized (by k ) Lagrange
multipliers at the certainty equilibrium, ck = (===> cfk > ===) is the normalized (by
0k @N) period 1 vector of Lagrange multipliers at the associate n-invariant
equilibrium.
In the£ economy H̄ (and after
¤ normalizing row k by 0k @N> each k), ] \ is
\0
given by N] ] === ] \1 \N > with ] \0 = [0], while ] \n =
 Sunspots, Indeterminacy and Pareto Ine!ciency 195
5 6
9 c11 @s11 ··· 0 c21 @s21 · · · ··· · · · cV1 @sV1 :
9 :
9 .. .. .. .. .. .. .. .. :
9 . . . . . . . . :
9 :
9 0 · · · c1F1 0 ··· 0 ··· 0 :
9 :>
9 0 ··· 0 0 ··· 0 ··· 0 :
9 :
9 .. .. .. .. .. .. .. .. :
9 . . . . . . . . :
9 :
7 0 ··· 0 0 · · · cv̂K1 ··· 0 8
c1K @s11 · · · c1K c2K @s21 · · · cv̂K V
· · · cK @s V1

and
55 6 5 c1 66
N 0 1 cv̂1
£ ¤ 9 9 .. ::
 (NE) Z = 7 7 · · · · · · 8 7 . 88=
N 0 c1K cv̂K
Evidently, there is no loss of generality in assuming that ] \ has maximal
rank K: Relabelling spots and agents, it su!ces to pick agent 1 and K so
0 0
that cv1 @cv1 h6= cvK @cvK > for some viand v0. With reference to the previous claim,
set  K =  0K >  1K > 0>  3K > ===>  VK ,  1K 6= 0> and  vK > v 6= 1> 2> chosen so that
£ ¤
U(t) K = 0> which can always be done because | n1 · · · | nV is in general
position, for each n.
Consider the following sequence of column operations (we identify columns
in the
£ obvious way), ¤ focusing, for the moment, on their eects on the columns
of ] \ (NE) Z \ :
a. For each v, v̂  v A 0, subtract from column ( nv ) column (snv1 )
multiplied by sv1 (i.e., get rid of the nonzero coe!cients of the matrix Z \ ).
0
b. Subtract column (snvf )> f = n, from column (sn vn ) to eliminate all the
n0 vn
(collinear) columns (s )> n0 6= n=
c. Use the linearly independent columns (s111 ) and (s121 ) to eliminate
columns (s1v1 )> v A 2> and the nonzero columns of the matrix E.

Step 2 :
Rearrange the columns of the (transformed) matrix to obtain
 ¸
G(s\ >t>)   Gs  
,
0 ]
where ]  has full rank K. The square matrix G(s\ >t>)   is given by the
columns (modified by the operations under a, b and c above) relative to (s0f )>
f ? F> (snvf ), each nvf 6= nvn and f ? F> and (t l )= The other K columns
relative to (snvn ), v  v̂> have been replaced by the columns¡ (
nv
)=¢
 \
We need to show that, generically, rank G(s\ >t>)  = J + L =


Let [G({k >ek >k ) I RFk ]\1 be the matrix obtained from
[G({k >ek >k ) I RFk ]1 deleting the rows associated with commodity F, each
> and the last ( + 1) rows. By the implicit function theorem,
196 Tito Pietra
\
G(s\ >t>) }k = [G({k >ek >k ) I RFk ]\1 G(s\ >t>) I RFk =
Let G(s\ >t>) I RFk be the matrix obtained from G(s\ >t>) I RFk with the
operations performed inP step 1 and the substitution of columns just described.
Then, G(s\ >t>)   =  k [G({k >ek >k ) I RFk ]\1 G(s\ >t>) I RFk =

Claim. Assume that [G({k >ek >k ) I RFk ]\1 G(s\ >t>) I RFk = 0k Pk > where
Mk is a square (J\ + L) dimensional matrix of full rank, which does not
depend upon 0k . Then, given any open set Y (H̄), there is H̄ 1 5 Y (H̄) such
that (s> t) 5 I 1 (0) and udqnG(s\ >t>)   |H̄ 1 = (J\ + L)=

Proof. Given the previous claim, set  k = k = Then, agent h’s vector of
Lagrange multipliers becomes k (1  ) : Hence, the normalized vector does
not change, while 0 0
k = k (1  )=
Then, G(s\ >t>)  |H̄  = G(s\ >t>)   |H̄  (0k )Pk and, therefore, (0k ) is



an eigenvalue of the matrix Pk1 G(s\ >t>)   |H̄ = The set of eigenvalues of any
matrix is finite, hence, for almost all , udqnG(s\ >t>)   |H̄  = (J\ + L)=
¡ ¢
Therefore, to conclude, we just need to show udqnPk = J\ + L , for
some k. Pick k = 2 (notice that, as it will become clear in the sequel, it needs
to be k 6= 1> K. That’s why we need K A 2) and let
 1 2W ¸
1 2 2
[G({2 >e2 >2 ) I RF2 ] = ,
22 32

where 1 2 is well known to be ¡a (J + L) dimensional

¢ square matrix of
rank (J\ + L)= Also, let 12 be the (J\ + L) × (J + L) dimensional matrix
obtained deleting rows F> each = It is well-known that udqn12 = (J\ + L)
(see, for instance, Balasko and Cass (1991)). Moreover,
£ ¤W
12  (s> ) U(t) = 0
£ ¤W
so that  (s> ) U(t) spans the null space of 12 = To conclude the proof,
first we observe that the last ( + 1) rows of G(s\ >t>) I RF2 are identically
zero and that G(s\ >t>) I RF2 is homogeneous of degree 1 in 02 , so that we
can write
 0 ¸
 2 G2
G(s\ >t>) I RF2 =

0

and, therefore, [G({1 >e1 >1 ) I RF2 ]\1 G(s\ >t>) I RF2 = 02 12 G2 = Finally,
we show that:
a. G2 has full column rank;
£ ¤W
b. vsdqG2 _ vsdq  (s> ) U(t) = {0}.
¡ ¢
It follows that E21 2 = P2 has full rank J\ + L , as required.
 Sunspots, Indeterminacy and Pareto Ine!ciency 197

Bear in mind that, at a n-invariant equilibrium, snv and cnv 2 are n-invariant,
so that we can omit the index n, to stress the symmetry.
Consider the changes induced in the matrix G(s\ >t>) I RF2 by the sequence
of operations on the columns and by their final rearranging. The first (J\ +L)
columns of the modified matrix (using the normalization by N@02 ) are given
by the matrix (02 @N)G2 > whose last ( + 1) rows are identically zero (in step
1, we got rid of all the non-zero terms of these rows but }211 and }221 and the
two corresponding columns have been replaced by the ones associated with
 11 5and  12 )= Indeed, G2 = 6
NL \ 0 ··· ··· ··· ··· ··· 0
9 0 D11 · · · · · · FV11 · · · · · · FL11 :
9 :
9 ··· 12 12
· · · D · · · FV · · · · · · FL12 :
9 :
9 .. .. .. . . .. .. .. :
9 . . . . . . . 0 :
9 :
9 ··· · · · · · · · · · D 1V
· · · W 1V
· · · :>
9 :
9 . . . . . . . . :
9 .. .. .. .. .. .. .. .. :
9 :
9 ··· ··· · · · · · · · · · · · · DNV 0 :
9 :
7 0 ··· ··· ··· ··· ··· 0 NLL 8
0 ··· ··· ··· ··· ··· ··· 0
where:
5 v 6
c2 0
9 .. :
9 . :
9 :
9 v nvn
c2 s :
Dnv = 9 9
:>
:
9 .. :
9 . :
7 cv2 8
cv2 0
£ ¤W
i.e., the vector 0 cv2 sv2 · · · cv2 replaces the column referred to commod-
ity f = n (remember that we subtracted from column  nv column (s1v1 ) mul-
tiplied by sv1 in step 1a and, then, replaced column (snvn ) with the (modified)
column  nv )=
The matrices W nv > v  v̂> are given by all zero coe!cients, but the coef-
ficient corresponding to commodity f = n, which is equal to cv2 (remember
step 1b above).
The matrices Fv1w and FL1w , w = 1> 2, v A 2, reflect the operations of step
1c. Then, for v A 2, w = 1> 2, Fv1w has all the coe!cients equal to zero, but
the ones on the first row and column, given by cw2 f1w v > where
 11 ¸  1 11 ¸1  v v1 ¸
fv c1 @s c21 @s21 c1 @s
= =
f12
v c 1
K @s 11
c2
K @s 21
cvK @sv1

Similarly, FL1w > w = 1> 2> has all the coe!cients equal to zero, but the ones
on the first row and column, given by cw2 f1w L >where
198 Tito Pietra
 ¸  1 11 ¸1  ¸
f11
L c1 @s c21 @s21 N
= =
f12
L c1K @s11 c2K @s21 N
Bear in mind that these matrices just appear on the blocks of rows of
states 11 and 12, given the particular role played by the value of the excess
demands of agents¡1 and K in these two¢ states.
Let G2 be the (J + L) × (J\ + L) dimensional matrix given by the first
(J + L) rows of G2 . Given that  = 1, for the purpose of this computation,
we can drop it from the notation.

Claim. D2 has full column rank.

\
Proof. Suppose that G2 dW = 0, and let d = (d0 > d1 > ===> dN > d= ) 5 <J +L >
with d= = (d=1 > ===> d=L )= Evidently, d0 = 0> d= = 0 and dnv = 0> for each
v A v̂. For dnv > n A 1, given the structure of Dnv > dnvn = 0, because for each
nvn there is a row (the one corresponding to commodity nvF) with one and
only one non zero coe!cient. This immediately implies that dnvf = 0, each
c. For n = 1, the same argument applies to blocks nv, v A 2. Taking into
account these facts and given that D11 has full rank, d11 = d12 = 0=

Claim. Let L ? V, and 3  K. Then, for each open set Y (H̄), there is
an open and dense subset Y 0(H̄)  Y (H̄) such that, for each H̄ 0 5 Y 0 (H̄)>
£ ¤W
vsdqG2 _ vsdq  (s) U(t) = {0} =

Proof. Openness follows by regularity of the equilibrium (if Y (H̄) is su!ciently

small). Hence we just need to show the density part.
£ ¤W
If e 5£ vsdq  (s) U(t) > then ¤
e = e0 s0 > (e11 s11 > ===> e1V s1V )> ===> (eN1 s11 > ===> eNV s1V )> e= ,
£ ¤W
with e= = U(t)W e0 > ===> eN =
£ ¤ \
As above, let d = d0 > d1 > dN > d= 5 <J +L and, £by contradiction, ¤ as-
sume that there is e 5 <J+L \ {0} such that e 5 vsdq  (s)W U(t)W and
e = G2 d> for some d= The structure of e imposes several restrictions on the
coe!cients of the vector d= We are going to show that, generically in x2 >
they are satisfied if and only if d = 0. Hence, e = 0 and the contradiction
establishes the claim.
Observe that: £ ¤
a. Given that e0 s0 =  d01 N> ===> d0F1 N> 0 > d0 = e0 = 0=
b. For
¡ =1states 11 and 12, givenPthat, for ¢w = 1> 2> e1w s1w =
P
[c2 d fL + vA2 d1v1 f1w
w 1w
v + nA1 d
nw1
> ===> cw2 (d1w1 swf + d1wf )> ===>
t 1w1 1w w 1w1 1wf
c 2d
¡ =1 ]> itPmust be e P = c2 d = ¢ Hence, d = 0> F A f A 1, and
d f1w L + vA2 d1v1 1w
f v + nA1 dnw1
= d 1w1 w1
s =
P
c. For e1v > v̂  v A 2, e1v s1v = [cv2 nA1 dnv1 > (cs2 d1v1 sv2 + cs2 d1v2 )> ===>
cs2 d1v1 ]= Hence, it must be e1v = cv2 d1v1 and, therefore, d1vf = 0> f A 1, and
P nv1
nA1 d = -d1v1 sv1 =
 Sunspots, Indeterminacy and Pareto Ine!ciency 199
£ ¤
h d. For n A 1, P and v  v̂> env snv =  cs2idnv1 > ==> cv2 dnvn svf > ===> cs2 dnvn +
0 0
0> cv2 dnv2 > ===> cv2 n0 6=n dn vn > ===> cv2 dnvF1 > 0 = Hence, env = cv2 dnvn , and,
P 0 0
therefore, dnvf = 0> for f 6= 1> n> dnv1 = dnvn sv1 and n0 6=n dn vn = 0=
e. For s  v̂> given that env snv = cv1 (dnv > 0)> it must be dnv = env = 0>
each n. £ ¤W £ ¤W
f. Given that N d= = e=W = U(t)W e0 > ===> eN , and taking into
consideration what we have just established, it must be
£ ¤W £ ¤W
N d= = U(t)W 0 c12 d11 · · · cS2 dNV =
Bear in mind that, given that (by d) dnvf = 0> for all f, but f = 1> n (each
v and n A 1 and each 1v> v A 2), and that dnv1 = dnvn sv1 > we can denote dnvn
as dnv , for each v> n A 1> and for n = 1> v A 2=
To summarize, we must have:
i. ¡d0 = dnv = 0> for P v A 1v v̂ and each P n (because ¢ of a and e);
ii. d=1 f1w L @s w1
+ vA2 d f1w w1
1 @s + nA1 d nw
= d1w
> w = 1> 2 (because
of b andPremembering that dnv1 = dnvn sv1 = dnv > for v A 2 and each nv> n A 1);
iii. nA1 dnv = d1v > v A 2 (because of c);
P 0
iv. n0 6=n dn v = 0> n A 1 and each v (because of d);
£ ¤W £ ¤W
v. N d= = U(t)W 0 c12 d11 · · · cV2 dNV (this is f).
Ignore the subset of states such that it must be dnv = 0= Translating these
restrictions
5 on the vector d into matrix form, it must be 6
Lv̂ +
Lv̂ ··· Lv̂ F
9 L 0 · · · L 0 :
9 v̂ v̂ :
9 .. .. .. .
.. .. : W
9 . . . . :d
9 :
9 L L · · · 0 0 :
7 v̂ ³ ´ v̂ ³ ´ ³ ´ 8
\W \ \W \ \W \
\ (1) G c2 \ (2) G c2 · · · \ (N) G c2 NLL
= Y dW = 0> ¡ n1 ¢
nv̂ = v̂N+L
where,
5 11 11 now, d = 6 (===> d > ===> d ,...,d )5 < , F is the (v̂ × L) matrix
fL @s1 · · · 0
7 f12
L @s
21
··· 08>
0 ··· 0
while
5
is the (v̂ × v̂) matrix6
0 0 f11 3 @s 11
· · · f11
v̂ @s
11
9 0 0 f12 3 @s
21
· · · f12 21 :
v̂ @s :
9
90 0 0 ··· 0 :
9 :>
9 .. .. .. . . .. :
7. . . . . 8
00 0 ··· 0
\
\
³ ´ (n) is the (n-invariant) matrix of events n payos (in states 1> ===> v̂) and
\ ¡ ¢
G c2 is the matrix having diagonal coe!cients c12 > ===> cv̂2 = To conclude the
proof, we need to show that Y dW = 0 if and only if d = 0, i.e. that Y has
full rank. Use the last block of columns to get rid of all the assets payos
but inside money and drop the corresponding rows and columns. Subtract
200 Tito Pietra

the block of columns of n = 1 from the blocks n A 1 and add the blocks
of columns so obtained, for n = 2> ===> N  1> to the block relative to n = 1.
Finally, divide the last column by N and subtract it (multiplied by cv2 ) from
the 5the first block of columns, obtaining 6
W    5 11 11 6
9 0 1 · · · 0 fL @s :
9 21 8 :
9 .. .. . . .. 
· · · 
7 f12 L @s :
9 . . . . :
9 ··· :
9 0 · · · 0 1 :
9 :=
9 0 Lv̂ · · · 0 0 :
9 :
9 . . . . . :
9 .. .. . . .. .. :
9 :
7 0 0 · · · Lv̂ 0 8
[0] [0] · · · [0] N
Hence, Y has full rank provided that the top left (2 × 2) submatrix W has
rank 2, where
 ¸
1  c12 f11 L @Ns
11
c22 f11
L @Ns
11
W = 21 =
c12 f12 L @Ns
21
1  c22 f12
L @Ns¡ ¢
By direct computation, ghwW = N  c12 f11 L @s
11
 c22 f12
L @s
21
@N=
Fix s11 > s21 > c11 > c21 > c1K > c2K and N=
11 2
¡ Then,
¡ 2 remembering
2 1 1
the definition
1 2
¢ 2 1 of f1L 2 and ¢ f2 , one obtains ghwW =
1  (c1  cK )c2 + (cK  c1 )c2 @(c1 cK  c1 cK ) =
By construction, either (c21  c2K ) 6= 0 or (c1K  c11 ) 6= 0 or both. If, for
instance,
h the second iterm is non zero, perturb agent 2’s utility function by
 2 =  02 > 0>  22 > ===>  VK ,  22 6= 0> and  v2 > v 6= 1> 2> chosen so that U(t) 2 = 0>
£ ¤
which can always be done because | n1 · · · | nV is in general position, for
each n.
Notice that this last perturbation of the utility function must be applied
to some k 6= 1> K=

References
1. Balasko, Y.: The set of regular equilibria= Journal of Economic Theory
58, 1-8 (1992).
2. Balasko, Y., D. Cass: Regular demand with several, general budget con-
straints, in M. Majumdar (ed.): Equilibrium and Dynamics: Essays in
Honor of David Gale. London: MacMillan 1991.
3. Cass, D.: Sunspot and incomplete financial markets: The leading example,
in G. Feiwell (ed.): The economics of imperfect competition and unem-
ployment: Joan Robinson and beyond. London: MacMillan 1989.
4. Cass, D.: Sunspots and incomplete financial markets: The general case.
Economic Theory 2, 341-358 (1992).
5. Cass, D., A. Citanna: Pareto improving financial innovation in incomplete
markets. Economic Theory 11, 467-494 (1998).
 Sunspots, Indeterminacy and Pareto Ine!ciency 201

6. Cass, D., K. Shell: Do sunspots matter? Journal of Political Economy 91,

193-227 (1983).
7. Citanna, A., A. Kajii, A. Villanacci: Constrained suboptimality in incom-
plete markets: A general approach and two applications. Economic Theory
11, 495-522 (1999).
8. Geanakoplos, J.D., H.M. Polemarchakis: Existence, regularity and con-
strained suboptimality of competitive allocations when the asset market
is incomplete. In W.P. Heller, R.M. Starr, D. Starrett (eds.): Uncertainty,
information and communication: Essays in honor of K.J. Arrow, Vol. III.
Cambridge: Cambridge University Press, 1986.
9. Gottardi, P., A. Kajii: Generic existence of sunspot equilibria: The role of
multiplicity= Review of Economic Studies, 66, 713-732 (1999).
10. Lisboa, M.B.: On indeterminacy of equilibria with incomplete financial
markets. Mimeo (1994).
11. Magill, M., M. Quinzii: Real eects of money in general equilibrium. Jour-
nal of Mathematical Economics 21, 301-342 (1992).
12. Mas-Colell, A.: The theory of general economic equilibrium. The dieren-
tiable approach. Cambridge: Cambridge University Press 1985.
13. Mas-Colell, A.: Three observations on sunspots and asset redundancy, in
P. Dasgupta, D. Gale, O.D. Hart, E. Maskin (eds.): Economic analysis of
markets and games. Essays in honor of Frank Hahn. Cambridge: MIT
Press 1992.
14. Pietra, T.: The structure of the set of sunspot equilibria in economies with
incomplete financial markets. Economic Theory 2, 321-340 (1992).
15. Pietra, T.: Sunspot equilibria and e!ciency in economies with incomplete
financial markets: A remark. Journal of Economic Theory 60, 181-190
(1993).
16. Pietra, T.: The structure of the set of sunspot equilibria in economies with
incomplete financial markets: Variable asset prices. Economic Theory 18,
649-660 (2001).
17. Pietra, T., P. Siconolfi: Equilibrium in economies with financial markets:
Uniqueness of expectations and indeterminacy. Journal of Economic The-
ory 71, 183-208 (1996).
18. Siconolfi, P.: Sunspot equilibria and incomplete financial markets. Journal
of Mathematical Economics 20, 327-339 (1991).
19. Smale, S.: Global analysis and economics III: Pareto optima and price
equilibria. Journal of Mathematical Economics 1, 107-117 (1974).
20. Suda, S., J.-M. Tallon, A. Villanacci: Real indeterminacy of equilibria in a
sunspot economy with inside money. Economic Theory 2, 309-320 (1992).
Growth in Economies With Non Convexities:
Sunspots and Lottery Equilibria, Theory and
Examples ?

Aldo Rustichini1 and Paolo Siconolfi2

1
Department of Economics, University of Minnesota, Minneapolis, MN 55455,
USA [email protected]
2
Graduate School of Business, Columbia University, New York, NY 10027, USA
[email protected]

Summary. We investigate the relation between lotteries and sunspot allocations

in a dynamic economy where the utility functions are not concave. In an intertem-
poral competitive economy, the household consumption set is identified with the
set of lotteries, while in the intertemporal sunspot economy it is the set of measur-
able allocations in the given probability space of sunspots. Sunspot intertemporal
equilibria whenever they exist are e!cient, independently of the sunspot space spec-
ification. If feasibility is, at each point in time, a restriction over the average value
of the lotteries, competitive equilibrium prices are linear in basic commodities and
intertemporal sunspot and competitive equilibria are equivalent. Two models have
this feature: Large economies and economies with semi-linear technologies. We pro-
vide examples showing that in general, intertemporal competitive equilibrium prices
are non-linear in basic commodities and, hence, intertemporal sunspot equilibria do
not exist.
The competitive static equilibrium allocations are stationary, intertemporal equi-
librium allocations, but the static sunspot equilibria need not to be stationary, in-
tertemporal sunspot equilibria. We construct examples of non-convex economies
with indeterminate and Pareto ranked static sunspot equilibrium allocations asso-
ciated to distinct specifications of the sunspot probability space. Furthermore, we
show that there exist large economies with a countable infinity of Pareto ranked sta-
tic sunspot equilibrium allocations with aggregate pro capita consumption invariant
both across equilibria as well as across realizations of the uncertainty. This implies
that these equilibria are equilibria of a pure exchange economy with non-convex
preferences.

Key words: Lottery Equilibria, Sunspot Equilibria, Non-convexities.

JEL classification numbers: D84, D90.
B
We thank Herakles Polemarchakis for helpful conversations on the topic. The
research of Aldo Rustichini was supported by the NSF grant NSF/SES-0136556.
204 Aldo Rustichini and Paolo Siconolfi

1 Introduction
Two Views of Sunspot Economies

The sunspot literature, originated by the Cass and Shell, [3], studied convex
economic environments where the First Welfare Theorem does not hold. In
these economies the introduction of ”extrinsic uncertainty” may change the
set of equilibrium allocations. Indeed the last assertion became a Folk theo-
rem, [16]. Most notably, there are economies where Pareto e!cient certainty
equilibria (i.e., equilibria where sunspots do not matter) coexist with ine!-
cient sunspot equilibria (e.g., [1]). This first research saw sunspot as a socially
undesirable, but possible, and to some extent, pervasive, equilibrium outcome
of competitive environments.
In the recent years, research on sunspots has taken quite a dierent route.
The objects of study are economies with non-convexities (see for instance
[10]). It was soon clear that non convexities in production or in preferences
had very dierent implications.
When non-convexities are present only in the production sets of the firms,
the introduction of sunspots is immaterial. Given the linearity of the profit
functions, a sunspot equilibrium is an equilibrium of the "concavified" econ-
omy and, thus, sunspots do not matter, [2].
The situation is very dierent when non-convexities are present in the
consumers’ preferences or consumption set. Most of the literature deals with
static economies with non convexities of the consumption set, generated by
either indivisibilities, [17], or incentive compatibility constraints, [8]. In these
environments randomness in allocations may improve welfare. The latter can
be generated in two dierent ways, either by introducing lotteries or by in-
troducing sunspots. In the sequel, we use the term competitive equilibrium to
refer to the equilibrium outcome of an environment where individuals face a
complete set of markets for lotteries and the term sunspot equilibrium when
they face, for given specification of the probability space of extrinsic uncer-
tainty, a complete set of (sunspots) contingent markets.

Competitive Equilibria and Sunspot Equilibria

Two major dierences between the concept of competitive equilibrium and

sunspot equilibria are clear, and both of them are crucial. The first dierence
concerns the nature of the consumption sets. With the competitive notion,
agents may choose any probability distribution over basic consumption bun-
dles, as in [11]. With the sunspot notion, they choose measurable maps with
respect to an exogenously given probability space of extrinsic uncertainty
( > A> 
ˆ ).
The second dierence concerns the nature of prices in the two concepts.
In the lottery economy prices are linear in lotteries and, hence, they can
be non-linear in commodities. In the sunspot economy, prices are linear in
 Growth in Economies With Non Convexities 205

sunspot contingent commodities. Thus, a first natural question is to investi-

gate whether competitive and sunspot equilibria are "equivalent." There are
two aspects to this question. The first is at the level of individual behavior
and it concerns the dierent nature of the consumption sets under the two
specifications. The second is at an equilibrium level and it concerns the nature
of equilibrium prices. Any individual sunspot allocation induces a probability
distribution over consumption bundles and, hence, it induces a lottery. How-
ever, the viceversa need not be true. A lottery induces a stochastic allocation,
but this may be either not measurable with respect to the given probability
space of extrinsic uncertainty, ( > A)> or incompatible with the given sunspot
probability law,  ˆ . When each lottery induces a (compatible) sunspot allo-
cation, lotteries and sunspot allocations are equivalent. Garret, Kreister, Qui
and Shell [5] show that equivalence is achieved by any ”continuous randomiz-
ing device”3 . When the probability space of extrinsic uncertainty is not rich
enough, sunspot equilibrium allocations, if they exist, may be Pareto subop-
timal. Dierent specifications of ( > A>  ˆ ) may yield dierent equilibria with,
in principle, Pareto ranked allocations ([17], [4]). This is not a contradiction
of the First Fundamental Theorem of Welfare Economics. Pareto optimality
is relative to a given consumption set, while in the previous statements we are
comparing allocations in dierent consumption sets. 4
However, when the extrinsic uncertainty space is ([0> 1)> E> O), under fairly
general assumptions, every sunspot equilibrium allocation induces an equilib-
rium lottery allocation. If equilibrium allocations of the lottery economy are
supported by linear prices, the viceversa is also true, ([5] and [6]). This equiv-
alence result is based on the assumed existence of sunspot equilibria and,
equivalently, of lottery equilibria supported by linear prices.
Thus we are led to the second aspect of the problem: singling out environ-
ments where competitive equilibrium prices are linear. In any economy where
feasibility is a restriction on the average values of (joint) lotteries, equilibrium
prices may be restricted (without loss of generality) to be linear. One such
environment is a large pure exchange economy (with finitely many types).
If each household of type l consumes i.i.d. lotteries  l , by the law of large
numbers, the aggregate average allocation is, with probability 1, determin-
istic and equal to the average (appropriately taken over types and lotteries)
consumption. Feasibility in a large exchange economy requires that average
endowment to be equal to average consumption. Hence, feasibility is just a re-
striction on the average value of the lotteries. Thus, in large economies lottery
and sunspot equilibrium allocations are equivalent, [8] and [5].
3
For instance, the probability space ([0> 1)> E> O), with E denoting the Borel sub-
sets of [0> 1) and O the Lesbegue measure.
4
In this literature as well as in this paper, a feasible (sunspot) allocation is e!cient
if it does not exists a feasible lottery that Pareto dominates it. Furthermore, it is
understood that preferences are extended over random allocations via expected
utility.
206 Aldo Rustichini and Paolo Siconolfi

The Model

We depart from the existing literature in two aspects. First, the lack of convex-
ity is generated by the preferences of the households and not by the nature
of their consumption set. In previous work, [12], we have studied the Lan-
caster model of characteristics, [9]. When the household production function
transforming commodities into characteristics is concave, but not linear, the
derived utility function over commodities (i.e., the composition of the concave
utility function over characteristics with the concave production function) can
be any continuous function. Furthermore, a long history of experimental ev-
idence indicates the lack of linearity of the household production function.
Thus, the Lancaster model of characteristics provides a sound justification for
the study of economies with non-convex preferences.
The second key feature of our model is time. We study a continuous time
growth economy. In this economy lotteries appear naturally because they are
the only sensible way to define the limit of allocation paths (over time) that
are in the limit achieving the supremum in the growth problem. Since the
utility function is not concave, an optimal deterministic path may not exist:
however, the sequence of paths yielding the supremum value is oscillating
among extreme points. The natural limit of these paths is a path taking
values in lotteries. Hence, in our model, the household consumption set is
identified with the set of lotteries and the feasibility requirement equates the
eective investment to the average production of investment goods at each
instant of time. The definitions of the consumption set and of the feasibility
requirement are, in an obvious sense, the outcome of the optimal use of time as
a convexifying device. The same type of mechanism is at work for competitive
versions of these growth economies. Households facing a budget constraint
reproduce, by varying their consumption over time, the distribution of their
consumption over goods induced by a lottery. In a way they would, by their
intertemporal choices, extend the commodity space. This extension is precisely
our “natural” choice of commodity space.

A Growth Economy

The growth economy provides thus a natural environment to study the rela-
tion between lotteries and sunspot allocations. We examine this relation by
using several specifications of the economy, dierentiated by time character-
istics and the number of households. The competitive equilibrium describes
the growth economy where the household consumption set and the feasibility
requirement are as previously described and there exist complete markets for
time contingent lotteries. In a sunspot equilibrium, for given (time invariant)
sunspot probability space ( > A>  ˆ ), the consumption set of the individuals
is identified, at each instant of time, with the set of ( > A)-measurable con-
tingent commodity allocations. Feasibility, conformably with the notion of
 Growth in Economies With Non Convexities 207

competitive equilibrium, requires, at each instant of time, the eective invest-

ment to be equal to the average production of investment goods. The average,
this time, is taken over the realization of the sunspot allocation according to
the sunspot probability measure  ˆ.
In a (finite) pure exchange economy with a representative agent, feasi-
bility requires the sunspot allocations to be equal to the sunspot invariant
endowments. Thus, sunspots do not matter. However, in the growth economy,
the adopted notion of feasibility allows for non trivial randomness of feasible
allocations even in the model with a single representative consumer.
An equilibrium can be either intertemporal or static and the economy can
be with a single representative household or with a large number of iden-
tical ones. The last distinction is immaterial in convex economies, but has
important consequences when preferences display non-convexities.
The intertemporal economy is just the standard competitive version of
the growth economy. We compare the intertemporal competitive equilibrium
allocations with the intertemporal sunspot equilibrium allocations. If, at each
competitive equilibrium, lotteries are priced linearly, a quite strong result
holds true: competitive allocations and sunspot allocations coincide. Also,
without loss of generality, probability adjusted prices are sunspot invariant.
Most importantly and in the contrast with the literature, independently of
the sunspot space specification, there cannot exists ine!cient sunspot equilib-
ria. The reason is quite simple. If a sunspot probability space does not provide
enough randomness to reproduce the desired lotteries, time supplies the ad-
ditional missing noise. This in turn implies that it cannot exists a solution
to the household programming problem or, equivalently, that there cannot
be an intertemporal sunspot equilibrium for "inadequate" specifications of
( > A> ˆ ). Hence, the key question is to single out environments guarantee-
ing price linearity of the lotteries in basic commodities. As already argued,
the large economy is always one of such environments. The second requires
a non-generic restriction on the technology that we call semi-linearity. With
a semi-linear technology, for each given vector of capital goods, the isoquant
curves of investment and consumption goods are linear (as in the text book
growth model). Thus, a lottery is feasible if its average value is feasible. The
latter is the key desideratum to obtain linear supporting prices.
However, with a representative household (or with finitely many), for gen-
eral specifications of the technology, lottery prices are non-linear and sunspot
equilibria do not exists. We provide examples of this fact by using the static
version of the economy. The notion of static equilibrium is the restriction of
the definition of intertemporal equilibrium to vectors of allocations and prices
that are constant over time. Thus, each static competitive (lottery) equilib-
rium is a stationary, intertemporal competitive equilibrium and viceversa.
Stationary allocations are important, simple to analyze and they are con-
structed, in standard economies, using our notion of static equilibria. Indeed
in standard, convex environments, static and stationary intertemporal equilib-
ria coincide. However, in our environment, sunspot (static) allocations behave
208 Aldo Rustichini and Paolo Siconolfi

dierently. There are non-convex economies with indeterminate and Pareto

ranked static sunspot equilibrium allocations associated with distinct spec-
ifications of probability space ( > A> ). By what previously said, ine!cient
static sunspot allocations can never be stationary, intertemporal sunspot equi-
librium allocations. Thus, when preferences are not convex, static sunspot
equilibria are an inadequate tool to analyze stationary equilibrium allocations
of growth economies.
Furthermore, we show that there exist large economies with a countable
infinity of Pareto ranked static sunspot equilibrium allocations with aggregate
pro capita consumption invariant both across equilibria as well as across real-
izations of the uncertainty. We select the technology to make sure that these
equilibria are equilibria of a large pure exchange economy with non-convex
preferences.

2 The Representative Household Economy

We model a growth economy with a representative household. The model
is standard except for the lack of convexity of the household’s preferences.
The particular notions of competitive and sunspot equilibria that we use are
much related to the solution of an optimal growth problem with, potentially,
a non-concave utility function. Hence, in order to clearly explain the various
adopted definitions of equilibrium we turn next (and quickly) to the optimal
growth problem. A detailed analysis of the growth problem is in Rustichini
and Siconolfi ([13]).

2.1 The Optimal Growth Problem

Consider a growth economy with a representative individual. The consump-

tion set in the basic commodity space is \> a non-empty, compact and convex
subset of <F + . The instantaneous utility function, X : \ $ <> is continuous,
but not necessarily concave. A vector of capital goods is { 5 <N + . The contin-
uous and jointly concave function I : <N + ×\ $ < N
+ describes the technology.
Let |ˆ : \ $ <+ denote a measurable map, |ˆ = (ˆ | (w))w0 . For given initial
condition {ˆ(0) = {0 , the optimal growth problem is:
Z 4
sup huw X (ˆ
| (w))gw (1)
|ˆ 0

subject to the constraint:

{(w)
˙ = I (ˆ ˆ(0) = {0 , |ˆ(w) 5 \=
{(w)> |ˆ(w))> { (2)

If X is not concave, an optimal solution to (1) may not exist.

The reason is simple. If preferences are not convex, a consumption path
(that is, the |ˆ path) can give, over some time interval, a strictly higher utility
 Growth in Economies With Non Convexities 209

than the path equal to the average over time of the original path. So variability
over time is, from the point of view of the allocation of consumption, desirable.
But the opposite is true from the point of view of production: a variable
consumption path can yield a strictly smaller asset accumulation than the
path equal to the average over time. Making the variability occur in shorter
and shorter time periods can reconcile the two desiderata. The utility is still
going to be roughly equal to the average of utilities in dierent points in
time. The loss in production is going to be smaller and smaller because the
resulting oscillations in the { path are smaller and smaller. But the limiting
measurable optimal path may not exist. The next example illustrates this in
a simple economy.
Example 1. An economy with no optimal growth path
The economy is described by F = N = 1, and I ({> |) = i ({)  |, with i
strictly concave and such that i 0 (1) = u and i (1) = 1. The initial condition
for the capital stock is {0 = 1.
The utility function X is monotonically increasing, but not concave. There
is however a concave function Y that satisfies Y (|) = X (|) for | 5 [0> 1@2] ^
[3@2> 4], and Y (|) = X (1@2) + (1  )X (3@2) A X (|)> for each | = (1@2) +
(1  )(3@2),  5 (0> 1)=
The optimal growth problem when the utility function is Y (rather than
X ) has, given the choice of our initial condition, the unique solution |(w) =
{(w) = 1, for all w. The value to the problem is 21 (X (1@2) + X (3@2)) = Y (1).
Furthermore, since Y (|)  X (|) , for all |, the value to the growth problem
(1) with the utility function Y is greater or equal to the value of the problem
with the utility function X .
Consider now the growth problem with X . Fix an arbitrary time interval.
Consider the consumption path |ˆ that alternates period by period the con-
sumption levels 1/2 and 3/2. For u close to zero, the value of this path X (ˆ| ) is
close to 12 (X (1@2) + X (3@2)) = Y (1)= Since the production function is strictly
concave, variability is costly. Thus, since the average (over time) consumption
of the path is 1, |ˆ is, given the initial condition, not feasible. However, by
making the time period arbitrarily small, the cost of the variability converges
to zero and the consumption path converges (in the appropriate topology, the
narrow topology: see below, and [18] for details) to the time invariant path
in the space of probability measures that assigns to both consumption levels,
1/2 and 3/2, a probability 1/2.

2.2 The Relaxed Problem

Let M+>1 (\ ) denote the set of Borel probability measures on \ . For any prob-
ability distribution  5 M+>1 (\ )> and any real valued function
R j(·> ·), j(·> )
denotes the expectation of j according to , i.e., j(·> ) = \ j(·> |)(g|).
Let ˆ : <+ $ M+>1 (\ ) be a measurable map. The relaxed (weak) problem
associated to 1 is:
210 Aldo Rustichini and Paolo Siconolfi
Z 4
sup ˆ
X ((w))gw (3)
ˆ
 0

subject to the constraint:

{(w)
˙ = I (ˆ ˆ
{(w)> (w))> ˆ 5 P+>1 (\ )
ˆ(0) = {0 > (w)
{ (4)

In the programming problem (3), the planner can choose a path of lotteries
over the consumption set and the feasibility constraints are written in average.
The value of (3) is greater or equal to the value of (1). Furthermore, in (3), the
utility function is linear in the control variable,  5 M+>1 (\ ). Hence, under
our specification, there exists an optimal solution (ˆ ˆ to (3), while (1) may
{> )
not have an optimal solution. A first known result shows that the supremum
of the problem (1) coincides with the value of the relaxed problem (3). Most
importantly, a second known result shows that the sequence of deterministic
trajectories with value approximating the supremum of (1) converges in an
appropriate topology (the narrow topology) to (ˆ ˆ the optimal solution to
{> )>
(3) (see [18] for details).
These two results justify our characterization of e!cient paths by means
of the programming problem (3). However, the interpretation should be clear.
The opportunity of selecting paths of lotteries over \ as well as the average
form of the feasibility requirements are not special assumptions. They emerge
from the optimal use of time as a convexifying device.

3 Intertemporal Competitive Equilibrium

The definition of competitive equilibrium is the natural extension of the con-
cept of competitive equilibrium for an economy with complete markets where
the household has a non-concave utility function. Faced with a budget con-
straint, the household will try to randomize consumption to achieve higher
utility. If a complete market for lotteries over consumption sets exists, he
will be able to do this. The randomization over consumption corresponds to
the randomization he can achieve in the dynamic economy by altering con-
sumption over time (chattering). The possibility of choosing lotteries over the
space of basic consumption goods allows him to randomize at each point in
time, rather than across time as he does with chattering. Hence, we identify
the consumption set, at each w, with M+>1 (\ ). The (lottery) price domain is
M+>1 (\ ) > the dual of M+>1 (\ ). It contains (an isomorphic copy of ) the set
of continuous functions on \ . We write ? ŝ>  ˆ A the dual pair. We denote by
O the Lebesgue measure on the real line. Firm and household allocations are
vectors of paths
(ˆ
{i > { ˆ >
ˆk > d̂i > d̂k >  ˆ )
i k

where, {ˆi (w) and { ˆk (w) 5 <N are, respectively, the supply and the demand
of capital , d̂i (w) and d̂k (w) 5 <N are demand and supply of investment,
 Growth in Economies With Non Convexities 211

ˆ (w) and 
and  ˆ (w) 5 M+>1 (\ ) are supply and demand of (lotteries over)
k i
consumption allocations. Prices are vectors of paths:

(t̂> ê> ŝ)

where ê(w) 5 <N is the rental price of capital, t̂(w) is the price of investment
good, and ŝ(w) 5 M+>1 (\ ) is the price of consumption allocations. The
space of allocations is a subset of the space of measurable functions from
<+ to <2N × M+>1 (\ ). The precise space is the set of measurable functions,
(ˆ ˆ such that the discounted O1 norm is finite. Prices are linear functional
{> d̂> )>
on this space, hence measurable functions with finite, discounted essential sup
norm. The inner product is defined in the natural way:
Z +4
? (t̂> ê> ŝ)(ˆ ˆ A
{> d̂> )) [t̂(w)ˆ ˆ A]gw=
{(w) + ê(w)d̂(w)+ ? ŝ(w)> (w)
0

In the intertemporal economy, the firm and the household solve the following
problems.

Definition 1. For a given vector of prices, the firm’s problem is:

Z +4
max [ê(w)ˆ
{(w)+ ? ŝ(w)> ˆ (w) A +t̂(w)d̂(w)]gw (5)
({>d>) 0

subject to
{(w)> ˆ (w)) O  almost every w;
d̂(w) = I (ˆ (6)
R +4
Let y(t̂> ê> ŝ) denote the value of the firm, i.e., y(t̂> ê> ŝ) = 0 [ê(w)ˆ
{(w)
+ ? ŝ(w)> ˆ (w) A +t̂(w)d̂(w)]gw, where (ˆ ˆ is a solution of the firm’s problem
{> d̂> )
(5) at (t̂> ê> ŝ);

Definition 2. For a given vector of prices and value of the firm y(·), the
consumer’s problem is:
Z +4
max huw X (ˆ
 (w))gw> (7)
({>d>)(w) 0
Z +4 Z +4
subject to ? ŝ(w)> ˆ (w) A gw  {(w)  t̂(w)d̂(w)]gw  y(t̂> ê> ŝ);
[ê(w)ˆ
0 0
and
ˆ(0) = {0 > {˙ = d̂(w)> O  a.e.
{

Although
R +4 X may be a non-concave function of | 5 \ , X> and, there-
fore, 0 huw X (·)gw, is a linear function of the lotteries,  5 P1>+ (\ ). Fur-
thermore, if at equilibrium, preferences satisfy local non satiation, the path
(ˆ
{k > d̂k ) is chosen to maximize the wealth of the household. Wealth maxi-
mization is independent of the particular shape X and it is always a linear
212 Aldo Rustichini and Paolo Siconolfi

problem. Hence, (7) is a concave programming problem. The nature of the

investment at time w, {(w)
˙ requires some clarification. Investment is a random
variable, dependent on the realization of the random variable input. However,
since this quantity enters linearly into the consumer’s problem, we may as-
sume that the consumer chooses a deterministic investment at each time. The
market clearing condition on investment is satisfied O- almost surely because
the market clearing condition on the stock of capital in each period is satisfied.
If we write the market clearing condition explicitly, it is stated as an equality
between the expected value of demand and supply of investment. We identify
an economy with an array (X> I> {0 ).

Definition 3. A competitive equilibrium of the economy (X> I> {0 ) is a vector

of prices and allocations

(t̂> ê> ŝ; {

ˆi > { ˆ >
ˆk > d̂i > d̂k >  ˆ )
i k

such that
1. (ˆ
{i > {
ˆk > d̂i ) is a solution of the firm’s problem (5);
2. (ˆ ˆ ) is a solution of the consumer’s problem (7);
{k > d̂k >  k
3. Markets clear:

 ˆ (w)> {
ˆ (w) =  ˆk (w)> d̂i (w) = d̂k (w)> O  almost every w;
ˆi (w) = {
i k

In the framework we have described, the economy is a neoclassical econ-

omy with concave (linear) preferences and convex technology. Hence under
technical assumptions a competitive equilibrium exists, is e!cient, and the
second welfare theorem holds.

3.1 Static Equilibrium

The notion of static equilibrium is the restriction of the definition of intertem-

poral equilibrium to vectors of allocations and prices that are constant over
time.

Definition 4. A static equilibrium is a vector

(t> d> s; {i > di >  i >  k )

where { 5 UN ,  i >  k 5 M+>1 (\ ); s 5 M+>1 (\ ) , t 5 <N , and y 5 U such

that
1.  k is a solution of the consumer’s problem:

max X ()> subject to ? s>  A y; (8)

5M+>1 (\ )
 Growth in Economies With Non Convexities 213

2. The firm’s value gross of the capital expenses is

y =? s>  i A +td0

3. ({i > di >  i ) is a solution of the firm’s problem:

max ut{0 + [? s>  A +td0 ] subject to d0  I ({0 > ) (9)

({0 >)

4. markets clear:
I ({>  i ) = 0>  k =  i > di = 0

The relation between the static and the intertemporal equilibrium notions
is fairly obvious. A stationary, intertemporal equilibrium is a static equilib-
rium. Viceversa, a static equilibrium ({> > s> t) is a stationary equilibrium of
the intertemporal economy, i.e., given the initial condition {(0) = {, the path,
(t̂> ê> ŝ> { ˆ
ˆ> d̂> )(w) = ({> 0> > huw (t> 0> s)), for all w, is an equilibrium of the in-
tertemporal economy. Hence, the stationary allocation ({> ) is an e!cient
allocation of the intertemporal economy. A static allocation ({> ) is e!cient
if it is an e!cient stationary allocation of the intertemporal economy, i.e., if
there exists a t 5 <N such that ({> ) 5 arg max X () + tI ({> ). (see, [13])=

4 Static and Intertemporal Sunspot Equilibria

In agreement with the definitions of competitive equilibria, we give intertem-
poral and static definitions of sunspot equilibria. The common element is the
existence of an exogenous (and, for sake of simplicity, time invariant) stan-
dard measurable space of extrinsic uncertainty (sunspots) ( > A) together
with an exogenous probability,  ˆ . The definitions of sunspot equilibria is a
simple restatement of the definitions of competitive equilibria in an environ-
ment where there is a complete set of contingent (on the realization $ 5 )
commodities, but where lotteries are absent. Capital and investment good
prices are restricted to be sunspot invariant. As already discussed, this is
a consequence (given the sunspot invariance of the prices t) of the concav-
ity in (ˆ{> d̂) of both the firm and the household programming problem. Let
B<+ denote the Borel sets of <+ . In the intertemporal economy, the space
of allocations is the space of paths of ( × <+ > A × B<+ )measurable func-
tions (ˆ{> d̂> |ˆ) : × <+ $ <2N+ × \ satisfying the restriction ”(ˆ {> d̂)($> w) is
R +4 uw
$invariant”, with norm k (ˆ {> |ˆ) k1>u  0 h k{ ˆ(w)> sup$5 |ˆ($> w) k gw,
finite. Prices are linear functional on this space,Rhence measurable functions
with finite norm k (t̂> ê> ŝ) k4>u  sup(w)5U+ huw k (t̂> ê> ŝ($))(w) k (g$).
Hence, ŝ($> w) is an element of <O + for each ($> w), and, thus, prices are linear
(in contingent commodities). In the static economy, the space of allocations
is the space of ( > A)measurable functions (ˆ {> d̂> |ˆ) : + $ <2N+ × \ satis-
fying the restriction that ”(ˆ {> d̂)($) is $invariant. ”, while prices are linear
214 Aldo Rustichini and Paolo Siconolfi

functionals over this space. As for the notion of competitive equilibrium, in a

sunspot equilibrium, the equivalent constraint of equation 6 in the definition
1 reads Z
d̂(w) = I (ˆ  (g$) O  almost every w;
{(w)> |ˆ($> w))ˆ (10)

while for a static sunspot equilibrium, feasibility requires, in agreement with
the definition of a static sunspot equilibrium:
Z
0=  (g$) O  almost every w;
I ({> |ˆ($)ˆ (11)

We do not explicitly write down the definitions of static and intertemporal

sunspot equilibria because, modulo the change in the allocation sets, are iden-
tical to the definitions of intertemporal and static competitive equilibria.
When, judging the e!ciency of sunspot allocations we use the following
definition of e!ciency:

Definition 5. A feasible sunspot allocation (ˆ {> d̂> |ˆ) of the intertemporal econ-
omy (X> I> {0 > ( > A>  ˆ )), is e!cient if it does not exists a feasible (lottery)
ˆ 0 ) of the economy (X> I> {0 ) such that
{0 > d̂0 > 
allocation (ˆ
Z Z Z
huw ( X (ˆ  (g$))gw ? huw (X (
| ($> w))ˆ ˆ 0 (w))gw=

5 Sunspot and Lottery Equilibrium Allocations

5.1 Sunspot Equilibria in Finite Economies

As already explained, two major dierences between the concept of competi-

tive equilibrium and sunspot equilibria are clear and crucial. The first dier-
ence concerns the nature of the consumption sets, while the second concerns
the nature of prices. We start by analyzing the first dierence.

Sunspot and Lottery Allocations

In economies with non convexities, there are two aspects to the link between
lottery and sunspot allocations. 1) The space of sunspots ( > A>  ˆ ) must be
rich enough to provide the randomness necessary to reproduce lotteries. 2)
Since ( > A>  ˆ ) is the common randomization device, individual sunspot al-
locations must be appropriately correlated to be >  ˆ  d=h=$, feasible. This
second aspect is obviously absent in our representative economy. Later, when
we study the large economy, 2) will manifest necessarily itself. The first aspect
is however present and it needs to be clarified.
In finite, complete markets economies dierent specifications of the sunspot
space may yield dierent and Pareto ranked sunspot equilibrium allocations
 Growth in Economies With Non Convexities 215

([17], [4]). In section 8, we work out an example of a large economy with

non convex preferences, which may describe a static economy as well a pure
exchange one, where distinct specifications of the space of sunspot yield a
countable infinity of Pareto ranked sunspot allocations. The existence of mul-
tiple sunspot equilibria provides the rationale for the search of ways to refine
sunspot equilibria ([7], [6]).
We identify the set of sunspots with the interval [0> 1), their probability
with the Lebesgue measure O and we allow for dierent specifications of the
algebra> Ê, with Ê  E, the Borel sets over the interval [0> 1). As shown
in [5], ([0> 1)> E> O) (or any probability space isomorphic to it) accomplishes
desideratum 1), in the following precise sense: Let F be a Borel subset of \
and observe that |ˆ1 (F) is a Borel subset of [0> 1), for |ˆ 5 \ ([0> 1)> E), the set
of ([0> 1]> E)  phdvxudeoh functions from [0> 1) to \ . A lottery  5 P1>+ (\ )
and an allocation |ˆ 5 \ ([0> 1)> E) are equivalent, |ˆ  , if for any F :
| 1 (F)) = O{$ : |ˆ($) 5 F}, and, therefore,  = O  |ˆ1 .
(F) = O(ˆ
[5], in Lemma 4, proves that for each  5 P1>+ (\ )> there are |ˆ 5
\ ([0> 1)> E) such that |ˆ   and viceversa. Also, by the same argument,
for each given sunspot space ( > A>  ˆ ) and each | 5 \ ( > A) there exists
|ˆ 5 \ ([0> 1)> E> O) such that |ˆ  |ˆ. Hence, {[0> 1)> E> O} is rich enough to
reconstruct lotteries as well as any sunspot allocation measurable with respect
to any arbitrarily given standard probability space ( > A>  ˆ ). From now on,
we identify a sunspot economy with the array (X> I> {0 > Ê), Ê  E. While, as
already stated, a lottery economy is an array (X> I> {0 ).

Constant Price Sunspot Equilibria

Again following [5], a key role in our analysis is played by a particular notion
of sunspot equilibria, that we call constant prices sunspot equilibria, hereafter,
CPSE. An intertemporal CPSE is an intertemporal sunspot equilibrium with
( > A> ˆ ) = ([0> 1)> E> O) and ŝ(w> $) = s(w), for all $ 5 [0> 1) and some s(w) 5
RO . The natural adaptation of the last definition to the static environment
provides the definition of CPSE for the static economy. Whenever we talk
about a CPSE of some economy, it is understood that Ê = E so that we can
talk of a CPSE of the economy (X> I> {0 ) without ambiguity.
In pure exchange economies with non-convexities, CPSE have a very pow-
erful property. Under fairly general conditions, these two results hold true.
1) To each lottery equilibrium supported by linear prices it corresponds an
equivalent CPSE and viceversa ([5], Theorem 3). 2) The sunspot invariance
price restriction is without loss of generality.
Thus, in finite economies, whenever CPSE exist, they perform as well as
linearly priced lotteries. However, the existence of CPSE (or, in general, of
any sunspot equilibrium) is related to the form of the feasibility constraint.
If the latter is a restriction on the average (sunspot or lottery) allocation (as
216 Aldo Rustichini and Paolo Siconolfi

in a large economy), CPSE exist, otherwise, their existence as well as the

existence of any sunspot allocation may fail. Equivalently, as we will clarify
later, CPSE exists if competitive (lottery) equilibria have linear supporting
prices, otherwise they do not. The example that follows illustrates this simple
point.

Example 2. A one period economy without sunspot equilibria.

We consider a one period, two commodities, pure exchange economy with

a unique individual. The feasibility restricts the realizations of lotteries and
sunspot allocations (as opposite to their average). In this economy, the unique
e!cient allocation can be supported in the lottery economy, but not in the
sunspot one. p
The consumer utility function is X (|) = j (|1 )2 + (|2 )2 , j A 0, and the
endowments are h = (1> 1). For all j, the unique e!cient allocation coincides
with h.

In the economy with lotteries, for given prices s 5 M+>1 (\ R ) , the consumer
seeks a lottery  5 arg{max5P+>1 (\ ) X () subject to |5\ s(|)(g|) 
s(h)}, while feasibility restricts lotteries into the set { 5 M+>1 (\ ) : {| 5
\ : |  h}} = 1. The e!cient lottery allocation is  h (i.e., the degenerate
lottery that assigns probability 1 to h) supported by, for instance, the price
s(|) = X (|) 5 M+>1 (\ ) .
In the economy with sunspot space ([0> 1)> E> O), we restrict, without loss
of generality, prices to be $  lqyduldqw and thus, Rfor given s 5 R2++ the
Rindividual seeks a sunspot allocation |ˆ 5 arg{max X (ˆ | ($))g$ subject to
s(ˆ| ($)  h)g$  0}. Feasibility restricts the sunspot allocations in the set
{ˆ| : O{ˆ | : |ˆ($)  h} = 1. If j  2, the optimal solutions are, for each $, the
consumption bundle ( s1s+s 1
2
> 0), if s1 ? s2 , (0> s1s+s
2
2
), if s2 ? s1 > or the pair
(2> 0) and (0> 2), if s1 = s2 . Thus a CPSE (and more generally, any sunspot
equilibrium) does not exist.
Furthermore, in the sunspot economy, for j 5 (0> 2), there is not a solu-
tion to the consumer problem. To make this point, suppose, without loss of
generality, that s1  s2 and sh = 1. Pick % A 0, set |ˆ() = 0, for  A % and
R j2
|ˆ() = (1@s1 > 0), for  5 [0> %). Then, X (ˆ | ($))g$ = (1@s1 )2@j (%) j which
diverges to infinity as % $ 0=
In the same economy, with a large number of identical individuals, sunspot
equilibria exist and are e!cient. The reason is simple. In the large lottery pure
exchange economy, by the law Rof large numbers, feasibility restricts the aver-
age lottery consumption, i.e., \ (|  h)(g|)  0= Apparently, the situation
is dierent for the sunspot economy, where feasibility restricts, realization by
realization, the average taken across individuals of the sunspot allocation.
However, the sunspot allocation can be appropriately correlated across indi-
viduals in order to make the average taken across individuals equal to the
average taken across sunspot realizations. This point is clarified in sections 7
and 8.
 Growth in Economies With Non Convexities 217

There are two important dierences between the growth economy and the
finite economy that change some aspects of the relation between lotteries and
sunspot. To these topics we turn next.

5.2 Intertemporal Sunspot Equilibria

As already said, in the growth economy, in addition to the probability space
([0> 1)> Ê> O), the representative household can use time as a convexifying de-
vice. The latter has two implications. First and in contrast to the finite econ-
omy results, intertemporal sunspot equilibria of the economy (X> I> {0 > Ê) are
e!cient (in the sense of the definition (5)) for any Ê  E. Second, all in-
tertemporal sunspot equilibria are, without loss of generality, CPSE. To avoid
confusions, we are not showing that sunspot equilibria exist, but rather that,
when they exist, they are e!cient and CPSE. The first property implies that
([0> 1)> Ê) provides, at each w, enough variability to reproduce the optimal lot-
tery and that the latter, by definition of sunspot, can be supported by linear
prices. This second property has a key implication. When the representative
agents maximize facing linear prices, they optimally select allocations yield-
ing the value of the concave regularizations of their objective functions. The
next Lemma makes precise this statement. Let Y = co X be the concave
regularization of the utility function X , since \ is compact, Y can be defined
as:
Y (|)  max5P1>+ (\ ) {X (), subject to |() = |}, for | 5 \ .

For given Ê  E, household wealth Z and price trajectory ŝ consider the

two following programming problems:
Z Z
(X ) max huw ( (X (ˆ
| ($> w))g$)gw,
|
ˆ(w)5\ ([0>1)>Ê) <+ [0>1)
Z Z
vxemhfw to ( | ($> w)g$)gw  Z
ŝ($> w)ˆ
<+ [0>1)

and Z Z
(Y ) max huw ( (Y (ˆ
| ($> w))g$)gw,
|ˆ(w)5\ ([0>1)>Ê) <+ [0>1)
Z Z
subject to ( | ($> w))g$)gw  Z=
ŝ($> w)ˆ
<+ [0>1)

The next lemma explains the relations between the optimal solutions to (X )
and (Y ). Its proof is deferred to the appendix.
Lemma 1. Any optimal solution to (X )> |ˆ , is an optimal solution to (Y ).
Furthermore,
Z Z Z Z
huw ( |  ($> w))g$)gw =
(X (ˆ huw ( |  ($> w))g$)gw=
(Y (ˆ
<+ [0>1) <+ [0>1)
218 Aldo Rustichini and Paolo Siconolfi

We now exploit Lemma 1 to show that each intertemporal sunspot equi-

librium of (X> I> {0 > E) is a sunspot equilibrium of (Y> I> {0 > E). To simplify
notation we denote by  ˆ prices and by ˆ allocations.

Proposition 1. Let (ˆ  > ˆ ) be an intertemporal sunspot equilibrium of the

economy (X> I> {0 > Ê). Then, (ˆ   > ˆ ) is a sunspot equilibrium of the economy
(Y> I> {0 > Ê). Furthermore,
Z Z Z Z
huw ( |  ($> w))g$)gw =
(X (ˆ huw ( |  ($> w))g$)gw=
(Y (ˆ
<+ [0>1) <+ [0>1)

Proof. The economies (Y> I> {0 ; Ê) and (X> I> {0 > Ê), when facing identical
price trajectories, have identical profit and household’s wealth maximization
problems. Thus, we just need to prove that |ˆ is an optimal solution to
Z Z
max huw ( (Y (ˆ
| ($> w))g$)gw
|ˆ(w)5\ ([0>1)>Ê) <+ [0>1)

subject to
Z Z Z Z
( | ($> w))g$)gw 
ŝ($> w)ˆ ( |  ($> w)g$)gw=
ŝ($> w)ˆ
<+ [0>1) <+ [0>1)

This follows immediately from Lemma 1 which implies as well the second part
of the proposition.
The economy (Y> I> u; Ê) is a concave economy. Therefore, sunspot do not
matter. This is the key property that we exploit in order to show both that
sunspot equilibria are e!cient and that they are, without loss of generality,
CPSE.
Proposition 2. Let (ˆ  > ˆ) be a sunspot equilibrium of the intertemporal econ-
omy (X> I> {0 > Ê). Then: 1) ˆ is an e!cient allocation of the economies
(X> I> {0 ) and (Y> I> {0 ); 2) (ˆ > ˆ) is, without loss of generality, a CPSE of
the economies (X> I> {0 ) and (Y> I> {0 ).
Proof. By proposition 1> (ˆ > ˆ) is a sunspot equilibrium of (Y> I> {0 > Ê). Since
the latter is a concave economy, by the First Fundamental Theorem of Welfare
Economics, ˆ is an e!cient allocation (according to definition 5) of the econ-
omy (Y> I> {0 ). Suppose that, by contradiction, there exists a feasible lottery
allocation (ˆ ˆ  ) of the economy (X> I> {0 ) such that:
{ > d̂ > 
Z Z Z

uw ˆ
h X ( (w))gw A uw
h ( (X (ˆ | ($> w))g$)gw (12)
<+ <+
Since,R by the definition
R Y (|)  X (|),R for all |R5 \ , and since, by propo-
sition 1, <+ huw ( (X (ˆ| ($> w))g$)gw = <+ huw ( (Y (ˆ | ($> w))g$)gw, 12 im-
R R R
plies that <+ huw Y (  (w))gw A <+ huw ( (Y (ˆ | ($> w))g$)gw. A contradic-
tion. Thus, 2) holds true. Since ˆ is e!cient for (Y> I> {0 ) and the latter is a
concave economy, the following equalities hold true:
 Growth in Economies With Non Convexities 219

Z Z
Y (ˆ
| ($> w))g$ = Y ( |ˆ($> w))g$)> and (13)
[0>1) [0>1)
R R
I (ˆ
{(w)> |ˆ($> w))g$ = I ( [0>1) (ˆ
[0>1)
{(w)> |ˆ($> w))g$)).
Thus, by 13, (ˆ {> d̂> |¯) is Ran e!cient allocation of the intertemporal econ-
omy (Y> I> {0 ), for |¯(w) = [0>1) |ˆ($> w)g$= Therefore, there exists a sunspot
invariant price (t̂  > ê > ŝ ) that supports (ˆ {> d̂> |¯) as a (degenerate) sunspot
equilibrium of the economy (Y> I> {0 > E  ), for any E   E. However, again by
13, (t̂  > ê > ŝ ) supports also the allocation (ˆ {> d̂> |ˆ) as an intertemporal sunspot
equilibrium of the economy (Y> I> {0 > E  ), for any Ê  E   E. Hence, to
conclude the argument we need to show that (t̂  > ê > ŝ ) supports (ˆ {> d̂> |ˆ) as
an intertemporal sunspot equilibrium of the economy (X> I> {0 > E  ), for any
Ê  E   E.
Since the economies (Y> I> u; E  ) and (X> I> u; E  ), when facing identical
price trajectories, have identical profit and household’s wealth maximization
problems, it su!ces to show that R |ˆ(w)w0Ris an optimal solution to
max|ˆ”(w)5\ ([0>1)>E  ) <+ huw ( [0>1) (X (ˆ | ”($> w))g$))gw
R R R R
vxemhfw wr <+ ŝ (w)( |ˆ”($> w))g$)gw  <+ ŝ (w)( [0>1) |ˆ($> w))g$)gw
Suppose by contradiction that there exists a budget feasible allocation |ˆ0
such that |ˆ0 (w) 5 \ ([0> 1)> E  ), for all w, and
Z Z Z Z
uw 0 uw
h ( (X (ˆ
| ($> w))g$)gw A h ( (X (ˆ
| ($> w))g$))gw
<+ [0>1) <+ [0>1)
Z Z
= huw ( (Y (ˆ
| ($> w))g$)gw=
<+ [0>1)

However, the definition of Y and the previous inequality imply that

Z Z Z Z
huw ( | 0 ($> w))g$)gw A
(Y (ˆ huw ( (Y (ˆ
| ($> w))g$)gw=
<+ [0>1) <+ [0>1)

A contradiction.

Thus, by the last proposition, if sunspot equilibria exist, they are "equiv-
alent" to competitive equilibria. As we stated in the introduction, there are
two economic environments, where this equivalence is guaranteed: the first is
defined by a restriction on the technology I , that we call it ”semi-linearity”,
and the second is the large economy. Next we turn to semi-linear economies.

5.3 Sunspot Equilibria and Semi Linear Economies

An economy (X> I> u) is semi-linear if I ({> |) = i ({) + E| for some (con-

N
cave) function i : U+ $ UN and matrix E of dimension O × N. Semi-linear
economies have an appealing feature: equilibrium prices are (without loss of
220 Aldo Rustichini and Paolo Siconolfi
R O
generality) linear, i.e., ? s>  A= s |(g|), for some s 5 U+ . The latter is
obviously a necessary feature to allow for the existence of sunspot equilibria.
First we show that prices are indeed linear in a semi-linear economy and than
we exploit this feature to construct equivalent CPSE. The next proposition
shows that lottery equilibrium allocations of a semi-linear economy are closely
related to the equilibrium allocations of the concavified version of the economy.
Most importantly, the supporting prices are identical and, therefore, linear.
Let (d̂ > {
ˆ > |ˆ ) be a feasible allocation of the semi-linear economy (Y> I> {0 )
and let (d̂> { ˆ be an allocation of the semi-linear economy (X> I> {0 ). The
ˆ> )
two allocations are equivalent if (d̂ > { ˆ )(w) = (d̂> { ˆ
ˆ)(w), X ((w)) |  (w))
= Y (ˆ
ˆ 
and |((w)) = |ˆ (w), for O  d=h=w.

Proposition 3. Let I (·) = i (·) + E|. The economies (X> I> u) and (Y> I> u)
have equivalent e!cient allocations. Furthermore, without loss of generality,
the supporting prices are in both economies equal and linear.

Proof. Let (d̂> { ˆ> |ˆ) be an e!cient allocation of the semi-linear economy de-
fined by (Y> I> {0 ). Let (w)ˆ 5 arg{max5P1>+ X () subject to |() = |ˆ(w)}.
By the definition of least concave, Y (ˆ ˆ
| (w)) = X ((w)) and by the semi-
linearity of the technology, (d̂> { ˆ is feasible. Thus, the allocations (d̂> {
ˆ> ) ˆ> |ˆ)
and (d̂ > { ˆ are equivalent. Since, (d̂> {
ˆ > ) ˆ> |ˆ) is e!cient for (Y> I> {0 ), the defin-
ition of Y implies that (d̂> { ˆ is e!cient for (X> I> {0 ). Conversely, let (d̂> {
ˆ> ) ˆ> )ˆ
be an allocation of the semi-linear economy (X> I> {0 ). Then, by the semi-
linearity of the technology, (w) ˆ 5 arg{max X () subject to |() = |((w))}. ˆ
ˆ
Thus, X ((w)) ˆ
= Y (|((w)). Then, the allocations (d̂> { ˆ and (d̂> {
ˆ> ) ˆ> |ˆˆ ),
ˆ
|ˆˆ (w) = |((w))> for all w, are equivalent and, quite obviously, (d̂> { ˆ> |ˆˆ ) is
 
an e!cient allocation of (Y> I> {0 ). Thus the first part of the proposition
holds true. Consider a pair of e!cient and equivalent allocations (d̂> { ˆ
ˆ> )
and (d̂> {
ˆ> |ˆ). By the concavity of (Y> I> {0 ), there exists a price (t̂> ê> ŝ)> with
ŝ(w) 5 UO , for all w, supporting (d̂> {ˆ> |ˆ) as a competitive equilibrium alloca-
tion. Thus, we need to show that (t̂> ê> ŝ) supports (d̂> { ˆ as an equilibrium
ˆ> )
allocation of the economy (X> I> {0 ). Since the firm profit and the household
wealth maximization problems are identical in theR two economies, we just
have to show that  ˆ is an optimal solution to: max huw X (ˆ
 (w)gw subject
R R <+
to <+ ŝ(w)|(ˆ  (w))gw  <+ ŝ(w)ˆ
| (w))gw Once again, the definition of Y implies
the claim.

The proposition allows for an immediate characterization of intertemporal

equilibria of the semi-linear economies in terms of CPSE sunspot equilibria.

Corollary 1. Let (t̂> ê> ŝ> { ˆ be a competitive equilibrium with linear prices
ˆ> d̂> )
of the semi-linear economy (X> I> u). Then, there exists a consumption sunspot
ˆ
allocation |ˆ, with |ˆ(w)  (w), for all w> such that the vector (t̂> ê> ŝ> {
ˆ> d̂> |ˆ) is
an intertemporal CPSE.
 Growth in Economies With Non Convexities 221

Proof. This is an immediate consequence of the equivalence relation .

It su!ces to observe that, for any given measurable function j and any
allocation
R |ˆ 5 \ ([0>R1)> E), by the standard change of variable formula,
[0>1) R
j(ˆ| ($))(g$) = \ j(|)O  |ˆ1 (g|). However, for |ˆ  ,  = O  |ˆ1 ,
R
hence \ j(|)  | 1 (g|) = \ j(|)(g|). It follows that at common prices
(t̂> ê> ŝ), equivalent paths (ˆ ˆ and (ˆ
{> d̂> ) {> d̂> |ˆ) yield the same value of profits,
overall utility and budget constraints. Since for each  5 P1>+ (\ ) there exists
equivalent |ˆ 5 \ ([0> 1)> E), and viceversa, the latter implies the claim.

Proposition 3 implies that every competitive equilibrium can be sup-

ported by linear prices. Thus, Corollary 1 implies that every competitive
equilibrium allocation can be supported as a sunspot equilibrium alloca-
tion. Furthermore, Proposition 2, imply that also of the converse of the
claim in Corollary 1 holds true. Hence, lotteries and sunspot are in the
semi-linear environment equivalent. Propositions 2 has another interesting
implication. Consider (ˆ {> d̂> |ˆ) an inter-temporal sunspot equilibrium allo-
cation of the economy (X> I> {0 > Ê), with I (not necessarily a semi-linear).
By Proposition 2 , (ˆ {> d̂> |ˆ) is an
R inter temporal sunspotR equilibrium al-
location of (Y> I> {0 > Ê). Hence, [0>1) Y (ˆ | ($> w)g$ = Y ( [0>1) |ˆ($> w)g$) and
R R
{0>1)
I (ˆ
{> |ˆ($))g$ = I (ˆ{> [0>1) |ˆ($)g$), for O  d=h=w= Thus, both Y and I
are linear in the in the relevant range. In other words, if there are intertempo-
ral sunspot equilibria, the representative agent economy is without essential
loss of generality, semi-linear.

6 Non Existence of Sunspot Equilibria

The converse of proposition 2 (point 1) is not true. There are intertemporal
competitive equilibria of economies (X> I> {0 ) that do not allow for an equiv-
alent CPSE. Equivalently, by proposition 2 (point 2), there are intertemporal
competitive equilibrium allocations of (X> I> {0 ) that cannot be supported as
intertemporal sunspot equilibria. In this section, we construct examples by
using the static notion of equilibrium. However, before doing that we need to
clarify the relation between static and intertemporal sunspot equilibria.

6.1 Static and Stationary Sunspot Equilibria

As already mentioned, e!cient static allocations are e!cient, stationary in-

tertemporal allocations. Also, static competitive equilibria are stationary, in-
tertemporal competitive equilibria. However, there might be static sunspot
equilibria that are not intertemporal (stationary) sunspot equilibria. The rea-
son is simple. In the static formulation of the model time is absent, while
in the intertemporal version, agents can use time to generate lotteries over
basic commodities. When prices are, as in the sunspot case, linear, the agents
222 Aldo Rustichini and Paolo Siconolfi

can reproduce lotteries that yield the least concave utility function. Hence,
only static sunspot equilibrium allocations yielding the least concave utility
can survive the test of time, i.e., they are stationary, intertemporal sunspot
equilibria. This is summarized in the next proposition.

Proposition 4. A static sunspot equilibrium of the economy (X> I> Ê) is a

stationary, intertemporal sunspot equilibrium only if it is a static sunspot
equilibrium of the “concavified” economy (Y> I> u; Ê) (as well as, a stationary,
intertemporal sunspot equilibrium of the concavified economy (Y> I> u; Ê)).

Proof. The argument is a trivial consequence of proposition 2.

6.2 Examples

We construct (open sets of) static economies (X> I ) that have:

1) a continuum of Pareto ranked static sunspot equilibria associated to
distinct specifications of the sunspot space;
2) e!cient allocations supportable by static sunspot equilibria, but not by
CPSE, and an empty set of intertemporal sunspot equilibria for some value
of the initial capital stock {0 and for all Ê  E;
3) an empty set of sunspot static equilibria.
By proposition 2, the Pareto ranked static sunspot allocation cannot be
intertemporal sunspot allocations and intertemporal sunspot equilibria are,
without loss of generality, CPSE. Thus, point 1) does not apply to the in-
tertemporal economy, while 2) implies that sunspot and lottery intertemporal
equilibrium allocations are not equivalent and that the set of intertemporal
sunspot equilibria may be empty. Thus, 1), 2) and 3) point out that, when
preferences are not convex, static sunspot equilibria are a inadequate tool to
analyze stationary equilibrium allocations of growth economies.

Proposition 5. Static sunspot equilibria may fail to exist. When they exist
they may be indeterminate and ine!cient. E!cient stationary allocations may
be decentralized as static sunspot equilibria, but not as static CPSE.

Example 3. An economy may have a continuum of static sunspot equilibria,

all Pareto ranked.

The static economy is described by F = N = 1, and I ({> |) = i ({)  |  ,

  1, with i strictly concave and such that i 0 (1) = u and i (1) = 1. The
utility function has X 0 (0) = +4, is monotonically increasing in an interval
[0> P ], and is such that co X (1) A X (1). For example the following utility
function satisfies all these conditions:

X (|) = min{max{|  > |}> 2P  |},

for  5 (0> 1) and P A 2 and arbitrarily large (14)
 Growth in Economies With Non Convexities 223

We restrict the space of extrinsic uncertainty to two points $ 5 {1> 2} with

(1) = . We set t = 1@. The first order conditions for the firm’s problem
(recall that this problem is concave) give that the equilibrium capital and the
firm supply are
1
{
ˆ = 1> |ˆi ($) = (ŝ($)) 1 > $ = 1> 2> (15)
when  A 1. When  = 1> the equilibrium prices are s($) = 1> $ = 1> 2. At the
equilibrium prices, the set of optimal |’s is the set of non negative quantities.
For arbitrarily given   1, an equilibrium is a sunspot allocation |ˆ =
| (1)> |ˆ (2)) such that:
(ˆ

1. the solution |ˆ of the consumer’s problem

X
max ($)X (ˆ| ($))> subject to
|
ˆ
$
X X
|  ($))1 |ˆ($) 
($)(ˆ |  ($))
($)(ˆ
$ $

coincides with |ˆ ,P
|  ($)) = 1=
2. the market clear: $ ($)(ˆ
Consider for example the utility function in equation (14). Note that the
equilibria are symmetric: to any equilibrium at (> 1  )) corresponds an
equilibrium at (1  > ), so we only consider   (1)  1@2. The Inada
condition that X 0 (0) = +4 insures that |ˆ À 0. Let, without loss of gen-
erality, s2 ? s1 . (Otherwise relabel the states) Since, lim|%1 X 0 (|)@s(1) =
@s(1) ? lim|&1 X 0 (|)@s(2) = 1@s(2), the budget feasible allocation |($) = 1>
1 1
$ = 1> 2 is not optimal. When max{ ŝ(1) > (1) ŝ(2) } ? P , the utility function
is monotonically increasing in the budget feasible set, so the solution of the
consumer problem will satisfy the budget constraint as an equality, and so we
only have to find |ˆ that coincide with the solution of
X
max ($)X (ˆ| ($))> subject to
(|1 >|2 )
$
X
($)ŝ ($)ˆ
| ($) = 1, for ŝ ($) = (ˆ
|  ($))1 =
$

The first order condition that marginal utilities are equal at the two choices
|m is necessary: note that the probabilities appear both in the utility and
in the budget constraint, and therefore cancel out. The sunspot equilibrium
allocation |ˆ and the supporting prices ŝ are a solution to the following
system of equations:

ŝ(2) 1 X
|ˆ(1) = ( ) > ($)ŝ($)ˆ | ($))1 > $ = 1> 2=
| ($) = 1> and ŝ($) = (ˆ
ŝ(1) $
(16)
224 Aldo Rustichini and Paolo Siconolfi

The latter is a system of 4 equations in 5 unknowns (ˆ | > ŝ> ). The solution
changes with  and this gives the continuum of equilibria. The value depends
on , and the equilibria are Pareto ranked.
The case  = 1 is simple:
1. there is an equilibrium for any  5 (1@P> 1@2];
2. the allocations are dierent for dierent :
1
1  (1  ) 1 1
1
|ˆ = ( > ); (17)

3. the value is strictly decreasing in , and given by:
 1
1 + (1  )( 1   1 )= (18)
The allocations described in 2. satisfies |ˆ(2) 5 (0> 1) and |(1) 5 (1> P ), for
 5 (1@P> 1@2]. The matrix of partial derivatives of the system of equations
| > ŝ) is for,  = 1 and for any  0 5 (1@P> 1@2]> invertible.
16 with respect to (ˆ
0
Hence, for  5 (1@P> 1@2], by the implicit function theorem, for each (> )
in a neighborhood of  = 1 and  0 , there exists a sunspot equilibrium.
The e!cient lottery of the static economy, in the next example, cannot
be supported by linear prices and, hence, by a CPSE. Quite surprising, it
can be supported by sunspot equilibria that display sunspot price variability.
As already explained, this implies that the set of sunspot static equilibria
is empty. This example is in contrast with the findings of [5] for finite, pure
exchange economies. The existence of sunspot equilibria which are (genuinely)
not CPSE is, in that context, possible only with finite sunspot state space and
discrete consumption sets.
Example 4. An e!cient allocation of a static economy may be supported as a
sunspot equilibrium, but not by a CPSE.
We consider the economy defined by technology of the previous example
and the utility function in equation (14), with parameter values (> ),  A 1
and  5 (0> 1)= As already explained, the optimal amount of capital is { = 1,
for all . The optimal lottery   is found by solving the following steps:
1) for given t A 0, find the set \  (t) = arg max| X (|)  t| 
2) find a price for the investment good t and a probability distribution over
\  (t) such that |  () = 1 for  5 P1>+ (\  (t)). Recall that, by definition,
for | and | 0 5 \  (t)> X (| 0 )  t(| 0 ) = X (|)  t(|) .
Since lim|%1 X 0 (|) =  ? lim|&1 X 0 (|) = 1, the deterministic allocation
  =  1 is suboptimal for all parameter values,  5 (0> 1) and  A 1 . The next
claim shows that, given the initial condition {0 = 1, the e!cient allocation is
a unique, stationary and non degenerate lottery. The proof of the Claim is in
the appendix.
Claim. The intertemporal economy (X> i ({)  |  > 1) has a unique e!cient
allocation (ˆ ˆ {
{> d̂> ), ˆ
ˆ(w) = 1, d̂(w) = 0 and (w) =   , for O  d=h=w= The

support of the optimal lottery  is {|1 > |2 }, with 0 ? |1 ? 1 ? |2  P .
 
 Growth in Economies With Non Convexities 225

In order to avoid unnecessary complications, we restrict the parameters

values in the set S (P ) = {(> ) :  5 (0> 1)>  A 1 and |2 ? P }, an
open subset of R2+ = To characterize the optimal lottery we seek values of the
four unknowns (|1 > |2 > > t) that satisfy four equations (market clearing, two
optimality conditions, |l 5 \ (t), and X (|1 )  t(|1 ) = X (|2 )  t(|2 ) ). By
straightforward computations, the unique solution is:
1 1  
|1 = ( 
1 )
(1) () (1) > |  = (  ) (1) () (1) >
2 1
(1)
()(1)
|2  1
t  = 1 ( 
1
) (1) () (1) and   = |2 |1

For given  A 1, both |1 and |2 are strictly increasing functions of = For

 = 0, |2 = 1 A 1 and |1 = 0, while for  = 1, |1 = 1 (and |2 = 4)=
In order to show that the e!cient stationary allocation can be supported
as a static sunspot equilibrium, we take = {1> 2} with  =   . Profit
maximization pins down the supporting prices ŝ as:
ŝ ($) = t  (|$ )1 , $ = 1> 2=
Since both |1 and |2 belong to regions where the utility function is concave,
s (1)
s (2) is equal to the value of the marginal rate of substitution computed at the

sunspot e!cient allocation. Furthermore, for each given P A 2, ss (2)
(1)
6= 1, for
an open and dense set of parameter values in S (P ). Hence, for (> ) in an
open and dense subset of S (P ), e!cient allocation are supported as sunspot
equilibria, but not as CPSE. Thus, by proposition 2, the uniqueness of the
e!cient allocation implies that, for these parameter values, the set of sunspot
equilibria is empty.
In Proposition 2, we were silent about the possibility of the sunspot equi-
libria of the static economy to support the e!cient allocations. The last ex-
ample shows that indeed CPSE may fail to accomplish this task. However,
it leaves open this possibility for dierent types of sunspots equilibria. Un-
fortunately, there is a fundamental reason that breaks the equivalence result:
sunspot equilibria may not exists. This is shown in the next example.
Example 5. An economy (X> I> u) may have neither sunspot equilibria nor in-
tertemporal sunspot equilibria for some initial condition {0 and any algebra
Ê  E.
Let F = N = 2, and
1
I1 ({1 > |1 ) = (i1 ({1 )  |12 ) (19)
2
I2 ({2 > |2 ) = i2 ({2 )  |2 (20)
while the preferences are defined by
X (|1 > |2 ) = |2 + x(|1 )> (21)
where for  ? 1 and  A 0 and small enough, x is defined by
226 Aldo Rustichini and Paolo Siconolfi
11
( 1 )|1  (10 + )> i ru |1 5 [0> 1  )>
x(|1 ) = |1 + (@2)[(|1  1)2  ()2 ]> i ru |1 5 [1  > 1 + ]
|1 > i ru |1 A 1 + =
The function x is continuous and piecewise dierentiable, and therefore so
is X ; for  small enough, X is strictly increasing.
Since the functions Il are, for l = 1> 2, separable in { and |, the optimal
capital stocks are independent of prices and are found by solving:

max dl il ({l )  u{l >

{l

for l = 1> 2 and d1 = 1@2 and d2 = 1= Since il are strictly concave the
optimizers {l are unique and we take them to satisfy:

il ({l ) = {l = 1> for l = 1> 2=

In this static economy, although the utility function is not concave, the
planner allocation, ({ > |  ), is deterministic, it is equal to (1> 1> 1> 1) and it is
supported by prices (t1 > t2 ) = (1> 1). Since the characterization of the e!cient
allocation plays a minor role in the proof of the non-existence, we postpone its
detailed construction to the appendix, and we just point out the only property
needed in the sequel, namely:
1
1 5 arg max x(|1 )  |12
2
In the static economy, we take ([0> 1)> E> O) as the probability space of
sunspots. We normalize prices by setting t1 = 1.
Let ({> |ˆi )(s> t) and |ˆk (t> s) denote, respectively, the firm’s profit maximiz-
ing and the household’s utility maximizing choices, for given price function
ŝ : [0> 1) $ R++ and t 5 RN ++ . Since the functions Il are, for l = 1> 2,
separable in { and |, the profit maximizing capital stocks are :

i ({l ) = {l = 1> for l = 1> 2=

Given the adopted specification of I , ({> |ˆi ) 5 ({> |ˆi )(s> t) if

|ˆ1>i (v) = ŝ1 ($), ŝ2 ($)  t2 and ŝ2 ($)ˆ

|2>i ($) = t2 |ˆ2>i ($)> O  d=h=$ (22)

Let |ˆk (s> t) denote the optimal consumption allocation chosen by the house-
hold at prices (s> t).
The market clearing conditions are:

Z
( |1i ($))2 > |ˆ2i ($))g$ = (i (ˆ
((ˆ {1 )> i (ˆ
{2 )) = (1> 1)
[0>1)
and |ˆi ($) = |ˆk ($), O  d=h=$ (23)
 Growth in Economies With Non Convexities 227

Now we give immediately the final step in the proof of the non-existence of
equilibrium, assuming first the following two preliminary steps. Later (see
equation (27)) we prove that at a sunspot equilibrium:
Z Z
ŝ1 ($)g$  ŝ1 ($)ˆ
|1 ($)g$= (24)
[0>1] [0>1]

We also prove (in lemma (2)) that any allocation |ˆ> with |ˆ1 ($) = 1, for
O  d=h=$, cannot be an equilibrium allocation. We assume for the moment
these two preliminary results, and show the conclusion.

Proposition 6. There are no sunspot equilibria of the static economy defined

by (19), (20), (21).

Proof. By contradiction, let |ˆ, |ˆ = |ˆi = |ˆk , be the equilibrium allocation.

Since by the firm first order conditions ŝ1 ($) = |ˆ1 ($), O  d=h=$> (24) can be
rewritten as:
Z Z Z
ŝ1 ($)g$ = |ˆ1 ($)g$  |1 ($))2 g$
(ˆ (25)
[0>1) [0>1) [0>1)

Hence, by the market clearing conditions (23) :

Z Z
|ˆ1 ($)g$  |1 ($))2 g$ = 1=
(ˆ (26)
[0>1) [0>1)

By Jensen’s inequality, the last inequality holds true if and only if |ˆ1 ($) =
1, O  d=h=$. However, lemma (2) below rules out that this latter can be a
sunspot equilibrium allocation.

To proceed in our argument, we show that any consumption allocation

|ˆ, with |ˆ1 ($) = 1, O  d=h=$, cannot be a sunspot equilibrium allocation.
The argument is based on a simple observation: at the consumption bundle
(1> |2 ), |2  0, the utility function is not concave and, hence, no linear price
can support that allocation.
Now the details. Arguing by contradiction, suppose that ({> |ˆ) is an equi-
librium
R allocation. Let (t> ŝ) be the equilibrium price function. Since { = (1> 1)
and [0>1] I ({> |ˆ($))g$ = 0, the value of the firm net of capital expenses is:
Z
y(ŝ> t) = s($)ˆ
|i ($)g$ = 1 + t2 =
[0>1)

Let |ˆ be any consumption allocation with |ˆ1 ($) = 1, Od=h=$. The proof
of the next Lemma is deferred to the appendix.

Lemma 2. |ˆ can not be a sunspot equilibrium allocation of the static econ-
omy.
228 Aldo Rustichini and Paolo Siconolfi

To conclude the construction we have to show that if there were a sunspot

equilibrium then inequality (24) would hold true. Evidently, by proposition 6,
the latter is a contradiction.
Suppose that there exists a sunspot equilibrium. Recall that (1> 1) is the
e!cient consumptionR bundle. Then, by 2 and by the definition of e!cient
allocation, X (1> 1)  [0>1] X (ˆ | ($))g$. Hence, by revealed preferences, it must
be that the constant allocation |ˆ , with |ˆ ($) = (1> 1), O  d=h=$> is at least
as expensive as the equilibrium allocation |ˆ. Hence, at the equilibrium prices
ŝ, it must be:
Z Z
(ŝ1 ($) + ŝ2 ($))g$  (ŝ1 ($)ˆ|1 ($) + ŝ2 ($)ˆ
|2 ($))g$ (27)
R R
However, by taking into account that ŝ2 ($)ˆ |2 ($)g$ = ŝ2 ($)g$ (see
23), the latter inequality simplifies to (24).
To conclude the analysis we show that given the initial condition {0 =
(1> 1), there cannot exist sunspot equilibria of the intertemporal economy.
Suppose, by contradiction, that there exists an intertemporal sunspot equilib-
rium. By propositions 2, the intertemporal sunspot equilibrium is e!cient and
, without loss of generality, a CPSE. Furthermore, there exists an equivalent
lottery equilibrium of the intertemporal economy (X> I> u). Hence, for given
initial condition {(0) = {0 , the economy (X> I> u) has two e!cient paths, one
stationary (ˆ ˆ with (1> 1>  (1>1) ) = (ˆ
{> ), ˆ
{> )(w), { > 
for all w, and one not, (ˆ ˆ  ).

Consider for  5 (0> 1), the path (ˆ { + (1  )ˆ ˆ + (1  )
{ >  ˆ ), where

(ˆ + (1  )ˆ )(w) is the compounded lottery that assigns with probability
ˆ
 the lottery (w) and (1  ) the lottery  ˆ  (w). Evidently, by e!ciency and
R R R
ˆ
linearity in lotteries, X ((w))gw = X ( ˆ  (w)gw = X ( ˆ + (1  )ˆ  )(w)gw.
However, by the concavity of I ,

I ((ˆ { > 
{ +(1)ˆ ˆ  )(w))  I (({> )(w))+(1)I
ˆ +(1) ˆ ˆ  )(w)) =
(({ > 
˙ + (1  ){˙  (w) = (1  ){˙  (w).
{(w)
If the inequality is strict for an O  positive measure set of R+ , the strict
monotonicity of X contradicts the e!ciency of both paths. Hence, for Od=h=w,
the previous inequality must be an equality. Then, the strict concavity of I
in { and of I1 in |1 , implies that { ˆ(w) = { ˆ  (w){| : |1 = 1} = 1, for
ˆ (w) and 
O  d=h=w=
But then given the form of I> |( ˆ  (w)) = 1, for O  d=h=w=, and since
X (and I ) is linear in |2 ,  ˆ  (w) and  ˆ =  1>1 yield the same utility and

ˆ
I (1> 1>  (w)) = I (1> 1> 1> 1), for Od=h=w= Hence, the path (ˆ ˆ  ) is supported
{ > 
by the supporting prices of the stationary allocation (ˆ ˆ Since the latter
{> ).
can not be supported by linear prices, the intertemporal economy does not
have sunspot equilibria.
 Growth in Economies With Non Convexities 229

7 The Large Economy

In this section we study the version of the economy with a large number of
identical households. For large numbers of individuals we mean (as usual) that
the law of large numbers holds true. Namely, if we assign to each household
an i.i.d. lottery  5 P1>+ (\ ) , theR pro-capita average allocation is equal
(with probability one) to |() = |(g|). The particular specification the
measurable space of households is not important, provided that the law of
large numbers holds. We take it to be {[0> 1)> E> O} (with the usual caveat
that applies to the validity of the Law of Large numbers with a continuum of
random variables). As for any large economy, we assume that feasibility is a
restriction over the random realizations of the pro-capita total output.
Since we are dealing with a market economy where identical individual
face identical choices, we limit attention to e!cient allocations yielding the
same level of utility to each household, ”equal utility allocations.” The set of
consumption allocations, at each instant, can be identified just with the set
P1>+ (\ ). To each  5 P1>+ (\ ) it is associated the pro-capita consumption,
|(). Hence, feasibility is just a restriction over average output. The reason
is obvious. If we assign to each household independent lotteries identically
equal to , by the law of large number, the average output is equal (with
probability 1) to |(). Considering more general random allocations does not
enhance e!ciency, [14].
A large economy has basically the same properties of a semi-linear econ-
omy. Since feasibility restricts the average values of the lotteries, the basic
properties of equilibrium and e!cient path of the economy (X> I> u) are closely
captured by those of the concavified economy (Y> I> u). This is the substance
of the arguments, which, are obviously very similar to the ones used for the
semi-linear case.

7.1 Equal Utility E!cient Paths

Equal utility e!cient paths are the solution to the following programming
problem:
Z
ˆ
max huw X ((w))gw subject to I (ˆ ˆ
{(w)> |((w)))  {(w)>
˙ {
ˆ(0) = {0 = (28)
(ˆ ˆ
{>)

The optimal solutions to (28) can be characterized in a simple way by making

reference to the ”concavified” economy (Y> I> u) Hence, consider the following
programming problem, where, given the concavity of Y , the planner optimizes
within deterministic allocations:
Z
max huw Y (ˆ {> |ˆ(w))  {(w)>
| (w))gw subject to I (ˆ ˙ {
ˆ(0) = {0 = (29)
(ˆ
{>ˆ
|)

Let Y ({0 ) be the value of (29) and X ({0 ) be the value to (28). Then:
230 Aldo Rustichini and Paolo Siconolfi

Proposition 7. (ˆ {> |ˆ) is an optimal solution to (28) if and only if (ˆ ˆ with

{> )
ˆ
X ((w)) = Y (ˆ ˆ
| (w)) and |((w)) = |ˆ(w), for all O  d=h=w> is an optimal solution
to (29). Hence, Y ({0 ) = X ({0 ).

Proof. The proof is an easy consequence of the definition of least concave

function.

7.2 Lottery and Sunspot Equilibria

The large competitive economy with a complete set of markets for lotteries
is identical to the one we have already described. The only dierence is in
the presence of a large number of households. In order to characterize the
equilibria of our large economy, we follow the same technique adopted for
the characterization of e!cient paths. We study the ”concavified” economy
where, rather than X , Y is the utility function of the households.
The e!cient allocations of the concavified economy and the e!cient allo-
cations of the large economy are equivalent in the sense of proposition 7. Fur-
thermore, for both economies, competitive allocations and e!cient allocations
coincide. Hence, the two economies have equivalent competitive allocations.
In the next proposition, we state that they have identical supporting prices.
Therefore, supporting prices of large economies are linear in commodities.
This is the key fact for sunspot decentralization of the e!cient paths.

Proposition 8. Let (ˆ {> d̂> |ˆ) be an e!cient path of the ”concavified” economy
supported by prices (t̂> ê> ŝ). Then the e!cient path (ˆ ˆ with X ((w))
{> d̂> ), ˆ =
Y (ˆ ˆ
| ) and |((w)) = |ˆ(w), of the large economy is supported by the same prices.

Proof. The proof is basically identical to the proof of proposition 3 and it is,
therefore, omitted.

Constant Prices Sunspot equilibria

Propositions 3, 7 and 8 establish a strong similitude between representative

economies with semi-linear technologies and economies with a large number
of households. It is not a surprise, therefore, that the results of Corollary 1
generalize to large economies. However, there is a dierence between this two
environments that makes the extension non immediate. In the representative
household semi-linear environment, aggregate and individual consumption are
obviously equal. Consumption may be random, although only its average value
is of importance for feasibility. In the large economy, individual and pro-capita
consumption are very dierent. The first may display randomness, while the
second is deterministic and equal to the average value (over lottery real-
izations) of the random individual consumptions. Hence, sunspot allocation
have in the large economy the additional task of creating enough correlation
among individual allocations in order to make the pro-capita consumption
 Growth in Economies With Non Convexities 231

sunspot independent and feasible. A sunspot consumption allocation of the

large economy is a measurable map D̂ : [0> 1)2 × U+ $ \ , where D̂(k> $> w) de-
notes the consumption bundle consumed by individual k at w when $ realizes.
Given an initial condition {0 , dq intertemporal FS VH of the large economy
is an $  lqyduldqw trajectory of prices (t̂> ê> ŝ), of investments, capital goods
and pro-capita consumption allocations (ˆ {> d̂> |ˆ) and a consumption alloca-
tion D̂ such that l) (ˆ {> d̂> |ˆ) is profit maximizing at (t̂> ê> ŝ); ll) (ˆ {> d̂> D̂(k))
R
is utility maximizing at (t̂> ê> ŝ), for O  d=h=k; lll) [0>1) D̂(k> $> w)gk = |ˆ(w),
O × O  d=h=($> w). ly) I (ˆ {(w)> |ˆ(w))  d̂(w)> O  d=h=w. We exploit propositions
7 and 8. Hence, for given initial condition, we take a competitive equilibrium
allocation of the concavified economy and we show that the same prices sup-
port a FS VH with equivalent allocation. This result implies, by propositions
7 and 8, that any e!cient (lottery) allocation of the large economy can be
equivalently supported by a FS VH.
The following observation, although not necessary (see for instance [5]),
greatly simplifies the argument. Consider the set  = {(|> x) : | 5 \
and x = X (|)}  UF+1 . Let FK()  UF+1 be the convex hull of
. Consider an equilibrium allocation of the ”concavified” economy. For
each w, by the definition of least concavity> (ˆ | (w))) 5 FK(). Then,
| (w)> Y (ˆ
Caratheodory’s theorem implies that there exists at most F + 2 consumption
bundles (ˆ ˆ 5 (F+2) such that  
|1 (w)> ====> |ˆF+2 (w)) and (w) ˆ (w)ˆ
| (w) = |ˆ(w)

ˆ
and    (w)X (ˆ | (w)) = Y (ˆ | (w)). Hence, when we analyze e!cient (lottery)
allocations of the large economy there is no loss of generality in restricting
attention to allocations (w) ˆ with support consisting of at most F + 2 points.

Proposition 9. Let (t̂> ê> ŝ; {

ˆ> d̂> |ˆ) be an intertemporal equilibrium of the con-
cavified economy (Y> I> {0 ). Then, there is a sunspot allocation D̂, with
Z
D̂(k> $> w)g$ = |ˆ(w)>

R
and [0>1)
X (D̂(k> $> w))g$ = Y (ˆ
| (w)), for O×Od=h=(k> w)> such that the vector
(t̂> ê> ŝ; {
ˆ> d̂> |ˆ> D̂) is an intertemporal CPSE of the large economy.

Proof. By proposition 3 and corollary 1> at (t̂> ê> ŝ) households maximize util-
ity by selecting the capital and investment trajectories (ˆ {> d̂) and an individual
sunspot allocation |ˆ such that
Z
|  (w> $))g$ = Y (ˆ
X (ˆ | (w))
[0>1)
R
and |ˆ (w> $)g$ = |ˆ(w). Hence, we just have to show, that there exists a
[0>1)
R
function D̂(=) such that D̂(k) = |ˆ , for all k, and [0>1) D(k> $> w)gk = |ˆ(w),
ˆ 5
O × O  d=h=(w> $). By Caratheodory’s theorem, for each |ˆ(w) there exist (w)
232 Aldo Rustichini and Paolo Siconolfi

ˆ
(F + 2) such that X (|((w))) = Y (ˆ ˆ
| (w)) and |((w)) = |ˆ(w). Denote by |ˆ (w),
ˆ
 = 1> ===> F + 2, the points in the support of (w)= Divide, for each w, the set of
1 ˆ  ˆ
sunspots [0> 1) in the F + 2 disjoint intervals, L (w) = [ m=0  m (w)> m=0  m (w)),
with the convention that  ˆ (w) = 0. Let +̂ denote addition modulo 1. Then D̂
0
is defined as:
D̂(k> $> w) = |ˆ (w), li $ +̂k 5 L (w)=
R
Evidently, D̂(k) = |ˆ , for all k, and [0>1) D̂(k> $> w)gk = |ˆ(w), since O{k :
D(k> $> w) = | (w)} = O{k : k+̂$ 5 L (w)} =  ˆ (w), for each pair (> w).


8 Indeterminacy of Sunspot Equilibria

We construct an example of a large static economy with a countable infin-
ity of sunspot equilibrium allocations, which are Pareto ranked. Once again,
ine!cient sunspot equilibria of the large static economy can never be station-
ary sunspot equilibria of the large economy. However, there is a significant
dierence between example 3 and this example, which justifies its construc-
tion. The static economy with a representative household bears no similarity
with any finite dimensional Arrow-Debreu economy. This is particularly evi-
dent when equilibrium calls for random consumption allocations. The latter
maps into randomness of the investment allocations, d = I ({> |). However,
feasibility requires that investment be equal to zero in average. It is the fea-
sibility requirement that makes the representative household economy very
dierent than anything else. Of course, its interest is justified by its relation
with the associated dynamic economy. The situation is quite dierent for a
large economy. Randomness in individual consumption allocations does not
translate (at e!cient allocations) into randomness of investment allocations.
When pro-capita output is deterministic, feasibility in the static large econ-
omy is identical to feasibility in any finite dimensional Arrow-Debreu economy
with a large number of identical households. In particular, it is immediate to
show that for specific choices of the map I , the equilibrium of the large static
economy is an equilibrium of a large pure exchange economy (with identical
individuals) with the same preferences and endowments equal to the e!cient
pro-capita aggregate consumption of the static economy. We make sure that
this feature is present in our example. Hence, the Pareto ranked multiple
equilibria of the large static economy are equilibria of the ”isomorphic” pure
exchange economy.

8.1 The Economy

There is a unique consumption good and a unique capital good. The technol-
ogy is I ({> |) = 1@2(i ({)| 2 ). i is chosen so that for { = arg max(i ({)@2)
u{, i ({ ) = 1. We identify the space of sunspots with a set of M points having
 Growth in Economies With Non Convexities 233

identical probability 1@M for some integer M A 1. The prices of the consump-
tion good are restricted to be sunspot invariant and normalized to one. The
form of the technology I and the restrictions on sunspots and prices imme-
diately imply that t = 1 and the output of the firm is equal to 1. Hence, the
economy is isomorphic to a standard pure exchange economy populated by
identical households with one unit of endowment of the only existing good.
There is a unique consumption good and a continuum of identical house-
holds, with generic index k in the space {[0> 1)> E> O}. The utility function X
is piecewise linear and non decreasing5 , defined as:
3 4
0, for | 5 [0> | ]>
X =C (|  | ), for | 5 [| > |e ]> D
(|e  | ) + (|  |e ), for | A |e =

We assume:

K1) | ? 1 ? |e > K2) 1 A  = (|e  | )@|e A  and K 0 2) (1  | )   A 0=

K1 allows for the existence of sunspot equilibria. By K2 there is a least

concave function Y associated to X . K2 and K 0 2 are not essential, but they
simplify the computations. The least concave function associated with X is:
µ ¶
|, for | 5 [0> |e ]>
Y =
(|e  | ) + (|  |e ), for | A |e =

There is a complete set of contingent commodity markets, where indi-

viduals buy contingent commodity bundle | M = (|1 > ===> |M ). In the next two
sections we drop the subscript k. Since states are equiprobable and prices
sunspot invariant and equal to 1, the programming problem of the households
is: X
(K M ) max m X (| m )> subject to (|m  1) = 0=
|M
m

Let |k (M) be the set of optimal solutions to (K ). An allocation DM is a

M
M
measurable map from [0> 1) to U+ . DM (k), k 5 [0> 1), is the vector of contingent
commodities of individual k prescribed by the allocation map DM =
An allocation DM is feasible if
Z
DM (k> m)O(gk) = 1> for all m = 1> ===> M=
[0>1]

Given M, a sunspot equilibrium is a feasible allocation such that DM (k) 5 |(M)>

for O  d=h=k=
5
It should be evident from the analysis that, at the cost of heavier computations,
all the conclusions can be generalized to strictly increasing and non concave func-
tions.
234 Aldo Rustichini and Paolo Siconolfi

8.2 Individual Optimization Problem

Facing several equiprobable states and sunspot invariant prices, households
optimize by selecting few consumption bundles and making endogenous their
probabilities by associating an endogenous number of states to each distinct
consumption bundle. This idea is formalized by the next lemma the proof
of which is in the appendix. Let V(| M ) = {| 5 U+ : |m = |, for some m},
| M 5 |(M)=
Lemma 3. #V(| M )  2, for all M and some | M 5 |(M)=
Because of Lemma 3> in order to get insights into the optimal solution to
(K M ), we can study, first, a fictitious and simple programming problem with
two states of uncertainty and with  5 (0> 1) denoting the probability of the
first state= The two prices are sunspot invariant and equal to 1. The household
solves:
(K  ) max X (|1 ) + (1  )X (|2 )> vxemhfw wr |1 + (1  )|2  1
The set of optimal solution to (K  ) is denoted by |ˆ() = (|1 > |2 )().
Evidently, (|1 > |2 ) 5 |ˆ() if and only if (|1 > |2 ) 5 |ˆ(1  ). Hence, we just
analyze the problem for  5 [1@2> 1).
Let  be such that
(1    )ˆ
|e = 1
By K1,   5 (0> 1) and since (1    )X (|e ) = Y ((1    )|e )> |ˆ( ) =
(0> |e ). Given the symmetry of the problem, we assume, without loss of gen-
erality, that
  A 1@2=
Let
1 1
|ˆK () = (0> ), XK () = (1  )X ( )>
1 1
and
(  )(|e  1)

¯= 5 (0> 1)=
|e (  )  |
By K2 and K20 , ¯ 5 (  > 1). The optimal solution and the value, Z ()>
of the programming problem, (K  ), are characterized by the next lemma the
proof of which is postponed to the appendix.
Lemma 4. The optimal solution to (K  ) is equal to
3 4
|ˆK ()> for  5 [1@2>  ¯ )>
|ˆ() = C |ˆK (¯ ) ^ (1> 1)> for  =  ¯> D
(1> 1), for  5 (¯  > 1]=
The value to K  , Z (), is equal to XK () and strictly increasing in [1@2>   ],
equal to XK () and strictly decreasing in [  > 
¯ )> and constant and equal to
 > 1]. Furthermore, Z (  ) A Z () A X (1> 1), for  5
X (1> 1)> for  5 [¯
[1@2>   ).
 Growth in Economies With Non Convexities 235

Lemma 4 implicitly characterizes the optimal solution to the program-

ming problem K M . Given an arbitrary M, let q(M) 5 arg max M qM Z ( Mq ).
2
(Remember that, by the symmetry in  of the programming problem K  >
Z ( Mq ) = Z ( Mq M
M ). Hence, the constraint 2  q  M is without loss of gen-
erality.) By Lemma 4> q(M)
M  > 1] only if Mq 5
5 [¯ ¯ ), for all M2  q  M.
@ [1@2> 
Evidently, if M is even or su!ciently high, q(M) M 5 [1@2>  ¯ ). Let M  be the
q
smallest odd integer such that M 5 (1@2>  ¯ ) for some q  M. Then:

Lemma 5. For M  M  > |(M) = |ˆK ( q(M)

M ).

8.3 Sunspot Equilibria

We construct a family of sunspot equilibrium allocations indexed by M. The

construction is basically identical to the one used in the proof of Proposition
9.
For M A M  , partition the interval [0> 1] in M subintervals, Lm = [m 
1@M> m@M), m = 1> ===> M. For each m and k, let k+̂( Mm ) be an addition modulo 1.
Then DM is defined as:
DM (k> m) = |K ( q(M) m q(M) M
M ) if k+̂( M ) A ( M )> while D (k> m) = 0 if
k+̂( Mm )  ( q(M)
M ).

The allocation map DM is feasible and DM (k) 5 |(M), for all k. Furthermore,
= Z ( q(M)
( M1 ) m X M (D(k> m)) M ).

Proposition 10. There exists a countable infinity of Pareto ranked sunspot

equilibria.

Proof. It su!ces to show that there exists a sequence of integers Mn such that
qMn
Z ( q(M n) 
Mn ) is monotonically increasing and Z ( Mn ) $ Z ( ). For M A M ,

 q  + q 
let q (M) 5 arg max 12  qM  ( M   ) and q (M) = arg max  qM ?¯ ( M   ).
Evidently, q (M) = q+ (M)  1 and q(M) M 5 arg maxq5{q (M)>q (M)} Z ( M )=
 +
q
 
If  is an irrational number, set Mn = n. If  is a rational number, let
q
(q > Q  ), q  Q  , be the pair of natural numbers satisfying a)   = Q  , and

b) Q + q A Q  + q , for each pair of natural numbers (q> Q ) 6= (q > Q  ) such

q
that   = Q . If Q  is even, consider a sequence Mn = 2n + 1, and if it is odd
Mn = (2)n . In both cases,    q(MM
n)
6= 0, for all n> and limn$4 (   q(M n)
M ) = 0.
By using if necessary subsequences, we get the claim.

Appendix
Proof of Lemma 1 The argument is based on some properties of Young
measures [18]. A Young measure on U+ × \ is a positive measure  such that
236 Aldo Rustichini and Paolo Siconolfi

for any Borel set D  U+ ,  (D×\ ) = O(D), where O is the Lebesgue measure.
The narrow topology on the set of Young measures is defined by the duality of
these measures with the Caratheodory’s integrands; equivalently a sequence
of Young measures converges narrowly if and only if the inner product with
any Caratheodory’s integrand converges.
Let |ˆ : U+ $ \ be a measurable map. The unique Young measure associ-
ated
R R that for any function X̂ : U+ ×\ $ Ū measurable anduw
to |ˆ, , is such  0>
U+ ×\
X̂ g = U+
X̂ (w> |
ˆ (w))gw= Since in our case, \ is compact, X̂ = h X
and X is continuous, the requirement X̂  0 is without loss of generality. A
Young measure  can be represented by its disintegration, which is a family
( w )w5U+ of probabilities over \ such that for any function X̂ : U+ × \ $ Ū
R R R
measurable and non negative, U+ ×\ X̂ g $ = U+ [ \ X̂ (w> |) w (g|)]gw= If  is
associated to |ˆ,  w =  |ˆ(w) , for  |ˆ(w) denoting the Dirac mass at |ˆ(w). The set of
Young measures associated to functions is dense (in the narrow topology) in
the set of Young measures. Thus, if (X ) has a solution |ˆ = (ˆ |  ($> =))$5[0>1) ,
the family of Young measures ( $ )$5[0>1) associated to |ˆ and, equivalently, its
disintegration  ˆ> ˆ  ($> w) =  |ˆ ($>w) > is an optimal solution to:
Z Z
(X 0 ) max huw ( ˆ
(X (($> w))g$)gw,
ˆ
(w)5P+>= ([0>1)>Ê) <+ [0>1)
Z Z
vxemhfw wr ( ˆ
ŝ($> w)|(($> w))g$)gw=
<+ [0>1)

ˆ
where P1>+ ([0> 1)> Ê) = {(w) : $ P+>= (\ ) : (w) is ([0> 1)> Ê) 
phdvxudeoh}=
The argument is concluded by the next claim that creates an obvious
parallelism between the programming problems (X 0 )and (Y ).
An optimal solution to (X 0 ), ˆ  and to (Y ), |ˆ > are equivalent if
R R 
ˆ ($> w))g$ and |ˆ ($> w) = |(
|  ($> w))g$ = X (
Y (ˆ ˆ  ($> w)), for O ×

O  d=h=($> w).

Claim. Optimal solution to (X 0 ) and to (Y ) are equivalent=

Proof. This is a trivial consequence of the linearity (in contingent commodi-

ties) of ŝ and the definition of Y=

Proof of Claim 6.2 We break the argument in two steps.

Step 1: The support of the optimal lottery   is {|1 > |2 }, with 0 ? |1 ? 1 ?
|2  P .

Proof. By the feasibility condition |  (  ) = 1 and by the Inada condition |1 A

0. Furthermore, as already observed 1 is never in the support of   . Thus, the
support of   contains at least two points. Suppose, by contradiction, that it
 Growth in Economies With Non Convexities 237

contains more than two points. Then, there exists at least two points |1 and |2
either both strictly less than one or strictly greater than 1. Thus, there exists
t  such that |1 and |2 5 arg max| |   t  |  or |1 and |2 5 arg max| |  t  |  .
However, both functions |   t  |  and |  t  |  are strictly concave and
therefore they have a unique maximizer.

Step 2: The intertemporal economy (X> |  > 1) has a unique e!cient allocation
(ˆ ˆ {
{> d̂> ), ˆ =   , for O  d=h=w=
ˆ(w) = 1, d̂(w) = 0 and (w)

Proof. We just prove that there exists a unique stationary allocation. A minor
modification of the argument implies the claim. By the form of the technology,
the optimal stock of capital is { = 1. By contradiction, suppose that there
exists two distinct optimal lotteries  and  with supports {|1> > |2> }, |1> ?
1 ? |2>  P .  = > . Denote with , in addition to the optimal lottery,
the probability of |1> ,  = > . Since, |  () = 1,  = > , and X () =
X (), {|1> > |2> } =
6 {|1> > |2> }, otherwise, the two lotteries are identical.
Define the lottery  which assigns probability   + 2 to the allocation
|1> +|1> 2(+)
|1 ()  + and probability 1   = 2 to the allocation |2 () 
(1)|2> +(1)|1>
2(+) . The lottery  is feasible since by convexity:
1 = 2 {| () + |  ()} = +
1     
2 (( + (|1> ) + + (|1> ) )+
2(+) 1 1
2 ( 2(+) (|2> ) + 2(+) (|2> ) )] A (|1> ) + (1  )(|2> ) =

| ()=
Furthermore, by concavity:
 2(+)
X () = +  
2 ( + |1> + + |1> ) + 2
1
( 2(+) 1
|2> + 2(+) |2> ) ] 
1  
2 [| () + | ()] = X () = X ().
Thus, the lottery  is weakly preferred to the optimal lotteries  and
 and frees up resources. Thus, |1 () can be increased (by an arbitrarily
small amount) delivering a lottery that Pareto dominates the optimal ones.
A contradiction.

The E!cient Stationary Allocation of Example 5

Bear in mind that i ({l ) = {l = 1, for l = 1> 2. Since X and I are
separable> |2 = i ({1 ) = 1 5 arg max|2 |2  t2 |2 , for t2 = 1.
For given t1 A 0, let \1 (t1 ) be the set of optimal solution to:

max x(|1 )  (t1 @2)(|1 )2 > (30)

|1

To pin down the e!cient allocation (for the first good) , we have to find
a price tR1 A 0 and a feasible lottery  over the \1 (t1 )> i.e., a lottery that
satisfies \1 (t1 ) |12 (g|1 ) = 1=

Lemma 6. The unique feasible and optimal solution to 30 is, for t  = 1,

|1 = 1=
238 Aldo Rustichini and Paolo Siconolfi

At this point, for sake of clarity, we break the argument in two steps. First
we show that, by the feasibility conditions, t  = 1. Then we show that for
t1 = 1, the optimal solution is |1 = 1 .
Step 1: t1 = 1.

Proof. Suppose by contradiction that t1 6= 1= If an optimal solution |1 6= 1 ±%

exists, it must satisfy the following first order conditions:
(11@1  %)  t1 |1 = 0> |1 5 [0> 1  %)
1  t1 |1 = 0> |1 A 1 + %
1 + (|1  1)  t1 |1 = 0> |1 5 (1  %> 1 + %)=
The first order conditions are satisfied by |1 = (1@t1 ) A 1 + %, if t1 ? ,
and by some |1 A 1> for t 5 [> 1). Furthermore, for t ? 1, |1 = 1  %
can not be an optimal solution, since lim|$(1%) [x0 (|1 )  t(|1 )} A 0 and
lim|$(1%)+ [x0 (|1 )  t(|1 )] = 1  t1 + %(t1  ) A 0. Hence, if t ? 1, the
consumption bundles that maximize the problem 30 are all greater than 1,
thereby violating the feasibility conditions. Hence, it must be 1  t1 . However,
if t1 A 1, the first order conditions imply that |1 ? 1. Furthermore, for t A 1,
|1 = 1 + % can not be an optimal solution, since limf$(1+%) [x0 (f1 )  t(f1 )] =
1  t1  %(t1  ) ? 0 and limf$(1+%)+ [x0 (f1 )  t(f1 )} ? 0. Once again,
feasibility conditions are violated. Hence, t = 1.

Step 2: |1 = 1

Proof. Observe that, for t1 = 1,

a) |1 = 1 satisfies the first and second order conditions
b) it does not exists |1 , |1 6= 1  % and |1 6= 1 + %, that satisfies the
first order conditions.
However, at |1 = 1  % and at |1 = 1 + %, the objective function x(|1 ) 
(|1 )2 @2 is not dierentiable, but it is locally concave. Hence, in order to rule
out 1  % and 1 + % as possible optimal solutions we compare the values of the
objective function at 1> 1 + % and 1  %. By direct computation:
[x(1  %)  (1  %)2 @2]  [x(1)  (1)2 @2] = (1@2)(1  )%2 ? 0>
dqg
[x(1 + %)  (1 + %)2 @2]  [x(1)  (1)2 @2] = (1@2)(1  )%2 ? 0.
Hence, the claim.

Proof of Lemma 2. By contradiction, suppose that there exists (t> ŝ),

such that |ˆ is (part of) a static sunspot equilibrium. Bear in mind that
|ˆ1 ($) = 1> ŝ2 ($)ˆ|2 ($) = t2 |ˆ2 ($), O  d=h=$, and x0 (1) = 1. Then, by the first
order conditions of the individual problem (K):
|1 ($)) = x0 (1) = 1 =  ŝ1 ($), 1  ŝ2 ($) and |ˆ2 ($) =  ŝ2 ($)ˆ
x0 (ˆ |2 ($), for
some  A 0, O  d=h=$=
Then, last conditions and equations 22 immediately imply that:
l) ŝR1 ($) = 1@> O  d=h=$, and R t2 = 1@; R
|2 ($)g$ = t2 [0>1) |ˆ2 ($)g$ = [0>1) (ˆ
ll) [0>1) ŝ2 ($)ˆ |2 ($)@)g$, and
 Growth in Economies With Non Convexities 239

Then
Z Z

ŝ($)ˆ
| ($)g$ = (t2 > t2 ) |ˆ ($)g$ = 2t2 = 1 + t2 , i.e., t2 = 1
[0>1) [0>1)

Consider the constant allocation |ˆ0 defined R by |ˆ0 ($) = (2> 0) for all $.
0
| 0 ($))O(g$) = 2. However:
Evidently, |ˆ is budget feasible at (t> s) and [0>1) X (ˆ
R R
2 A [0>1) X (ˆ|  ($))g$ = x(1) + [0>1) |ˆ2 ($)g$ = 1 + x(1) = 2  (1@2)%2

Proof of Lemma 3 This is consequence of Jensen’s inequality. More

precisely, the utility function X is concave in [0> | ] and in [| > 4)= For each
| M 5 | M (M), let | M1 = (|1 > ===> |M1 ) be the vector of components of | M contained
in the interval [0> | ] , i.e., | M1 5 [0> | ]M1 . There is no loss of generality in
assuming that the entries of | M1 are the first M1 entries of | M . Let M2 = M  M1 .
Evidently, by construction, | M2 = (|M1 +1 > ===> |M ) 5 (| > 4)M2 . Then:

M1 X (|m ) M X (|m )
m X (|m ) = M1 m=1 + M2 m=M 1 +1

M1 M2
M1 M
[(M1 X ( m=1 (|m @M1 )) + M2 X ( m=M 1 +1
(|m @M2 )]=

Proof of Lemma 4 Given the shape of X , for  5 [1@2> 1), the optimal
solutions to (X  ) is either i) the constant bundle 1̄ = (1> 1) and/or ii) |ˆ 6= 1̄,
with min{|1 > |2 } = 0 and ˆ |1 + (1  )ˆ
|2 = 1.
There are just two candidates that satisfy ll). They are obtained by setting
equal to 0 either the first or the second entry of the consumption bundle. The
first candidate |ˆK () has already been defined, the second, |ˆO ()> is defined
as follows:
1 1
|ˆO () = ( > 0) and XO () = X ( )
 
1
Let () = XK ()  XO (). For  5 [  > 1), 1  |e A 1 , and hence:
() = (1  )[(  )|e  | ] +   (1  | )=
Then:

(  ) = (  (1    ))|
(1) =   (1  | )=
Since  A 1@2 and  A , (  ) A 0= Furthermore, by X 3, (1) A 0, Hence:


XK () A XO ()> i ru  5 [  > 1)>

1 1
For  5 (1@2>   ], 1 A  A | and, by direct computation:
() = (1 + (2  1)| ) A 0
Hence, XK () A XO (), for all  5 [1@2> 1). Furthermore:
240 Aldo Rustichini and Paolo Siconolfi
½ ¾
| , for  5 [1@2>   )
XK ()  X (1̄) = .
(  )((1  )|e  1) + | , for  5 [  > 1)

Therefore, by trivial computations, XK (¯  ) = X (1̄), Z () = XK (), for

 5 [1@2> 
¯ ), while Z () = X (1̄),  5 [¯
 > 1). Finally, by taking into account
assumption K2),
; <
? | A 0, for  5 [1@2>   ) @
gZ ()@g = (  )|e + | ? 0, for  5 (  >  ¯) .
= >
0, for  5 [¯> 1)

References
1. Cass, D. (1989), ”Sunspots and incomplete financial markets: The leading ex-
ample,” in G. Feiwell, ed., ”The economics of imperfect competition and em-
ployment, Joan Robinson and beyond,” (Macmillan, London).
2. Cass, D. and Polemarchakis, H, (1990), "Convexity and Sunspots: A Remark,"
Journal of Economic Theory, 52, 433-439.
3. Cass, D. and Shell, K., (1983), ”Do sunspot matter?”, Journal of Political Econ-
omy, 91 (1983), 193-227.
4. Garrat, R., (1995), "Decentralizing lottery allocations in markets with indivisi-
ble commodities," Economic Theory, 5, 295-313.
5. Garrat, R., Keister, D., Qi, C. and Shell, K., (2003), “Equilibrium Prices When
the Sunspot Variable is Continuous”, Journal of Economic Theory, 107, 11-38.
6. Garratt, R. and Keister, T., (2002) ”A characterization of robust sunspot equi-
libria,” Journal of Economic Theory, 107, 136-144..
7. Goenka, A. and K. Shell, K., (1997) ”Robustness of sunspot equilibria,” Eco-
nomic Theory, 10, 79-98.
8. Kehoe, T., Levine, D. and Prescott, E., (2001), ”Lotteries, Sunspots and Incen-
tives Constraints”, Journal of Economic Theory, 107, 39-69.
9. Lancaster, K. L., (1966), “A New Approach to Consumer Theory”, Journal of
Political Economy, 74, 2, 132-157.
10. "Sunspots and Lotteries," Prescott, E., and Shell, K., eds. Journal of Economic
Theory, (November 2002), Volume 107, Issue 1, Pages 1-157.
11. Prescott, E. and Townsend, R., (1984), ”Pareto Optima and Competitive Equi-
libria in an Economy with Adverse Selection and Moral Hazard,” Econometrica,
52(1), 21-45
12. Rustichini, A. and Siconolfi, P., (2003), " Non-convexities in preferences over
commodities and the Lancaster’s model of characteristics," mimeo.
13. Rustichini, A. and Siconolfi, P., (2002), "Growth with non-convex preferences,"
mimeo.
14. Rustichini, A. and Siconolfi, P.(2002), "Large Economies with non-convex pref-
erences," mimeo.
15. Rustichini, A. and Siconolfi, P.(2004), "Growth in economies with non convex-
ities: sunspots and lottery equilibria," forthcoming Economic Theory.
 Growth in Economies With Non Convexities 241

16. Shell, K., (1987), "Sunspot Equilibrium," in The New Palgrave: General Equi-
librium, Eatwell, J., Milgate, M., and Newman, P., eds. (W.W. Norton, New
York, London).
17. Shell, K., and Wright, R., (1993), "Indivisibilities, Lotteries and Sunspot Equi-
libria," Economic Theory, 3, 1-17.
18. Valadier, M., (1993), "A course on Young Measures," mimeo.
Shaking the Tree: An Agency-Theoretic Model
of Asset Pricing⋆

Jamsheed Shorish1 and Stephen E. Spear2

1
Department of Economics and Finance, Institute for Advanced Studies, Vienna
Austria [email protected]
2
Graduate School of Industrial Administration, Carnegie Mellon University,
Pittsburgh PA [email protected]

1 Introduction
The conditional heteroskedasticity and long-memory persistence apparent in
financial times series is one of the most thoroughly studied but least under-
stood of empirical regularities in financial economics. These regularities were
first noted formally in the mid-1960s by researchers examining the long-run
statistical behavior of stock prices, interest rates, and foreign exchange rates
(see e.g. [8], [12], or [21]). In the early 1980’s, the availability of high-speed
computers made it possible for researchers to begin modeling the time-varying
behavior of these time series explicitly. Early work on these topics includes
the original Autoregressive Conditional Heteroskedasticity (ARCH) model of
[7], and work by [17] and by [10] on fractional differencing (which built on
the seminal analysis of fractional Brownian motions by [22]). Since that time,
a number of extensions and elaborations of these models have appeared (see
[2] for a detailed review of this literature). Application of these models to the
stock price, interest rate, and foreign exchange data generally yields results
which are highly statistically significant. It is no surprise, then, that ARCH
and fractionally integrated time-series models have become increasingly more
popular as tools for analyzing financial data.
What is surprising, however, is the relative dearth of theoretical results
which might explain the observed time-varying volatility and/or persistence
in the data. Standard asset pricing models of the type originally formulated
by [20] or [3] are simply too stationary to deliver the desired effects. In the
[20] framework, for example, the price of asset i is given by

pi (yt ) = βEt {Mi (yt , yt+1 ) (yt+1 + pi (yt+1 )) | yt }

⋆
We would like to thank Beth Shorish for her patience and guidance during this
project, as well as conference participants at the 1998 North American Econometric
Society Summer Meetings, Montreal, and the 53rd Econometric Society European
Meetings, Berlin for their many useful comments.
244 Jamsheed Shorish and Stephen E. Spear

where β is the discount rate, Mi is the pricing kernel, and yt is the vector
of dividends paid at time t. With a stationary pricing kernel, only the as-
sumption of conditional heteroskedasticity in the dividend process will deliver
ARCH effects in this framework. In models with production, convergence to
steady-state equilibrium has the effect of eliminating non-stationarities in the
model which might generate time-varying conditional volatilities, unless, of
course, one assumes that the exogenously given shock process is itself ARCH.
This approach has been examined explicitly in work by [1] and [11]. Each of
these papers allows for underlying dividend processes with time-varying condi-
tional variances and examines the implications of changing risk on the general
equilibrium behavior of the model. While such studies can yield interesting
insights, they hardly constitute a theoretical explanation of ARCH.
There have been several hypotheses put forth to explain ARCH effects. One
such hypothesis holds that serially correlated news arrival drives increases in
variance (see, for example, [5] or [9] for details). While this is plausible on its
face, is poses the obvious question as to why the information arrival process
should be serially correlated. Other researchers ([26], [27]) have examined time
deformations that can occur when calendar time and market time proceed at
different rates. In these models, trade can occur more or less frequently in the
same calendar period. This can lead to observed volatilities which vary with
calendar time even though the asset is covariance stationary when measured
in “operational time”. While these models do exhibit the desired conditional
heteroskedasticity, they do not explain what might drive the fluctuations in
trading rates that give rise to the time deformation to begin with.
A third hypothesis has been put forward by [14] to explain the observed
ARCH effects in interest rate time series. In their model, market incomplete-
ness together with constraints on borrowing leads to differences in savings
behavior between rich and poor agents. This implies that the distribution
of wealth in the economy influences the degree to which otherwise nicely be-
haved stochastic endowment shocks affect the interest rate in the model. Time
variation in the distribution of income translates into time variation in the
volatility of interest rates. While this mechanism may be at work in generat-
ing ARCH effects in interest rates, it seems implausible that variations in the
wealth distribution could occur with sufficient frequency to drive the ARCH
effects observed in stock prices.
The literature on theoretical explanations of the observed persistence in
economics time-series is even sparser. The earliest result is work by [13] show-
ing how aggregation of independent AR(1) processes can generate a process
which is fractionally integrated. Since many financial and economic decision
problems in stochastic environments lead to decision rules and pricing rela-
tions which are autoregressive, this would seem a plausible explanation for
why economic aggregates exhibit persistence. One problem with this result,
however, is that it is driven by the requirement that there exist AR(1) pro-
cesses which have autoregressive parameters very close to 1. This requirement
is generally not consistent with observed microeconomic data. A second prob-
 Shaking the Tree: An Agency-Theoretic Model of Asset Pricing 245

lem is that the result isn’t applicable to much of the financial data which is
available in disaggregate form. More recent work on this issue has been done
by [6] who examines a model based on the Ising models of physics. In Durlauf’s
model, firms are located on a lattice, and the activities of neighboring firms
”spill over” and influence the original firm’s production activities.
In this paper, we examine a very different mechanism based on an agency
theoretic model of the internal dynamics of the firm. One of the weaknesses
of exchange models of asset pricing is the need to exogenously specify the
dividend process. This weakness is remedied somewhat in models of capital
accumulation, since production is modeled explicitly. But the neoclassical view
of the firm as a shell housing the basic technical processes that transform
inputs into outputs fails to capture several essential aspects of real firms that
may be relevant in determining the value of the firm. A key feature, which
we focus on here, is the ability of a firm’s managers to respond to favorable
opportunities by expanding output (or sales from inventory) and to reduce
output (or restock inventory) in response to unfavorable shocks.
We examine this feature using a standard principal-agent model together
with a particular reduced form for production which allows the agent to con-
trol, at some cost, the rate of growth of the firm’s output. To keep the analysis
relatively simple, we embed this part of the model in a simple variant of the
[20] asset-pricing model. As in Lucas, we have a single representative agent
who owns an orchard full of trees (the assets) which bear stochastic amounts
of completely perishable fruit (dividends) in every period. Unlike Lucas, how-
ever, we do not assume that the process of fruit production is completely
exogenous. Instead, we include an agent, who one may think of as the gar-
dener, who can influence the production of fruit by exerting effort in the
orchard. Specifically, we assume that the agent can cause the amount of fruit
produced to grow by exerting effort. In the context of the tree model, we can
think of our gardener’s fixed amount of time being allocated between routine
crop tending and innovations in fertilizers, root stocks, and hybridization. The
latter facilitate growth, which in turn demands more of the former activities,
making it more difficult to engage in activities that promote innovation.
In order to keep the analysis relatively simple, we will focus on the case
of a single asset or firm. The owner of the firm contracts with a manager
to operate the firm and pays him a portion of the output as compensation.
Again, for simplicity, we assume that the principal’s interest is in maximizing
the value of the firm, while the agent is risk averse with respect to his income.
We consider a simple repeated relationship between the principal and agent
without commitment. Under these assumptions, we show that the optimal
contract generates a time-series for the firm’s price which exhibits significant
conditional heteroskedasticity and long-memory persistence, even when the
underlying innovations are i.i.d.
In Section 2 we lay out the formal model. In Section 3 we look at a specific
parametrization of the model and indicate how to solve this model numeri-
cally. Section 4 compares the first-best and second-best contract equilibria for
246 Jamsheed Shorish and Stephen E. Spear

the model, using both analytical conclusions and the numerical solution. Sec-
tion 5 examines simulated time-series data generated by the model, and tests
the data for evidence of long-memory persistence and of ARCH effects. We
find that for certain specifications of the manager’s preferences, both of these
phenomena occur. Section 6 presents conclusions, while an Appendix contains
the solution of the first-best contract, as well as the existence proof for the
second-best contract equilibrium and details of the numerical simulation.

2 The Model
The model is based on the stochastic asset pricing model of [20], in which
the firm is comprised of two agents, an owner and a manager. The manager’s
action is defined as an effort level which contributes to the production of
the firm. Following Lucas, we associate the firm’s production with a simple
specification of a dividend process:
xt ∼ G(xt | xt−1 , at ) (1)
where xt and xt−1 are the current and previous dividends, respectively, at is
the manager’s action, and G is a continuous conditional distribution function
over a compact support X—for simplicity we specify a first order dividend
process.
Each period the manager chooses an action which maximizes the expected
utility for the upcoming period. We assume that the manager is accepting a
series of one-period contracts, and cannot commit to a longer contract. The
manager’s preferences are separable and given by
U (wt , at ) ≡ u(wt ) − h(at ) (2)
where wt is the manager’s wealth, defined as the wage paid by the owner, u
is an increasing, continuous concave function, and h is a strictly increasing,
continuous convex function. The h function measures the disutility of working
to the manager. Finally, we assume that the agent possesses outside employ-
ment opportunities, which sets a lower bound ū for the utility the agent must
receive from employment.
The owner of the firm wishes to maximize the firm’s value, i.e. the asset
price. The owner must pay the manager a wage wt in return for the manager’s
labor at in production (which is summarized by the dividend process). We
suppose that the manager’s labor contribution is unobserved when the wage
is specified–thus the owner can only condition the wage upon the current
dividend xt . The owner seeks to
max Et−1 (pt ), (3)
at ,wt

where Ej denotes the conditional expectation operator with respect to infor-

mation known at time j.
 Shaking the Tree: An Agency-Theoretic Model of Asset Pricing 247

As in [20] the price of the firm is given by the discounted expected future
dividends, net of the labor wage. This can be succinctly written as
pt = βEt pt+1 + xt − wt (xt ), (4)
where β ∈ (0, 1) is the owner’s discount factor.
To complete the model, some mechanism for the owner’s expectation of the
future price must be defined. Suppose that the owner uses prior values of the
state variables, in this case the dividend, in order to form her expectations. We
assume that only the expected current dividend value is used in the forecast,
and that the forecast function v : X → R+ is continuous and time invariant:
Et (pt+1 ) = v(xt ). (5)
The owner’s problem is to select an action at and a wage wt to maximize
the value of the firm, given that the action of the agent will be unobservable.
Using equations (1) to (5) we can restate the owner’s problem as
 
max (x − wt (x)) dG(x |xt−1 , at ) + β v(x) dG(x |xt−1 , at ) (6)
at ,wt X X
such that 
at ∈ arg max u(wt )dG(x | xt−1 , a) − h(a), (7)
a X

u(wt ) dG(x |xt−1 , at ) − h(at ) ≥ ū.
X
Note that in this form the owner faces a typical second-best repeated static
moral hazard problem–the dividend process is conditioned on the manager’s
choice of labor, but this is not incorporated into the contracting institution by
the agents. While the single-period contract specification is used for reasons
of tractability, it may be helpful to think of this as describing a firm which
hires and fires labor of the same type every period. A newly-hired manager
may observe that previous labor has affected the dividend process (as the
distribution function for dividends is common knowledge) but has no control
over the previous manager’s labor choice.
We would like to examine the solution to the owner’s problem without
having to worry about the learning dynamics associated with the forecast
function v. That is, we will assume that the owner has already learned the
rational expectations equilibrium (REE) price function, and uses this function
to forecast future prices. In other words, the owner knows the actual function
v ∗ which takes the observed dividend and returns the expected value of the
firm, i.e. Et−1 pt ≡ v ∗ (xt−1 ). This means that the REE price function must
satisfy
 
∗
v (xt−1 ) = max Et−1 [xt |xt−1 , at ] − wt (x) dG(x |xt−1 , at ) (8)
at ,wt X
 %
+β v ∗ (x) dG(x |xt−1 , at ) .
X
248 Jamsheed Shorish and Stephen E. Spear

such that

at ∈ arg max u(wt ) dG(x |xt−1 , a) − h(a), (9)
a X

u(wt ) dG(x |xt−1 , at ) − h(at ) ≥ ū. (10)
X

In the Appendix it is shown that under the current assumptions on the

dividend process and some additional restrictions on preferences, there ex-
ists a function v ∗ which defines at each point in time the value of the firm.
However, in order to more fully characterize the qualities of the value func-
tion (and the associated optimal labor and wage functions, respectively) it is
necessary to introduce functional form assumptions on the dividend process
and on preferences. These assumptions also allow for numerical simulations
to take place. In selecting these functional forms we have attempted to main-
tain a balance between generality and tractability–even with the specifications
given below, the rational expectations price function is still general enough
that an analytical solution cannot be found, and numerical analysis must be
performed.

Definition 1. The dividend process is an AR(1) process with innovation in

the mean, i.e.

G(xt | at , xt−1 ) ⇔ xt = at + ρxt−1 + εt , |ρ| < 1, εt ∼ N (0, 1) ∀t.

Definition 2. U (wt , at ) ≡ − exp(−wt ) − exp(at ).

With Definition 2 we may adopt the first-order-approach (see e.g. [25], [18])
to replace the argmax operator in equation (7) with the associated first-order
condition: &
∂

&
u(wt (x)) dG(x | a, xt−1 ) − h(a)&& = 0. (11)
∂a X a=at

3 Numerical Approximation
The specifications for the dividend process and preferences are not enough
to generate an analytical solution for the value, wage or labor functions. The
approach taken in this paper is to leave the remaining functional forms as
general as possible, and to instead focus on numerical solutions to the REE
condition. The aim here is to identify and analyze the dynamics of the price
and dividend processes given the numerical solution, instead of concentrating
solely upon the analytical results of e.g. a local quadratic approximation of
the value function v ∗ .
However, we would like to be able to compare the resulting owner-manager
contract and, ultimately, the dynamics of the price process with a benchmark
 Shaking the Tree: An Agency-Theoretic Model of Asset Pricing 249

case. We can then see what this type of environment is ‘bringing to the table’
when compared with other contracting forms. In particular, it is interesting to
compare the ‘second-best’ model outlined above with the ‘first-best’ problem,
in which the manager’s action is set by the owner without considering the
manager’s optimal choice. In this case the incentive compatibility condition
is absent from the owner’s optimization, and the only thing the owner need
worry about is giving the manager enough utility (in this case ū) to choose
employment.
In the first-best case, the problem facing the owner is
 
max Et−1 [xt |xt−1 , at ] − wt (x) dG(x |xt−1 , at )
at ,wt X
 %
+β v(x) dG(x |xt−1 , at ))
X

such that 
u(wt ) dG(x |xt−1 , at ) − h(at ) ≥ ū.
X
In the Appendix it is shown that when Definitions 1 and 2 hold the wage func-
tion has the usual property that the manager’s risk is entirely smoothed away–
regardless of the observed dividend, he always receives a utility of ū. Both the
wage function and the effort function are constant, and the expected price
function v ∗ is linear and increasing. Further details on the first-best contract
will be presented when compared with the second-best contract below.
Unfortunately, in the second-best case it is not possible to find a closed-
form solution. So there remains the problem of finding the expected price
function (or ‘value function’) v ∗ of the firm. We adopt here a numerical ap-
proximation technique to identify the value and policy functions. The method
of obtaining the value function used here is by iterating on a functional oper-
ator T , defined by

T (v n ) (xt−1 ) = Et−1 [xt |xt−1 , ant ] − wtn (x) dG(x |xt−1 , ant ) (12)
X

+β v n (x) dG(x |xt−1 , ant ).
X

where wtn and ant are the optimal choices given the candidate value function
v n . Since T is a contraction (see Appendix) it follows that
lim T n (v0 ) = v ∗ ∀v0 ∈ CX
1
. (13)
n→∞

This condition simply states that for any initial continuous (bounded) function
taking dividends into prices, iterations of the operator T on the initial function
will converge to the rational expectations price function.
In the approximations these iterations are derived from a class of functions
known as universal approximators. A universal approximator has the prop-
erty that given a finite collection of points from the domain and range of an
250 Jamsheed Shorish and Stephen E. Spear

unknown function, the approximator can update its parameters such that it
converges almost everywhere to the unknown function. Members of the class
of universal approximators include neural networks, which specify a ‘general’
functional form up to a finite set of parameters. These parameters are then
updated by iteration using the collection of points from the unknown func-
tion until they approach a set of ‘true’ parameters which in principle allow the
neural network to arbitrarily approximate the unknown function. (See [16] for
a discussion of universal approximation and neural networks, and [29], [19] for
a general discussion of the applicability of neural networks to both functional
approximation and regression analysis.) The properties of neural networks are
by now well established–for the purpose of this paper, they are essentially a
convenient method of implementing nonlinear least squares regression in a
deterministic setting.
We are now in a position to outline the algorithm for numerically comput-
ing the rational expectations value function v ∗ :
1. Select an initial value function v 0 . Using neural networks, this amounts
to defining a network whose parameters are randomized.
2. Specify a finite grid over the dividend space X. Given v 0 and a distribution
for the error process εt , numerically compute for each point in the grid
the optimal values for wt0 and a0t (i.e. carry out the optimization on the
right-hand side of equation 12, where n = 0). For each grid point, the
value for T ◦ v 0 is computed.
3. Iterate the value function to v 1 = T ◦ v 0 , for which there is a finite col-
lection of values given by step 2. These values define a neural network f 1
which approximates the function v 1 .
4. Use the neural network ' n approximation
' f 1 in place of v 0 in step 2. Re-
n−1 '
peat steps 2-3 until f − f
' < η, where η is some predefined error
tolerance level. Call f n the rational expectations value function, or v ∗ .
5. Given each point in the dividend grid, perform the optimization on the
RHS of equation (12) using v ∗ . This gives in the optimal values for the
wage and effort level for each grid point. These values serve to define
a neural network [wt∗ , a∗t ] = [w∗ (xt , xt−1 ), a∗ (xt−1 )], which is the ’policy
function’ for the economy. Given the state of the economy (i.e., a real-
ization of the current dividend, plus the previous dividend) the policy
function tells the principal what wage should be paid, and how hard the
manager should work in the current period.
The rational expectations value function is the law of motion for the econ-
omy. Once found, v ∗ defines the optimal a∗t –this, combined with the previous
level of the dividend xt−1 and a new realization of the error process εt , gen-
erates the current dividend xt . Having observed xt , v ∗ also defines wt∗ –this
is the wage paid to the agent by the owner and depends only upon [xt , xt−1 ]
since the owner cannot observe the manager’s effort. The actual price of the
asset pt is then
pt = βv ∗ (xt ) + xt − w∗ (xt , xt−1 ). (14)
 Shaking the Tree: An Agency-Theoretic Model of Asset Pricing 251

With the current dividend xt the next effort level a∗t+1 can be defined, and
the process continues. Thus, the economy can be simulated and sequences of
dividends and prices can be generated and analyzed.
Particulars on the specification of the networks v ∗ (xt−1 ) and [w∗ (xt , xt−1 ),
∗
a (xt−1 )], including the number of hidden units, number of iterations, con-
vergence criteria, and other parameter values may be found in the Appendix.

4 First-Best and Second-Best Comparison

We start with the analytical ’benchmark’ results for the first-best case. In the
Appendix it is shown that the optimal value and policy functions are:

a∗
 
1 ∗ ρ
Et−1 pt = v ∗ (xt−1 ) = + ln(|u| − ea ) + xt−1 , (15)
1 − β 1 − βρ 1 − βρ
 
|u|
a∗ = ln ,
2 − βρ
  
∗ 1 − βρ
w = − ln |u| .
2 − βρ
Figure 1 displays the optimal value function for the case where ū = −2,
β = 0.9, and ρ = 0.9. These are the identical parameter values used for the
second-best approximation.

Fig. 1. First-Best Contract Expected Price

Clearly, in the first best case we will have no conditional heteroskedasticity

or long memory–the actual price follows the process
252 Jamsheed Shorish and Stephen E. Spear

a∗
    
1 − βρ 1 ∗ ρ
pt = xt +ln |u| +β + ln(|u| − ea ) + xt ⇔
2 − βρ 1 − β 1 − βρ 1 − βρ
1 − βρ + ρ
pt = c + xt , (16)
1 − βρ
where we have grouped constant terms into c for convenience. Since

xt = a∗ + ρxt−1 + εt

we can lag and substitute (16) into the dividend relation to yield
 
1 − βρ + ρ
pt = ρpt−1 + (1 − ρ)c + (a∗ + εt ).
1 − βρ

The price thus follows an AR(1) process with the same autoregressive param-
eter as the dividend process, but with a different mean and variance. This
result supports the intuition that because in the first-best case there is no
response by the agent to changes in production, there should be no resultant
correlation between production (or price) volatility from period to period. In
the first-best case, the sole connection between the present and the past is the
autoregressive dividend process.
In the second-best case, however, things are markedly different. Figures 2-4
present the numerical results for the second-best value function, wage function
and effort function respectively. Figure 2 shows the numerical approximation
of the value function given by the program (8)-(10), using Definitions 1 and 2
and the parameter values ū = −2, β = 0.9, and ρ = 0.9. The value function is
strictly concave, and for high levels of the dividend the expected future price is
nearly constant. This reflects the fact that for very high levels of the dividend
the manager shirks a great deal, absorbing any expected future gains from
the dividend. Figure 3 presents the manager’s effort function–note that for
low or negative dividend levels the manager wishes to work a (small) positive
amount, but that as dividends rise the manager rapidly works less and less.
Figure 4 presents the optimal wage as a function of the current and past
dividends. From this we see that the manager receives positive compensation
when there is a large positive gain in the dividend process (indicating hard
work by the agent). Compensation then falls as the difference between the
two dividends falls, and becomes sharply negative when the current dividend
is far below the previous one.

5 Simulation of Time Series

Once the value function and optimal wage and effort functions have been
approximated it is possible to simulate the economy and generate synthetic
time series for analysis. The simulation of time series data is important because
the second-best model has no analytical solution. Thus, analytical tools such
 Shaking the Tree: An Agency-Theoretic Model of Asset Pricing 253

Fig. 2. Second-Best Contract Expected Price

Fig. 3. Second-Best Contract Manager Effort

as comparative statics must be traded for tools which take advantage of the
large number of simulated data points that can be generated from the model.
Tools such as regression analysis do, of course, make the tacit assumption
that the model is an accurate representation of the salient features of the
owner-manager relationship in a real economy. Nonetheless, the complexity of
the model and the relative paucity of relevant data in the real world are both
strong incentives to use simulated time series data as a proxy for economic
data generated from actual owner-manager behavior.
The owner-manager model is by its nature a nonlinear model of pric-
ing. Thus, one might expect the dynamics of observed price and dividend
sequences to reflect this nonlinearity. Of particular interest is the extent to
254 Jamsheed Shorish and Stephen E. Spear

Fig. 4. Second-Best Contract Wage

which the nonlinearity in the model is capable of generating time-series data

with properties observed in real stock price data. Thus, in analyzing the sim-
ulated time-series, we look for evidence of conditional heteroskedasticity and
long-memory persistence.
The tools used for the analysis of the simulated time series are the general-
ized ARCH (or GARCH) model of conditional heteroskedasticity (see [7] and
[2]), the estimated power spectrum of the simulated time-series, and a test
for long-memory persistence. The results of this analysis indicate that 1) the
price series exhibits strong ARCH-like behavior, with the GARCH (1,1) model
demonstrating significant correlated volatility, and 2) the power spectrum and
the long-memory test indicate some long-memory persistence.
The second-best value function and the optimal wage and action functions
were used to generate time series of both prices and dividends for one hundred
thousand periods. A typical example of the dividend time series, along with
the associated wage compensation and effort levels for 200 periods are given
in Figures 5 and 6. A sample path of the second-best price for the same
200 periods is presented in Figure 7. Since it is the value of the firm which
empirically demonstrates ARCH-like behavior and long-memory persistence,
only the time series for the price of the firm was tested in what follows.

5.1 Spectral Analysis

To answer the question of whether the simulated price series exhibits per-
sistence, we first examine the empirical power spectrum. We calculated the
Welch-averaged spectral density of the simulated price series with 100,000
data points, with a Hanning window 100 units wide. The smoothed spectrum
with 95% confidence bands is shown in Figure 8.
 Shaking the Tree: An Agency-Theoretic Model of Asset Pricing 255

Fig. 5. Sample Dividend Series from Synthetic Data, 200 Periods

Fig. 6. Wage and Manager Effort Series from Synthetic Data

The spectrum indicates considerable power in the lower frequencies, a hall-

mark of long-memory processes. However, it does not appear that the spec-
trum demonstrates strong long-memory persistence. Rather, it appears at first
blush that the spectrum of the second-best price series is still dominated by
the AR(1) structure of the dividend process. In order to ascertain if long-
memory has any impact we turn to a ’rough and ready’ test of persistence
given by [10]. This test attempts to verify whether or not the spectrum bears
a some similarity to an inverse power law in frequency. In the literature on
fractionally integrated time-series, this is often referred to as ”1/f ” noise.
To test the hypothesis that the observed data behave as 1/f noise, we
regressed the log of the power spectrum against the log of the (sin of) fre-
256 Jamsheed Shorish and Stephen E. Spear

Fig. 7. Sample Price Series from Synthetic Data, 200 Periods

Fig. 8. Smoothed Price Spectrum with 95% Confidence Bands, 100,000 Obs.

quency, and the results are presented in Table 1. As shown in the table, the
estimated long-memory exponent d = 0.61 is significant beyond the 99% con-
fidence level. We conclude, then, that the simulated price series exhibits some
long-memory persistence.
This results raises the obvious question as to what causes the observed
persistence. One way of approaching this is to consider an experiment first
performed in the 1940’s by the hydrologist Harold E. Hurst. Hurst’s exper-
iment involved generating random sequences of biased random walks, and
analyzing the properties of the resulting time-series. Subsequences were char-
acterized by a fixed bias in the step size of the random walk. A second random
variable determined when the bias would be changed, generating a new sub-
 Shaking the Tree: An Agency-Theoretic Model of Asset Pricing 257

Table 1. Spectral Regression (Geweke & Porter-Hudak) Estimate Results

Variable Coefficient Std. Error t-statistic

intercept 2.2096 0.0705 31.349∗∗
d 0.6088 0.0441 13.816∗∗
2
R 0.7578

sequence. When Hurst examined large data sets generated in this fashion, he
found significant evidence of long-memory persistence. Subsequent work on
time-series models with exogenous structural breaks has confirmed the long-
memory properties of the time-series generated by such models (see e.g. [24]).
In our model, structural breaks in the trend of the stock price time series oc-
cur when the drift parameter swings from negative to positive (or vice-versa)
in response to increased (or decreased) effort by the agent. These breaks oc-
cur endogenously (but randomly, given their dependence on the innovation
process of the dividend) and, we believe, generate the observed persistence.
As we will see in the following section, the trend breaks may also generate
the significant conditional heteroskedasticity observed in the simulated price
series.

5.2 Estimation of the GARCH (1,1) Model

The GARCH (1,1) model for conditional heteroskedasticity is defined as

pt | Ψt−1 ∼ N βpt−l , σt2 ,

σt2 = α0 + α1 ε2t−1 + α2 σt−1

2
,
εt = pt − βpt−l ,
′
where Ψt−1 is the information set available at time t−1, pt−l ≡ [1, pt−1 , . . . pt−l ]
is an (l + 1)−dimensional vector of lagged endogenous variables, εt is the
residual of the mean regression, and β, α ≡ [α0 , α1 , α2 ] are (1 × l + 1) and
(1 × 3) −dimensional vectors, respectively. This specification allows for the
possibility of ‘booms’ and ‘busts’ (or alternatively, ‘fads’ and ‘bubbles’) in the
price sequence, and has (along with its variants E-GARCH and N-ARCH)
been implemented extensively on economic data (see e.g. [4], [15] and [23]).3
The GARCH (1,1) model of the second-best price series was estimated
using Maximum Likelihood with 100,000 observations, with a first-order au-
toregressive process for the conditional price expectation. Before estimation,
3
Veronesi (1996) also used the GARCH (1,1) specification to test for ARCH, in
a model which uses ’regime shifts’ (similar to the trend breaks of Perron [1989]) to
generate correlated volatility in asset returns.
258 Jamsheed Shorish and Stephen E. Spear

the Lagrange Multiplier (LM) test for ARCH ([7]) and the Jarque-Bera nor-
mality test were applied to the residuals of the mean equation. The results
of the estimation and tests are presented in Table 2. The LM test strongly
rejected the i.i.d. residual hypothesis at the 99% confidence level, while the
Jarque-Bera test rejected the normality hypothesis beyond the 99% confidence
level. These are strong indications that the time series contains some measure
of correlated volatility.

Table 2. GARCH(1,1) Estimation Results

Variable Coefficient Std. Error z-statistic

Mean Equation:
Et−1 pt = β0 + β1 pt−1
Constant 2.5293 0.0266 95.015∗∗
pt−1 0.8309 0.0018 453.23∗∗
LM ARCH Test: T ∗ R2 = 488.27, Pr(i.i.d.) < 10e−6
Jarque-Bera Statistic: 3815.3, Pr(normal) < 10e−6
Variance Equation
Et−1 σt2 = α0 + α1 ε2t−1 + α2 σt−1
2

Constant 0.4802 0.0243 19.739∗∗

ε2t−1 0.0631 0.0025 25.561∗∗
2
σt−1 0.7738 0.0097 79.328∗∗

** refers to significance at the 99% confidence level.

The coefficients of the GARCH (1,1) model were all statistically significant
beyond the 99% confidence level. In addition, the conditional variance process
is strongly persistent (with α2 = 0.774). This provides reasonable grounds for
acceptance of the GARCH (1,1) model as demonstrating the existence of
conditional heteroskedasticity in price data which is generated by second-best
contracting. Naturally, fitting the GARCH model to the data is an a priori
model misspecification, since the price data are actually generated by (14).
The GARCH estimation is in this case simply used to demonstrate the strong
presence of correlated volatility in the price data.

6 Conclusion
We have seen that the simple model presented here yields a price sequence
which can have hidden nonlinear behavior. In addition, it appears that exam-
ples of this sequence are fit well by an ARCH-like specification, which has been
noted to exist in empirical data. These results support the conclusion that en-
dogenous correlated volatility and persistence is possible when the manager
 Shaking the Tree: An Agency-Theoretic Model of Asset Pricing 259

influences the production process and is in turn affected by his contractual

relationship with the owner.
The model of owner-manager behavior in an asset-pricing model presented
here has been simplified in many key ways. First and most important, the
type of contract that the manager is able to make with the owner is essen-
tially static–commitment only occurs over the single period the contract is in
force, and the same contract is assumed to be accepted in perpetuity. Further
investigation would see whether a more general model, incorporating a multi-
period contract with commitment, would be substantively different from the
simplified model presented here. It would be interesting to see, for example,
whether the correlation between dividends and prices is less strong under a
fully dynamic model, since it is empirically observed that dividend time series
is less volatile than (hence, less strongly correlated with) the asset price series.
In addition, it is not clear whether the manager affects the mean of the
production process (as assumed here), the variance of the process (as would be
true, for instance, if the manager could influence the impact of the exogenous
shock upon the dividend), or both. Future research will develop a production-
based model of manager effort which can then suggest a dividend process
such as equation (1) as a direct conclusion. Of course, given the analytical
complexity of the simplified case, it is not at all clear that a more general
model with the above extensions would yield analytically testable conclusions.
Rather, it would appear from this presentation that the tools of numerical
approximation and simulation of time series would be just as valuable for
these cases.

7 Appendix
7.1 Solution of the First-Best Contract

Consider the first-best economy:

   %
∗ ∗
v (xt−1 ) = max a + ρxt−1 − w(x)dG(x|xt−1 , a) + β v (x)dG(x|xt−1 , a)
w,a


s.t. e−w(x) dG(x|xt−1 , a) + ea ≤ |u| ,

where we have used the functional forms from Definitions 1 and 2. From
the optimization with respect to w we know that the wage payment will be
independent of the observed output x, so that
∗
e−w + ea = |u| ⇒

w∗ = − ln(|u| − ea ),
given a particular level of managerial effort a.
260 Jamsheed Shorish and Stephen E. Spear

The owner’s optimization is now

  %
∗ a
v (xt−1 ) = max a + ρxt−1 + ln(|u| − e ) + β v(x)dG(x|xt−1 , a) .
a

The objective function of the owner is concave in a. Posit a candidate policy

function a = a∗ , where a∗ is some constant, and a candidate value function
v ∗ (xt−1 ) = c1 + c2 xt−1 . Then Bellman’s equation at the optimal level of effort
is
∗
c1 + c2 xt−1 = a∗ + ρxt−1 + ln(|u| − ea ) + βc1 + βc2 (a∗ + ρxt−1 ).

Matching coefficients yields

∗
c1 = a∗ + ln(|u| − ea ) + βc1 + βc2 a∗

c2 = ρ + βρc2
or
a∗
 
1 ∗
c∗1 = + ln(|u| − ea ) ,
1−β 1 − βρ
ρ
c∗2 =
1 − βρ
(note that this can also be directly verified by appealing to the equivalent
sequence problem–cf. [28]).
The owner thus wishes to solve

∗
max a∗ + ρxt−1 + ln(|u| − ea ) (17)
a∗ ∈(−∞,ln(|u|))

a∗
  %
β ∗ βρ
+ + ln(|u| − ea ) + a∗ .
1 − β 1 − βρ 1 − βρ

This problem has (after some rewriting) the associated first-order condition
∗
ea 1
= ⇒
|u| − ea∗ 1 − βρ
 
∗ |u|
a = ln .
2 − βρ
This implies that the optimal wage payment is
  
∗ 1 − βρ
w∗ = − ln(|u| − ea ) = − ln |u| .
2 − βρ

In the first-best equilibrium, then, the value of the firm is a linear function
of the previous level of the dividend, while the wage and manager effort are
constant. As either the rate of discounting β or the autoregressive parameter ρ
 Shaking the Tree: An Agency-Theoretic Model of Asset Pricing 261
 
converge to zero the effort level converges to ln |u|
2 , while the wage converges
 
to − ln |u|
2 . In these cases either the owner does not care about the future,
or the manager does not contribute to future dividends through the past
dividend level, and the manager’s effort level is low. As β or ρ rise, however,
the effort level rises and the wage rises to compensate.

7.2 Existence of the Second-Best Value Function

We seek to prove that a function v ∗ exists which solves:

  %
∗ ∗
v(x) = max (y − w(y))g(y |x, a )dy + β v(y)g(y |x, a )dy
w,a
∗
X X

such that 
a∗ ∈ arg max u(w(y))g(y | x, a)dy − h(a),
a X

u(w(y))g(y |x, a∗ )dy − h(a∗ ) ≥ ū,
X
where X is the dividend space, y is the current (unobserved) dividend, x is
the previous dividend, w is the wage paid to the manager, a∗ is the manager’s
optimal action, and ū is the manager’s reservation utility. Note that for expo-
sition our notation here differs slightly from the form in the text–in addition,
we have replaced the current dividend density function dG with the density
function gdy.
We assume that the first-order approach is valid; that is, the manager’s
preferences are such that the incentive compatibility condition (ICC)

∗
a ∈ arg max u(w(y))g(y | x, a)dy − h(a)
a X
may be replaced with (cf. [18])

∂g(y | x, a∗ )

u(w(y)) dy − h′ (a∗ ) = 0. (ICC)
X ∂a
Assuming that the first-order approach is valid implies that an interior so-
lution to the problem ICC exists, i.e. that the second-order condition satisfies
∂ 2 g(y | x, a∗ )

′′
U (w(y)) 2
dy − h (a∗ ) < 0.
X ∂a
when the second-order condition exists. In order to ensure this, we must add
the condition that the density function of current dividends g (y | x, a) be at
least twice-continuously differentiable in a.
From this we know immediately that the Implicit Function Theorem (IFT)
applies around a∗ ; next we assume that g (y | x, a) is at least twice-continuously
differentiable in x so that we may write
262 Jamsheed Shorish and Stephen E. Spear

∂g(y | x, a∗ (x; w))


U (w(y)) dy − h′ (a∗ (x; w)) = 0.
X ∂a

Note that since w takes as its argument the current dividend y, the optimal
action a∗ will not depend parametrically upon w, but rather functionally. By
the IFT, however, we know that this functional dependence is one-to-one.
Thus, the incentive compatibility condition defines the optimal action given
the previous dividend and the wage function.
We may now write the dynamic programming problem of the owner as
  %
∗ ∗
v(x) = max (y − w(y))g(y |x, a (x; w))dy + β v(y)g(y |x, a (x; w))dy
w X X
(18)
such that 
u(w(y))g(y |x, a∗ (x; w))dy − h(a∗ (x; w)) ≥ u (PC)
X
where PC is the participation constraint of the manager.
This problem has a straightforward solution. The current-period return
function is bounded in w by assumption, and we also suppose( that the
conditional expected value of the current dividend is finite (i.e., X yg(y |x,
a∗ (x; w)dy < ∞). The constraint PC is compact-valued, non-empty and con-
1 1
tinuous. As before (see equation 12) we define an operator T : CX → CX
by
 
T (v)(x) = (y − wv (y))g(y |x, a∗ (x; wv ))dy + β v(y)g(y |x, a∗ (x; wv ))dy
X X
(19)
where wv is the optimal solution to the problem (18) + (PC) for a given
function v.
From the above considerations we know that the Theorem of the Maximum
obtains–the T operator takes bounded continuous functions into bounded con-
tinuous functions. Furthermore, we can use Blackwell’s sufficiency conditions
to show that T is a contraction. Recall that if T is an operator taking bounded
continuous functions into bounded continuous functions, then it is a contrac-
tion mapping if
1. T is monotonic, i.e. T (v)(x) ≤ T (w)(x) whenever v(x) ≤ w(x) ∀x
2. T is ’sublinear’, i.e. T (v + c)(x) ≤ T (v)(x) + βc ∀v, β ∈ (0, 1) , c < 0.
It is clear that the operator defined by (19) satisfies Blackwell’s conditions,
so that T is a contraction mapping. Hence, we know that a continuous function
v ∗ exists which solves the owner’s problem, and that

lim T n (v0 ) = v ∗ ∀v0 ∈ CX

1
.
n→∞
 Shaking the Tree: An Agency-Theoretic Model of Asset Pricing 263

7.3 Second-Best Approximation Details

The second-best value function v ∗ was estimated using a single-layer feedfor-

ward neural network of 6 hidden units. Following the procedure outlined in
Section 3, neural networks were fit to dividend grid data and then used to
calculate the next iteration of the value function. Each neural network fit the
grid data so that the sum-of-squared-error (SSE) between the estimated data
and the actual data was less than 10−10 .
The dividend space X ranged from -2 to 20, and was divided into a grid
{xi } of 30 equidistant points to be used as network input. These points were
used to calculate the value function points v n (xi ). These points constituted the
target vector for fitting the neural network. The actual grid data used to esti-
mate the neural network varied from iteration to iteration. An adaptive grid
was used to focus the neural network’s attention on those points which were
particularly hard to estimate, i.e. those grid points whose absolute errors were
above the mean absolute error of the current iteration’s estimate. Convergence
of the value function was assumed when the largest absolute error between
consecutive value function estimates was less than 10−3 . As mentioned in the
text, the owner’s discount factor β was set to 0.9, the autoregressive param-
eter ρ of the dividend process was also 0.9, and the reservation utility ū was
-2.
After the value function iteration had converged, grid data for the optimal
wage function and the optimal effort function were generated. The optimal
wage function was estimated using a single-layer feedforward neural network
of 16 hidden units, and the SEE between the estimated and actual wage values
was less than 10−4 . The optimal effort function required 6 hidden units, and
the SSE between the estimated and actual effort values was less than 10−12 .
264 Jamsheed Shorish and Stephen E. Spear

8 References

1. A. B. Abel. Stock prices under time-varying dividend risk. Journal of Monetary

Economics, 22:375–393, 1988.
2. T. Bollerslev et al. Arch modeling in finance. Journal of Econometrics, 52:5–59,
1992.
3. W. A. Brock. Asset prices in a production economy. In J. J. McCall, editor,
The Economics of Information and Uncertainty. University of Chicago Press
and NBER, 1982.
4. T. E. Day and C. M. Lewis. Stock market volatility and the information content
of stock index options. Journal of Econometrics, 52:267–287, 1992.
5. F. X. Diebold and M. Nerlove. The dynamics of exchange rate volatility: A
multivariate latent factor arch model. Journal of Applied Econometrics, 4:1–21,
1989.
6. S. Durlauf. Locally interacting systems, coordination failure and the behavior
of aggregate activity. Technical report, Department of Economics, Stanford
University Working Paper, 1990.
7. R. F. Engle. Autoregressive conditional heteroskedasticity with estimates of the
variance of united kingdom inflation. Econometrica, 50(4):987–1007, 1982.
8. E.F. Fama. The behavior of stock market prices. Journal of Business, 38:34–105,
1965.
9. A.R. Gallant et al. On fitting a recalcitrant series: The pound/dollar exchange
rate 1974-83. Technical report, Department of Economics, Duke University
Working Paper, 1989.
10. J. Geweke and S. Porter-Hudak. The estimation and application of long memory
time series models. Journal of Time Series Analysis, 4(4):211–238, 1983.
11. A. Giovannini. Uncertainty and liquidity. Journal of Monetary Economics,
23:239–258, 1989.
12. C.W.J. Granger. The typical spectral shape of an economic variable. Econo-
metrica, 34:150–161, 1966.
13. C.W.J. Granger. Long memory relationships and the aggregation of dynamic
models. Journal of Econometrics, 14:227–238, 1980.
14. W. J. Den Haan and Scott Spear. Credit conditions, cross-sectional dispersion
and arch effects in a dynamic equilibrium model. Technical report, Department
of Economics, University of California at San Diego Working Paper, 1994.
15. M.L. Higgens and A. K. Bera. A class of nonlinear arch models. International
Economic Review, 33(1):137–158, 1992.
16. K. Hornik et al. Universal approximation of an unknown mapping and its
derivatives using multilayer feedforward networks. Technical report, University
of California-San Diego Working Paper, 1990.
17. J.R.M. Hosking. Fractional differencing. Biometrika, 68(1):165–176, 1981.
18. I. Jewitt. Justifying the first-order approach to principal-agent problems. Econo-
metrica, 56(5):1177–1190, 1988.
19. K.L. Judd. Numerical Methods in Economics. MIT Press, 1998.
20. R.E. Lucas. Asset prices in an exchange economy. Econometrica, 46(6):1429–
1445, 1978.
 Shaking the Tree: An Agency-Theoretic Model of Asset Pricing 265

21. B. Mandelbrot. The variation of certain speculative prices. Journal of Business,

36:392–417, 1963.
22. B. Mandelbrot and J.W. van Ness. Fractional brownian motions, fractional
noises and applications. SIAM Review, 10(4):422–437, 1968.
23. D.B. Nelson. Conditional heteroskedasticity in asset returns: A new approach.
Econometrica, 59(2):347–370, 1991.
24. P. Perron. The great crash, the oil price shock, and the unit root hypothesis.
Econometrica, 57(6):1361–1402, 1989.
25. W.P. Rogerson. The first-order approach to principal-agent problems. Econo-
metrica, 53(6):1357–1367, 1985.
26. J.H. Stock. Measuring business cycle time. Journal of Political Economy,
95:1240–1261, 1987.
27. J.H. Stock. Estimating continuous time processes subject to time deformation.
Journal of the American Statistical Association, 83:77–85, 1988.
28. N.L. Stokey and R.E. Lucas. Recursive Methods in Economic Dynamics. Cam-
bridge: Harvard UP, 1989.
29. H. White. Learning in artificial neural networks: A statistical perspective. Tech-
nical report, University of California-San Diego, Working Paper, 1989.
Central-Bank Interest-Rate Control in a
Cashless, Arrow-Debreu Economy

Neil Wallace

Department of Economics, The Pennsylvania State University, 608 Kern,

University Park, Pa., 16802, USA [email protected]

Summary. A pure-exchange, competitive economy with date-specific units of ac-

count is studied. It diers from the standard pure-exchange model in having unit-
of-account endowments. For such an economy, conditions are given for an outside
agent, a central bank, to be able to control the nominal rate of interest. It can do so
provided it has positive unit-of-account endowments. A more general outside agent
that has a positive endowment of some good and positive unit-of-account endow-
ments can select the time path of the price level.

1 Introduction
Money is often described as having three functions: (i) a unit-of-account func-
tion, (ii) a medium-of-exchange function, and (iii) a store-of-value function.
In a cashless economy, the third is not operative and, probably, neither is the
second. To lay people, the notion of a cashless economy may seem strange.
To economists, it is not at all strange. After all, the theory of relative prices,
the modern competitive version of which is called the Arrow-Debreu model,
has always been the developed part of economics. It is monetary theory that
has always been undeveloped. And, as it happens, the Arrow-Debreu model
has room for a unit of account–what is called an abstract unit of account,
abstract in the sense of not having a physical counterpart. This note describes
conditions under which an activity that resembles central-bank open-market
operations determines nominal interest rates in a simple Arrow-Debreu model.

2 A two-date model
The model is of a 2-date, pure-exchange economy with Q people and O goods
at each date. In order to make the model fit the unit-of-account vision that
people seem to have in mind for a cashless economy, my version has date-
specific abstract units of account called date-1 ecus and date-2 ecus. Goods
268 Neil Wallace

are indexed by o 5 {1> 2> ===> O} and people by q 5 {1> 2> ===> Q }= Let $ qw 5 RO
++
denote person q0 s endowment vector of date-w goods. Although somewhat non
standard, I will also endow people with ecus at each date and let hqw 5 R denote
person q0 v endowment of date w ecus. (I permit hqw to be positive or negative.
A negative holding is a debt denominated in date-w ecus.) Note that all the
objects are perishable. For goods, that is what pure exchange means. For ecus
that is one of the meanings of cashless: there is no technology that permits a
person to convert ecus at one date into ecus at another date.
Let fqw 5 RO + be q’s consumption vector of date w goods. Person q has a
utility function, a function of consumption of goods, xq : R2O + $ R> which
is strictly increasing. Also, let {qw denote person q0 v consumption or end-of-
date-w holdings of date-w ecus. As implied by the function x, people do not
value consumption of ecus, just as they do not value consumption of cash in
a cash economy.

Definition 1. Person q can aord non negative (fq1 > fq2 > {q1 > {q2 ) at the prices
(s1 > s2 > U) if

(s1 $ q1 + hq1  s1 fq1  {q1 ) + U(s2 $ q2 + hq2  s2 fq2  {q2 )  0= (1)

The restriction {qw  0 should be interpreted as a no-reneging condition. If

{qw? 0> then a person leaves date w owing date-w ecus. That is not permitted.
Notice that the nominal interest rate is U1  1= Given that definition of what
a person can aord, we can define a competitive equilibrium (CE) as follows.

Definition 2. A CE is (fq1 > fq2 > {q1 > {q2 ) for each q and (s1 > s2 > U) such that (i)
(fq1 > fq2 > {q1 > {q2 ) maximizes xq subject
P to beingP
aordable atP(s1 > s2 > U),
P and (ii)
the allocation is feasible; that is, q fqw  q $ qw and q {qw  q hqw for
w = 1> 2=

This definition does not include equilibria in which either or both of date-
1 ecus and date-2 ecus are worthless. In general, there are such equilibria.
Therefore, when we describe necessary conditions for a CE that satisfies def-
inition 2, they are necessary conditions for a CE in which date-1 and date-2
ecus have value. Consideration of such equilibria is similar to consideration
of valued-cash equilibria in models with cash. Often, those models also have
equilibria in which cash is not valued.
If there are no endowments of the unit of account–that is, if hqw  0–then
this is (a special case of) the standard pure-exchange model. And it has the
standard zero-degree homogeneity property.

Claim. If (fq1 > fq2 ) for each q and (s1 > s2 > U) is a CE, then (fq1 > fq2 ) for each q
and ( 1 s1 >  2 s2 >  1 U@ 2 ) is a CE for any ( 1 >  2 ) 5 R2++ =

Proof. Obvious.
Central-Bank Interest-Rate Control in a Cashless, Arrow-Debreu Economy 269

Notice that this implies that there is a sense in which U 5 R+ can be

selected arbitrarily. Such a selection can be interpreted as influencing the
inflation rate between dates 1 and 2 because U has the units date-1 ecus per
unit of date-2 ecus.
Now suppose that there are endowments of ecus. Then, for some prices,
some people may have negative wealth. That may cause problems for existence
of a CE. If a CE exists for given ecu endowments, and if those endowments
are held fixed, then there would seem to be what has been called real indeter-
minacy because as prices vary the wealth distribution varies. I will duck the
existence question here and deal with properties of a CE.

Claim. In any definition 2 CE, (i) prices are positive (s1 A 0> s2 A 0> and
q
P A q0) and {w  0 and (ii) the feasibility conditions hold at equality and
U
q hw = 0=

Proof. Part (i) follows from monotonicity of x= As for part (ii), monotonicity
of x also implies that (1) holds at equality, and, therefore, that the sum of
(1) over q holds at equality. Then, because prices are positive, the sum of
(1)) over q implies
P that the feasibility conditions hold at equality. That and
{qw  0 imply q hqw = 0.

P
The necessary condition, q hqw = 0, is reassuring. It says that total en-
dowments of ecus at each date are zero. In other words, claims on date-w ecus
must be oset by debts of date-w ecus.
And, as is standard, if we adjust ecu endowments appropriately, then the
zero-degree homogeneity property holds.

Claim. If (fq1 > fq2 ) for each q and (s1 > s2 > U) is a CE for ecu endowments hqw ,
then (fq1 > fq2 ) for each q and ( 1 s1 >  2 s2 >  1 U@ 2 ) is a CE for ecu endowments
 1 hq1 and  2 hq2 for any ( 1 >  2 ) 5 R2++ =

Proof. Obvious.

3 A Central Bank
Subject to the above proviso about existence, it seems that the nominal in-
terest rate can be anything. Therefore, there would seem to be scope for an
outside entity, the central bank, to choose the nominal interest rate. One way
to think of a central bank as enforcing a particular magnitude for the nominal
interest rate is to have it oer to buy date-1 ecus for date-2 ecus, and vice
versa. This would seem to be the analogue of open-market operations.
To explore this, let us label the central bank agent 0= Assuming, as is
natural, that the central bank does not deal in goods and does not have
270 Neil Wallace

endowments of goods, we can regard it as choosing non negative {01 and {02
subject to the following special case of (1):

h01  {01 + U(h02  {02 ) = 0= (2)

We can interpret h0w  {0w as the central bank’s sale (purchase if negative) of
date-w ecus. Then, (2) says only that the central bank trades at a price.
Because the central bank does not maximize utility in any ordinary sense,
we do not assume that it necessarily chooses {0w = 0= However, we do assume
that it is also subject to {0w  0: it cannot leave date w owing date-w ecus.
Because the central bank does not deal in goods and does not maximize
utility, the only amendment needed in the definition of a CE is to replace
the feasibility condition for ecus to include agent 0= And because (1) holds at
equality for each private agent and (2) holds, it follows that such feasibility
P PQ
implies Q q
q=0 {w =
q q
q=0 hw for w = 1> 2. And because {w = 0 for q 6= 0 in
any CE, a necessary condition for a CE is
Q
X
{0w  h0w = hqw for w = 1> 2= (3)
q=1

Notice that
P for qgeneral private endowments of ecus, endowments that do
not satisfy Q h = 0> (2) and (3) are inconsistent. Therefore, we continue
PQw
q=1
to assume that q=1 hqw = 0> the necessary condition for a CE when there is
no central bank. Under that assumption and the assumption that the central
bank has positive endowments of ecus, an assumption discussed below, it can
be shown that the central bank can make an arbitrarily chosen U a necessary
condition for a CE.
PQ
Claim. Assume q=1 hqw = 0 and h0w A 0 for w = 1> 2. For any Û 5 R++ >
the central bank (agent 0) can behave as a price taker in such a way that a
necessary condition for a CE is U = Û=

Proof. We have to construct behavior for agent 0. Let the function i (U) deter-
mine the choice of h01  {01 > with h02  {02 given by (2). Let i : R+ $ (Ûh02 > h01 )
be strictly monotone, continuous, and satisfy i (Û) = 0= The bounds on the
range of i are chosen to satisfy {0w  0= Now, suppose, by way of contradiction,
that there is a CE with U 6= Û= By the choice of i> this implies {0w  h0w 6= 0.
But this violates (3), a necessary condition for a CE.

There seems to be nothing more to this proof than the idea that if the
central bank chooses to supply something for which private excess demand is
perfectly inelastic at the quantity 0, then equilibrium requires that the price
be such as to make the central bank willing not to supply it.
Positive endowments for the central bank play an important role. In par-
ticular, if the central bank has no endowments, then its budget constraint,
Central-Bank Interest-Rate Control in a Cashless, Arrow-Debreu Economy 271

(2), and {0w  0 imply {0w  h0w = 0 at all U= How do we interpret positive
PQ
endowments for the central bank simultaneously with q=1 hqw = 0? I am not
sure. Perhaps the positive central-bank endowments are the analogue of the
central bank’s power in a cash economy to print cash at any time.
It may seem odd to claim that the central bank can determine the nominal
interest rate without providing a proof that there exists such an equilibrium.
The known existence results are for versions without endowments of the date-
specific units of account, versions in which the magnitude of the nominal
interest rate does not matter in the above model. Almost certainly, those
existence results can be extended. After all, by choosing s1 and s2 to be
su!ciently large, given unit-of-account endowments can be made arbitrarily
small in real terms. And setting a nominal interest rate does not prevent s1
and s2 from being arbitrarily large.

4 Fiscal and monetary policy

Notice, of course, that control of U as just described leaves one degree of
indeterminacy. It is tempting, therefore, to describe a somewhat more general
policy, one that would eliminate all indeterminacy. Now, I assume that agent
0 has, in addition to endowments of ecus, some of one of the goods. Without
loss of generality, let it have an endowment of good 1 at date 1, denoted $ 011 .
I will not describe where it obtained this endowment. One can think of it as
coming from a bit of lump-sum taxation, which is why I call the policy being
described fiscal and monetary policy. Just as the agent 0 ecu endowments
only had to positive, not large, I only require that $ 011 is positive.
Now I can show that there is a way for agent 0 to behave, behave in
the sense of having an excess demand function, that allows it to determine
both s11 , the price of good rqh  rqh, and U= Before doing that, I have to
again amend the definition of a CE. Once again, agent 0 does not maximize
anything. Hence, we need only amend the feasibility conditions to include
agent 0. I assume that agent 0 consumes at most some of good rqh  rqh= As
a convention, I define its consumption, denoted f011 , to be the amount of good
rqh  rqh it has after trading. Then the feasibility conditions hold at equality.
PQ
Claim. Assume q=1 hqw = 0 and h0w A 0 for w = 1> 2 and $ 011 A 0. For any
(s11 > U ) 5 R2++ > agent 0 can behave as a price taker in such a way that a
necessary condition for a CE is (s11 > U) = (s11 > U )=

Proof. We construct behavior for agent 0. Let the function i (s> U) : R2O+1
+ $
R2O+2 denote the excess demand for agent 0= We suppose that i (s> U) =
0 0 0 0
i (s > U ) if s11 = s11 and U = U . Moreover, i (s> U)  0 except for good
rqhrqh and ecus. Aside from continuity, we let the sign of excess demand for
the vector (i11 > i2 ), excess demand for good rqhrqh and for date-2 ecus be as
indicated in figure 1 with i1 , excess demand for date-1 ecus, determined by the
272 Neil Wallace

(-,+) (-,0) (-,-)

(0,+) (0,-) p11

(p*11,R*)

(+,+) (+,0) (+,-)

Fig. 1. Sign of (i11 > i2 )

agent-0 budget constraint at equality. In addition, let i be continuous. Now,

suppose, by way of contradiction, that there is a CE with (s11 > U) 6= (s11 > U ))=
By the choice of i> this implies {0w  h0w 6= 0 for some w= But this violates (3),
a necessary condition for a CE.

The idea is the same as in the proof of claim 4. The private sector cannot be
either a net supplier or demander of ecus at either date. Therefore, given that
agent 0 is a net supplier or demander at prices that do not satisfy (s11 > U) =
(s11 > U )> satisfaction of this equality is necessary for a CE. By the way, while
the behavior posited in the proof of claim 4 is consistent with a CE in which
ecus at both dates are worthless, that is not the case for the behavior posited
in the proof of claim 5. As should come as no surprise, the willingness of agent
0 to sell goods for ecus prevents ecus from being worthless.
Finally, it is obvious that nothing in these arguments hinges on there being
only two dates. The results apply for any finite number of dates.

5 Concluding remarks
The above model is disarmingly simple. It can be because it does not have to
confront the hard problems of modeling cash economies. Because no one holds
cash from one date to the next, there is no need to explain why people hold
non interest-bearing cash when they seem to have available assets with higher
rates of return. Also, there is no zero lower bound on nominal interest rates.
This bound arises in an economy with cash because people have available a
Central-Bank Interest-Rate Control in a Cashless, Arrow-Debreu Economy 273

technology for converting cash at the current date into cash in the future;
namely, storing it under their mattresses. In a cashless economy, there is no
such opportunity and, therefore, no lower bound on nominal interest rates.
Also, for cashless economies, there is no terminal condition problem concerning
why cash has value at a last date.
But simplicity it not an end in itself. Suppose we were to require not only
that the model be cashless, but that it be a limit of a cash economy, a limit as
something in the economy makes cash disappear. If we take that sensible view,
then we have to face all the hard problems of monetary theory: we have to
decide what cash (monetary) model we like and what limit to take to produce
cashlessness.
In that regard, some recent work on monetary models takes a mechanism-
design point of view and requires that money play a benefical role relative to all
feasible mechanisms. That work concludes that such a monetary model must
have several frictions relative to Arrow-Debreu. For example, in [1], individuals
cannot commit to future actions and there is imperfect public monitoring of
past individual actions in the form of an updating lag of the public record of
individual actions. Remove either and you get a cashless economy. However, if
you only remove the imperfect monitoring by letting the lag get short, which
is what seems to be happening in actual economies, then you get a cashless
economy that is not an Arrow-Debreu model. Thus, it should not be taken
for granted that the cashless limit of interest is an Arrow-Debreu model.

References
1. Kocherlakota, N., and N. Wallace, Optimal allocations with incomplete record-
keeping and no commitment. J. of Economic Theory, 1998, 272-89.