(Lecture Notes in Economics and Mathematical Systems 649) Milan Horniacek (Auth.) - Cooperation and Efficiency in Markets-Springer-Verlag Berlin Heidelberg (2011)

Lecture Notes in Economics
and Mathematical Systems 649

Founding Editors:
M. Beckmann
H.P. Knzi
Managing Editors:
Prof. Dr. G. Fandel
Fachbereich Wirtschaftswissenschaften
Fernuniversitt Hagen
Feithstr. 140/AVZ II, 58084 Hagen, Germany
Prof. Dr. W. Trockel
Institut fr Mathematische Wirtschaftsforschung (IMW)
Universitt Bielefeld
Universittsstr. 25, 33615 Bielefeld, Germany
Editorial Board:
H. Dawid, D. Dimitrow, A. Gerber, C-J. Haake, C. Hofmann, T. Pfeiffer,
R. Slowinski, W.H.M. Zijm
For further volumes:

http://www.springer.com/series/300
Milan Horniacek
Cooperation and Efficiency

in Markets
123
Milan Horniacek
Comenius University
Faculty of Social and Economic Sciences
Institute of Public Policy and Economics
Mlynsk luhy 4
82005 Bratislava
Slovakia
milan.horniacek@fses.uniba.sk
ISSN 0075-8442
ISBN 978-3-642-19762-8 e-ISBN 978-3-642-19763-5
DOI 10.1007/978-3-642-19763-5
Springer Heidelberg Dordrecht London New York
Library of Congress Control Number: 2011929346
c Springer-Verlag Berlin Heidelberg 2011

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. Violations
are liable to prosecution under the German Copyright Law.
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: eStudio Calamar S.L.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Preface
This book is a result of my research in game theory and oligopoly theory. It has
also been influenced by my teaching experiences. Countable infinite repetition of
strategic form noncooperative games with discounting of future payoffs has been
the focus of my research interest for about 20 years. In most of my work, I have
tried to analyze solution concepts that also take into account the deviations by the
coalitions of players (if possible, by all coalitions). The emergence of cooperation
between patient players, who are forced to be competitors in a static setting (or even
in a dynamic setting with a finite horizon), is a fascinating phenomenon from both
the intellectual and human points of view. From the intellectual point of view, the
development of strategy profiles that are immune to the deviations by coalitions is a
challenging and interesting work. From the human point of view, it is nice to know
that when there is a sufficiently long shadow of the future selfish behavior, aimed
at prospering at the expense of another individual, need not be rational. It is a good
news that when all players agree on a cooperation scheme, no one can gain by dis-
torting it unilaterally. If it turns out that no coalition can make any of its members
better off without making another member worse off by distorting a cooperation
scheme, it is a pleasant news. My sensitivity to the human side of results on infi-
nite repeated games increased in last three years during which I was teaching the
course Models of competition and cooperation at the Faculty of Social and Eco-
nomic Sciences, Comenius University, Bratislava. This course includes the issues of
the evolution and stability of cooperation. This motivated me to think more deeply
about the eternal struggle between those who want to preserve mutually beneficial
cooperation and those who want to distort or abuse it to achieve their selfish goals.
My interest in cooperative behavior in infinite repeated games led me to think
about an important economic phenomenon of our world at present trading between
(relatively small) farmers and other producers of foodstuffs (or producers of other
consumer goods) and chain stores. The producers of foodstuffs are weaker partners
in this relationship; chain stores have considerable market power. They take advan-
tage of myopic price taking behavior of their suppliers. The resulting static supply
functions narrow the space of attainable contracts. Obviously, an interaction with
an infinite horizon enlarges this space. This motivated me to study a class of games
that includes the models of infinite repeated interactions between chain stores and
their suppliers as a special case infinite countably repeated games between firms
v
vi Preface
on both sides of a market with discounting of future profits. This book deals with the
analysis of such games that takes into account coordinated actions of all coalitions.
A cooperation between firms on the same side of a market is vulnerable to a
legal challenge on the basis of antitrust laws. In my model, some of the cooperating
firms are on the same and some are on the different sides of a market. This raises a
question as to whether or not the equilibrium behavior in it can be defended against
the charges of the violation of antitrust laws. Therefore, I identify (some) conditions
under which the equilibrium behavior in my model has socially desirable proper-
ties it minimizes the cost on the levels of economic activity repeated along the
equilibrium path and increases consumer welfare.
The research underlying this book was supported by the grant VEGA 1/0138/09.
The purchase of the Scientific WorkPlace software, which made the preparation of
the book much easier, was financed from this grant.
Bratislava, Milan Horniacek

March 2011
Contents
1 Introduction . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 1
1.1 Choice of Solution Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 3
1.2 Choice of Non-collusive Benchmarks . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 5
1.3 Mathematical Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 6
Reference . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 7
2 Model . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 9
2.1 Stage Game .. . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 9
2.2 Repeated Game. . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 15
2.3 Solution Concepts .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 17
Reference . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 19
3 Existence of an SRPE and an SSPE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 21

3.1 Existence of an SRPE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 21
3.2 Existence of an SSPE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 26
4 Efficiency of an SRPE and an SSPE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 53

4.1 Natural Oligopoly .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 54
4.2 Natural Oligopsony . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 58
4.3 Impact on Consumer Welfare.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 61
4.3.1 Comparison with a Monopsonist Choosing
the Traded Quantities on the Demand Side
and Price Taking Behavior on the Supply Side
of the Analyzed Market . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 62
4.3.2 Comparison with Price Taking Behavior
on the Demand Side and a Cournot Oligopoly
on the Supply Side of the Analyzed Market .. . . . . . . . . . . . .. . . . . . . . 70
Reference . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 83
5 Afterword .. . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 85
Reference . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 90
Index . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 91
vii
Chapter 1
Introduction
This book analyzes countable infinite repetition of interaction between firms on both
sides of a market, with discounting of future profits.1 Our first aim is to prove the
sufficient conditions for the existence of equilibria, which take into account the devi-
ations by the coalitions, in it. We explain the reasons for our choice of solution
concepts in Sect. 1.1 of Chap. 1.
Our second and third aims relate to the efficiency properties of the equilibria in
our model. The second aim (covered in Sects. 4.1 and 4.2) is to show that when the
sum of the firms expected average discounted profits is maximized in the equilib-
rium there is no waste of resources in the equilibria. The aggregate production cost
of the equilibrium output of each type of good is minimized along the equilibrium
path. If the firms on the demand side of the analyzed market are retailers (i.e., they
resell the purchased goods in a retail market), then the aggregate selling cost of the
equilibrium vector of sales in the retail market is minimized along the equilibrium
path. The third aim (covered in Sect. 4.3) concerns the efficiency of the equilibria
in our model from the point of view of welfare economics. We identify the condi-
tions under which consumer welfare in each period along the equilibrium path in
our model exceeds consumer welfare in a benchmark non-collusive equilibrium. In
Sect. 1.2 of the Introduction, we explain our choice of benchmark non-collusive
equilibria. In the last section of the Introduction, we explain the mathematical
notations used in this book.
The stage game in our model is a strategic form noncooperative game. It models
the analyzed market with a finite number of producers and a finite number of buy-
ers.2 All firms simultaneously make contract proposals to all firms on the other side
of the analyzed market (i.e., each producer makes a contract proposal to each buyer
and each buyer makes a contract proposal to each producer). A contract proposal
1
Throughout the book we use the well-established term repeated game for what [Mertens
(1989)] refers to as the supergame.
2
Throughout the book, we use the term analyzed market for the market modeled by the stage
game (of the repeated game that is the object of our study). Besides this market, the firms payoffs
are influenced by the market in which the buyers in the analyzed market earn their revenue. The
latter market is either a retail market for the goods traded in the analyzed market or, if buyers are
themselves producers, a market for their products.
M. Horniacek, Cooperation and Efficiency in Markets, Lecture Notes in Economics 1

and Mathematical Systems 649, DOI 10.1007/978-3-642-19763-5_1,
2 1 Introduction
contains a proposed traded quantity and a proposed price.3 A firms decision to not
make a contract proposal to some firm on the other side of the analyzed market is
modeled by the zero contract proposal, containing zero proposed quantity and zero
proposed price. A pair of a producer and a buyer trades if and only if their con-
tract proposals match and differ from the zero contract proposal. The quantities of
the goods purchased by a buyer in the analyzed market determine his/her revenue
in a market in which he/she is a seller. The firms can also withdraw from the ana-
lyzed market. A firm indicates its withdrawal from the analyzed market by making
zero contract proposals to all firms on the other side of that market. In such a case,
its payoff in the stage game is zero. The stage game payoff of a firm that did not
withdraw from the analyzed market equals its profit.
The described choice of the stage game has an underlying economic reason and
an underlying game-theoretic reason. The contracts between the firms have to result
from some process involving proposals and reactions to them.4 Such processes can
be modeled by finite horizon extensive form noncooperative games. It is not possible
to find one such process that would be superior from the point of view of realism or
for some economic reasons. Therefore, instead of choosing one particular process of
concluding contracts, we use a shorthand for all such processes. It does not spec-
ify the order of moves in the conclusion of contracts but merely expresses the basic
fact that two firms can conclude a contract if and only if they agree on it. This short-
hand can be modeled by a strategic form noncooperative game. This is important
from the game-theoretic point of view. Suppose that the process of the conclusion
of contracts was modeled by an extensive form game without simultaneous moves.
Then, some unilateral single period deviations from the profile of equilibrium strate-
gies in the repeated game would have to be punished by the rejection of a contract
proposal. (Otherwise, a deviator can increase his/her payoff in the repeated game
by proposing a contract increasing his/her stage game payoff at the expense of a
trading partner.) Nevertheless, in some cases (for moderate deviations), the rejec-
tion would make both the proposer and the responder worse off than the acceptance
of a deviators contract proposal. Therefore, the grand coalition can weakly Pareto
improve the vector of payoffs in the repeated game5 by abandoning the punishment.
Thus, the equilibrium strategy profile would not be renegotiation-proof.
3
Throughout the book, the term price always implies the unit price.
4
This is certainly not unrealistic. Moreover, traded quantities and prices cannot be given by a
supply function or a demand function in the analyzed market. These functions are derived under
the assumption of price taking behavior on one side of the market. Such assumption would limit
the possibilities for cooperation in the repeated game.
5
If there was at least one producer trading with all buyers along the equilibrium path and the stage
game payoff functions were continuous, the grand coalition can also strictly Pareto improve the
vector of payoffs in the repeated game. The method of doing so would be analogous to the one in
Step 4 in Sect. 4.3.2.
1.1 Choice of Solution Concepts 3
1.1 Choice of Solution Concepts
A subgame perfect equilibrium (first defined in [Selten (1965)] in German and

in [Selten (1975)] in English) is a standard solution concept for countable infi-
nite repeated games with discounting of future stage game payoffs. Nevertheless,
it requires only the immunity of an equilibrium strategy profile to unilateral devia-
tions. The continuation equilibrium payoff vectors need not be (even weakly) Pareto
efficient.6 That is, it does not take into account the possibility that the grand coalition
reaches an agreement to abandon the punishment of a previous unilateral deviation
in order to increase the continuation payoff of each player. Moreover, a subgame
perfect equilibrium does not require immunity to deviations by non-singleton proper
coalitions (i.e., coalitions with at least two members that do not contain all players).
In repeated games between firms it is plausible that players having closer business
links or very intensive effects on one another will coordinate on a deviation.
Thus far (to our best knowledge), there is only one solution concept for infinite
horizon noncooperative games (of which the countable infinite repeated games with
discounting of future stage game payoffs are a special case) that takes into account
the deviations by all coalitions without imposing restrictions either on the strategies
of individual players or on the strategy profiles to which a coalition can deviate.7
This solution is a strong perfect equilibrium developed by [Rubinstein (1980)]. It
is an application of Aumanns concept of a strong Nash equilibrium for strategic
form noncooperative games ([Aumann (1959)])8 to infinite horizon noncooperative
games, and requires that no coalition in no subgame can strictly Pareto improve the
vector of the continuation payoffs of its members by a deviation. Since we want
our main solution concept to be immune to all deviations by all coalitions, we have
to base it on a strong perfect equilibrium. Nevertheless, we strengthen the latter
concept by requiring that no coalition in no subgame can weakly Pareto improve
the vector of the continuation payoffs of its members by a deviation. We call the
resulting solution concept a strict strong perfect equilibrium (henceforth, SSPE
see Definition 2.2). If a deviation only weakly Pareto improves the vector of con-
tinuation payoffs of the members of a deviating coalition, then its members, whose
continuation payoff does not change, are indifferent between participating and not
participating in the deviation. Nevertheless, they can decide to deviate (perhaps, for
some other reasons not captured by the model). Therefore, the replacement of a
strong perfect equilibrium by an SSPE strengthens our results.
6
A continuation equilibrium in a subgame is a restriction of an equilibrium strategy profile to
that subgame. The continuation equilibrium payoff vector is the payoff vector generated by the
continuation equilibrium.
7
[Bernheim et al. (1987)] study only finite horizon games. Various concepts of renegotiation-
proofness for infinite horizon discrete time games with two players (mentioned below in the text)
impose restrictions on the strategy profiles to which the grand coalition can deviate, or, in the case
of [Maskin & Tirole (1988)], restrict attention to Markov strategies.
8
Despite the title of the paper a strong Nash equilibrium is a solution concept for noncooperative
games. See also [Bernheim et al. (1987), p.2-3].
4 1 Introduction
In many infinite horizon games of economic interest a strong perfect equilib-

rium (and hence, also an SSPE) fails to exist. Namely, in countable infinite repeated
games with discounting of future stage game payoffs (unless the stage game has a
strong Nash equilibrium), a strong perfect equilibrium fails to exist when the weak
Pareto efficient frontier of the set of individually rational stage game payoff vec-
tors has no sufficiently large flat portion (i.e., a sufficiently large portion that is a
subset of a hyperplane of the vector space with a dimension equal to the number
of players). In our model, we do not face such a problem. The firms can coordinate
(under the assumptions of Proposition 3.4) to play a stage game strategy profile that
maximizes the sum of their stage game payoffs, and deviations can be punished by
changing only the prices in the analyzed market. Then, the continuation payoffs dur-
ing punishments have the same sum as the continuation payoffs in subgames where
no firm is punished. Thus, they are strictly Pareto efficient. (In this case, the stage
game strategy profiles, which lead to the same vector of traded quantities in the ana-
lyzed market and differ only in prices, generate a flat portion of the weak Pareto
efficient frontier of the set of stage game payoff vectors.)
A strong perfect equilibrium is sometimes criticized for being too strong. Such
criticism is based on the criticism of a strong Nash equilibrium for strategic form
noncooperative games (see [Bernheim et al. (1987), p. 3]). It is claimed that the devi-
ations by coalitions, which are themselves not immune to a deviation by some strict
subcoalition, should not be taken into account. Extending this claim to countable
infinite repeated games, it can be argued that the deviations by coalitions, which
are themselves not immune to a deviation by some strict subcoalition, need not
be punished. Nevertheless, as [Horniacek (1996), Sect. 6] shows, in countable infi-
nite repeated games, this claim is not justified for deviations that last forever. The
point here is that a deviation by a strict subcoalition of a deviating coalition can be
punished by the remaining members of the deviating coalition.
There are solution concepts for infinite horizon discrete time games that take
into account (besides unilateral deviations) only the deviations by the grand coali-
tion. There are various versions of renegotiation-proofness. With the exception of
[Farrell (1993)], all of them are defined for games with two players. All of them
impose restrictions on the strategy profiles to which the grand coalition can renego-
tiate (i.e., deviate). In a weakly renegotiation-proof equilibrium of [Farrell & Maskin
(1989)] (which coincides with the concept of an internally consistent equilibrium of
[Bernheim & Ray (1989)]), only renegotiation to another continuation equilibrium
is allowed. [Farrell (1993)] imposes the same restriction. Moreover, he requires that
all players, who did not deviate, block renegotiation i.e., such players would not
gain by renegotiation from a continuation equilibrium triggered by a deviation back
to the initial continuation equilibrium. In a strong renegotiation-proof equilibrium
of [Farrell & Maskin (1989)] (which coincides with the concept of a strongly consis-
tent equilibrium of [Bernheim & Ray (1989)]), only renegotiation to a continuation
equilibrium of some weakly renegotiation-proof equilibrium is allowed. [Maskin &
Tirole (1988)] restrict attention to Markov strategies.
Besides an SSPE, we also use a concept that we refer to as a strict renegotiation-
proof equilibrium (henceforth, SRPE). This equilibrium is a subgame perfect
1.2 Choice of Non-collusive Benchmarks 5
equilibrium with the property that the grand coalition cannot weakly Pareto improve
the vector of continuation payoffs in any subgame (see Definition 2.1). Thus, unlike
the solution concepts mentioned above, it does not impose any restriction on the
strategy profiles to which the grand coalition can renegotiate. Clearly, the grand
coalition cannot abandon a punishment of any unilateral deviation (because it
would make at least one of its members strictly worse off). Of course, each SSPE
is an SRPE but not vice versa. The results in Chap. 4 hold also for an SRPE. Thus,
a collusive scheme, which can be sustained as an SRPE but not as an SSPE, has
positive properties discussed there.
1.2 Choice of Non-collusive Benchmarks
The economics literature on cartels and tacit collusion assumes that all participating
firms act on the same side of a market. Usually all of them are on the supply side of
a market and produce goods that are identical or close substitutes. This assumption
is made sometimes implicitly also in official documents of antitrust (or, in the
European Union, competition policy) authorities (e.g., [EU (2010b)]). Thus, it is not
surprising that the attitude of these authorities to cartels and collusion is negative.
There are some authors who recognize that in certain circumstances cartels
and collusion can have a positive impact, or at least, criticize the approach of the
antitrust policy to prices.
Some economic historians study the role that cooperation between firms played
in economic development of Germany prior to World War I, where cartels were
legal. [Chandler (1990), part IV, especially Chapter 10] coined the term cooperative
managerial capitalism for the institutional structure of the German economy at that
time.
Spatial economists point out that antitrust laws do not take into account the char-
acteristics of spatial oligopolistic competition. For example, according to [Greenhut
et al. (1987)], the antimerger and price policies of the United States, as evidenced
by the Celler Antimerger Act and the Robinson-Patman Act, are based on space-
less economic theory but are applied to spatial economic behavior [Greenhut
et al. (1987), p. 359]. They also point out that collusion between duopolists in the
noncoincident-location case need not increase price discrimination [Greenhut et al.
(1987), p. 149]. Moreover, some forms of spatial competition generate a higher price
than that charged by a spatial monopolist. The Lschian competition, under which
the firm presumes that its rivals will react identically to any proposed price change,
is an example (see [Greenhut et al. (1987), p. 20]).
We compare collusive schemes sustainable in an SRPE with two non-collusive
benchmarks.9 Since in most real world markets, at least one side is characterized
9
Both benchmarks are non-spatial. Nevertheless, this does not reduce the value of our results
because the comparison of various forms of competitive behavior with collusion is more favorable
for the former in a non-spatial setting.
6 1 Introduction
by imperfect competition (or even monopoly), our first benchmark model involves
monopsony on the demand side and the second one, Cournot oligopoly on the supply
side of the analyzed market. Nevertheless, in both, we assume price taking behavior
on the other side of the market. The standard model of monopsony requires a supply
or an inverse supply function. The model of Cournot oligopoly requires (an) inverse
demand function(s). The former exists only in the case of price taking behavior
on the supply side of the market. The latter exists only in the case of price taking
behavior on the demand side of the market.10
The restriction to the monopsonistic market in the first benchmark model allows
us to obtain results with fewer assumptions than we would need if we assumed
oligopsony on the demand side. Moreover, this model is a plausible approxima-
tion of the relations between a chain-store and its suppliers. (As pointed out in the
Preface, the problems in these relations constituted one of the motivations for the
analysis in this book.) The purchased quantities are the monopsonists choice vari-
ables. This corresponds to the standard treatment of monopoly in the literature, in
which quantity is the choice variable.
If we replaced Cournot oligopoly by Bertrand oligopoly with differentiated
products in the second benchmark model, we would obtain (under additional
assumptions on demand curves) the same qualitative results. Working with Bertrand
oligopoly, in which some groups of firms produce identical products, would lead to
mathematical problems stemming from the discontinuity of the firms demand func-
tions. Moreover, taking into account the result of [Kreps & Scheinkman (1983)], in
the case of homogeneous goods, the Cournot oligopoly model is more general than
the Bertrand oligopoly model.
1.3 Mathematical Notations
Throughout the book, N denotes the set of positive integers and < the set of real
numbers.
For n 2 N; <nC is the nonnegative orthant of <n , i.e., <nC D 0; 1/n . For n 2 N ,
z 2 <n , and v 2 <n z v implies that zj vj for each j 2 f1; : : : ; ng, z > v
implies that z v but z v, and z v implies that zj > vj for each j 2 f1; : : : ; ng.
For n 2 N 4n is the simplex in <n , i.e.,
8 9
< n
X =
4n D z 2 <nC j zj D 1 :
: ;
j D1
For a set Z <n ; conZ denotes its convex hull.
10
In the case of imperfect competition on both sides of a market, we would need a non-collusive
model of bargaining in order to determine the prices and quantities. There is no generally known
model of this type. Therefore, we prefer to work with a monopsony model and a Cournot oligopoly.
References 7
We endow each finite dimensional vector space with the Euclidean topology and
each infinite dimensional Cartesian product of finite dimensional vector spaces with
the product topology. Each subspace of any topological vector space Z inherits
topology from Z.
For a finite set K; # .K/ denotes its cardinality and 2K is the set of all subsets of
K, including the empty set.
For a twice differentiable function f W < ! <; the symbol f 0 .x/ denotes its
first derivative at x and the symbol f 00 .x/ its second derivative at x.
References
Aumann, R.J. Acceptable Points in General Cooperative n-person Games, Annals of Mathemati-
cal Studies Series 40 (1959), 287-324.
Bernheim, B.D., B. Peleg, and M.D. Whinston: Coalition-Proof Nash equilibria I. Concepts,
Journal of Economic Theory 42 (1987), 1-12.
Bernheim, B.D. and D. Ray: Collective Dynamic Consistency in Repeated Games, Games and
Economic Behavior 1 (1989), 295-326.
Chandler, A.D. Jr.: Scale and Scope. The Dynamics of Industrial Capitalism. Cambridge, MA: The
Belknap Press of the Harvard University Press, 1990.
Commission Action Against Cartels Questions and Answers. MEMO 10/290, 30 June 2010.
http://europa.eu/rapid/pressReleasesAction.do?reference=MEMO/10/290&format=HTML&
aged=0&language=EN&guiLanguage=en
Farrell, J.: Renegotiation in Repeated Oligopoly Games. Mimeo. University of California at
Berkeley, 1993.
Farrell, J. and E. Maskin: Renegotiation in Repeated Games, Games and Economic Behavior 1
(1989), 327-360.
Greenhut, M.L., G. Norman, and C. Hung: The Economics of Imperfect Competition. Cambridge:
Cambridge University Press, 1987.
Horniacek, M.: The Approximation of a Strong Perfect Equilibrium in a Discounted Supergame,
Journal of Mathematical Economics 25 (1996), 85-107.
Kreps, D.M. and J.A. Scheinkman: Quantity Precommitment and Bertrand Competition Yield
Cournot Outcomes, Bell Journal of Economics 14 (1983), 326-337.
Maskin, E. and J. Tirole: A Theory of Dynamic Oligopoly II: Price Competition, Kinked Demand
Curves, and Edgeworth Cycles, Econometrica 56 (1988), 571-599.
Mertens, J.-F.: Supergames, in The New Palgrave: Game Theory, ed. by J. Eatwell, M. Milgate,
and P. Newman. New York: W.W. Norton, 1989, 238-241.
Rubinstein, A: : Strong Perfect Equilibrium in Supergames, International Journal of Game
Theory 9 (1980), 1-12.
Selten, R.: Spieltheoretische Behandlung eines Oligopolmodells mit Nachfragetrgheit,
Zeitschrift fr die Gesamte Staatswissenschaft 121 (1965), 301-324 and 667-689.
Selten, R.: Reexamination of the Perfectness Concept for Equilibrium Points in Extensive
Games, International Journal of Game Theory 4 (1975), 25-55.
Chapter 2
Model
In this chapter we describe our model of a market with strategically behaving agents
on both sides. First, we characterize the stage game between the firms in the model.
Then, we proceed to the formulation of the countable infinite repeated game with
discounting of payoffs. In Sect. 2.3 we define the solution concepts that we apply to
the repeated game: SRPE and SSPE.
2.1 Stage Game
We denote the stage game by G. The stage game is a strategic form noncooperative
game.
There is a nonempty finite set J of producers and a nonempty finite set I of
buyers. We have J D f1; : : : ; # .J /g and I D f# .J / C 1; : : : ; # .J / C # .I /g.
Thus, J [ I is the set of players in G. (We use the terms player(s) and firm(s)
interchangeably.) Buyers buy goods from producers. They either use the purchased
goods to produce new good(s) (which they sell in the market(s) where customers
are price takers) or sell them in the retail market to final consumers. Each producer
produces one type of good, and can sell to any number of buyers in I: Therefore,
we use the symbol J also for the set of goods in the model. We do not exclude the
possibility that some or even all goods in J are identical.
A coalition is a nonempty subset of J [ I . Thus, the set of all coalitions equals
2J [I n fg. A proper coalition is a nonempty strict subset of J [ I . In subscripts
and superscripts we write C instead of .J [ I / nC for each coalition C J [ I ,
and k instead of .J [ I / n fkg and k instead of fkg for each k 2 J [ I .
For each j 2 J; there exists j > 0; which is the upper bound on the output of
good j . These
upper
bounds can stem, for Q example, from capacity constraints. We
let Yj D 0; j for each j 2 J and Y D j 2J Yj . Thus, Yj is the set of feasible
outputs of producer j .
Producer j 2 J has cost function cj W Yj ! <C .
Assumption 2.1. For each j 2 J; cj is (i) continuous, (ii) strictly increasing, and
(iii) cj .0/ > 0.

10 2 Model
Let
( )
X
#.I /
X D x D .xi /i 2I D xj i j 2J 2Y j xj i 2 Yj 8j 2 J : (2.1)
i 2I
i 2I
Here, xj i is the quantity of good j 2 J purchased by buyer i 2 I . We will

call an element of X a vector of traded quantities. Of course, in each vector of
traded quantities, the sum of the purchases of all buyers from each producer j 2 J
has to be a feasible output of j . Clearly, X is a nonempty and compact subset of
#.J /#.I /
<C . (It contains, for example, the zero vector in R#.J /#.I / . It is a subset of
compact set Y #.I / : It is closed because it is defined by linear constraints in the finite
dimensional space.) We define function W X ! Y by
!
X
.x/ D xj i : (2.2)
i 2I j 2J
That is, .x/ is the output vector, in which each producers output equals the
sum of his/her deliveries to buyers specified by x.
We set Xi D Y for each i 2 I . It is the set of feasible vectors of the purchases of
buyer i from the producers in J . The cost function of buyer i 2 I , ci W Xi ! <C ,
assigns to each vector xi 2 Xi of the quantities of goods purchased from the produc-
ers in J his/her cost associated with their processing or cost associated with their
sale in the retail market. These costs include, for example, transport cost, wages,
depreciation allowance, expenditure on the maintenance of buildings, expenditure
on the maintenance of machines (especially if the purchased goods are processed
into new goods), and storage and handling cost (especially if the purchased goods
are sold in the retail market). These do not include the expenditure on the purchase
of goods from producers. Thus, the cost expressed by function ci is the expenditures
on the inputs used by firm i 2 I that are complementary to the goods purchased
from the producers in J .
Assumption 2.2. For each i 2 I , (i) ci is continuous and (ii) ci .0/ > 0.
For each i 2 I , Ui W X ! <C is buyer i s revenue function. It assigns to each

x 2 X , buyer i s revenue from selling the output produced from the inputs given
by vector xi , or (if he/she is a retailer) his/her revenue from selling the quantities of
the purchased goods given by xi in the retail market.P For example, if all buyers pro-
duce n 2 N goods, Ui has the form Ui .x/ D nmD1 Pmi ..fk .xk //k2I / fmi .xi /,
where fk W Xk ! <nC is buyer ks (vector) production function1 (with fk .xk / D
n#.I /
.fmk .xk //m2f1;:::;ng ) and for each m 2 f1; : : : ; ng, Pmi W <C ! <C is the
1
Only the quantities of the inputs purchased from the producers in J are the arguments of fk . We
assume that the other inputs are fixed unless buyer k decides to leave the analyzed market.
2.1 Stage Game 11
inverse demand function for the m-th P output of buyer i . If all buyers are retail-
ers, then Ui has the form Ui .x/ D j 2J P e j . .x// xj i , where for each j 2 J ,
e j W Y ! <C is the inverse demand function for the sale of good j in the retail
P
market. (We can view the latter situation as a special case of the former when we set
n D # .J /, fk .xk / D xk for each k 2 I , and Pj i .x/ D Pe j . .x// for each j 2 J
and every i 2 I .)
Assumption 2.3. For each i 2 I , Ui is (i) continuous and (ii) xi D 0 implies

Ui .x/ D 0.
The following assumption ensures that the zero vector cannot solve the maxi-
mization problem in (2.3) given below.
Assumption 2.4. There exists x C 2 X such that

X X
Ui x C ci xiC > cj j x C :
i 2I WxiC >0 j 2J Wj .x C />0
Let
8 9
<X X X =
X max D arg max Ui .x/ cj j .x/ ci .xi / j x 2 X
: ;
i 2I j 2J Wj .x/>0 i 2I Wxi >0
(2.3)
Thus, each x max 2 X max is a vector of traded quantities that maximizes the sur-
plus from the trade in the analyzed market. That is, it maximizes the difference
between the sum of the buyers revenue (from the sale of goods produced from
the inputs purchased in the analyzed market or from the sale of purchased goods
in the retail market) and the sum of the production costs of active producers (i.e.,
assuming that the producers who did not withdraw from the analyzed market have
positive outputs) and the costs of active buyers (i.e., assuming that the buyers who
did not withdraw from the analyzed market purchase a positive amount of at least
one good). Since the sum of the producers revenue in the analyzed market equals
the sum of buyers expenditure in it, x max also maximizes the sum of the profits in
the analyzed market.
X max is nonempty despite the discontinuity of the objective function in the max-
imization program in (2.3) caused by the fact that the firms who withdraw from the
analyzed market do not incur any fixed cost. We can solve the latter maximization
problem in two steps as follows. In the first step, we solve for each D 2 2J [I with
D \ J and D \ I , the maximization problem
8 P P 9
< Ui .x/ ci .xi / cj j .x/ j =
D D max i 2I \D j 2J \D : (2.4)
: x 2 X; x D 08 .j; i / .D \ J / .D \ I / ;
ji
12 2 Model
We denote the set of its solutions by X .D/ max . Here, D is the maximal surplus from
the trade in the analyzed market among the firms in D (i.e., when all firms outside
D withdraw from the analyzed market). In this auxiliary problem, we assume that
no firm in D withdraws from the analyzed market (i.e., each firm in D incurs a fixed
cost). Since X is a nonempty and compact set and the objective function in the max-
imization problem in (2.4) is continuous, X .D/ max is nonempty and compact. The
following statement holds for each x .D/
.D/ 2 X
max .D/ max
: if there exists a nonempty
C D such that for each j 2 J , j x > 0 if and only if j 2 C , and for each
.D/
i 2 I , xi > 0 if and only if i 2 C , then
X h i X
C Ui x .D/ ci xi.D/ cj j x .D/ > D ;
i 2C \I j 2C \J
where the second inequality follows from the fact that each firm has a positive
fixed cost (see part (iii) of Assumption
2.1 and part (ii) of Assumption 2.2).
There-
fore, for each F 2 arg max D j D 2 2J [I ; D \ J ; D \ I , we have
X .F / max X max . As such, in the second step, we set
.max/ X .F / max j :
X D
F 2 arg max D j D 2 2J [I ; D \ J ; D \ I
(2.5)
Since X .F / max is nonempty and compact for each

n o
F 2 arg max D j D 2 2J [I ; D \ J ; D \ I
and the set of such coalitions F is nonempty and finite, X max is nonempty and
compact. We use to denote the maximal value of the objective function in the

maximization problem in (2.3). Assumption 2.4 implies that > 0. For each
x 2 X , we let EJ .x/ D j 2 J j j .x/ > 0 , EI .x/ D fi 2 I j xi > 0g, and
E .x/ D EJ .x/ [ EI .x/.
It is plausible to assume that the prices in the analyzed market cannot be arbitrar-
ily high. We denote the upper bound on a price of good j 2 J by pjmax .2 We assume
that pjmax > 0 for each j 2 J and
pjmax xjmax
i max fUi .x/ j x 2 X g ; 8x
max
2 X max ; 8j 2 EJ .x max / ;
n o
8i 2 k 2 I j xjk
max
>0 : (2.6)
2
Of course, from the modeling point of view, we need the upper bounds on the prices in the
analyzed market in order to ensure that the stage game payoffs are bounded and the repeated game
(described in Sect. 2.2) is continuous at infinity.
2.1 Stage Game 13
Thus, (taking into account part (ii) of Assumption 2.2) for each x max 2 X max ,
every j 2 EJ .x max /, and each i 2 I with xjmaxi > 0, no purchase, in which the
buyer i buys the quantity of good j equal to xjmax i at price pj , allows him/her
max
Q h i#.I /
to earn a nonnegative profit. We let P D j 2J 0; pj
max
. P is the set of
feasible vectors
of the
prices in the analyzed market. Each of these vectors has the
form p D pj i j 2J .
i 2I
Now we describe the sets of pure strategies and payoff functions in stage game
G D hJ [ I; .Ak /k2J [I ; .gk /k2J [I i. We have
( )
#.I / X
Aj D pj i ; qj i i 2I 2 0; pjmax Yj j qj i 2 Yj ; 8j 2 J (2.7)
i 2I
and #.J /
Ai D 0; pjmax Yj ; 8i 2 I: (2.8)
A pure strategy of producer j 2 J in G is a collection of contracts (one for each
buyer) that he/she proposes to the buyers. Each contract proposal specifies a pro-
posed price and a proposed quantity.3 Of course, the sum of the proposed quantities
cannot exceed a producers capacity. There is no need to treat a producers decision
not to propose a contract to some buyer as a special case. We identify such decision
with the contract proposal .0; 0/. For each j 2 J , we identify the producers pure
strategy .0; 0/#.I / with his/her decision to withdraw from the analyzed market. For
such a decision, the producer does not incur a fixed cost and his/her payoff in G
equals zero.4
A pure strategy of buyer i 2 I in G is a collection of contracts (one for each
that he/she proposes to the producers. Thus, each ai 2 Ai has the form
producer)
ai D pij ; qij j 2J . Again, we identify the contract proposal .0; 0/ made to j 2 J
with the decision not to trade with producer j . Further, for each i 2 I , we identify
the buyers pure strategy .0; 0/#.J / with his/her decision to withdraw from the ana-
lyzed market. For such a decision, the buyer does not incur a fixed cost and his/her
payoff in G equals Qzero.
We let A D k2J [I Ak . For each a 2 A and every .j; i / 2 J I trade
between producer j and buyer i takes place if and only if j s contract proposal
to i differs
from .0; 0/ and coincides
with i s contract proposal to j , i.e., if and
only if pj i ; qj i D pij ; qij .0; 0/. For every a 2 A, x .a/ is the vector of
3
The producers can sell their product to different buyers at different prices. This enables us to
construct a punishment for a deviation by a proper coalition of buyers, which does not harm the
buyers who did not deviate, in the repeated game.
4
We assume here that a firm can leave the analyzed market in one period. In the repeated game
(described in the following section), we assume that a firm can enter the analyzed market in one
period and the entry requires only paying the fixed cost. Our qualitative results hold if the exit from
and entry into the analyzed market took more than one period and entry required the incurring of a
sunk cost (exceeding the single period fixed cost).
14 2 Model
traded quantities
generated
by strategy profile a. We define
it by xj i.a/ D qj i if
pj i ; qj i D pij ; qij and by xj i .a/ D 0 if pj i ; qj i pij ; qij . Further, for
each a 2 A, y .a/ D .x .a// is the output vector generated by strategy profile a.
Similarly, for each a 2 A, ep .a/ 2 P is the vector of prices at which the quantities
given by x .a/ are traded. We use the convention that a zero price is assigned to each
zero traded quantity. Thus, e
pj i .a/ D pj i D pij if xj i .a/ > 0 and e p j i .a/ D 0 if
xj i .a/ D 0.
For k 2 J [ I , player ks payoff function in G, gk W A ! <, is defined by
X
8j 2 J W gj .a/ D p j i .a/ xj i .a/ cj yj .a/ ;
e
i 2I

if 9i 2 I with pj i ; qj i .0; 0/ ; (2.9)

8j 2 J W gj .a/ D 0; if pj i ; qj i D .0; 0/ 8i 2 I; (2.10)
X
8i 2 I W gi .a/ D Ui .x .a// e
p j i .a/ xj i .a/ ci .xi .a// ;
j 2J

if 9j 2 J with pij ; qij .0; 0/ ; (2.11)

8i 2 I W gi .a/ D 0; if pij ; qij D .0; 0/ 8j 2 J: (2.12)
We define function g W A ! <#.J [I / by g .a/ D .gk .a//k2J [I and let
n o
V D v 2 <#.J [I / j 9a 2 A such that g .a/ D v : (2.13)
Clearly, taking into account (2.9)(2.12), we have

8 P 9

v 2 <#.J [I / j vj D pj i xj i cj j .x/ 8j 2 J nD; > >

>
>

i 2I
P >
>
[ < vi D Ui .x/ ci .xi / pj i xj i 8i 2 I nD; vk D 08k 2 D; =
V D j 2J :

>
>
D22J [I
x 2 X; xj i D 08 .j;h i / 2 .Ji \ D/ I [ J .I \ D/ ; >
>
>
>
: pj i 2 0; pjmax 8 .j; i / 2 J I ;
(2.14)
That is, V is the set of payoff vectors in G that can result from the pure strategy
profiles. It follows from (2.13) that V is nonempty. Clearly, V is a compact subset
of <#.J [I / . (For each D 2 2J [I the corresponding set in the union in (2.14) is
compact because X is a compact set, function cj , j 2 J is continuous, and functions
Ui and ci , i 2 I are continuous.) Therefore, V has a strict Pareto efficient frontier
that we denote by } .V /.
Since V contains payoff vectors generated by the pure strategy profiles in G, it
need not be convex. Thus, a vector in } .V / can be weakly (or even strictly) Pareto
dominated by a convex combination of other vectors in } .V /. We let V C D conV.
2.2 Repeated Game 15
In order to enable the firms to achieve payoff vectors in V C nV , we also allow objec-
tively correlated strategies (see [Aumann (1974)]; henceforth, we use only the term
correlated strategies) in G. We assume that all firms can observe the signals of
a public randomizing device generating uniformly distributed signals from interval
0; 1. A correlated strategy of firm k 2 J [ I is the mapping k W 0; 1 ! Ak .
It assigns to each signal ! 2 0; 1, firm ks pure strategy in G. We Q denote the set
of correlated
Q strategies of firm k 2 J [ I by k and let D k2J [I k and
C D k2C k for each C 2 2J [I n fg. Of course, a pure strategy is a special
case of a correlated strategy. (That is, ak 2 Ak is from the point of view of the
outcome of G identical to k 2 k as defined by k .!/ D ak for each ! 2 0; 1.)
Taking into account the Carathodory theorem (see [Hildenbrand (1974), p. 37]), in
order to obtain any vector in V C , it is enough to use a correlated strategy profile that
uses at most # .J [ I / C 1 pure strategy profiles with a positive probability.
With a slight abuse of the notation, we will use the symbol gk , k 2 J [ I also
for the payoff functions in G defined on and the symbol g for the function that
assigns to each 2 , the vector of expected payoffs in G.
C
Since V is nonempty and compact, V is also nonempty and compact.
We use
} V C to denote its strict Pareto efficient frontier and let } V C D } V C \
#.J [I /
<C . Thus, } V C is the set of individually rational strictly Pareto efficient
payoff vectors in G. (Recall that each firm can guarantee itself a zero payoff by
withdrawing from the analyzed market. The firms on the other side of the analyzed
market can as a group prevent it from achieving a positive payoff when they
refuse to trade with it, i.e., when each of them proposes the contract

C.0;
0/ to it.
Thus, each firms minimax payoff in G equals zero.) We use } V to denote
the subset of } V C with the following property: if v 2 } V C and vk D 0,
then there exists 2 such that v D g ./ and k assigns to each ! 2 0; 1, firm ks
pure strategy in G leading to its withdrawal from the analyzed market (i.e., k .!/ D
.0; 0/#.I / for each ! 2 0; 1 if k 2 J , and C
k.!/ D .0; 0/#.J / for each ! 2 0; 1

if k 2 I ). Thus, the payoff vectors in } V , besides being individually rational
and strictly Pareto efficient, are also strictly individually rational for firms that do
not withdraw from the analyzed market.
We assume that when a firm contemplates (either unilaterally or as a member
of a coalition of deviating firms), a deviation from its correlated strategy in G; it
does so before the signal from the public randomizing device is observed. Thus, the
deviation strategy of a firm has to assign its pure strategy to each ! 2 0; 1.
2.2 Repeated Game
The repeated game is a countable infinite repetition of G with discounting of future

profits (without discounting of current profit). That is, G is played in periods num-
bered by positive integers. All firms use the common discount factor 2 .0; 1/.
We denote the repeated game with the discount factor by ./ and its game form
by . The actions in are observable. That is, at the end of each period t 2 N ,
16 2 Model
every firm k 2 J [ I observes for every firm r 2 .J [ I / n fkg rs pure

strategy in G in period t. We denote the set of histories in by H and the set of
histories leading to period t 2 N by H .t / . Each element of H .t / contains decisions
made by the players and signals of the public randomizingS device prior to period t.
H .1/ contains only the empty history. We set Hf D t 2N H .t / . Thus, Hf is the
set of nonterminal (i.e., finite) histories. We use H1 to denote the set of termi-
infinite)histories. Hence, H D Hf [ H1 . Each h 2 H1 has the form
nal (i.e.,
h D a.t / ; ! .t / t 2N , where a.t / 2 A and ! .t / 2 0; 1 for each t 2 N . For each

h 2 H1 , x .h/ D x a.t / t 2N is the sequence of vectors of traded quantities gen-

erated by terminal history h, and y .h/ D y a.t / t 2N is the sequence of output
vectors generated by h.
The behavioral strategy of firm k 2 J [ I in is the function sk W Hf ! k . It
assigns to each nonterminal history h 2 Hf , one of ks correlated strategies in G.
We denote Qthe set of the behavioralQ strategies of firm k 2 J [ I in by Sk and
let S D k2J [I Sk and SC D k2C Sk for each C 2 2J [I n fg. For s 2 S
and C 2 2J [I n fg we let sC D .sk /k2C . We use the term strategy profile for
s .C / 2 SC .5 For every k 2 J [ I , the terms strategy of k and strategy
profile of
.C /
fkg are equivalent. For C 2 2J [I n fg with # .C / 2, s .C / D sk 2 SC ,
k2C
.C / .C /
and D 2 2C n fg, we let sD D sk .
k2D
We express the firms payoffs in ./ (computed before they implement their
behavioral strategies and a signal of the public randomizing device in the first period
is observed) as their expected average discounted profits (i.e., their expected average
discounted stage game payoffs). For each k 2 J [ I , function k W S !n< isofirm
ks payoff function in ./.6 That is, when s 2 S generates a sequence v.t k
/
t 2N
of expected stage game payoffs of firm k 2 J [ I , we have
X
k .s/ D .1 / t 1 v.t
k
/
: (2.15)
t 2N
We define function W S ! <#.J [I / by .s/ D .k .s//k2J [I . With this for-

mulation of the payoff functions, the set of the vectors of the firms payoffs in ./
equals V C .
For each h 2 H .t / with t 2 N n f1g, we use h to denote the subhistory of h
leading to period t 1. The subhistory contains all information contained in h except
for the firms decisions and the signal of the public randomizing device in period
5
We use the symbol sC for a profile of the behavioral strategies of the members of a coalition C
determined by a previously mentioned profile of the behavioral strategies (of all firms) s 2 S and
the symbol s .C / for a profile of the behavioral strategies of the members of a coalition C that is not
determined by any previously mentioned s 2 S.
6
Of course, the functional values of k depend on . Nevertheless, in order to avoid unnecessary
notational complication, we use the symbol k instead of k; .
2.3 Solution Concepts 17
t 1 t 2
t 1. (That is, if t > 2 and h D a.n/ ; ! .n/ nD1 , then h D a.n/ ; ! .n/ nD1 . If
h 2 H .2/ , then h D .)
For each nonterminal history h 2 Hf , .h/ ./ is the subgame of ./ follow-
ing h. .h/ is its game form. Since ./ is a game with observable actions, each of
its subgames is a proper subgame.
For any set B defined for ./ and any h 2 Hf , the symbol B.h/ stands for the
restriction of B to subgame .h/ ./ (e.g., Hf .h/ is the set of nonterminal histories
in .h/ .//. Similarly, for any function f defined for ./ and any h 2 Hf , the
symbol f.h/ stands for the restriction of f to subgame .h/ ./ (e.g., k.h/ is firm
ks payoff function in .h/ ./). For h 2 Hf and h0 2 Hf .h/ , the subgame of
(the subgame) .h/ ./ following history h0 in .h/ ./ is identical to the subgame
.h;h0 / ./ of ./. We use the latter symbol to denote this. For h 2 Hf , C 2
.C /
2J [I n fg, s .C / 2 SC .h/ , and h0 2 Hf .h/ , s.h 0 / is the restriction of strategy profile
s .C / of coalition C in subgame .h/ ./ to subgame .h;h0/ ./.

We assume that when a firm contemplates (either unilaterally or as a member
of a coalition of deviating firms) a deviation from its behavioral strategy in a sub-
game .h/ ./ with h 2 H .t / ; it does so before the signal of the public randomizing
device for period t is observed. This assumption corresponds to the excluding of
a signal of the public randomizing device in period t from h 2 H .t / and defining
a firms behavioral strategy as a mapping from the set of nonterminal histories to
the set of its stage game correlated strategies (instead of as a mapping from the
Cartesian product of the set of nonterminal histories with 0; 1 to the set of its stage
game pure strategies). This allows the grand coalition to use any profile of stage
game correlated strategies and hence, (if V is not convex) to achieve a stage game
payoff vector belonging to V C nV in the first period of a subgame. (If the firms
would have contemplated a deviation by the grand coalition only after observing
the signal of the public randomizing device in the first period of a subgame, they
would not have been able to use a profile of stage game correlated strategies with
non-singleton support in the first period of a subgame.) The ability of the grand
coalition to weakly Pareto improve the vector of current period stage game payoffs
by a deviation contemplated after the current period signal of the public randomiz-
ing device is observed implies its ability to do so also by a deviation contemplated
before the current period signal of the public randomizing device is observed, but
not vice versa. An analogous comment holds for the deviations by any non-singleton
coalition.
2.3 Solution Concepts
As already stated in the Introduction, the SRPE and the SSPE are the solutions
concepts that we apply to ./.
Definition 2.1. A strict renegotiation-proof equilibrium of ./ is a strategy profile

s 2 S with the following properties.
18 2 Model
(i) There do not exist a nonterminal history h 2 Hf , a firm k 2 J [ I , and its

strategy sk 2 Sk.h/ such that

k.h/ sk ; sk.h/ > k.h/ s.h/ : (2.16)
(ii) There do not exist a nonterminal history h 2 Hf and a strategy profile s 2 S.h/
such that

k.h/ .s/ k.h/ s.h/ ; 8k 2 J [ I (2.17)
and

9 k 2 J [ I with k.h/ .s/ > k.h/ s.h/ : (2.18)
It follows from part (i) of Definition 2.1 that an SRPE is a subgame perfect equi-
librium. Part (ii) implies that all continuation equilibrium payoff vectors7 are strictly
Pareto efficient.
Definition 2.2. A strict strong perfect equilibrium of ./ is a strategy profile s 2

S with the following properties.
(i) There do not exist a nonterminal history h 2 Hf , a coalition C 2
2J [I n f; J [ I g, and its strategy profile s .C / 2 SC .h/ such that

k.h/ s .C / ; sC

.h/

k.h/ s.h/ ; 8k 2 C (2.19)
and
9 k 2 C with k.h/ s .C / ; sC

.h/

> k.h/ s.h/ : (2.20)
(ii) There do not exist a nonterminal history h 2 Hf and a strategy profile s 2 S.h/
such that

k.h/ .s/ k.h/ s.h/ ; 8k 2 J [ I (2.21)
and

9 k 2 J [ I with k.h/ .s/ > k.h/ s.h/ : (2.22)
Thus, a strategy profile s is an SSPE if no coalition in no subgame can increase

the expected average discounted profit of at least one of its members without
decreasing the expected average discounted profit of any other member by a devia-
tion from the prescriptions of s . That is, no coalition in no subgame has a deviation
leading to a vector of expected average discounted profits of its members that weakly
Pareto dominates the vector of their expected average discounted profits generated
by s . Part (ii) of Definition 2.2 implies that all continuation equilibrium payoff
7
Let s 2 S be an equilibrium strategy profile and let h 2 Hf . Then, the continuation
equilibrium

in subgame .h/ ./ is s.h/ , and the continuation equilibrium payoff vector is .h/ s.h/ .
References 19
vectors in an SSPE are strictly Pareto efficient. Part (i) also holds for singleton coali-
tions. Therefore, it implies that each SSPE is a subgame perfect equilibrium. Thus,
each SSPE is an SRPE.
Definition 2.2 does not exclude the possibility that a deviation by a coalition from
its continuation equilibrium strategy profile in some subgame leaves the payoffs of
all its members unchanged. Thus, taking into account the meaning of the term strict
Nash equilibrium (which is a Nash equilibrium where each unilateral deviation
by any player decreases his/her payoff), we should use the term semi-strict strong
perfect equilibrium instead of SSPE. Nevertheless, in order to avoid terminological
complexities, we use the latter term.
References
Aumann, R.J.: Subjectivity and Correlation in Randomized Strategies, Journal of Mathematical

Economics 1 (1974), 67-96.
Hildenbrand, W.: Core and Equilibria of a Large Economy. Princeton, N.J.: Princeton University
Press, 1974.
Chapter 3
Existence of an SRPE and an SSPE
In this chapter, we prove the sufficient conditions for the existence of an SRPE and
an SSPE in the model described in Chap. 2. We let Assumptions 2.12.4 hold.
3.1 Existence of an SRPE
Let
vmax
i D max fUi .x/ ci .xi / j x 2 X g ; 8i 2 I; (3.1)

vmax
j D max pjmax yj cj yj j yj 2 Yj ; 8j 2 J; (3.2)
8 9
< X =
vmin D U .x/ c .x / p max
x j x 2 X ; 8i 2 I; (3.3)
i
: i i i j ji
;
j 2J

vmin
j D cj j ; 8j 2 J: (3.4)
Here, (taking into account that pjmax is the upper bound on the price of good j 2 J )
vmax
k
(vmin
k
) is the highest (lowest) stage game profit of firm k 2 J [ I .
P
Proposition 3.1. For each v 2 } V C with k2J [I vk D there exists 2
.0; 1/ such that for every 2 .; 1/, ./ has an SRPE with equilibrium payoff
vector equal to v :
P
Proof. A. Preliminaries. Take (arbitrary) v 2 } V C with k2J [I vk D .

Using the definition of } V C (see the second-last paragraph in Sect. 2.1), v
is a strictly Pareto efficient payoff vector in ./. Further, there exists 2 ,
n 2 f1; : : : ; # .J [ I / C 1g, a.r/ 2 A, r 2 f1; : : : ; ng, and D .r /r2f1;:::;ng 2
4n such that 0, g . / D v , and for each r 2 f1; : : : ; ng, leads to
the occurrence of a.r/ with probability r .r/
. Moreover,
for each j 2 J with
vj D 0, we have xj i a.r/ D 0 and e pj i a D 0 for every i 2 I and each
r 2 f1; : : : ; ng. Similarly, for each i 2 I with vi D 0, we have xj i .a.r/ / D 0

22 3 Existence of an SRPE and an SSPE
P
p j i .a.r/ / D 0 for every j 2 J and each r 2 f1; : : : ; ng. As k2J [I vk D ,
and e
.r/
x a 2 X max for each r 2 f1; :::ng. Since the firms have a positive fixed cost, for
each r 2 f1; : : : ; ng, a.r/ prescribes the withdrawal from the analyzed
.r/market
for
every j 2 J with yj a.r/ D 0 as well as for every i 2
I with x
i a D 0. We
can assume without loss of generality that pj.r/ .r/
i ; qj i D pij.r/ ; qij.r/ D .0; 0/

for each .j; i / 2 J I and every r 2 f1; : : : ; ng with xj i a.r/ D 0. Let
S S
EJ D nrD1 EJ x a.r/ , EI D nrD1 EI x a.r/ , and E D EJ [ EI .
Thus, vk > 0 for each k 2 E . From this and condition (2.6), it follows that for each

i 2 EI , there exists r 2 f1; : : : ; ng such that xi a.r/ > 0 and e p j i a.r/ < pjmax
for each j 2 J . Further, for each j 2 Ej , there exist r 2 f1; : : : ; ng and i 2 I such

that xj i a.r/ > 0 and e p j i a.r/ > cj yj a.r/ =yj a.r/ .

We set .r/ D ! 2 0; 1 j .!/ D a.r/ for each r 2 f1; : : : ; ng, and let
function 1 W 0; 1 ! f1; : : : ; ng assign to every ! 2 0; 1, r 2 f1; : : : ; ng that
satisfies ! 2 .r/.P P
Let j D h nrD1 r i 2I e p j i a.r/ xj i a.r/ i for each j 2 EJ and
P P
i D nrD1 r j 2J e p j i a.r/ pjmax xj i a.r/ for each i 2 EI . Clearly,
k < 0 for each k 2 E . For k 2 E and T 2 N , consider inequality

.1 / vmax
k vk C 1 T k 0: (3.5)
For the limit case D 1, weak inequality (3.5) holds as equality. Differentiating
itsleft-hand side with respect to and evaluating the derivative at D 1, we get
1

vmax
k v k
T k . This expression is positive if and only if T > vk vmax
k k .
We set
1
T D max min n 2 N j n > vk vmax
k k j k 2 E : (3.6)
Then, for each k 2 E , there exists k1 2 .0; 1/ such that for every 2 k1 ; 1/,
(3.5) holds for firm k. We set 1 D max fk1 j k 2 E g. Further, for each k 2 E ,
there exists k2 2 .0; 1/ such that for every 2 .k2 ; 1/,

1 T vk C k C T vk > 0: (3.7)
(For the limit case D 1, (3.7) holds. Its left-hand side is continuous in .) We set
2 D max fk2 j k 2 E g and D max f1 ; 2 g.

For each D 2 2E satisfying the condition

i 2 I \ D & j 2 J \ D H) xj i a.r/ D 08r 2 f1; : : : ; ng (3.8)
and every r 2 f1; : : : ; ng, let a.r/ j D 2 A differ from a.r/ in only two
p j i a.r/ j D D 0 for each j 2 J \ D and every i 2 EI with
respects: e
3.1 Existence of an SRPE 23

xj i a.r/ > 0, and e p j i a.r/ j D D pjmax for each i 2 I \ D and every j 2 EI
.r/
with xj i a > 0. (The other prices are equal to those prescribed by a.r/ and
.r/
x a j D D x a.r/ .) Note that a.r/ j D a.r/ for each r 2 f1; : : : ; ng.

For each D 2 2E satisfying the condition (3.8), we define j D 2 by
j D .!/ D a j D, where r D 1 .!/, for every ! 2 0; 1.
.r/
B. Description of strategy profile s . For each h 2 Hf n fg let L .h/ be the set

of firms in E that according to h, deviated from the prescriptions of s in the
.t /
last period recordedn in h. Thus, for each t 2 N n f1g o and every h 2 H , we
.t 1/

have L .h/ D k 2 E j ak sk .h / ! .t 1/ . (Here, sk .h / is firm ks
stage game correlated strategy prescribed by sk after the subhistory h of h, and

sk .h / ! .t 1/ is ks stage game pure strategy prescribed by sk .h / for signal
! .t 1/ of the public randomizing device.) In the description of s , we use for each
k 2 E , function k W Hf ! f0; 1; : : : ; T g ; which we define recursively below. For
every t 2 N , each h 2 H .t / , and every k 2 E , k .h/ is the number of remaining
periods, including period t, for which the punishment of firm k (that started before
period t or will start in period t) has to last.
We set k ./ D 0 for each k 2 E . Next, suppose that S there exists t 2 N n f1g
t 1
such that we have already defined k .h0 / for each h0 2 mD1 H .m/ and every
.t /
k 2 E . Consider h 2 H . For each j 2 EJ , we proceed as follows. If j L .h/

.r/
and there do not existi 2 L .h/ \ EI and r 2 f1; : : : ; ng with xj i a > 0, then
j .h/ D max j .h / 1; 0 . If j 2 L .h/ and there do not exist i 2 L .h/ \ EI

and r 2 f1; : : : ; ng with xj i a.r/ > 0, then j .h/ D T . If there exist i 2

L .h/ \ EI and r 2 f1; : : : ; ng with xj i a.r/ > 0, then j .h/ D 0. For each
i 2 EI , we proceed as follows. If i L .h/ and there do not exist j 2 L .h/ \ EJ

and r 2 f1; : : : ; ng with xj i a.r/ > 0, then i .h/ D max f i .h / 1; 0g. If
i 2 L .h/, then i .h/ D T . If i L .h/ and there exist j 2 L .h/ \ EJ and

r 2 f1; : : : ; ng with xj i a.r/ > 0, then i .h/ D 0.
Now we describe the strategy profile s . For each t 2 N and every h 2 H .t / , we
let D .h/ D fk 2 E j k .h/ > 0g and set s .h/ D j D .h/.1
Clearly,
.s/ D v . For each r 2 f1; : : : ; ng, the sum of the firms stage game

profits given by a.r/ j D is the same for each D 2 2E satisfying the condition
(3.8). The probabilities with which stage game pure strategy profiles a.r/ j D are
used under s depend only on r, and not on D. Therefore,
X X

k.h/ s.h/ D vk D ; 8h 2 Hf : (3.9)
k2J [I k2J [I
C. Strategy profile s is an SRPE of ./ for 2 .; 1/. Since ./ is continuous

at infinity,2 with regard to unilateral deviations, it suffices to show that there do not
1
Note that the definition of function k , k 2 E ensures that D .h/ satisfies (3.8).
2
The firms stage game payoffs are bounded, and the future stage game payoffs are discounted.
exist k 2 J [ I and h 2 Hf such that k can increase his/her expected average

discounted profit in .h/ ./ by a unilateral single period deviation in its first period.
C1. Unilateral deviations by producers. A unilateral deviation by j 2 J nEJ cannot
lead to him/her trading with any buyer . Thus, for each ! 2 0; 1, his/her unilateral
deviation to any stage game pure strategy decreases his/her single period profit from
zero to cj .0/ < 0, and does not affect future play. Therefore, the producers
unilateral deviation to any stage game correlated strategy decreases his/her expected
average discounted profit in the subgame of ./ at the beginning of which the
deviation takes place.
Now, take (arbitrary) j 2 EJ and h 2 Hf . We distinguish between the two
cases.
First, j .h/ D 0. Without a deviation, j s expected average discounted profit
in .h/ ./ is at least vj . A deviation gives him/her an expected stage game profit
in the first period of .h/ ./ that is bounded from above by vmaxj , and then he/she
is punished for T periods and has the expected stage game profit vj C j . Thus, a
deviation gives him/her an expected average discounted profit in .h/ ./ that is not
higher than

T
.1 / vmax
j C 1 vj C j C T C1 vj .1 / vj C vj D vj ; (3.10)
where the inequality follows from (3.5). Therefore, a deviation does not increase
j s expected average discounted profit in .h/ ./.
Second, j .h/ > 0. Without a deviation, j s expected average discounted profit
in .h/ ./ equals

1 j .h/ vj C j C j .h/ vj 1 T vj C j C T vj > 0: (3.11)
(The latter payoff is positive by (3.7).) A deviation gives j a nonpositive expected

stage game profit in the first period of .h/ ./. (A withdrawal from the analyzed
market gives him/her zero stage game profit. His/Her failure to withdraw from the
analyzed market at ! 2 0; 1, for which sj.h/ prescribes a withdrawal, gives him/her
stage game profit cj .0/ < 0. Any other pure strategy in G gives him/her a negative
profit because the price of his/her output equals zero and he/she has a positive fixed
cost.) Then, the punishment (giving him/her the stage game expected profit vj C j
for T periods) is resumed. Thus, a deviation gives him/her the expected average
discounted profit in .h/ ./ that is not higher than

1 T vj C j C T C1 vj < 1 T vj C j C T vj : (3.12)
Therefore, a deviation decreases j s expected average discounted profit in .h/ ./.

C2. Unilateral deviations by buyers. A unilateral deviation by i 2 I nEI cannot
lead to him/her trading with any producer. Thus, for each ! 2 0; 1, the buyers
unilateral deviation to any stage game pure strategy decreases his/her single period
3.1 Existence of an SRPE 25
profit from zero to ci .0/ < 0 and does not affect future play. Therefore, his/her
unilateral deviation to any stage game correlated strategy decreases his/her expected
average discounted profit in the subgame of ./ at the beginning of which the
deviation takes place.
Now, take (arbitrary) i 2 EI and h 2 Hf . We distinguish between the two cases.
First, i .h/ D 0. Without a deviation, i s expected average discounted profit in
.h/ ./ is at least vi . A deviation gives him/her an expected stage game profit in
the first period of .h/ ./ that is bounded from above by vmaxi , and then he/she is
punished for T periods and provided the expected stage game profit vi C i . Thus,
a deviation gives him/her an expected average discounted profit in .h/ ./ that is
not higher than

T
.1 / vmax
i C 1 vi C i C T C1 vi .1 / vi C vi D vi ; (3.13)
where the inequality follows from (3.5). Therefore, a deviation does not increase i s
expected average discounted profit in .h/ ./.
Second, i .h/ > 0. Without a deviation, i s expected average discounted profit
in .h/ ./ equals

1 i .h/ vi C i C i .h/ vi 1 T vi C i C T vi > 0: (3.14)
(The latter payoff is positive by (3.7).) A deviation gives him/her a nonpositive

expected stage game profit in the first period of .h/ ./. (A withdrawal from the
analyzed market gives him/her zero stage game profit. His/Her failure to withdraw
from the analyzed market at ! 2 0; 1, for which si.h/ prescribes a withdrawal,
gives him/her the stage game profit ci .0/ < 0. Any other pure strategy in G gives
him/her a nonpositive profit because he/she pays price pjmax as defined in (2.6) for
each good j 2 EJ that he/she buys, he/she has a positive fixed cost, and Ui .0/ D 0.)
Then, the punishment (giving him/her the stage game expected profit vi C i for T
periods) is resumed. Thus, a deviation gives him/her an expected average discounted
profit in .h/ ./ that is not higher than

1 T vi C i C T C1 vi < 1 T vi C i C T vi : (3.15)
Therefore, a deviation decreases i s expected average discounted profit in .h/ ./.

C3. Payoff vector in each
subgame
is strictly Pareto efficient. This claim follows
from the fact that
.h/ s.h/ 2 V C for each h 2 Hf and from (3.9). t
u
The strategy profile described in the proof of Proposition

3.1 is based on the use
of the same vectors of the traded quantities (x a.r/ , r 2 f1; : : : ; ng) with the cor-
responding probabilities (r , r 2 f1; : : : ; ng) in each period, after any history. They
are used even in the punishments for unilateral deviations by firms with a positive
equilibrium payoff, and only the prices, at which a deviator trades in the analyzed
market, change in a way that harms him/her.3 (Firms with zero equilibrium payoff
cannot increase their stage game payoff by any unilateral deviation.) These modified
prices are in effect long enough to wipe out any single period gain from a deviation.
Since the sum of the firms expected stage game profits is maximized in each period
(even during punishments), the sum of their expected average discounted profits
(which equals the average discounted sum of their expected stage game profits) is
maximized in each subgame. Thus, the continuation equilibrium payoff vector is
strictly Pareto efficient in each subgame.
3.2 Existence of an SSPE
In this section, we first deal with a monopsonistic market, where I D f# .J / C 1g.

In this case, under the conditions given below in Lemma 3.1, the stage game G has
a strict strong equilibrium (henceforth, SSE). An infinite repetition of the latter is an
SSPE of ./ for each 2 .0; 1/.
Definition 3.1. A correlated strategy profile 2 is an SSE of G if

(i) there do not exist C 2 2J [I n fJ [ I; g and .C / 2 C such that

gk .C / ; C

gk ; 8k 2 C; (3.16)
and
9k 2 C with gk .C / ; C

> gk ; (3.17)
(ii) there does not exist 2 such that

gk ./ gk ; 8k 2 J [ I; (3.18)
and
9k 2 J [ I with gk ./ > gk : (3.19)
Thus, a correlated strategy profile is an SSE of G if no coalition of players can,

by a deviation, increase the expected payoff of at least one of its members without
decreasing the expected payoff of another member. Part (i) of Definition 3.1 when
applied to singleton coalitions implies that an SSE of G is a Nash equilibrium of G.
3
For simplicity, the strategy profile described in the proof of Proposition 3.1 prescribes for each
punished producer trading at zero prices. When the assumptions of the latter proposition are satis-
fied, we can construct an SRPE in which all trading takes place at positive
prices.
For example,
a
punished producer j with v j > 0 can sell his/her output at price cj yj a.r/ =yj a.r/ when

the vector of traded quantities x a.r/ , r 2 f1; : : : ; ng, is prescribed during his/her punishment.
Nevertheless, this could lead to an increase in the duration of punishment, as well as to an increase
in the lower bound on the values of the discount factor for which a strategy profile is an SRPE.
3.2 Existence of an SSPE 27
Part (ii) implies that the expected equilibrium payoff vector in an SSE is strictly
Pareto efficient with respect to V C .
An SSE satisfying Definition 3.1 would remain an SSE even if we allowed the
firms mixed strategies in G (i.e., randomizations over their pure strategies based
on the signals of their private randomizing devices, where the randomizations of
the different firms are independent) and the firms strategies that assign their mixed
strategies to the signals of the public randomizing device. A mixed strategy of firm
k 2 J [ I with a finite support is a best response against ak 2 Ak if and only if
each pure strategy in its support is a best response against ak . If a mixed strategy of
firm k 2 J [ I with an infinite support is a best response against ak 2 Ak , then
each subset of its support with a positive measure contains ks pure strategy that is a
best response against ak . Suppose that there exist an SSE of G , k 2 J [ I , a 2
A, and a mixed strategy k of firm k such that a occurs with a positive probability
when all firms follow and k is a better response against ak than ak . Then,
there exists ak 2 Ak that (belongs to the support of k and) is a better response
against ak than ak . Thus, a correlated strategy of firm k, which differs from k
only in that it prescribes ak for each ! 2 0; 1 for which prescribes a, is a

better response against k than k . This contradicts the assumption that is an
SSE of G. Next, suppose that there exists an SSE of G , a coalition C with
# .C / 2 2; # .J [ I / 1 \ N , aC 2 AC , and a profile .C / of the mixed
strategies of the firms in C such that aC occurs with a positive probability when

the firms in J nC follow C and .C / gives against aC a vector of the expected
payoffs of the firms in C that weakly Pareto dominates the one generated by the
prescriptions of C . Then, there exists a profile .C / of correlated strategies of firms
in C that assigns a positive probability to at most # .C / C 1 elements of AC and
gives against aC the same vector of expected payoffs as .C / . Thus, a profile of
the correlated strategies of the firms in C , which differs from C only in that it
prescribes .C / instead of the prescription of C for the subset of 0; 1 for which

C prescribes aC , violates condition (i) of Definition 3.1. Of course, it can happen
that .C / does not weakly Pareto improve the vector of the expected payoffs of the
firms in C against any aC that occurs with a positive probability when the firms in

J nC follow C , but it weakly Pareto improves the vector of the expected payoffs

of the firms in C against C . Then, .C / changes the vector of the expected payoffs
of the firms in C against at least two profiles of the pure strategies of the firms in

J nC that occur with a positive probability when the latter firms follow C . For
each such aC , there exists a profile .C / .aC / of the correlated strategies of the
firms in C that gives the same vector of the expected payoffs of the firms in C
against aC as .C / (and it assigns a positive probability to at most # .C / C 1
elements of AC ). Thus, a profile of the correlated strategies of the firms in C , which
differs from C only in that it prescribes .C / .aC / instead of the prescription of
C for the subset of 0; 1 for which C
prescribes aC , violates condition (i) of
Definition 3.1. As far as the deviations by the grand coalition are concerned, any
profile of mixed strategies gives a vector of the expected payoffs belonging to V C .
Any element of V C can be generated by a profile of the correlated strategies.
Lemma 3.1. Assume that # .I / D 1, v 2 V C , and that there exist 2 , n 2

f1; : : : ; # .J [ I / C 1g, a.r/ 2 A, r 2 f1; : : : ; ng , and 2 4n with 0 such
that v D g . /, and for each r 2 f1; : : : ; ng ; a.r/ occurs with probability r
when the firms follow ,

gk a.r/ 0; 8k 2 J [ I; 8r 2 f1; : : : ; ng ; (3.20)
and
X
D gk a.r/ ; 8D 2 2J [I with f# .J / C 1g D; 8r 2 f1; : : : ; ng :
k2D
(3.21)
Then, there exists an SSE of G with the equilibrium vector of expected payoffs
equal to v .
Proof. Consider (arbitrary) v 2 V C , 2 , n 2 f1; : : : ; # .J [ I / C 1g, a.r/ 2

A, r 2 f1; : : : ; ng, and 2 4n that satisfy the assumptions of Lemma 3.1. We
can assume without loss of generality that for each r 2 f1; : : : ; ng, aj.r/ D .0; 0/

and a#.J /C1 prescribes the making of the contract proposal .0; 0/ to producer j if
.r/
xj #.J /C1 a D 0. (If does not have the latter property, we can replace it
by another profile of correlated strategies that has the latter property and satisfies
all assumptions of Lemma 3.1.4 This follows from two facts. First, prices proposed
by the firms, which do not trade together, for their bilateral trade do not affect the
payoffs. Second, if producer j and buyer i do not trade together, then replacing the
traded quantity proposed by j to i and replacing the traded quantity proposed by
i to j by zero does not change the vector of the traded quantities or payoffs.) We
show that is an SSE of G.
Note that (3.20) implies v 0 and (3.21) implies
X
D vk ; 8D 2 2J [I with # .J / C 1 2 D & # .D/ > 1: (3.22)
k2D
Take (arbitrary) x max P2 X max . Since

P (3.21) also holds for D D E .x / and
max
v 0, we have D k2D vk D k2J [I vk . That is, v maximizes the sum
ofthe firms
payoffs over V C .This further implies that for each r 2
f1; :: : ; ng,
.r/
x a 2 X max and gj a.r/ D 0 for each j 2 J with xj #.J /C1 a.r/ D 0.
This also implies that the grand coalition cannot, by a deviation, increase the
expected payoff of at least one of its members without decreasing the expected
payoff of some other member.
4
The fact that we do not include the latter property among the assumptions of Lemma 3.1 is the
only reason why we do not formulate the last sentence of the latter lemma as Then, is an SSE
of G.

Consider (arbitrary) j 2 J , his/her unilateral deviation to j 2 j n j , and
! 2 0; 1 with j .!/ j .!/. Suppose that all other firms stick to the prescrip-

tions of j . Then, j .!/ either leads to j s withdrawal from the analyzed market
(that gives him/her payoff 0 gj . .!//) or to no trade between him/her and
the monopsonist without his/her withdrawal from the analyzed market (thatgives
him/her payoff cj .0/ < 0 gj . .!//). Therefore, gj j ; j
vj , and
hence, j s unilateral deviation does not increase his/her expected payoff. The same
conclusion holds when he/she deviates as a member of any coalition of producers,
because deviations by other producers cannot affect his/her expected payoff from
his/her deviation strategy.
Next consider a monopsonists unilateral deviation to
n o

#.J /C1 2 #.J /C1 n #.J /C1

and consider (arbitrary) ! 2 0; 1 with #.J /C1 .!/ #.J /C1
.!/ and

g#.J /C1 #.J /C1 .!/ ; #.J /C1 .!/ g#.J /C1 .!/ :
Take r 2 f1; : : : ; ng satisfying .!/ D a.r/ . Suppose that all produc-

ers stick to the prescriptions of #.J /C1
. Then, #.J /jC1 .!/ either leads to
the monopsonists withdrawal
from the analyzed market (that gives him/her
payoff 0 g#.J /C1 a.r/ ) or to no trade between him/her and any pro-
ducer without his/her withdrawal from the analyzed market (that gives him/her
payoff c#.J /C1 .0/ < 0 g#.J /C1 a.r/ ) or to trade with each pro-

ducer in a nonempty set K EJ x a.r/ . (He/She cannot trade with any
.r/
j 2 J nEJ x a because such j sticks to the prescription of j and makes
the contract proposal .0;
.r/ 0/ to the monopsonist.) .r/ In the latter case, he/she buys
quantity
xj /C1 a for price e
p j /C1 a from each j 2 K. Therefore,
#.J #.J
.r/

gj #.J /C1 .!/ ; #.J /C1 .!/ D gj a for each j 2 K. Thus, setting
D D K [ f# .J / C 1g and using (3.21), we have

g#.J /C1 #.J /C1 .!/ ; #.J /C1 .!/
X

D gj #.J /C1 .!/ ; #.J /C1 .!/
j 2K
X
D D gj a.r/ g#.J /C1 a.r/ :
j 2K

This further implies that g#.J /C1 #.J ;
/C1 #.J /C1
v#.J /C1 . Thus, a monop-
sonists unilateral deviation does not increase his/her expected payoff.
Finally, consider a deviation by a coalition
C J [ I with # .J / C 1 2 C
and C \ J to .C / 2 C n C . Suppose that the firms in J nC stick

to the prescriptions of C . Take (arbitrary) r 2 f1; : : : ; ng and ! 2 0; 1 with
.r/ .C /
.!/ D a . If .!/ does not lead to the monopsonists trade with any
producer,
P it gives the firms in C the sum of the payoffs that is not higher than
.r/ .C /
0 k2C gk a . If .!/ leads to the monopsonists trade with each
producer in a nonempty set J , we set K D M nC . (K can be empty.
M
Clearly, K EJ x a.r/ because the monopsonist cannot trade with any

j 2 J n EJ x a.r/ [ C each such j sticks to the prescription of j and

makes the contract proposal .0; 0/.) Then, gj .C / .!/ ; C
.!/ D gj a.r/
.C /

for each j 2 K. Further, gj .!/ ; C .!/ 0 for every j 2 .C \ J / nM
because such j does not trade. Thus, setting D D M [f# .J / C 1g and using (3.20)
and (3.21) (and noting that C nD D .C \ J / nM and DnC D K), we have
X X
gk .C / .!/ ; C

.!/ gk .C / .!/ ; C

.!/
k2C k2C \D
X
D gj .C / .!/ ; C

.!/
j 2DnC
X X
D D gj a.r/ gk a.r/
j 2DnC k2C \D
X X
gk a.r/ D gk .!/ ;
k2C k2C
where
P weomit the
.C /
P DnC
sum over

if D C . This further implies that
k2C gk ; C v
k2C k . Thus, a deviation by C cannot increase
the expected payoff of at least one of its members without decreasing the expected
payoff of some other member. t
u
The sufficient condition for the existence of an SSE in G given in Lemma 3.1 is
easy to understand. Each pure strategy profile that occurs under a candidate equi-
librium correlated strategy profile (i.e., under a correlated strategy profile which we
want to be an SSE of G) with a positive probability has to be individually rational.
Moreover, it has to give each coalition containing the monopsonist and at least one
producer (i.e., each coalition that can trade without cooperation of any firm that
does not belong to it) a sum of payoffs that is not lower than the sum of payoffs it
can achieve without trading with any producer who does not belong to it. The latter
condition ensures that no such coalition can increase the expected payoff of at least
one of its members without decreasing the expected payoff of another member. This
also ensures that (for any signal of the public randomizing device), the monopsonist
cannot increase his/her payoff by a unilateral deviation that leads to his/her trade
with at least one producer. (Each producer, with whom he/she would trade, would
earn the same payoff as in the case when no deviation takes place. Thus, a deviation
would have to increase the sum of the payoffs of the firms that participate in the
trade.) Individual rationality implies that the monopsonist cannot increase his/her
payoff by a unilateral withdrawal from the analyzed market (or any other unilateral
deviation that prevents him/her from trading with any producer) and no producer
can increase his/her payoff by a unilateral deviation or as a member of a deviating
coalition of producers.
The following lemma shows that the sufficient condition in Lemma 3.1 is with
the exception of the requirement that the equilibrium profile of the correlated strate-
gies uses with a positive probability at most # .J [ I / C 1 pure strategy profiles in
G also necessary for the existence of an SSE of G.
Lemma 3.2. Assume that # .I / D 1 and v 2 V C is the vector of the equilib-

rium expected payoffs in an SSE of G. Then, v 0, (3.22) holds, and there exists
2 such that v D g . / and each a 2 A that occurs with a positive
probability
P when the firms follow satisfies (3.20), (3.21), x .a / 2 X max , and

k2E .x max / gk .a / D for every x 2 X max . Further,
max
X
vk D ; 8x max 2 X max : (3.23)
k2E .x max /
Proof. Let v 2 V C be the equilibrium vector of the expected payoffs in an SSE

of G. Since v 2 V C and the firms can use only pure or correlated strategies (and
a pure strategy is a special case of a correlated strategy),5 there exists 2 with
g . / D v that is an SSE of G. Take (arbitrary) a 2 A that occurs with a positive
probability when the firms follow . If we had gk .a / < 0 for some k 2 J [I , firm
k can increase its expected payoff by (unilateral deviation prescribing) withdrawal
from the analyzed market when prescribes a and by sticking to the prescriptions
of k when prescribes any other pure strategy profile. Therefore, g .a / 0.
Since this holds for each a 2 A that occurs with a positive probability when the
firms follow , we have v 0.
Suppose P that there exists a non-singleton coalition D such that # .J / C 1 2 D
.D/
and D > k2D gk .a /. Then, there exists v 2 <#.D/ satisfying v.D/

P .D/
.gk .a //k2D and k2D vk D D : Moreover, (using the definition of D in
(2.4)) there exists x 2 X such that P xj #.J /C1 D 0 for each j 2 J nD and
D D U#.J /C1 .x/ c#.J /C1 .x/ j 2D\J cj xj #.J /C1 . Let
h i
.D/
pj D cj xj #.J /C1 C vj =xj #.J /C1
5
The arguments in the proof also apply to a profile of mixed strategies after the following modifi-
cations. First, we replace a firms payoff from a profile of pure strategies occurring with a positive
probability in the equilibrium by its expected payoff from its pure strategy that belongs to the sup-
port of its equilibrium mixed strategy. Second, we replace the sum of the payoffs of the members
of a non-singleton coalition containing the monopsonist from a profile of pure strategies occurring
with a positive probability in equilibrium by the sum of their expected payoffs from a profile of
their pure strategies belonging to the Cartesian product of the supports of their mixed strategies.
These remarks hold under the assumption that members of a deviating coalition can communicate
the signals of their private randomizing devices to other members.

and aj D pj ; xj #.J /C1 for each j 2 J \ D. Further, let a#.J /C1 prescribe the
contract proposal pj ; xj #.J /C1 to each j 2 J \ D and the contract proposal

.0; 0/ to every j 2 J nD. Then, gk .am /m2D ; aD
D v.D/
k
for each k 2 D. Let
.D/ .D/
2 D satisfies .!/ D .am /m2D for each ! 2 0; 1 with .!/ D a ,
.D/
and .!/ D .!/ for every ! 2 0; 1 with .!/ a . Then, we have

.D/ D
gk ; D > vk for each k 2 D. This contradiction along with the fact that
is an SSE of G proves that (3.21) holds for each a 2 A that occurs with a positive
probability when the firms follow . Then, (3.22) follows from (3.21). For each
x max 2 X max and every a 2 A that occurs with a positive probability when the
firms follow , applying (3.21) to D D E .x max / and using the fact that g .a / 0
yields
X X
gk a D gk a D :
k2E .x max / k2J [I
This implies (3.23), and also that x .a / 2 X max for each a 2 A that occurs
with a positive probability when the firms follow : t
u
The following proposition shows that the payoff vector satisfying the conditions
stated in Lemma 3.1 is the equilibrium vector of the expected average discounted
payoffs in an SSPE of ./ for any discount factor.
Proposition 3.2. Assume that #.I / D 1 and v 2 V C ; that there exists 2 ,

n 2 f1; : : : ; # .J [ I / C 1g, a.r/ 2 A, r 2 f1; : : : ; ng, and 2 4n with 0
such that v D g . /, and that for each r 2 f1; : : : ; ng, a.r/ occurs with proba-
bility r when the firms follow and it satisfies (3.20) and (3.21). Then, for each
2 .0; 1/ ; there exists an SSPE of ./ with the equilibrium vector of expected
average discounted payoffs equal to v .
Proof. Consider (arbitrary) v 2 V C , 2 , n 2 f1; : : : ; # .J [ I / C 1g, a.r/ 2

A, r 2 f1; : : : ; ng, and 2 4n satisfying the assumptions of Proposition 3.2. The
latter assumptions coincide with the assumptions of Lemma 3.1. As already noted
in the proof of Lemma 3.1, we can assume without loss of generality that for each
r 2 f1; : : : ; ng, aj.r/ D .0; 0/ and a#.J

/C1
prescribes the making of the contract
proposal .0; 0/ to producer j if xj #.J /C1 D 0. We have shown in the proof of
Lemma 3.1 that is an SSE of G.
We define s 2 S by s .h/ D for each h 2 Hf . Thus, s leads to the infinite
repetition of in each subgame of .
Since is a (correlated) Nash equilibrium of G, no firm can increase its
expected average discounted profit in any subgame of ./ by a unilateral deviation.
No producer can increase his/her expected average discounted profit in any subgame
of ./ as a member of a deviating coalition of producers because the deviations
by other producers cannot affect his/her expected average discounted profit from
his/her deviation strategy.
As shown in the proof of Lemma 3.1, v maximizes the sum of the firms payoffs
over V C . Therefore, in any subgame of ./ and for any 2 .0; 1/, a devia-
tion by the grand coalition (regardless of the number of periods it lasts) does not
increase the sum of the firms expected average discounted profits. Thus, it can-
not increase the expected average discounted profit of at least one firm without
decreasing the expected average discounted profit of some other firm.
Finally, consider (arbitrary) non-singleton coalition C J [ I with # .J / C 1 2
C . As shown in the proof of Lemma 3.1, a deviation by such coalition C does not
increase the sum of the stage game expected profits of its members. Therefore, in
any subgame of ./ and for any 2 .0; 1/, a deviation by such C (regardless of
the number of periods it lasts) does not increase the sum of the expected average dis-
counted profits of the members of C . Thus, it cannot increase the expected average
discounted profit of at least one firm in C without decreasing the expected average
discounted profit of some other firm in C . t
u
The following proposition shows that in each SSPE of ./ and in every

subgame .h/ of ./, the vector of the expected average discounted profits is indi-
vidually rational, maximizes the sum of the firms expected average discounted
profits, and satisfies (3.23). Moreover, for each non-singleton coalition containing
the monopsonist, the sum of the expected average discounted profits of its mem-
bers is not lower than the maximal sum of their stage game profits that they can
achieve without trading with the producers outside the coalition. The latter property
implies that each profile of stage game pure strategies, which occurs with a positive
probability in .h/ ./ when the firms follow the prescriptions of their equilibrium
strategies in it, generates a vector of traded quantities that maximizes the sum of the
firms stage game profits. All these properties would continue to hold (and the argu-
ments in the proof of the following proposition would remain valid) if we allowed
mixed strategies in the stage game (along with the assigning of stage game mixed
strategies to nonterminal histories by the behavioral strategies in the repeated game).
Therefore, the results in Chap. 4 that build on Proposition 3.3 would continue to hold
even if we allowed mixed strategies in the stage game.
Proposition 3.3. Assume that # .I / D 1. Let 2 .0; 1/, s 2 S be an SSPE of

P
./, h 2 Hf , and v D
.h/ s.h/ . Then, v 0, k2J [I vk D , and v

satisfies (3.22) and (3.23). Moreover, x .a / 2 X max for each a 2 A that occurs

with a positive probability in .h/ ./ when the firms follow the prescriptions of s.h/ .

2 .0; 1/ ; s 2 S that is an SSPE of ./, and h 2 Hf .
Proof. Take (arbitrary)
Let v D
.h/ s.h/
. If we have vk < 0 for some k 2 J [ I , firm k can increase
its expected average discounted profit in .h/ ./ by withdrawing from the analyzed
market with probability one in each period after every history in .h/ ./ leading
to it. Therefore, v 0. If v does not satisfy (3.22) for some non-singleton
coalition
D with # .J / C 1 2 D, there exists v.D/ 2 <#.D/ satisfying v.D/ vk k2D and
P .D/
k2D vk D D . Thus, D can increase the expected average discounted profit of
each of its members in .h/ ./ by using strategy ak , k 2 D described in the proof
of Lemma 3.2, with probability one in each period after every history in .h/ ./
leading to it. Thus, v satisfies (3.22). For each x max 2 X max , applying (3.22) to
D D E .x max / and using the fact that v 0 yields
X X
vk D vk D : (3.24)
k2E .x max / k2J [I
The second equality in (3.24) implies that the claim in the last sentence of Proposi-
tion 3.3 holds. t
u
Now we give two examples of a stage game in which a vector of payoffs satisfy-
ing the assumptions of Proposition 3.2 exists. In the first, all firms have to participate
in trade in order to maximize the sum of the stage game profits of all firms (that
results from the unique vector of the traded quantities belonging to X max /. In the
second example, X max has four elements and each producer can be replaced by
another (that is identical to him) in the set of firms that participate in trade in the
maximization of the sum of the stage game profits of all firms. Therefore, (the stage
game payoff vector maximizing the sum of the profits of all firms) v assigns zero
profit to each producer (i.e., the monopsonist captures the whole surplus from trade).
Despite this, v is sustainable as the vector of the equilibrium expected average dis-
counted profits in an SSPE of ./ for any 2 .0; 1/. Since there exists only one
buyer, a unilateral deviation by any producer, who is (according to the prescriptions
of the equilibrium strategy profile) supposed to trade, can lead only to his/her non-
participation in the trade. Thus, this cannot give him/her a positive stage game profit,
and hence, he/she need not be punished.
Example 3.1. J D f1; 2; 3; 4g and I D f5g. Producers 1 and 3 produce the same
good, which we refer to as type one good; producers 2 and 4 produce
the same2 good,
which we refer to as type two
P good. We have Yj D 0; 5 and c j yj D 10yj C 25
for each j 2 J , c5 .x5 / D j 2J xj 5 C 15. The monopsonist is a retailer and
U5 .x/ D P1 .x/ .x15 C x35 / C P2 .x/ .x25 C x45 / ; (3.25)
where P1 W X ! 0; 121 defined by
P1 .x/ D max f121 10 .x15 C x35 / 5 .x25 C x45 / ; 0g (3.26)
is the inverse demand function for the type one good, and P2 W X ! 0; 121 defined
by
P2 .x/ D max f121 5 .x15 C x35 / 10 .x25 C x45 / ; 0g (3.27)
is the inverse demand function for the type two good. We let pjmax D 326 for each
j 2 J.
In this example, the sum of the firms stage game profits is maximized when each
producer supplies 1:5 unit of his/her output to the monopsonist. That is, X max D
f.1:5; 1:5; 1:5; 1:5/g. The maximized sum of profits is 245 financial units. Thus,
D f1;2;3;4;5g D 245.
When each type of good is produced and supplied to the monopsonist by only
one producer (and the remaining two producers withdraw from the analyzed mar-
ket), the maximal sum of the firms stage game profits equals 223 financial units.
Thus, f1;2;5g D f1;4;5g D f2;3;5g D f2;4;5g D 223. When one type of good is
produced and supplied to the monopsonist by both of its producers and the other
type is produced and supplied to the monopsonist by only one of its producers
(and the other producer withdraws from the analyzed market), the maximal sum
of the firms stage game profits equals (approximately) 237:27 financial units. Thus,
f1;2;3;5g D .1;3;4;5/ D f1;2;4;5g D f2;3;4;5g D 237:27. When one type of good
is produced and supplied to the monopsonist by both of its producers and the other
type is not produced at all (with the withdrawal of both its producers from the ana-
lyzed market), the maximal sum of the firms stage game profits is 175 financial
units. Thus, f1;3;5g D f2;4;5g D 175. When only one type of good is produced and
supplied to the monopsonist by only one of its producers (and the remaining three
producers withdraw from the analyzed market), the maximal sum of the firms stage
game profits equals 140 financial units. Thus, fj;5g D 140 for each j 2 J .
Consider the payoff vector v D .6:5; 6:5; 6:5; 6:5; 219/. It is easy to verify that
it satisfies (3.22). We have v D g .a /, where aj D .36; 1:5/ for each j 2 J
and a5 D .36; 1:5/j 2J . In this case, v is generated by a pure strategy profile in
G, and as such, (3.22) implies (3.21). Thus, v and a satisfy all assumptions of
Proposition 3.2.
Example 3.2. J D f1; 2; 3; 4g and I D f5g. Producers 1 and 3 produce the same
good, which we refer to as type one good; producers 2 and 4 produce the same

good, which we refer to as type two good. WePhave Yj D 0; 100 and cj yj D
10yj C 100 for each j 2 J and c5 .x5 / D j 2J xj 5 C 300. The monopsonist
uses the goods purchased from the producers in J to produce a good that he/she
sells in a market, in which he/she is a monopolist. His/Her production function
f W X ! 0; 200 has the form f .x/ D min fx15 C x35 ; x25 C x45 g. The inverse
demand function for his/her product, P W 0; 200 ! 0; 102 has the form P .Q/ D
max f102 Q; 0g, where Q is his/her output. This yields
U5 .x/ D maxf102 minfx15 C x35 ; x25 C x45 g; 0g minfx15 C x35 ; x25 C x45 g:
(3.28)
Let pjmax D 66 for each j 2 J .

Clearly, since both producers of each type of good have the same constant
marginal cost and positive fixed cost, the maximization of the sum of the firms stage
game profits requires that each type of good is produced by only one producer (and
it does not matter which one). Further, given the form of the monopsonists produc-
tion function, the outputs of both types of goods have to be equal. The maximization
of the sum of the firms stage game payoffs under these constraints gives
X max D f.40; 40; 0; 0/ ; .40; 0; 0; 40/ ; .0; 40; 40; 0/ ; .0; 0; 40; 40/g (3.29)
and D f1;2;5g D f1;4;5g D f2;3;5g D .3;4;5/ D 1100. When two producers

produce one or both types of good, the optimal output of each type of good (that
maximizes the sum of the firms stage game profits for a given set of producers
who do not withdraw from the analyzed market) equals 40 but the sum of the firms
stage game profits is reduced by the fixed costs of the redundant producer(s). Thus,
D[f5g D 1000 for each D J with # .D/ D 3, D \ f1; 3g , and D \ f2; 4g
, as well as f1;2;3;4;5g D 900. Since the monopsonist needs a positive amount of
each type of good in order to produce something and obtain a positive revenue, we
have fj;5g D 400 for each j 2 J and f1;3;5g D f2;4;5g D 500.
For each element of X max , there exists another with a different set of active
producers. Therefore, there is only one payoff vector that satisfies (3.22), namely,
v D .0; 0; 0; 0; 1100/. It can be generated by any profile of correlated strategies in
G that assigns a positive probability only to the profiles of pure strategies that sat-
isfy x .a/ 2 X max and g .a/ D v . Each such a prescribes the trading between the
monopsonist and both active producers at price equal to 12:5 financial units. (An
example of such a 2 A is given by a1 D a2 D .12:5; 40/, a3 D a4 D .0; 0/,
and a5 D ..12:5; 40/ ; .12:5; 40/ ; .0; 0/ ; .0; 0//.) Clearly, each such profile of
correlated strategies satisfies (3.20) and (3.21).
Now we proceed to a sufficient condition for the existence of an SSPE of ./

for 2 .0; 1/ close enough to one when there is more than one buyer.
For each C 2 2J [I n fJ [ I g with C \ J and C \ I , let
8 " #9

< P U .x/ c .x / C P P >
i i i pjmax xjk cj j .x/ =
C D max
vmax i 2I \C j 2J \C k2I nC
: >
;
jx2X
(3.30)
and
8 " # 9

< P U .x/ c .x / P p max x P >
=
i i i k kj cj j .x/
C D min
vmin i 2I \C k2J nC j 2J \C ;
: >
;
jx2X
(3.31)
where we omit the sum over I nC if I C and h the sum i over J nC if J C . Since
the price of good j 2 J is from the interval 0; pj , vC (vC ) is the maximal
max max min
(minimal) sum of the stage game payoffs of the firms in coalition C .
Proposition 3.4. Assume that there exists v 2 V C with. v 0, 2 , n 2

f1; : : : ; # .J [ I / C 1g, a.r/ 2 A, r 2 f1; : : : ; ng, and 2 4n with 0 such
.r/
v D g . /, for each r 2 f1; : : : ; ng, a
that occurs with r
S probability and
x a.r/ 2 X max . Further, assume that vi > 0 for each i 2 nrD1 EI x a.r/ ;

gj a.r/ > 0; 8r 2 f1; : : : ; ng ; 8j 2 EJ x a.r/ ; (3.32)

and for each r 2 f1; : : : ; ng and every .j; i / 2 EJ x a.r/ EI x a.r/ ,

there exists a price pj.r/
i such that pj.r/
i p j i a.r/ if xj i a.r/ > 0 and
2 0; e

pj.r/
i D 0 if xj i a.r/ D 0,
X
.r/
pj i xj i a.r/ cj yj a.r/ ;
i 2EI .x .a.r/ //

8r 2 f1; : : : ; ng ; 8j 2 EJ x a.r/ ; (3.33)
!
X X
pj.r/
i xj i a .r/
cj xj i a .r/
i 2K i 2K
X
.r/
pj i xj i a.r/ cj yj a.r/ ;
i 2EI .x .a.r/ //

8r 2 f1; : : : ; ng ; 8j 2 EJ x a.r/ ;
n o
8 K i 2 I j xj i a.r/ > 0 ; (3.34)
and
8 " # 9

P P >
>

U i .x/ ci .x i / p max
xki >
>

k >
>

i 2I \M
" k2J nC # >
>

>
>

P P >

C Ui .x/ ci .xi / e
p ki a .r/
xki >>
>

>
>

i 2.C \I /nM Wxi >0 " k2J nC # >
>

>
>
< P P .r/ =
v.r/ D max C p jk
x jk c j j .x/
C;M

j 2J \M k2I nC
" # >>
>

P P >
>

C ep jk a .r/
xjk cj j .x/ j > >

>
>

j 2.C \J /nM Wj .x/>0 k2I nC >
>

>
>

x 2 X; x 2 x a .r/
; 0 >
>

ji ji >
>

8 .j; i / 2 ..J \ C / nM / I [ J ..I \ C / nM / ; >
>
: .r/ >
;
xj i D xj i a 8 .j; i / 2 .J nC / .I nC /
X
gk a.r/ ;
k2C
8r 2 f1; : : : ; ng ; 8C 2 2J [I n fJ [ I g with
C \ J & C \ I ; 8M 2 2C ; (3.35)
with strict
inequality
if M and there exists either i 2 M \ I and j 2 J nC
with xj i a.r/ > 0 or j 2 J \ M and i 2 I nC with xj i a.r/ > 0.6 The sum
over J nC (I nC , I \ M , .C \ I / nM , J \ M , .C \ J / nM ) is omitted here if
J C (I C , I \ M D , C \ I M , J \ M D , C \ J M .)
Then, there exists 2 .0; 1/such that for each 2 .; 1/, ./ has an SSPE with
the vector of the equilibrium expected average discounted payoffs equal to v .
Proof. A. Preliminaries. Take (arbitrary) v 2 V C , 2 , n 2 f1; : : : ; #.J [ I / C

1g, a.r/ 2 A, r 2 f1; : : : ; ng, and 2 4n satisfying the assumptions of Proposi-
tion 3.4. We can assume without loss of generality that for each r 2 f1; : : : ; ng and

every .j; i / 2 J I with xj i a.r/ D 0, aj.r/ prescribes the making of the contract
proposal .0; 0/ to buyer i and ai.r/ prescribes

the making of the contract proposal

S S
.0; 0/ to producer j . Let EJ D nrD1 EJ x a.r/ , EI D nrD1 EI x a.r/ ,
E D EJ [ EI ,
max
vC j k 2 C 2 2J [I n fJ [ I g ;
wmax
k D max vmax
k ; max ; (3.36)
C \ J ; C \ I
8k 2 E ;
J [I

D min C j k 2C 22
vmin n fJ [ I g ;
wmin
k k ; min
vmin ;
C \ J ; C \ I
8k 2 E ; (3.37)
2 3
n
X X
i D 4r p j i a.r/ pjmax xj i a.r/ 5 ; 8i 2 EI ;
e (3.38)
rD1 j 2J
8 9
P
n
< r ep j i a.r/ pjmax xj i a.r/ j =
i D max .r/ ; 8i 2 EI ; (3.39)
: rD1 ;
j 2 EJ ; 9 r 2 f1; : : : ; ng with xj i a >0
" #
Xn X
.r/

.r/ .r/
j D r e
pj i a pj i xj i a ; 8j 2 EJ ; (3.40)
rD1 i 2I
and
8 9
< P n .r/ .r/ .r/ =
r e pj i a pj i xj i a j
j D max .r/ ; 8j 2 EJ : (3.41)
: rD1 ;
i 2 EI ; 9 r 2 f1; : : : ; ng with xj i a >0
6
As in the case of the maximum in (2.3), the maximum in (3.35) is well defined despite the dis-
continuity of the objective function caused by the disregarding of the fixed costs of the buyers in
.C \ I / nM with zero vector of purchases and the fixed costs of the producers in .C \ J / nM
with zero output. We compute the maximum for each possible subset of C nM (including the empty
set), assuming that all firms in it withdraw from the analyzed market, and then take the maximum
of all such maxima.
It follows from the fact that vi > 0 for each i 2 EI , part (ii) of Assump-
tion 2.3, and (2.6) that for each i 2 EI , there exists j 2 J and r 2 f1; : : : ; ng
with ep j i .a.r/ / < pjmax and xj i .a.r/ / > 0. This implies that i i < 0 for
.r/
each i 2 EI . It follows from the properties of price pj i that j j < 0 for each
j 2 EJ :
For k 2 E and T 2 N consider inequality

.1 / wmax
k wmin
k C 1 T k 0: (3.42)
For the limit case D 1, (3.42) holds as equality. Differentiatingits left hand side

with respect to and evaluating the derivative at D 1, we get wmax k
wmin
k

Tk . This expression is positive if and only if
1
T > wmin
k wk
max
k :
We set
1
T D max min n 2 N j n > wmin
k wk
max
k j k 2 E : (3.43)

Then, for each k 2 E , there exists k 2 .0; 1/ such that for every 2 k ; 1 ,

(3.42) holds for firm k. We set D max k j k 2 E . Further, for each k 2 E ,
there exists k 2 .0; 1/ such that for every 2 k ; 1 ,

1 T vk C k C T vk > 0: (3.44)
(For the limit case D

1, inequality
(3.44) holds. Its left-hand side is continuous in
.) We set D max k j k 2 E .
It follows from (3.35) that for each C 2 2J [I n fJ [ I g with C \ J and
C \ I and every M 2 2C , we have
n
X X
vC;M D r v.r/
C;M vk ; (3.45)
rD1 k2C
with strict inequality if M and there exist either i 2 M \ I , j 2 J nC , and

r 2 f1; : : : ; ng with xj i a.r/ > 0, or j 2 J \ M; i 2 I nC , and r 2 f1; : : : ; ng

with xj i a.r/ > 0.
For each C 2 2J [I n fJ [ I g with C \ J and C \ I and every M 2
2 , which contains either i 2 M \ I such that xj i a.r/ > 0 for some j 2 J nC
C

and r 2 f1; : : : ; ng, or j 2 M \ J such that xj i a.r/ > 0 for some i 2 I nC and
r 2 f1; : : : ; ng, there exists C;M 2 .0; 1/ such that for each 2 C;M ; 1/,
!
X X X
vC;M 1 T vk C k C T vk : (3.46)
k2C k2M \E k2C
(Using (3.45) for M with the above described properties, for the limit case D 1,
(3.46) holds as a strict inequality. Its right-hand side is continuous in .) We set
8 9
< C;M j M 2 2C ; =
C D max 9 .j; i / 2 .J \ M / .I nC / [ .J nC / .I \ M / ;
: ;
& r 2 f1; : : : ; ng with xj i a.r/ > 0
8C 2 2J [I n fJ [ I g with C \ J & C \ I :
We set
n n oo
D max ; ; max C j C 2 2J [I n fJ [ I g ; C \ J ; C \ I :
(3.47)
As in the proof of Proposition 3.1, we set
n o
.r/ D ! 2 0; 1 j .!/ D a.r/
for each r 2 f1; : : : ; ng, and let function 1 W 0; 1 ! f1; : : : ; ng assign to every
! 2 0; 1, r 2 f1; : : : ; ng that satisfies ! 2 .r/.
E
.r/D 2 2 and
For each every r 2 f1; : : : ; ng, let a.r/ j D differ from a.r/
.r/ only in
that ep j i a j D D pj for each .j; i / 2 J .I \ D/ with xj i a
max
> 0 and
.r/ .r/ .r/
e
p j i a j D D pj i for each .j; i / 2 .J \ D/ .I nD/ with xj i a > 0.

(The other prices are those prescribed by a.r/ and x a.r/ j D D x a.r/ .) Note
that a.r/ j D a.r/ for each r 2 f1; : : : ; ng.

For each D 2 2E , define j D by j D .!/ D a.r/ j D, where r D 1 .!/.
Clearly, j D .
It follows from (3.33) (and the fact that for each r 2 f1; : : : ; ng and every j 2

J nEJ x a.r/ , aj.r/ prescribes the withdrawal from the analyzed market) that

gj a.r/ j D 0; 8D 2 2E ; 8j 2 J; 8r 2 f1; : : : ; ng (3.48)
and

gj . j D/ 0; 8D 2 2E ; 8j 2 J: (3.49)
Further, (3.34) implies that

P P
p j i a.r/ j D xj i a.r/ cj
e xj i a .r/
gj a.r/ j D
i 2K i 2K

8r 2 f1; : : : ; ng ; 8j 2 EJ x a.r/ ; 8 K i 2 I j xj i a.r/ > 0 ;

8D 2 2E : (3.50)
B. Description of strategy profile s . We say that a producer j and a buyer i deviated

in a coordinated way in a single period from s 2 S (that will be defined together
with the function k , k 2 E recursively below) when (for an actually observed
signal ! of the public randomizing device) j made to i the same contract proposal
as i made to j and the latter contract proposal differed from the one prescribed by
s and proposed a positive traded quantity.
For each h 2 Hf n fg, let LI .h/ be the set of buyers in EI who, according to h,
deviated (unilaterally or as members of a deviating coalition) from the prescriptions
of s 2 S in the last period recorded in h. Thus, for each t 2 N n f1g and every
h 2 H .t / , we have
n o
LI .h/ D i 2 EI j ai.t 1/ si .h / ! .t 1/ :
(Here, si .h / is buyer i s stage game correlated strategy prescribed by si after

subhistory h of h and si .h / ! .t 1/ is i s stage game pure strategy prescribed
by si .h / for signal ! .t 1/ of the public randomizing device.) Further, for each
h 2 Hf n fg, let LJ .h/ be the set of producers in EJ who according to h, deviated
in a coordinated way with at least one buyer from the prescriptions of s in the last
period recorded in h. We let L .h/ D LJ .h/ [ LI .h/.
In the description of s , we use for each k 2 E , function k W Hf !
f0; 1; : : : ; T g that we define recursively below. For every t 2 N , each h 2 H .t / ,
and every k 2 E , k .h/ is the number of remaining periods, including period t,
for which the punishment of firm k (that started before period t or will start in
period t) has to last.
We set k ./ D 0 for each k 2 E . Next, suppose that there exists t 2
N n f1g such that we have already defined k h for each k 2 E and every
S 1
h 2 tmD1 H .m/ . Consider (arbitrary) h 2 H .t / . Take (arbitrary) k 2 E . If
k L .h/ then k .h/ D max f k .h / 1; 0g. If k 2 L .h/ then k .h/ D T .
Now, we describe the strategy profile s . For each h 2 Hf , we let
D .h/ D fk 2 E j k .h/ > 0g
and set s .h/ D j D .h/.

Clearly,
.s / D v . We have x a.r/ j D 2 X max for each r 2 f1; : : : ; ng and

every D 2 2E . Therefore,
X X

k.h/ s.h/ D vk D ; 8h 2 Hf : (3.51)
k2J [I k2J [I
C. Strategy profile s is an SSPE of ./ with 2 .; 1/ :

C1. Deviations by coalitions of producers. When only producers deviate, a devia-
tion by any k 2 J in any subgame cannot affect the expected average discounted
profit of any j 2 J n fkg in any subgame. (A deviation by k cannot affect j s
single period profit in the period in which it takes place and it has no impact on
the values of function m , m 2 E .) Therefore, in order to show that a coalition

of producers cannot weakly Pareto improve the vector of the expected average dis-
counted profits of its members in any subgame, it is enough to show that no producer
can increase his/her expected average discounted profit in any subgame by a unilat-
eral deviation. As ./ is a game continuous at infinity (see part C of the proof
of Proposition 3.1), the single period deviation principle applies to unilateral devi-
ations. For a unilateral deviation by j 2 J nEJ , the arguments in part C1 of the

proof of Proposition 3.1 also hold here. Thus, consider (arbitrary) D 2 2E and
a unilateral single period deviation by a producer j 2 EJ in the first period of
a subgame .h/ ./ with s .h/ D j D. Since this unilateral deviation does not
change the functional value of function j , it suffices to show that it cannot increase
j s expected profit in the first period of .h/ ./. In order to do so, it is enough
to show that j cannot increase his/her stage game profit by a unilateral
deviation
from a.r/ j D for any r 2 f1; : : : ; ng. If j EJ x a.r/ j D , the argument is
the same as that for the producers in J nEJ . Thus, take (arbitrary) r 2 f1; : : : ; ng

with j 2 EJ x a.r/ j D . Then, j s unilateral deviation from a.r/ j D (which
changes the set of buyers with whom he/she trades otherwise there would be
no change in his/her stage game payoff) either leads to j s withdrawal from the
analyzed market, to no trade between him/her and any buyer without his/her with-
drawal from the analyzed
market, or to
his/her trade with the buyers in a nonempty
set K i 2 I j xj i a.r/ j D > 0 . (He/She cannot trade with any i 2 I with

xj i a.r/ j D D 0 because such i makes the contract proposal .0; 0/ to him/her.)
His/Her profit resulting from the deviation
(when all other firms stick to the prescrip-
tions of a.r/ j D) is 0 gj a.r/ j D in the first case, cj .0/ < 0 gj a.r/ j D

in the second case,and using (3.50), it is not higher than gj a.r/ j D in the third
case.
C2. Deviations by coalitions of buyers. In order to show that a coalition of buyers
cannot weakly Pareto improve the vector of expected average discounted profits of
its members in any subgame, it is enough to show that no buyer can increase his/her
expected average discounted profit in any subgame by a unilateral deviation. (For
each subgame, if no buyer can increase his/her expected average discounted profit
in it by a unilateral deviation, then no coalition of buyers can increase the sum of
their expected average discounted profits. This follows from the summing of the
inequalities in (3.52) and (3.54) over all buyers in the intersection of a deviating
coalition and EI . The deviations by the buyers outside EI cannot affect the stage
game payoffs of the buyers in EI . Further, the deviations by the buyers in EI cannot
affect the stage game payoffs of the buyers outside EI .) As already noted, the single
period deviation principle applies to unilateral deviations. (This principle applies
also to the sum of the expected average discounted profits of a coalition because
such a sum is a scalar.7 ) For a unilateral deviation by i 2 I nEI the arguments in
7
The argument for this claim is analogous to the one for unilateral deviations. Of course, in general
the single period deviation principle does not apply to the deviations by coalitions. Nevertheless, it
applies to the sum of the payoffs of the members of a deviating coalition because the latter sum is
a scalar.
part C2 of the proof of Proposition 3.1 also hold here. Thus, consider a unilateral
single period deviation by a buyer i 2 EI in the first period of a subgame .h/ ./ :
We consider two distinguished cases.
First, i .h/ D 0. Without a deviation, i s expected average discounted profit in
.h/ ./ is at least vi > wmin
i . A deviation gives him/her expected single period profit
in the first period of .h/ ./ that is bounded from above by vmax
i wmax
i and then
he/she is punished for T periods by the single period expected profit vi C i vi C
i . Thus, a deviation gives him/her an expected average discounted profit in.h/ ./
that is not higher than

T
.1 / vmax
i C 1 vi C i C T C1 vi

.1 / wmax
i C 1 T vi C i C T C1 vi

T
.1 / wmin
i C 1 vi C T C1 vi < vi ; (3.52)
where the second weak inequality follows from (3.42). Therefore, a deviation
decreases i s expected average discounted profit in .h/ ./.
Second, i .h/ > 0. Without a deviation, i s expected average discounted profit
in .h/ ./ equals

1 i .h/ vi C i C i .h/ vi 1 T vi C i C T vi > 0: (3.53)
(The right-hand side of (3.53) is positive by (3.44).) A deviation gives him/her a

nonpositive single period expected profit in the first period of .h/ ./. (His/Her
withdrawal from the analyzed market leads to zero single period profit. When he/she
does not withdraw from the analyzed market but does not trade with any producer,
his/her single period profit is ci .0/ < 0. Taking into account (2.6), any stage
game action that leads to him/her trading with at least one producer in EJ gives
him/her a nonpositive single period profit because he/she pays the price pjmax for
each good j 2 EJ that he/she buys.) Then, the punishment (giving him/her single
period expected profit vi C i for T periods) is resumed. Thus, a deviation gives
him/her expected average discounted profit in .h/ ./ that is not higher than

1 T vi C i C T C1 vi < 1 T vi C i C T vi : (3.54)
Therefore, a deviation decreases i s expected average discounted profit in .h/ ./.

C3. Deviations by a coalition C J [I with C \J and C \I . Consider
(arbitrary) h 2 Hf and coalition C J [ I with C \ J and C \ I .
It is enough to show that a deviation by C cannot increase the sum of the expected
average discounted profits of the firms in C in .h/ ./. ./ is a game continuous
at infinity and the sum of the expected average discounted profits of the firms in C
in .h/ ./ is a scalar. Therefore, it suffices to show that for any h 2 Hf , a single
period deviation by C in the first period of .h/ ./ cannot increase the sum of the
expected average discounted profits of the members of C in .h/ ./.8
Of course, a change in prices, at which the producers in C trade with the buy-
ers in C in the first period of .h/ ./, does not change the sum of the single
period profits of the members of C . Moreover, a punishment that it triggers can-
not increase the sum of the expected average discounted profits of the members of
C in .h/ ./. Hence, it is enough to concentrate on the other changes brought about
by the deviation by C .
Let D D fk 2 E j k .h/ > 0g. Then, when the members of C contemplate a
single period deviation in the first period of .h/ ./, they know that without their
deviation, a.r/ j D will occur with probability r . (Recall that they contemplate
a deviation before they observe the signal of the public randomizing device in the
first period .h/ ./.) Let Z be a nonempty subset of C whose members actually
deviate in the first period of .h/ ./. That is, firm k 2 C belongs to Z if and
only if for at least one r 2 f1; : : : ; ng and some interval .r/ of .r/ with a
positive length, it intends to use a stage game pure strategy different from ak.r/ j D.
(A deviation at only one element of .r/ has no affect on the single period profit
in the current period or in the following periods because it takes place with zero
probability.) If (for the given equilibrium strategies of the firms outside C and the
deviation strategies of the firms in C n fi g) a deviation strategy of a buyer i 2 C \ I
leads (for each ! 2 0; 1) to the same quantities purchased by him/her and the same
prices for all goods traded by him/her in positive quantities as the prescriptions of
si , then we do not include i in Z. (Such deviation by i does not increase the sum of
the expected single period profits of the members of C in the first period of .h/ ./.
Thus, even without a punishment of i by the producers outside C , it cannot increase
the sum of the expected average discounted profits of the members of C in .h/ ./.)
We consider three distinguished cases.
First, either there exists no coordinated deviation and Z \ I \ D .h/ D , or for
eachpair of j 2 Z \ J and i 2 Z \ I who deviated in a coordinated way,we have
xjk a.r/ D 0 for each k 2 I nC and every r 2 f1; : : : ; ng and xki a.r/ D 0 for
each k 2J nC and every r 2 f1; : : : ; ng; further, for each i 2 Z \ I \ D .h/, we
have xki a.r/ D 0 for each k 2 J nC and every r 2 f1; : : : ; ng. This corresponds

to the situation in (3.45) with either M D or M but xjk a.r/ D 0 for

each .j; k/ 2 .M \ J / .I nC / and every r 2 f1; : : : ; ng and xki a.r/ D 0 for
each .k; i / 2 .J nC / .M \ I / and every r 2 f1; : : : ; ng. It follows from (3.45)
8
When a deviation by a coalition C lasts for several periods or is infinite, in any single period, in
which it takes place, it can happen that only the firms in a subcoalition Z of C intend to behave
in a way that differs from the prescriptions of their equilibrium strategies. Nevertheless, in order
to show that a deviation by C cannot increase the sum of the expected average discounted profits
of the members of C (in a subgame in which it takes place), we have to consider all single period
deviations by C (in the first period of the analyzed subgame), including those in which only the
firms in the subcoalition Z of C intend to behave in a way that differs from the prescriptions of
their equilibrium strategies.
that a deviation does not increase the sum of the expected single period profits of
the members of C in the first period of .h/ ./. (If D .h/ and according to s ,
the firms in C n .Z \ I \ D .h// are punishing the firms outside C or the firms in
C n .Z \ I \ D .h// are punished by the firms outside C , then such punishments
which are triggered by the deviations contained in h have the same effect on the
sum of the expected profits of the members of C both with and without a deviation.
The punishments of the producers in C by the buyers in C and the punishments of
the buyers in C by the producers in C do not affect the sum of the expected profits
of the members of C .) If a deviation triggers a punishment of a buyer in Z by the
producers outside C in the following T periods, then such a punishment decreases
the sum of the expected average discounted profits of the members of C in .h/ ./.
Thus, a deviation cannot increase the sum of the expected average discounted profits
of the members of C in .h/ ./.
Second, there exists either j 2 .Z \J / nD .h/, who deviated in a coordinated
way with some buyer and satisfies xjk a.r/ > 0 for some k 2 I nC and some

r 2 f1; : : : ; ng or i 2 .Z \ I / nD .h/ who satisfies xki a.r/ > 0 for some
k 2 J nC and some r 2 f1; : : : ; ng. Let Z be the set of all producers and buyers
in Z who satisfy the condition described above. Take (arbitrary) k 2 Z . A single
period gain (in terms of the sum of their expected profits) of the members of C from
a deviation cannot exceed wmaxk
wmin
k
. It follows from (3.42) that this gain is wiped
out by the punishment of k in the following T periods. (The punishment of any
k 2 Z by at least one firm outside C is sufficient for this. Note that we use k , and
not k , in (3.42). If D .h/ , the punishment of the firms in Z affects neither
the punishments of the firms outside C by the firms in C nZ nor the punishments of
the firms in C nZ by the firms outside C triggered by the deviations contained in h.
If a deviation by a firm in ZnZ leads to the starting or resuming of its punishment
by the firms outside C (as described in the analysis of the following case), then such
a punishment further decreases the sum of the expected average discounted profits
of the members of C in .h/ ./.) Thus, a deviation cannot increase the sum of the
expected average discounted profits of the members of C in .h/ ./.
Third, there exists either j 2 Z \ J \ D .h/ with xjk a.r/ > 0 for some

k 2 I nC and some r 2 f1; : : : ; ng or i 2 Z \ I \ D .h/ with xki a.r/ > 0 for
some k 2 J nC and some r 2 f1; : : : ; ng. Let Z C be the set of all producers and
buyers in Z who satisfy the condition described above. Moreover, in this case, the
set Z ; described in the analysis of the second case, is empty. This case corresponds
to the situation in (3.45) with M , which is equal to the union of Z C with the set of
producers in ZnZ C who coordinated their deviation with some buyer in Z C and
the set of buyers in ZnZ C who coordinated their deviation with some producer
in Z C . (Since Z D , we have xjk a.r/ D 0 for each j 2 M nZ C , every

k 2 I nC , and each r 2 f1; : : : ; ng, as well as xki a.r/ D 0 for each i 2 M nZ C ,
every k 2 J nC , and each r 2 f1; : : : ; ng.) Ignoring the punishments of the firms in
C nZ C by the firms outside C and the punishments of the firms outside C by the
firms in C nZ C triggered by the deviations contained in h,9 the sum of the expected
average discounted profits of the members of C without a deviation is
X X X X
vk C 1 k .h/ k vk C 1 T k ; (3.55)
k2C k2Z C k2C k2Z C
where
2 3
n
X X
i D r 4 p j i a.r/ pjmax xj i a.r/ 5 ; 8i 2 I \ Z C (3.56)
e
rD1 j 2J nC
and
2 3
n
X X
.r/

j D r 4 ep j i a.r/ pj i xj i a.r/ 5 ; 8j 2 J \ Z C :
rD1 i 2I nC
(3.57)
We use zC to denote the sum of the expected stage game profits of the members
of C in the first period of .h/ ./ resulting from a deviation. Using (3.46), zC is
not higher than the right-hand side of (3.55). The punishment of the firms in Z C is
resumed from the second period of .h/ ./. Thus, the sum of the expected average
discounted profits of the firms in C resulting from a deviation (again, ignoring the
punishments of the firms in C nZ C by the firms outside C and the punishments of
the firms outside C by the firms in C nZ C triggered by the deviations contained in
h, as well as new punishments of the buyers in ZnZ C by the firms outside C that
further decrease the sum of the expected average discounted profits of the firms in
C in .h/ ./) is
2 3
X X X X
.1 / zC C 4 vk C 1 T k 5 vk C 1 T k :
k2C k2Z C k2C k2Z C
(3.58)
It follows from (3.55) and (3.58) that a deviation cannot increase the sum of the
expected average discounted profits of the firms in C in .h/ ./.
C4. Payoff vector in each subgame is strictly
Pareto efficient. This argument follows

from (3.51) and the fact that
k.h/ s.h/ 2 V C for each h 2 Hf . t
u
Let v be the equilibrium vector of the expected average discounted profits in

an SSPE of ./. Of course, the requirement of individual rationality implies that
9
The punishments of the firms in C nZ C by the firms outside C and the punishments of the firms
outside C by the firms in C nZ C triggered by the deviations contained in h have the same effect
on the sum of the expected average discounted profits of the members of C both with and without
a deviation.
all components of v are nonnegative. Since v 2 V C , this can be obtained from a

profile of stage game correlated strategies under which no more than # .J [ I /C1
profiles of stage game pure strategies occur with a positive probability. The other
conditions in the statement of Proposition 3.4 are not necessary for the existence of
an SSPE of ./ with the equilibrium vector of the expected average discounted
profits equal to v but (taken together) they are sufficient for it. They make possible
the use of our method of proof of Proposition 3.4. Now, we give a brief intuitive
outline of the latter method.
Unilateral deviations by buyers with a positive equilibrium expected average dis-
counted profit, as well as their deviations when they are members of a deviating
coalition of buyers, are punished in the same way as in the proof of Proposition 3.1
by increasing their purchasing prices for a finite number of periods. This number of
periods is high enough to wipe out any gain from a deviation. Unilateral deviations
by these buyers, as well as their deviations as members of a deviating coalition of
buyers, during their punishment are punished by resuming the punishment. (Buyers
with zero equilibrium expected average discounted profit withdraw from the ana-
lyzed market with probability one along the continuation equilibrium path in any
subgame. No producer proposes a contract to them. Therefore, they cannot increase
their expected stage game profit by a unilateral deviation or by taking part in a
deviation by a coalition of buyers.)
All producers, who along the equilibrium path trade a positive quantity of their
product with some buyer, have a positive equilibrium expected average discounted
profit. We denote the set of such producers by EJ . Similarly, all buyers, who along
the equilibrium path purchase a positive quantity of at least one good, have a positive
equilibrium expected average discounted profit. We denote the set of such buyers
by EI .
The equilibrium strategy profile s described in the proof of Proposition 3.4 pre-
scribes the same quantity traded between producer j 2 EJ and buyer i 2 EI in
each period and after every history leading to it. It can prescribe one of the three
prices for a trade between them. It prescribes the highest possible price when j is
punishing i for the latters deviation. It prescribes a normal (intermediate) price
when neither i nor j is being punished. It prescribes the lowest price when j is
punished for a deviation coordinated with some buyer k 2 I n fi g. (A deviation by
producer j with buyer k is coordinated if j made to k the same contract proposal
as k made to j and the latter proposal was different from the one prescribed by
s and proposed a positive traded quantity.) Such coordinated deviation is the only
reason for the punishment of a producer in EJ . As explained below, other types of
a deviation by a producer in EJ cannot increase his/her expected stage game profit.
(The producers with zero equilibrium expected average discounted profit i.e., the
producers outside EJ withdraw from the analyzed market with probability one
along the continuation equilibrium path in any subgame. No buyer proposes a con-
tract to them. Therefore, they cannot increase their expected stage game profit by a
unilateral deviation or by taking part in a deviation by a coalition of producers.)
Conditions (3.32) and (3.33) imply that (when the firms follow s ) each producer
in EJ earns a nonnegative stage game profit in each period for every signal of the
public randomizing device. Therefore, he/she cannot increase his/her expected stage
game profit by withdrawing from the analyzed market or by a deviation that leaves
him/her with zero sales without withdrawing from the analyzed market. Condition
(3.34) implies that he/she cannot increase his/her expected stage game profit by a
deviation that (is unilateral or is a part of a deviation by a coalition of producers
and) for some signals of the public randomizing device, reduces the set of buyers
with whom he/she trades (while keeping unchanged the quantities traded with those
buyers with whom he/she still trades and the prices at which he/she trades with
them). Condition (3.34) is satisfied, for example, if the producers cost functions are
differentiable and convex and for each producer in EJ , the lowest price of his/her
product (that can be prescribed by s ) is not lower than his/her marginal cost at his
total output in each period (prescribed by s /.
The statement of Proposition 3.4 contains also the requirement that each profile
of stage game pure strategies that occurs with a positive probability when the firms
follow generates a vector of traded quantities that maximizes the sum of the
firms stage game profits. This requirement implies that v maximizes the sum of
the firms payoffs over V C (i.e., over the set of all vectors of the expected average
discounted profits). Therefore, v is strictly Pareto efficient. During the punishment
only prices and not traded quantities are changed. Thus, the sum of the firms contin-
uation equilibrium expected average discounted profits is the same in each subgame.
Therefore, the continuation equilibrium vector of the expected average discounted
profits in each subgame is strictly Pareto efficient.
The condition (3.35) ensures that no coalition C (other than the grand one),
which contains at least one producer and at least one buyer, can (in any subgame)
increase the expected average discounted profit of at least one of its members with-
out decreasing the expected average discounted profit of another member. (It also
takes into account the possibility of withdrawal from the analyzed market by some
firm(s) in C .) It contains as special cases three conditions. (See also (3.45) and
(3.46). First, C cannot increase the sum of the expected stage game profits of its
members by a deviation that only replaces some positive quantities traded by its
members with zero. (This is the case of empty M .10 This is the first subcase of the
first case considered in part C3 of the proof of Proposition 3.4.) Second, C cannot
increase the sum of the expected stage game profits of its members by any deviation
if it does not contain a buyer (a producer) who (according to the prescriptions of
s ) trades a positive quantity with a producer (a buyer) outside C with a positive
probability. (This is the case when M is nonempty but no firm in it trades accord-
ing to the prescriptions of s with a firm outside C with a positive probability. It is
the second subcase of the first case considered in part C3 of the proof of Proposi-
tion 3.4.) Third, a single period deviation cannot give the members of C the sum of
the expected stage game profits equal or higher than the sum of their expected aver-
age discounted profits generated by s if at least one buyer (producer) in C trades
10
In deviations considered in (3.35), only a pair of a producer and a buyer, who belong to M , can
trade a positive quantity different from the one prescribed by s . If either the producer or the buyer
does not belong to M , a deviation can only replace a traded quantity prescribed by s with zero.
(according to the prescriptions of s ) a positive quantity with a producer (a buyer)

outside C , and when a deviation takes place, each such buyer (producer) is supposed
(according to s ) to trade with the producers (buyers) outside C at the maximal pos-
sible prices (minimal possible prices) that can be prescribed by s . (This is the case
of nonempty M containing at least one firm that trades according to the prescrip-
tions of s with a firm outside C with a positive probability. It corresponds to the
third case considered in part C3 of the proof of Proposition 3.4.) In this case, C
also satisfies the following condition: no buyer (producer) k 2 C , who is supposed
(according to s ) to trade with at least one producer (buyer) outside C and is cur-
rently not being punished, deviates (in a way coordinated with some buyer). When
the latter condition is not satisfied C may be able to increase the sum of the expected
stage game profits of its members by a deviation. (This corresponds to the second
case considered in part C3 of the proof of Proposition 3.4.) The strategy profile s
handles such deviation by meting a punishment to firm k that wipes out the increase
in the sum of the expected stage game profits of the firms in C resulting from the
deviation.
Now, we give an example of a stage game that satisfies the requirements of
Proposition 3.4.11
3.3. We have J D f1; 2; 3g, I D f4; 5g ; Yj D 0; 20 for each j 2 J ,

Example
cPj yj D 10yj C 50 for every j 2 f1; 2g, c3 .y3 / D 10y3 C 79, and ci .xi / D
j 2J xj i C 20 for each i 2 I . Every i 2 I produces one type of good and its pro-
duction function fi W Xi ! 0; 20 has the form fi .xi / D min fx1i C x3i ; x2i g. The
inverse demand functions for the buyers outputs, P4 W <2C ! <C and P5 W <2C !
<C , have the form P4 .q4 ; q5 / D max f42 2q4 C q5 ; 0g and P5 .q4 ; q5 / D
max f42 C q4 2q5 ; 0g. We have Ui .x/ D Pi .f4 .x4 / ; f5 .x5 // fi .xi / for each
i 2 f4; 5g.
In this example, X max is a singleton. Its unique element x max has the following
components: x14 max
D x15
max
D 10, x24 max
D x25
max
D 10, and x34 max
D x35max
D 0.

This gives D 60. Thus, n D 1 and reduces to (it assigns probability one
to) a 2 A with x .a / D x D x max . We set e p 14 .a / D e
p 15 .a / D 13:9,

e
p 24 .a / D e p 25 .a / D 13:9, and e p 35 .a / D 0. Then v D g .a / D
p 34 .a / D e

.28; 28; 0; 2; 2/. Thus, a1 prescribes the making of the contract proposal .13:9; 10/
to each buyer, a2 prescribes the making of the contract proposal .13:9; 10/ to each
buyer, and a3 prescribes the making of the contract proposal .0; 0/ to each buyer.
For every i 2 I , ai prescribes the making of the contract proposal .13:9; 10/ to
producers 1 and 2 and the making of the contract proposal .0; 0/ to producer 3.
Here, EJ D f1; 2g and EI D f4; 5g. As such, vi D 2 for each i 2 EI and a
satisfies (3.32).
In this example, (approximately)
maxfU4 .x/ j x 2 X g D maxfU5 .x/ j x 2 X g D 320:03:
11
Example 4.3 in the following chapter also contains a stage game satisfying the assumptions of
Proposition 3.4.
Thus, the maximal prices of the producers products given by p1max D p2max D

p3max D 32:03 satisfy (2.6). We set p14 D p15 D p24 D p25 D 12:5. Since this
price is equal to the average costs of producers 1 and 2 when their production is 20
units (and 1 .x / D 2 .x / D 20), (3.33) is satisfied. As this price exceeds the
constant marginal cost of each producer, (3.34) holds. It remains to show that (3.35)
holds.
Consider first a deviating coalition containing one buyer and producers 1 and 2.
Since the buyers are identical, we can take C D f1; 2; 4g without loss of generality.
As 5 C , each of the producers 1 and 2 can deliver to buyer 5 either 10 units or
nothing. On the other hand, they can agree with buyer 4 on any feasible (i.e., not
exceeding their production capacity) deliveries. If both of them agree with buyer 4
on the delivery of a positive quantity other than 10 units, we have M D f1; 2; 4g. If
only one of them delivers 10 units to firm 5, the latter firm produces nothing and the
maximum sum of the profits of the firms in C equals
X
f1;2;4g C 12:5 10 10 10 D 70 C 25 D 45 < 58 D gk a : (3.59)
k2C
(In this case, each of the producers 1 and 2 delivers 5 units to buyer 4.) If both of
them do so, firm 5 produces 10 units and optimal traded quantities within C result
from the solution to the maximization program12
max .52 2x14 / x14 22x14 70 subject to 0 x14 10 (3.60)
(The sum of the fixed costs of the members of C is 120. The sum of the variable
costs of producers 1 and 2 when their production is 10 units for buyer 5 equals 200.
The revenue from the sale of these quantities to buyer 5 at price 12.5 equals 250.
Subtracting the above given cost from this revenue gives 70.) The solution of this
maximization programP is x14 D 7:5, giving the sum of the profits of the firms in C
equal to 42:5 < 58 D k2C gk .a /. If none of the producers 1 and 2 trades with
buyer 5 and no firm in C withdraws from the analyzed market, the maximal sum
of the profits of the members of C is f1;2;4g D 70. If buyer 4 withdraws from
the analyzed market (and hence, M D ), the maximal sum of the profits of the
firms in C is 0. (This sum is obtained when producers 1 and 2 too withdraw from
the analyzed market. Trading only 10 units with buyer 5 at price 13.9 gives each
producer profit 11.) If one of the producers 1 and 2 withdraws from the analyzed
market, the maximal sum of the profits of the firms in C equal 0. (Again, this sum
is obtained by the withdrawal of all firms in C from the analyzed market. Without
trading with both producer 1 and producer 2, firm 4 cannot produce anything.)
The result for the deviating coalition f2; 3; 4g is worse than that for C D f1; 2; 4g.
Producer 3 has a higher fixed cost than producer 1 by 29 financial units, whereas
12
Taking into account the form of production function f4 , the maximization of the sum of the
profits of the members of C requires x14 D x24 .
g3 .a / D g1 .a / 28. He/She cannot trade with buyer 5 C . The result for
the deviating coalition f1; 2; 3; 4g is not better than that for C D f1; 2; 4g. Producer
3 cannot trade with buyer 5. Taking into account the form of production function
f4 , producing the sum of the outputs of firms 1 and 3 exceeding 20 units cannot
increase the sum of the profits of the members of coalition f1; 2; 3; 4g. The sum of
the outputs of firms 1 and 3 not exceeding 20 units is produced with lower cost when
only firm 1 produces it. Similarly, the best result for coalition f1:3:4g is not better
than the best result for coalition f1; 4g given below.
Next, consider a deviating coalition containing one buyer and one of the produc-
ers 1 and 2. We take C D f1; 4g without loss of generality.13 Firm 4 can buy either
10 units or nothing from producer 2. If it buys nothing, its output is zero and the
maximal sum of the profits of the firms in C equals zero. (This sum is obtained
by the withdrawal of both firms in C from the analyzed market. Selling 10 units
of the output of firm 1 to buyer 5 at price 13.9 financial units gives profit 11.) If
it buys 10 units from producer 2 and buys a positive amount other than 10 units
from producer 1 (and hence, M D f1; 4g), it has to buy from producer 2 at price
p2max D 32:03. Further, in this case, firm 1, if it sells 10 units to buyer 5, has to

sell them at price p15 D 12:5. Taking into account (2.6), the difference between
the revenue and the costs of firm 4 including the expenditure on the purchase of
10 units from firm 2 but excluding the expenditure on the purchase from firm 1 is
negative. The profit of firm 1 without the revenue from the sales to buyer 4 is lower
than 12:5 10 10 10 50 D 25. Hence, the sum of the profits of the members
of C is negative. If firm 4 buys 10 units from producer 2 and also 10 units from
producer 1 (and hence, M D ), then a deviation can occur only in the refusal of
producer 1 to trade with firm 5. Then, firm 5s output equals zero and the sum of the
profits of the members of C is 109.
Now, consider a deviating coalition containing both buyers and one of the pro-
ducers, 1 or 2. We take C D f1; 4; 5g without loss of generality.14 If C buys nothing
from producer 2, the output of the firms 4 and 5 is zero and the maximal sum of
the profits of the members of C equals zero (obtained when all of them withdraw
from the analyzed market). If one of the firms 4 and 5 buys nothing from producer 2
and the other one buys 10 units from him without loss of generality assuming that
5 buys nothing from 2 firm 5s withdrawal from the analyzed market is optimal
for C . Then, the result for C is not better than that for coalition f1; 4g when producer
1 sells nothing to buyer 5 and buyer 4 buys 10 units from producer 2. As already
shown, the sum of profits for f1; 4g is negative in this case. Hence, the sum of the
profits of the members of C is negative. Next, suppose that both firms 4 and 5 buy 10
units each from producer 2. Then, both will buy a positive quantity from producer 1
13
The best result for a deviating coalition containing one buyer and producer 3 is worse than that
for coalition f1; 4g : Producer 3 cannot trade with the buyer outside the deviating coalition. Further,
his/her fixed cost exceeds the fixed cost of producer 1 by 29 > g1 .a / g3 .a /.
14
The best result for C is better than that for coalition f3; 4; 5g. The fixed cost of producer 3
exceeds the fixed cost of producer 1 by 29 > g1 .a / g3 .a /. Taking into account the form of
production functions f4 and f5 , the best result for coalition f1; 3; 4; 5g is not better than the best
result for C .
and at least one of them buys a quantity different from 10 units. (If both buy 10 units
from producer 1, there would be no deviation. If any of them was buying nothing
from producer 1, then its output would be zero and the best result for coalition C
would be worse than the best result when such a buyer withdraws from the analyzed
market. In this case, the requirement (3.35) cannot be violated because, as is already
shown, it is not violated for coalition f1; 4g and gi .a / > 0 for each i 2 f4; 5g.)
First, suppose (without loss of generality) that firm 5 buys a positive quantity other
than 10 units from firm 1 and firm 4 buys 10 units from firm 1. As such, M D f1; 5g
and firm 5 buys from producer 2 at price p2max . Taking into account (2.6), the dif-
ference between the revenue and the cost of firm 5 (excluding the expenditure on
the purchase from firm 1 but including the expenditure on the purchase from firm
2) is negative. Thus, the sum of the profits of the members of coalition C is lower
than that in the case of the withdrawal of firm 5 from the analyzed market. We have
already shown that the sum of the profits of the members of C is negative in such
a case. Second, suppose that both firms 4 and 5 buy a positive quantity other than
10 units from firm 1. Then, M D f1; 4; 5g and both firms 4 and 5 buy at price p2max
from producer 2. Taking into account (2.6), for both firms 4 and 5, the difference
between the revenue and the cost (excluding the expenditure on the purchase from
firm 1 but including the expenditure on the purchase from firm 2) is negative. Thus,
the sum of the profits of the members of C is negative.
Finally, consider deviating coalition C D f2; 3; 4; 5g. If producer 3 produces
nothing and C does not trade with producer 1, then firms 4 and 5 produce nothing
and the maximal sum of the profits of the members of C is zero (obtained when all
members of C withdraw from the analyzed market.) When firm 3 withdraws from
the analyzed market, the result for coalition C is not better than the corresponding
result15 for coalition f2; 4; 5g (which is the same as the corresponding result for
coalition f1; 4; 5g analyzed above). Thus, suppose that producer 3 has a positive
output and C does not trade with producer 1. (This gives a sum of the profits of the
members of C higher than the positive output of producer 3 and the trade between
C and producer 1. The reason for this is that the constant marginal cost of producer
3 is lower than price 13.9 the minimum price at which buyers in C would have to
buy from producer 1.) Taking into account the unique element of X max and the fact
that producer 3 differs from producer 1 only in terms of the fixed cost, the maximal
sum of the profits of the members of C in this case is
.c3 .0/ c1 .0// D 60 29

X
D 31 < 32 D gk a : (3.61)
k2C
15
By the corresponding result we mean the result obtained from the same trades between the
members of f2; 4; 5g and firms outside C .
Chapter 4
Efficiency of an SRPE and an SSPE
In this chapter, we deal with the efficiency and welfare properties of an SRPE. Since
each SSPE is also an SRPE, the results also hold for the former.
In Sects. 4.1 and 4.2, we analyze the cost (or transformational) efficiency of an
SRPE and an SSPE. First, in Sect. 4.1, we partition the set of goods into groups of
identical goods, which we call types of goods. We show that in an SRPE, in each
subgame in which the sum of the firms expected average discounted profits is max-
imized, in every period along the equilibrium path, the active producers (i.e., those
with a positive output) of each type of good can produce their total output prescribed
by the equilibrium strategy profile with lower or the same total production cost as
any other group of producers of the same type of good. Thus, in a SRPE, in each
such subgame of ./, in every period along the equilibrium path, for each type
of good, the group of its active producers forms a natural oligopoly (as defined by
[Baumol (1982)]) or (if there is only one active producer) a natural monopoly.
Second, in Sect. 4.2, we concentrate on the case when all buyers are retailers.
We show that in an SRPE, in each subgame in which the sum of the firms expected
average discounted profits is maximized, in every period along the equilibrium path,
the active buyers (i.e., those with a nonzero vector of purchased quantities) can sell
the vector of their sales prescribed by the equilibrium strategy profile with lower
or the same total selling cost as any other group of buyers. Thus, in an SRPE, in
each such subgame of ./, in every period along the equilibrium path, the group
of active buyers forms a natural oligopsony or (if there is only one active buyer) a
natural monopsony.
In Sect. 4.3, we deal with the properties of an SRPE in our model with respect
to consumer welfare. We compare them with the welfare properties of the static
(i.e., single period) non-collusive equilibria under imperfect competition. Namely,
we use two benchmarks: (1) a monopsony choosing the purchased quantities on the
demand side of the analyzed market coupled with price taking behavior on its supply
side, and (2) Cournot oligopoly on the supply side of the analyzed market coupled
with price taking behavior on its demand side. We think that as many real world
industries are oligopolies or are formed by monopsonists with their own networks
of suppliers and markets with perfect competition (or something close to it) on
both sides are rare this is a relevant comparison. We identify the conditions under

54 4 Efficiency of an SRPE and an SSPE
which a movement from the benchmark to a collusive outcome in an SRPE increases

consumer welfare.
We keep Assumptions 2.12.4 in this chapter.
4.1 Natural Oligopoly

In this section, we assume that there exists a partition J .1/ ; : : : ; J .m/ of J with
the property that for each k 2 f1; : : : ; mg, J .k/ contains goods that are identical and
that differ from the goods contained in J .r/ for each r 2 f1; : : : ; mg n fkg. We call
each k 2 f1; : : : ; mg a type of good and view the firms in J .k/ as the industry
producing type k of good. It is natural to assume that the identity of the suppliers of
any type of good does not affect the buyers revenue and cost functions. Therefore,
in Sect. 4.1, we make, besides Assumptions 2.12.4, the following assumption.

Assumption 4.1. P (i) For each iP2 I , we have ci .xi / D ci xi for every xi ; xi 2
Xi2 satisfying j 2J .k/ xj i D j 2J .k/ xj i for each k 2 f1; : : : ; mg. (ii) For each
P
i 2 I , we have Ui .x/ D Ui x for every x; x 2 X 2 satisfying j 2J .k/ xjr D
P
j 2J .k/ xjr for each k 2 f1; : : : ; mg and every r 2 I .

First, we give the definition of a natural oligopoly in a single product industry

(tailored to our model).
P i
Definition 4.1. Consider k 2 f1; : : : ; mgand output Qk 2 0; j 2J .k/ max j .

A set of firms J .k/ J .k/ with # J .k/ 2 is a natural oligopoly in the indus-
try producing
Q type k of good for output level Qk if there exists an output vector
y .k/ 2 j 2J .k/ Yj such that
X
yj.k/ D Qk (4.1)
j 2J .k/
and
X X Y X
cj yj.k/ cj yj ; 8y 2 Yj with yj D Qk :
j 2J .k/ j 2J .k/ Wyj >0 j 2J .k/ j 2J .k/
(4.2)
Thus, a group of firms in the industry

P producing i type k of good is a natural
oligopoly for output level Qk 2 0; j 2J .k/ j max
if it can produce output Qk
with a lower or the same cost as any other group of firms (not excluding a single-
ton group) in the industry. The difference between our and the standard ([Baumol
(1982)]) definition of a natural oligopoly lies in allowing the firms in an industry
to have different cost functions, as well as in taking into account the capacity con-
straints and an upper bound on the number of firms capable of producing a given
4.1 Natural Oligopoly 55
type of good.1 Even if the firms in an industry are identical (i.e., if they have the
same cost function and the same capacity), the capacity constraints can lead to the
situation where the firms in a natural oligopoly do not have the same output (com-
ponents of vector y .k/ are not equal). For example, suppose that all firms in the
industry producing type k of good are identical and without capacity constraints
one firm would produce output Qk with lower cost than any group of two or more
firms because of the strictly concave cost function but the capacity of each firm in
the industry is from the interval .0:5Qk ; Qk /. Then, there exists a natural duopoly
for output Qk , in which one firm produces an output equal to its capacity and the
other an output equal to the difference between Qk and its capacity.
Next, we give the definition of a natural monopoly in a single product industry
(tailored to our model).
P i
Definition 4.2. Consider k 2 f1; : : : ; mgand output Qk 2 0; j 2J .k/ max j .
A firm r 2 J .k/ is a natural monopoly in the industry producing type k of good
for output level Q.k/ if
Q.k/ r (4.3)
and
X Y X
cr Q.k/ cj yj ; 8y 2 Yj with yj D Qk : (4.4)
j 2J .k/ Wyj >0 j 2J .k/ j 2J .k/
Thus, a firm in the

industry
P i type k of good is a natural monopoly for
producing
output level Qk 2 0; j 2J .k/ j max
if it can produce output Qk with a lower or
the same cost as any other firm or group of firms in the industry.
First, we show that the maximization of the sum of the firms stage game profits
leads to the production of each type of good with a positive output by a natural
oligopoly or a natural monopoly.
Lemma 4.1. Let x max 2 X max . For each k 2 f1; : : : ; mg with

# J .k/ \ EJ .x max / 2
J .k/ \ EJ .xP
max
/ is a natural oligopoly in the industry producing type k of good for
output level j 2J .k/ j .x max /. For each k 2 f1; : : : ; mg with

# J .k/ \ EJ .x max / D 1
the only firm in J .k/ \ EJ .x max

P/ is a natural monopoly in the industry producing
type k of good for output level j 2J .k/ j .x max /.
1
An analogous comment applies to the definition of a natural monopoly (Definition 4.2) below.
Proof. Take (arbitrary)

P x max 2 X max and k 2 f1; : : : ; mg with J .k/ \ EJ .x max /
. Let Qk D j 2J .k/ j .x max /. Suppose that the claim of Lemma 4.1 does not
.k/C
hold. Then, P J
Q there exists a nonempty J .k/ with J .k/C J .k/ \ EJ .x max /
and y 2 j 2J .k/C Yj such that j 2J .k/C yj D Qk and
X X
cj yj < cj j .x max / :
j 2J .k/C j 2J .k/ \EJ .x max /
Therefore, there exists x 2 X satisfying xj i D xjmax for each .j; i / 2

.k/
.k/ .k/C i
J nJ I , xj i D 0 for every .j; i / 2 J nJ

I , j x D yj for
P P
each j 2 J .k/C , and j 2J .k/C xj i D j 2J .k/ \EJ .x max / xj i for every i 2 I .
max
These properties of y and x imply that EI .x/ D EI .x / and

max
X X X
Ui x ci xi cj j x
i 2EI .x / i 2EI .x / j 2EJ .x /
X X
> Ui .x max / ci ximax
i 2EI .x max / i 2EI .x max /
X
cj j .x max / : (4.5)
j 2EJ .x max /
This contradiction with the definition of X max in (2.3) completes the proof. t
u
The basic idea of the proof of Lemma 4.1 is simple. Suppose that the produc-
tion of type k of good takes place neither in a natural oligopoly nor by a natural
monopoly. Then, some group of its producers, which differs from J .k/ \ EJ .x max /,
can produce the same quantity with a lower cost. Thus, after a change in the set of
the active producers of type k of good, it is possible to reduce the cost of its pro-
duction while keeping unchanged its total output, outputs of all firms capable of
producing other types of goods, and buyers purchases of each type of good. This
increases the sum of the stage game profits of all firms (which equals the difference
between the sum of the buyers revenues and the sum of the production cost and
the buyers costs determined by their cost functions). Thus, we have a contradiction
with the fact that the latter sum is maximized for each element of X max .
Now, we show that an SRPE in ./ leads in each period along the equilibrium
path in every subgame, in which the sum of the firms expected average discounted
profits is maximized to the production being carried out in a natural oligopoly
or by a natural monopoly in each industry with a positive output. This result is the
consequence of the preceding lemma.
Proposition 4.1. Consider an SRPE s of ./, h 2 Hf , and the sequence

x D fx .t/gt 2N 2 X 1 of the vectors of the actually traded quantities in .h/ ./
4.1 Natural Oligopoly 57

generated by the prescriptions of s.h/ . Assume that
X

r.h/ s.h/ D : (4.6)
r2J [I
Then, for each k 2 f1; : : : ; mg and every t 2 N with

n o
J .k/ .t/ D j 2 J .k/ j j .x .t// > 0 ;
J .k/ .t/ is
Pa natural oligopoly in the industry producing type k of good for out-
put level j 2J .k/ j .x .t// if # J .k/ .t/ 2 or the only firm in J .k/ .t/ is
a natural monopoly
P in the industry
producing type k of good for output level
.k/
j 2J .k/ j .x .t// if #J .t/ D 1. t
u
Proof. Consider (arbitrary) SRPE s of ./, h 2 Hf , the sequence x 2 X 1 of

the vectors of the actually traded quantities in .h/ ./ generated by the prescrip-

tions of s.h/ , and k 2 f1; : : : ; mg. Note that the sum of the firms expected average
discounted profits (which equals the average discounted sum of their stage game
expected profits) along the equilibrium path in .h/ ./ is maximized if and only
if the sum of their stage game expected profits is maximized in each period. Thus,
(4.6) implies that x .t/ 2 X max for each t 2 N . Therefore, the claim of Proposi-
tion 4.1 follows from Lemma 4.1. t
u
Propositions 4.1 and 3.3 have an obvious corollary.
Corollary 4.1. Assume that # .I / D 1. Consider an SSPE s of ./, h 2 Hf ,

and the sequence x D fx .t/gt 2N 2 X 1 of the vectors of the actually traded

quantities in .h/ ./ generated by the prescriptions of s.h/ . Then, for each k 2
.k/
.k/

f1; : : : ; mg and every t 2 N with J .t/ D j 2 J j j .x .t// > 0 ,
J .k/ .t/ isPa natural oligopoly in the industry producing type k of good for out-
put level j 2J .k/ j .x .t// if # J .k/ .t/ 2 or the only firm in J .k/ .t/ is
a natural monopoly in the industry
P producing type k of good for output level
.k/
j 2J .k/ j .x .t// if # J .t/ D 1.
Proof. It follows from the assumptions of Corollary 4.1 and from Proposition 3.3
that (4.6) holds. Since s is an SSPE of ./, it is also an SRPE of ./. Thus, the
claim of Corollary 4.1 follows from Proposition 4.1. t
u
Thus, when the analyzed market is monopsonistic, in each SSPE of ./, in

every period along the equilibrium path in each subgame, each type of good with a
positive output is produced in a natural oligopoly or by a natural monopoly.
4.2 Natural Oligopsony
A natural oligopsony is an analogue of a natural oligopoly on the demand side of

a market. Similarly, a natural monopsony is an analogue of a natural monopoly on
the demand side of a market. We apply these concepts to the buyers in our model
because they operate on the demand side of the analyzed market. We define them (in
a way tailored to our model) for a given vector of quantities of the goods purchased
from the producers in our model. In this section, we restrict our attention to the
buyers in the analyzed market who are retailers. (In this case, the evaluation of the
efficiency of a market structure of the retail market on the basis of the sum of selling
costs, without taking into account the expenditure on the goods that are being sold,
is justified.) Therefore, besides keeping Assumptions 2.12.4, in this section, we
make the following assumption.
Assumption 4.2. All buyers in I are retailers. For each i 2 I , function Ui has the
form X
Ui .x/ D Pj . .x// xj i ;
j 2J
where Pj W Y ! <C is the inverse demand function for good j 2 J in the retail
market.
The definitions of a natural oligopoly and a natural monopoly are based on the
minimization of the sum of the production costs for a given industry output. The
producers cost functions are defined under the assumption that they are price takers
in their input markets. Thus, a structure of an industry that minimizes the sum of the
production costs is not affected by any market power. Similarly, (taking into account
Assumption 4.2), our definitions of a natural oligopsony and a natural monopsony
are based on the minimization of the sum of the costs of selling goods purchased
from the producers in J in the retail market for a given vector of the purchased
quantities. These costs are the expenditure on the inputs of the selling process in the
retail market that are not purchased from the producers in J . (Labor and transport
services or vehicles are obvious examples of such inputs.) The buyers cost func-
tions are defined under the assumption that they are price takers in the markets in
which they buy inputs other than the goods for sale purchased from the producers in
J (e.g., labor market, market for transport services, market for cash registers). Thus,
their cost functions are independent from any market power that they may have in
the analyzed market (and independent from a bargaining process about the prices in
this market, which is affected by their market power). Therefore, a structure of the
retail market in our model, which minimizes the sum of the buyers costs of selling
the goods purchased from the producers in J , is independent of any market power of
the buyers. Although it affects the outcome of an interaction between the producers
in J and the buyers in I , it depends only on the technology available to the buyers
and the prices in the markets that they cannot influence. Thus, it is independent of
the buyers desire to have a market power in relation to the producers in J .
4.2 Natural Oligopsony 59
Now, we give the definitions of a natural oligopsony and a natural monopsony in

the analyzed market.
Definition 4.3. A set of buyers I I is a natural oligopsony in the analyzed mar-
ket
for a vector of the purchased quantities z 2 Y if # .I / 2 and there exists
#.I /
xi i 2I 2 Y such that
X
xi D z (4.7)
i 2I
and
X X
ci xi ci .xi / ;
i 2I i 2I Wxi >0
X
8x 2 X such that xi D z: (4.8)
i 2I

It follows from (4.7) and (4.8) that xi i 2I minimizes the cost of selling the
vector of quantities z in the retail market by the buyers in the analyzed market. Nev-
ertheless, a natural oligopsony is defined as a subset of the set of all buyers rather
than a pair of a subset of the set of all buyers and a collection of vectors of the quan-
tities purchased by them. Thus, when I is the unique natural oligopsony in the
analyzed market for a vector of purchased quantities z, there can still be more than
one collection of vectors of the quantities purchased by the firms in I that mini-
mizes the cost of selling the vector of quantities z in the retail market. This approach
to the definition of a natural oligopsony corresponds to both our and Baumols (see
[Baumol (1982)]) approach to the definition of a natural oligopoly.
Definition 4.4. A buyer i 2 I is a natural monopsony in the analyzed market for
a vector of the purchased quantities z 2 Y if
X
ci .z/ ci .xi / ;
i 2I Wxi >0
X
8x 2 X such that xi D z: (4.9)
i 2I
It follows from (4.9) that the sale of the quantities of the goods specified by z in
the retail market by buyer i minimizes the cost of selling them.
A natural monopsony and (in general) also a natural oligopsony are a result of
both the economies of scale and the economies of scope.2 The former implies that
it is not possible to decrease the sum of selling costs by dividing a vector of the
quantities purchased by some buyer between two or more buyers3 in such a way
2
See [Baumol (1977)] for the characterization of the functions that exhibit both the economies of
scale and the economies of scope.
3
The information about a natural oligopsony (natural monopsony) I (i ) in the analyzed market
has the greatest value when the latter remains a natural oligopsony (natural monopsony) even after
that the composition of their sales (i.e., the ratios between the quantities of any two
goods) would be the same. In the case of a natural oligopsony, the economies of
scale are already exhausted. It is not possible to decrease the sum of selling costs by
assigning the sum of the vectors of the quantities purchased by two or more buyers
with the same composition to just one (or any lesser number) of the buyers. The
economies of scope imply that it is not possible to decrease the sum of the selling
costs by dividing a vector of the quantities purchased by some buyer between two
or more buyers in such a way that each of them would specialize in sale of only
some (or even only one) of the goods that their predecessor was selling. In the
case of a natural oligopsony, the economies of scope are already exhausted. It is not
possible to decrease the sum of selling costs by assigning the sum of the vectors of
the quantities purchased by two or more buyers, which buy different goods, to just
one (or any lesser number) of the buyers.
First, we show that the maximization of the sum of the firms stage game profits
leads to a natural oligopsony or a natural monopsony in the analyzed market.
Lemma 4.2. Let x max 2 X max . If # .EI .x max // 2, then EI .x max / is a natu-
ral
P oligopsony in the analyzed market for the vector of the purchased quantities
i 2I xi
max
. If EI .x max / D fi g, then buyer i is a natural
P monopsony in the
analyzed market for the vector of the purchased quantities i 2I ximax .
P
Proof. Take (arbitrary) x max 2 X max and let z D i 2I ximax . Suppose that the claim
of LemmaP4.2 does not hold. Then, we can find a nonempty I C I and x 2 X
such that i 2I C xi D z, xi D 0 for each i 2 I nI C , xi > 0 for every i 2 I C , and
X X
ci xi < ci ximax :
i 2I C i 2EI .x max /
P P
Note that i 2I C xi D i 2EI .x max / ximax D z implies that j x D j .x max /

for each j 2 J and EJ x D EJ .x max /. Therefore, it follows from the properties
of x that
X X X
Pj x j x cj j x ci xi
j 2J j 2EJ .x /
i 2I C
X X
> Pj . .x // j .x max /
max
cj j .x max /
j 2J j 2EJ .x max /
X
ci ximax : (4.10)
i 2EI .x max /
This contradiction with the definition of X max in (2.3) completes the proof. t
u
adding any finite number of replicas of all firms in I (i ) to I . In this case, the sum of the costs
of selling the quantities of goods given by z cannot be decreased by any change in the number of
buyers.
4.3 Impact on Consumer Welfare 61
The basic idea of the proof of Lemma 4.2 is simple. Suppose that either EI .x max /
has at least two members but it is not a natural oligopsony in the analyzed market for
the sum of the vectors of the quantities purchased by its members, or EI .x max / is
a singleton but its only member is not a natural monopsony in the analyzed market
for his/her vector of the purchased quantities. Then, for another group of buyers, it
is possible to reduce the sum of selling costs while keeping the vector of total pur-
chased quantities, as well as the output of each producer, unchanged. This increases
the sum of the stage game profits of all firms (which equals the difference between
the sum of the revenues from the sales in the retail market and the sum of the produc-
tion costs of the active producers and the selling costs of the active buyers). Thus, we
have a contradiction with the fact that the latter sum is maximized for each element
of X max .
Now we show that an SRPE in ./ leads in each period along the equi-
librium path in every subgame, in which the sum of the firms expected average
discounted profits is maximized to a natural oligopsony or a natural monopsony
in the analyzed market. This is a consequence of the preceding lemma.
Proposition 4.2. Consider an SRPE s of ./, h 2 Hf , and the sequence

x D fx .t/gt 2N 2 X 1 of the vectors of the actually traded quantities in .h/ ./ gen-

erated by the prescriptions of s.h/ . Assume that (4.6) holds. Then, for each t 2 N ,
I .t/ D fi 2 I j xi .t/ > 0g is a natural oligopsony if # .I .t// 2 or the only

buyer in it is a natural monopsonyP if # .I .t// D 1 in the analyzed market for the

vector of the purchased quantities i 2I xi .t/. t
u
Proof. Consider (arbitrary) SRPE s of ./, h 2 Hf , and the sequence x D

fx .t/gt 2N 2 X 1 of the vectors of the actually traded quantities in .h/ ./ gen-

erated by the prescriptions of s.h/ . As noted in the proof of Proposition 4.1, (4.6)
implies that x .t/ 2 X max for each t 2 N . Therefore, the claim of Proposition 4.2
follows from Lemma 4.2. t
u
4.3 Impact on Consumer Welfare
In this section, we describe the impact of a collusion in an SRPE (and hence, also
of a collusion in an SSPE) on consumer welfare. In this comparison, we restrict
our attention to the pure strategies in ./. Such an approach is sufficient when V
contains a payoff vector that is strictly Pareto efficient with respect to V C , it is the
equilibrium payoff vector in an SRPE, and it weakly Pareto dominates a stage game
payoff vector given by a benchmark used in the comparison.
Besides Assumptions 2.12.4, we make the following assumption in Sect. 4.3.
Assumption 4.3. (i) For each j 2 J , function cj is twice continuously differen-

tiable. (ii) For every i 2 I , function Ui is twice continuously differentiable. (iii) For
every i 2 I , function ci is twice continuously differentiable.
In Sect. 4.3.2, we relax part (ii) of Assumption 4.3.

The results in this section are based on the fact that in an SRPE in each subgame,
the vector of the firms equilibrium average discounted profits is strictly Pareto effi-
cient with respect to V C . This implies that in each subgame along the equilibrium
path, in every period, the vector of the firms stage game profits is strictly Pareto
efficient with respect to V C .
4.3.1 Comparison with a Monopsonist Choosing the Traded

Quantities on the Demand Side and Price Taking Behavior
on the Supply Side of the Analyzed Market
In this section, we compare the collusive outcome in an SRPE with the following
benchmark: the producers in J are price takers in the analyzed market4 and there
is only one buyer on its demand side. That is, I D f# .J / C 1g and x#.J /C1 D x
for each x 2 X . We assume that the purchased quantities of the goods in J are
the monopsonists decision variables and the unit prices he/she pays for them are
determined by the inverse supply functions in the analyzed market. Of course, the
monopsonist is also theh monopolist
i in his/her product markets.
1
Let j W Yj ! 0; pj max
be the inverse supply function for good j 2 J . It
assigns to each quantity qj of good j purchased by the monopsonist, the unit price
of good j (i.e., the unit price at which producer j wants to produce and bring to the
analyzed market quantity qj ). The following assumption (which we use only in the
current section) ensures together with Assumption 4.3 that the inverse supply
functions can be defined in such a way that they are continuously differentiable on
their domains.
Assumption 4.4. For each j 2 J , function cj is strictly convex on Yj and pjmax
cj0 .j /.
h i
For each j 2 J , producer j s supply function j W 0; pjmax ! Yj is defined
1 h i
by j pj D 0 if pj < cj0 .0/, j pj D cj0 pj if pj 2 cj0 .0/ ; cj0 j ,
4
Following the microeconomics textbooks (e.g., [Varian (1992), p. 215]) that define a competitive
firm as a price taking firm, we can use the term perfect competition on the supply side of the
analyzed market. Nevertheless, according to Assumption 4.5 below, all goods in J are different.
Hence, each type of good is sold in the analyzed market by only one firm (although it can be sold
in other markets by other firms that are potential competitors of the firm in J selling it in the
analyzed market). Since perfect competition is associated with a large number of firms producing
a homogeneous good, we prefer the term price taking behavior on the supply side of the analyzed
market.
In the analysis of non-collusive behavior (similarly as in the game ./), we allow the firms
to leave the analyzed market. Nevertheless, we do not consider the entry of firms (that could drive
the economic profits of the price taking firms down to zero). This is another reason why we do not
use the term perfect competition on the supply side of the analyzed market.

and j pj D j if pj > cj0 j . (The third part of the definition of j follows

from the capacity constraint.) Then, it is straightforward that j1 qj D cj0 qj

for each qj 2 0; j . In order to ensure that j1 is well defined, continuous, and
has a right-hand side continuous first derivative at zero and a left-hand
side
contin-

uous first derivative at j , we set j1 .0/ D cj0 .0/ and j1 j D cj0 j . This
choice of j1 .0/ is innocuous because it affects neither the revenue of producer j
nor expenditure of the monopsonist.
In order to obtain the results in this section, we also need the following assump-
tion.
Assumption 4.5. No two producers in J produce identical goods; further, we have
@2 U#.J /C1 .x/

0; 8j 2 J; 8k 2 J n fj g ; 8x 2 X (4.11)
@xj#.J /C1 @xk#.J /C1
and
@2 c#.J /C1 .x/
0; 8j 2 J; 8k 2 J n fj g ; 8x 2 X: (4.12)
@xj #.J /C1 @xk#.J /C1
Thus, we assume that the monopsonists marginal revenue from using any input
j 2 J is nondecreasing in his/her use of any other input. Further, his/her marginal
costs of using any input j 2 J is nonincreasing in his/her use of any other input.
Assumption 4.5 is satisfied, for example, if the monopsonist is a retailer, all goods
in J are complements, the inverse demand functions for them have nonnegative sec-
ond partial derivatives, and the monopsonists cost function is additively separable.5
Example 4.1 contains a stage game satisfying Assumption 4.5.
Let b
x 2 X be the vector of the traded quantities in a long-run equilibrium in
the benchmark considered in this section in which all firms are present in the ana-
lyzed market. (Since it is a long-run equilibrium, each firm earns a nonnegative
profit in it. Since all firms have positive fixed costs, we have b x j#.J /C1 > 0 for each
j 2 J .) Then, each producer j 2 J produces output b x j#.J /C1 and sells it for price
1

j bx j#.J /C1 . In order to avoid an uninteresting case (in which our results in this
section would fail to hold because of the producers capacity constraints), we restrict
our attention to the case when there is j 2 J with b x j#.J /C1 < j .
Clearly, b x solves the maximization problem
5
Our results in this subchapter continue to hold when function U#.J /C1 is not differentiable on
some subset of its domain but has right-hand side and left-hand side partial derivatives satisfying
Assumption 4.5 on it. A stage game, in which the monopsonist is a retailer and each inverse demand
function is linear on the subset of its domain on which it has a positive functional value, is an
example of this. With a slight abuse of language, when subsets, at which Assumption 4.5 does not
hold, have zero measure, we say that Assumption 4.5 is satisfied.
max U#.J /C1 .x/ c#.J /C1 .x/

X
1
j xj#.J /C1 xj#.J /C1
j 2J

subject to xj#.J /C1 2 0; j 8j 2 J: (4.13)
Thus, b
x satisfies the first-order conditions
@U#.J /C1 .b
x/ @c#.J /C1 .b
x/
cj0 bx j#.J /C1 xj#.J /C1 cj00 bx j#.J /C1 D 0;
@xj#.J /C1 @xj#.J /C1

x j#.J /C1 2 0; j ;
8j 2 J with b (4.14)
x / @c#.J /C1 .b
@U#.J /C1 .b x/
cj0 b x j#.J /C1 xj#.J /C1 cj00 b
x j#.J /C1 0;
@xj#.J /C1 @xj#.J /C1
x j#.J /C1 D j ;
8j 2 J with b (4.15)
@U#.J /C1 .b
x / @c#.J /C1 .b
x/
cj0 bx j#.J /C1 0;
@xj#.J /C1 @xj#.J /C1
8j 2 J with b
x j#.J /C1 D 0: (4.16)
It follows from (4.14), (4.15), and Assumption 4.4 that
@U#.J /C1 .b
x / @c#.J /C1 .b
x/
cj0 bx j#.J /C1 > 0;
@xj#.J /C1 @xj#.J /C1

8j 2 J with b x j#.J /C1 2 0; j : (4.17)

p D cj0 b
Let b vj D b
x j#.J /C1 and b pj bx j#.J /C1 cj b x j#.J /C1 for each j 2 J ,
j P
b
pD b p j j 2J , and b x / c#.J /C1 b
v#.J /C1 D U#.J /C1 .b x #.J /C1 j 2J b pj b
x j#.J /C1 .
Consider the maximization problem
X
max U#.J /C1 .b
x C z/ c#.J /C1 .b
x C z/ bp j C rj b x j #.J /C1 C zj (4.18)
j 2J
subject to

bp j C rj bx j#.J /C1 C zj cj b
x j #.J /C1 C zj b
vj ; 8j 2 J; (4.19)
zj b
x j#.J /C1 ; 8j 2 J; (4.20)
zj j b
x j#.J /C1 ; 8j 2 J; (4.21)
rj b
p j ; 8j 2 J; (4.22)
rj pjmax b
pj ; 8j 2 J: (4.23)
Thus, in the maximization problem (4.18)(4.23), we seek a vector of the

changes in the traded quantities given by vector b x and a vector of the changes in the
prices given by vector b p that maximize the monopsonists profit without decreasing
the profit of any producer j 2 J to below b vj . Taking into account the constraints
(4.20)(4.23), the set of feasible solutions of the latter maximization problem is
bounded. It is clearly nonempty it contains 0 2 <2#.J / . Since function cj , j 2 J is
continuous and the constraints have the form of weak inequalities, the set of feasible
solutions is also closed. Hence, it is compact. The objective function is continuous.
Therefore, the maximization problem (4.18)(4.23) has a solution.
Let .b r/ be a solution to the maximization problem (4.18)(4.23). We have
z;b
b
p j Cb r j > 0 and b x j#.J /C1 Cb zj > 0 for each j 2 J . (A producer, for whom at least
one of these inequalities fails, earns a negative profit.) Taking this into account, the
constraint in (4.19) is binding for each j 2 J . (If this was not the case for some
j 2 J , then a slight decrease in rj would increase the monopsonists profit without
violating any constraint.)
Suppose that b z D 0. Then, b r D 0. (If we had b r j < 0 for some j 2 J , pro-
ducer j s profit would fall below b vj . If we had b r j > 0 for some j 2 J , j s profit
would exceed b vj :) Take (arbitrary) j 2 J with b x j#.J /C1 < j . At .b r / D .0; 0/
z;b
the monopsonists expenditure on good j equals the sum of producer j s cost of
producing it and his/her profit b vj . Keeping this principle in mind, an infinitesimal
increase in zj (while keeping the purchased quantities of and the expenditure on all
other goods in J unchanged) increases the monopsonists expenditure on good j
by cj0 bx j#.J /C1 and his/her profit by the left-hand side of (4.17), which is positive.
Thus, it is possible to increase the monopsonists profit without violating any con-
straint. This contradicts the assumption that .b z;b r/ is a solution of the maximization
problem (4.18)(4.23) and proves that we haveb z 0:
We cannot have b
z < 0. Since all constraints in (4.19) are binding, this would
imply j1 b x j #.J /C1 Cb zj < b pj C b r j for each j 2 J with b zj < 0 and6 b rj D 0
for every j 2 J with b zj D 0. Thus, the replacement of .0; 0/ by .b r / decreases
z;b
the monopsonists profit. (If it does not thenthe replacement of b x by bx Cb z and the
replacement of b p j by j1 b x j #.J /C1 Cb zj for each j 2 J with b zj < 0 would
increase his/her profit. This would contradict the fact that b x solves (4.13)).
6
The profit of each j 2 J from its competitive supplyis strictly
increasing in price on the set of
prices that are higher than cj0 .0/ and no higher than cj0 j . For

cj0 .0/ < pj0 < pj cj0 j
we have
h i
pj j pj cj j pj > pj j pj0 cj j pj0
h i
> pj0 j pj0 cj j pj0 :
n o
z D zC C z , zC > 0, z < 0, J C D j 2 J j zC
Suppose that b j > 0 , J D
n o
j 2 J j z
j < 0 , and J
C
\ J D . We already know that the replacement of b
x
by bx C z (with the changes in prices that ensure that each producers profit remains
unchanged) decreases the monopsonists profit. From this and Assumption 4.5, it
follows that the replacement of b x CzC by b x CzC Cz (with the changes in prices that
ensure that each producers profit remains unchanged) decreases the monopsonists
profit.7 Therefore, we haveb z > 0.
The replacement of .b x; b
p / by .b
x Cb z; b
p Cbr / leads to a vector of the firms profits
that is strictly Pareto efficient with respect to V and8 it weakly Pareto dominates b v.
(There exists > 0 such that the replacement of b r j by b
r j C for each j 2 J gives a
vector of the firms profits that strictly Pareto dominatesb v.) This is achieved without
decreasing the output of any good while the output of at least one good rises.
Assume that the firms are using pure strategies in game ./ and that V con-
tains a vector of the firms stage game profits v that weakly Pareto dominates b v, is
strictly Pareto efficient with respect to V C , and is the vector of the firms equilib-
rium expected average discounted profits in an SRPE. Let .b x C z; bp C r/ 2 X P
generate v. Then, there exists r r such that .z; r/ solves (4.18)(4.23).9 This
implies that z > 0. Thus, a movement from a non-collusive long run equilibrium to
a collusive outcome in an SRPE does not lead to a decline in the output of any good
7
The changes in the monopsonists expenditure on goods in J are the same in both replacements.
Assumption 4.5 implies that

@ U#.J /C1 b x C zC c#.J /C1 b x C zC
@zj #.J /C1

@ U#.J /C1 b x c#.J /C1 bx
; 8j 2 J :
@zj #.J /C1
We can view the replacement of bx C zC by bx C zC C z (as well as the replacement of bx by

b
x C z ) as a sequence of small changes. Then, for each of them, we can approximate a change
in the functional value of the function U#.J /C1 c#.J /C1 by the scalar product of its gradient and
a vector of the changes in its arguments.

8
This can be seen as follows. Let v be thevector of the firms profits generated by b x Cb z; b
p Cb
r .
Suppose that there exists .e z;e
r/ b z;br such that b x Ce z; b
p Ce r gives a vector of the firms
payoffs e v that weakly Pareto dominates v. Clearly, there
exists j 2 J with evj > vj . (Otherwise,
we would have a contradiction with the fact that b z;b
r solves (4.18)(4.23).) Thus, by decreasing
e
r j for each j 2 J withe vj > vj , we obtain a vector offirms profits v with vj D vj for each j 2 J
and v#.J /C1 > v#.J /C1 . This contradicts the fact that b z;br solves (4.18)(4.23) and completes the
argument.
Let r result from the changes in the prices that each j 2 J with profitb vj . Suppose that .r; z/
9
leave
does not solve (4.18)(4.23). Take solution b z;b
r of (4.18)(4.23).
Then, the sum of the compo-
nents of the vector of the profits generated by b
x C b
z; b
p C b
r exceeds the sum of the components
of the vector of the profits v generated by bx C z; bp Cr (which equals the sum of the components
of the vector of the profits generated by b x C z; bp C r ). Therefore, we can change the prices given
by bp Cb r in such a way that we obtain a vector of the profits that weakly Pareto dominates v. This
contradiction with the assumption that v is strictly Pareto efficient completes the argument.
and yields an increase in the output of at least one good. This is a consequence of
two facts. First, in a movement to a collusive outcome, an increase in the quantity
of a good purchased by the monopsonist does not lead to a higher price given by the
inverse supply curve, but only to a higher price that compensates for the increase in
a producers costs. (Equation (4.17) used in the arguments above reflects this.) Sec-
ond, (we assumed that) goods are complements. Thus, an increase in the quantity of
one good purchased by the monopsonist does not make a decrease in the purchased
quantity of another good profitable.
First, suppose that the monopsonist uses the purchased goods to produce one type
of new good. Then, z > 0 implies that his/her output rises. (An increase in the pur-
chased quantities of the goods in J leads to higher expenditure for the monopsonist
on them. Therefore, z > 0 can be optimal only if it leads to an increase in his/her
output.) This leads (under the assumption that the good produced by the monopson-
ist is not a Giffen good) to lower price. Thus, the consumers of the good produced
by the monopsonist are clearly better off.
Next suppose that the monopsonist is a retailer. In order to assess the conse-
quences of the movement from .b x; b
p / to .b
x C z; b
p C r/ for consumer welfare, we
have to also take into account prices. When the goods in J are complements (as
we assumed in the justification of Assumption 4.5), the prices of some goods can
rise. If none of them rises (as in Example 4.1), consumer welfare clearly increases.
Now, we give the sufficient conditions under which consumer welfare increases even
when some prices rise. Suppose that there is a finite set B of consumers who buy the
goods sold by the monopsonist in the retail market but they can also buy other con-
sumptions goods from other sellers, who also sell to other buyers. We assume that
the prices of these goods do not change after collusion takes place.
Each
consumer
has strictly convex preferences. For consumer b 2 B, let b qb D b q bj j 2J be his/her
vector of the consumption of the goods purchased from the monopsonist before col-
lusion. Then, each consumer, whose consumption vector after collusion differs from
his/her consumption vector before collusion, is better off after collusion if
X X
Pj . .b
x Cb q bj
z//b Pj . .b
x //b
q bj ;
j 2J j 2J
8b 2 B: (4.24)
That is, for each consumer, his/her expenditure on the original quantities of the
goods in J (purchased before collusion) at new prices (that prevail after collusion)
is not higher than that at the original prices (that prevailed before collusion). If (4.24)
holds, then each consumer can buy at new prices, the same quantities of all goods
(including those that he/she does not buy from the monopsonist) as before collu-
sion. Thus, as his/her preferences are strictly convex (which implies that his/her
Marshallian demand vector is unique), his/her Marshallian demand vector at new
prices gives him/her a higher utility than his/her Marshallian demand vector at orig-
inal prices. Since collusion increases the sale of at least one good in J in the retail
market, there exists at least one consumer whose consumption vector after collusion
differs from his/her consumption vector before collusion. If (4.24) holds for each
such consumer, then collusion increases consumer welfare.
Now, we give an example of a stage game satisfying Assumption 4.5.
Example 4.1. Let J D f1; 2g, I D f3g, the monopsonist be a retailer, Y1 D 0; 5,
Y2 D 0; 5, c1 .y1 / D y12 C 2, c2 .y2 / D y22 C 2, c3 .x13 ; x23 / D x13 C x23 C 1,
p1max D 29, and p2max D 29. The monopsonists revenue function has the form
U3 .x13 ; x23 / D P1 .x13 ; x23 / x13 C P2 .x13 ; x23 / x23 ; (4.25)
where P1 W Y ! 0; 18 is the inverse demand function for good 1 in the retail
market, P2 W Y ! 0; 18 is the inverse demand function for good 2 in the retail
market,
P1 .x13 ; x23 / D max f13 2x13 C x23 ; 0g ; (4.26)
and
P2 .x13 ; x23 / D max f13 C x13 2x23 ; 0g : (4.27)
This example satisfies Assumptions 2.12.4, Assumption 4.3, and Assump-
tions 4.44.5 (U3 is not differentiable on the subset of its domain with the zero
measure see footnote 5).
We have 1 .p1 / D 0:5p1 for p1 2 0; 10 and 1 .p1 / D 5 for p1 2 10; 29,
1
2 .p2 / D 0:5p2 for p2 2 0; 10 and 2 .p2 / D 5 for p2 2 10; 29, 1 .q1 / D
1
2q1 , and 2 .q2 / D 2q2 . In the long-run non-collusive equilibrium (obtained by
solving the maximization problem (4.13)), b x 13 D 2 and b
x 23 D 2. This gives bp1 D
4, b
p 2 D 4, bv1 D 2, bv2 D 2, b
v3 D 23, and the prices in the retail market P1 .2; 2/ D
11 and P2 .2; 2/ D 11.
Consider a collusive outcome based on the maximization of the sum of the firms
profits. In this example X max D fx max g (i.e., X max is a singleton), where x13max
D

3 >b x 13 and x23 D 3 > b
max
x 23 . This gives D 31 and the prices in the retail market
P1 .3; 3/ D 10 < P1 .2; 2; / and P2 .3; 3/ D 10 < P2 .2; 2/. Letting (for example)

p13 D 4:9 and p23 D 4:9,10 profits are v1 D v2 D 3:7 and v3 D 23:6.11 Then,
the firms can agree on moving from the long-run non-collusive equilibrium to the
collusive outcome that we have just described. Such movement increases the profit
of each firm. It also increases consumer welfare because the consumed quantity of
each good increases and the price of every good decreases.
The following example shows that a collusive outcome can increase the con-
sumers welfare in comparison with the long-run non-collusive equilibrium in some
cases even when Assumption 4.5 is not satisfied (namely, when the goods in J are
substitutes or identical).
10
A reader may find this choice of prices strange. It is motivated by the desire to have prices that
are not periodic numbers and have at most two non-zero decimal points (and hence, they can be
paid by the banknotes and coins in circulation).
11
In this example, f1;3g D f2;3g D 9. Thus, the firms profits in the collusive outcome satisfy
(3.22). Therefore, the described collusive outcome can be sustained in an SSPE of ./ for any
2 .0; 1/. (See Proposition 3.2.)
Example 4.2. Consider the stage game from Example 3.1 with one modification:
producers fixed costs are 10 financial units instead of 25, i.e., cj .yj / D 10yj2 C 10
for each j 2 J .12 This leaves X max unchanged - it has the unique element
x max D .1:5; 1:5; 1:5; 1:5/. This also increases the maximal sum of the profits
to D 305. (The new values of C for non-singleton coalitions containing the
monopsonist can be computed from the original ones given in Example 3.1 by
adding the product of 15 and the number of producers in a coalition.) We have
P1 .x max / D P2 .x max / D 76. We will use the collusive arrangement based on x max
with the vector of profits v D .5; 5; 5; 5; 285/.This results from price pj D 25 for
each j 2 J . (It is easy to verify that v satisfies condition (3.22).)
This example satisfies Assumptions 2.12.4, Assumption 4.3 (U3 is not differ-
entiable on the subset of its domain with the zero measure see footnote 5), and
Assumption 4.4.
For each j 2 J , we have j pj D pj =20 for pj 2 0; 100 and j pj D 5
for pj 2 100; 326. The aggregate supply function for the type one good is
1 .p1 / D 0:1p1 for p1 2 0; 100 and 1 .p1 / D 10 for p1 2 100; 326.
The aggregate supply function for the type two good is 2 .p2 / D 0:1p2 for
p2 2 0; 100 and 2 .p2 / D 10 for p2 2 100; 326. We use Q1 to denote
the aggregate output of the type one good (the sum of the outputs of producers
1 and 3) and Q2 to denote the aggregate output of the type two good (the sum
of the outputs of producers 2 and 4). Then, the inverse supply function for type
one good has the form 11 .Q1 / D 10Q1 , and the inverse supply function for
the type two good has the form 21 .Q2 / D 10Q2. The monopsonist decides
only on the aggregate amount of each type of good that he/she wants to buy. The
price of each type of good is then determined by the inverse supply function for it
defined
P above.Thus, in the maximization problem (4.13) we have to replace the term
j 2J j1 xj#.J /C1 xj#.J /C1 by 11 .Q1 / Q1 21 .Q2 / Q2 and the con-
straints by Q1 2 0; 10 and Q2 2 0; 10. Then, using the inverse supply functions
for both types of goods and the producers supply functions, we have
x D .0:5Q1 ; 0:5Q2 ; 0:5Q1 ; 0:5Q2 / :
The solution of this modified maximization problem gives the long-run non-
collusive equilibrium: b x j D 1:2 < xjmax , b
p j D 24, b vj D 4:4 for each j 2 J ,
b
v5 D 273, P1 .b x / D 85 > P1 .x max /, and P2 .b x / D 85 > P2 .x max /. Since
v b v, the firms can agree on moving from the long-run non-collusive equilibrium
to the collusive outcome described above. Such movement increases the output and
decreases the price of each type of good. Thus, it (irrespective of the distribution of
12
With the original fixed cost a long-run non-collusive equilibrium would not exist. The producers
profits in the solution of the maximization program (4.13) would be negative. Nevertheless, the
stage game in example 3.1 with original fixed cost can be used as an example that a collusion can
make viable production that is not viable under a non-collusive arrangement.
the aggregate consumption vector between consumers) clearly increases consumer

welfare.
4.3.2 Comparison with Price Taking Behavior on the Demand

Side and a Cournot Oligopoly on the Supply Side
of the Analyzed Market
In this section, we compare the collusive outcome in an SRPE with the following
benchmark: the producers compete in the Cournot manner in the analyzed market
and the buyers are price takers in it, the buyers are themselves producers, each buyer
can produce one type of good, and the buyers compete in the Cournot manner in the
markets for their products.13 (We can formally model the buyers behavior in their
output markets in this way even if some or all of these markets are separated, i.e.,
if the quantities brought to the market by some buyers do not affect the prices for
which the other buyers sell their outputs.)
We also identify I with the set of goods that are produced by the firms in I .
For each i 2 I , there exists a production
Q function fi W Y ! <C of firm i and
an inverse demand function Pi W k2I 0; max ffk .xk / j xk 2 Y g ! <C for the
good produced by i such that Ui .x/ D Pi ..fk .xk //k2I / fi .xi /.
As in Sect. 4.1, we also assume here that there exists a partition
n o
J .1/ ; : : : ; J .m/
of J with the property that for each k 2 f1; : : : ; mg, J .k/ contains goods that
are identical and they differ from the goods contained in J .r/ for each r 2
f1; : : : ; mg n fkg. We also let Assumption 4.1 hold in this section. We assume (with-
out loss of generality) that k 2 J .k/ for each k 2 f1; : : : ; mg. We also assume
that the upper bounds on the prices of the goods of the same type are equal (i.e.,
2
for each k 2 f1; : : : ; mg and every .j; n/ 2 J .k/ , we have pjmax D pnmax ).
Q h P i
Let Y D m kD1 0; j 2J .k/ j . That is, Y is the set of the feasible vectors
of the outputs of the types of goods. For each y D .yk /k2f1;:::;mg 2 Y and every
k 2 f1; : : : ; mg, yk is the output of type k of good. We let Xi D Y for every
i 2 I . For each i 2 I , there exist functions fi W Xi ! <C and ci W Xi ! <C
.i / .i /
such that for each P x i 2 X i , we
have fi .x i / D f i Q and ci .xi / D ci Q ,
where Q.i / D j 2J .k/ xj i . That is, for every firm i 2 I , its output and
k2f1;:::;mg
cost (expressed by the functional value of function ci ) depend only on the total used
quantity of each type k 2 f1; : : : ; mg of input, and not on the used quantities of the
individual inputs in J .k/ .
13
I am grateful to Martin Gregor for his comments on this section.
We restrict our attention to the situation where an efficient use of inputs by each
firm in I requires fixed proportions between the types of inputs (i.e., for each i 2 I ,
fi is a Leontieff production function).14This is formally stated in the following
assumption and it justifies the introduction of function Fi , i 2 I .
Assumption 4.6. For each i 2 I , there exists Q.i / 2 Xi such that Q.i / 0,

fi Q.i / D 1, and for every Q.i / 2 Xi , fi Q .i / D fi i Q.i / , where

i D mink2f1;:::;mg Qk.i / =Q.i
k
/
.
P
.i /
For each i 2 I , let max
i D min k2f1;:::;mg j 2J .k/ j =Q k
and define

function Fi W 0; max
i ! <C by Fi . i / D fi i Q.i / .
Assumption 4.7. For each i 2 I , function Fi (i) is twice continuously 00

differen-
tiable
on 0; max
i , (ii) is strictly increasing on its domain, and (iii) Fi . i 0 on
/
0; max
i .
Assumption 4.8. For each i 2 I , function ci is (i) twice continuously differen-

tiable at each Q.i / 0 from its domain, (ii) is strictly increasing in each argument

on its domain, and (iii) @2 ci Q .i / =@Qk.i / @Qn.i / 0 for each .k; n/ 2 f1; : : : ; mg2
at each Q.i / 0 from its domain.

.i /
For each i 2 I , define function Ci W 0; max i ! < C by C i . i / D ci i Q .
It follows from Assumption
4.8 that for every i 2 I , function Ci is twice continu-
ously differentiable on 0; max i , strictly increasing on its domain, and15

Ci00 . i / 0; 8 i 2 0; max
i : (4.28)
14
We can extend the analysis (at the expense of making it more cumbersome) to the situation where
for each firm i 2 I and every k 2 f1; : : : ; mg, a positive output requires a positive input of at least
one good in J .k/ but the goods in the latter group are close substitutes instead of being identical
and their quantities are aggregated using the fixed coefficients (in general, different from one) that
express the fixed (i.e., independent of their quantities) marginal rates of substitution between them.

15
For i 2 I and i 2 0; max i , we have

m @c .i/
X i i Q
Ci0 . i / D Q.i/
.i/ k
kD1 @ i Q
k
and, using part (iii) of Assumption 4.8,

0 1
m
X m
X @2 ci i Q.i/
Ci00 . i / D @Q.i/ Q.i/ A 0:
k .i/ .i/ n
kD1 nD1 @ i Q @ i Qn
k
Assumption 4.9. (i) For each i 2 I , function Pi is twice continuously differentiable

at every element of its domain at which it has a positive functional value,
@Pi .q/
0; 8k 2 I n fi g ;
@qk
Y
8q 2 0; max Fn . n / j n 2 0; max
n
n2I
with Pi .q/ > 0 (4.29)
and
@Pi .q/
< 0;
@qi
Y
8q 2 0; max Fn . n / j n 2 0; max
n
n2I
with Pi .q/ > 0: (4.30)
(ii) For each i 2 I ,
@2 Pi .q/
0;
@2 qi
Y
8q 2 0; max Fn . n / j n 2 0; max
n
n2I
with Pi .q/ > 0: (4.31)
Assumption 4.10. For each j 2 J , function cj is convex.
Part (i) of Assumption 4.9 is satisfied when (besides the twice continuous dif-
ferentiability of inverse demand functions), the products of the firms in I are
complements (i.e., not only the inputs used by the firms in I belonging to differ-
ent elements of the partition of J are complements, but also the outputs of the firms
in I are complements) and none of them is a Giffen good. Since the producers in
J are Cournot competitors in this section, the convexity of their cost functions is
sufficient (we do not need their strict convexity) for obtaining our results.
In Example 4.3, we deal with a stage game that satisfies all assumptions made in
this section and illustrates the results obtained in it.16
The following analysis is valid when a benchmark long-run non-collusive equi-
librium exists. Such an equilibrium consists of two interdependent Cournot equi-
1
libria. Its requires the existence of an inverse demand function n W
existence
Y ! 0; pn max
for each type n 2 f1; : : : ; mg of good: (Since the goods of the
same type produced by different firms are identical, they have the same prices in a
16
The stage game in Example 3.3 satisfies all assumptions made in this section. Nevertheless, (due
to the same form of buyers production functions and symmetry of inverse demand functions for
their products) inverse demand functions for the analyzed market (functions 1 1
1 and 2 in the
notation used below in the text) do not exist in it.
non-collusive benchmark equilibrium.)

Q The direct
demand functions in the analyzed
market for types of inputs,
P P n W k2f1;:::;mg 0; p max
k ! <C , n 2 f1; : : : ; mg are
defined by n .p/ D i 2I j 2J .n/ e x j i .p/, where e x .p/ 2 X is a Cournot equi-
librium in the market for the products of the firms in I when the vector of the prices
of the types of inputs (that the buyers take as given) is p. It is the first Cournot
equilibrium contained in the benchmark long-run non-collusive equilibrium. (The
firms in I choose the quantities of goods in J that they use as inputs. Nevertheless,
these inputs uniquely determine their purchases of each type of input as well as their
outputs.) For each i 2 I , ex i .p/ maximizes buyer i s profit for the vectors of the
inputs purchased by all other buyers given by e x k .p/, k 2 I n .i /. That is, it solves
the maximization problem

max Pi fi .xi / ; .fk .e
x k .p///k2I nfi g fi .xi /
0 1
m
X X
ci .xi / @pn xj i A
nD1 j 2J .n/
X
subject to xi 0 & xi C e
x k .p/ 2 Y: (4.32)
k2I nfi g
In the remainder of this section, we restrict our attention to the case where
for each n 2 f1; : : : ; mg, 1 n is twice continuously differentiable with respect
to all arguments and strictly decreasing and concave in the output of type n of
good.17Then, the second Cournot equilibrium contained in the benchmark long-run
non-collusive equilibrium a Cournot equilibrium in the analyzed market e y 2 Y
exists. For every
2 f1; : : : ; mg and each j 2 J . / , e
y j solves the maximization
problem
0 0 1 1
X X
max 1

@yj C yk ; @
e ykA
e A yj
k2J . / nfj g k2J .n/ n2f1;:::;mgnf g

cj yj subject to yj 2 Yj : (4.33)
Thus, the benchmark long-run non-collusive equilibrium is defined by
00 00 1 11 1
BB B X CC C
x @@n @@
e yj A
e AA A
j 2J . / 2f1;:::;mg n2f1;:::;mg
17
This ensures that in the Cournot oligopoly in the analyzed market, each producers payoff func-
tion is continuous in the whole vector of arguments and concave (and hence, quasi-concave) in
his/her own output.
and 0 00 1 11
B B@ X CC
@n @ yj A
e AA ;
j 2J . / 2f1;:::;mg n2f1;:::;mg
i.e., by a vector of traded quantities and a vector of prices that have one compo-
nent for each type of good. We carry out the analysis in this section without giving
the sufficient conditions for the existence of the benchmark long-run non-collusive
equilibrium. These conditions are quite complicated.18 If a benchmark long-run
non-collusive equilibrium does not exist, then the non-collusive behavior tends to
be erratic. This reinforces the case for allowing collusion between firms not limited
to one side of a market. If a benchmark long-run non-collusive equilibrium exists,
then the results in this section identify the conditions under which it is inferior to a
collusive outcome fromQ the pointof viewof consumer welfare.
Let .b
x; b
p / 2 X n2f1;:::;mg; 0; pnmax be the benchmark long-run non-collusive
equilibrium, in which all firms are active (i.e., none of them withdraws from the
analyzed market), andb v be the vector of the firms profits in this equilibrium.
The following analysis is valid when the upper bounds on the prices in the
analyzed market are high enough.19 It is enough to fix

2 0; min b
x j i j .j; i / 2 J I;b
xj i > 0 (4.34)
and set

max fUi .x/ j x 2 X g .n/

e max max
p max
n xj i > 0 j j 2 J
j i 2 I;b ;
b
xj i
8n 2 f1; : : : ; mg ; (4.35)
and
.n/
pjmax D e
p max
n ; 8n 2 f1; : : : ; mg ; 8j 2 J : (4.36)
18
Given the assumptions already made in this section, a sufficient condition for the existence of
a Cournot equilibrium in the market for the buyers products is that Ui is concave in xi and ci is
convex in the whole vector of arguments for each i 2 I . Nevertheless, in order to obtain continuous
inverse demand functions for the types of goods in the analyzed market, we need for each vector
of the prices of the types types of goods either Cournot equilibria in the market for the buyers
products that give the same vector of aggregate demands for the types of goods or continuous
selection from the equilibrium correspondence. Moreover, in order to apply the standard sufficient
condition for the existence of a pure strategy Nash equilibrium of a strategic form noncooperative
game (see, e.g., [Osborne & Rubinstein (1994), Proposition 20.3, p. 20]) to a Cournot oligopoly
on the supply side of the analyzed market, the inverse demand functions for the analyzed market
would have to be such that each producers profit is quasi-concave in his/her output.
19
In a non-collusive benchmark equilibrium, the upper bounds on the prices in the analyzed market
are determined by the production capacities of the firms in J and the inverse demand functions for
the types of goods. Nevertheless, these upper bonds need not be high enough for the validity of the
following analysis (see Steps 14 below).
This ensures that when for .j; i / 2 J I with b x j i > 0, the quantity of his/her
product that producer j 2 J supplies to buyer i is not lower than b x j i and all firms
earn positive profits the price of no type n 2 f1; : : : ; mg of good exceeds pnmax .
Set X
b .i / D Q
Q b .i

/
D b
x j i ;
. / j 2J
2f1;:::;mg 2f1;:::;mg

b
i D min 2f1;::;mg Q b .i /
=Q
.i / b D Q
for each i 2 I , Q b .i / ,b
D b i ,
P i 2I i 2I
Q D j 2J . / j .b x / for every
2 f1; : : : ; mg, and Q D Q 2f1;:::;mg . That
b .i /
is, Q is the equilibrium input vector of firm i 2 I , with one component for each
type of input, Q is the equilibrium aggregate output of type
of input, and Q is
the equilibrium vector of the aggregate outputs of the inputs used by firms in I ,
with one component for each type of input. Since .b x; b
p / is a long run equilibrium,
we have b v 0. This together with the fact that all firms are active and they
have positive fixed costs further implies that j .b x / >0 and bp
n > 0for each
b b
n 2 f1; : : : ; mg, as well as Fi i > 0, i > 0, and Pi Fk k b > 0 for
k2I
every i 2 I . It follows from (4.32), Assumption 4.6, and part (ii) of Assumption 4.8
b .i / D b
that Q i Q.i / for each i 2 I . In order to avoid an uninteresting case (in which
our results in this section would fail to hold because of the producers capacity
constraints), we restrict our attention to the case where j .b x / < j for each j 2 J .

b
This implies that i < i for each i 2 I . Of course, 1
max
Q Db p for each

b
2 f1; : : : ; mg. Moreover, for every i 2 I , i solves the maximization problem

max Pi Fi . i / ; Fk b
k Fi . i /
k2I nfi g
m
X
Ci . i / i p Q.i
b
/
: (4.37)
D1
Thus, b
satisfies the following first-order conditions:
2 3
@Pi Fk b
k
4Pi Fk b
k C Fi . i / k2I 5 Fi0 b
i
k2I @qi
X m
.i /
Ci0 b
i D 1 Q Q ;

D1
8i 2 I: (4.38)
From (4.33), we get first-order conditions:


@1
Q
1
Q C j .b
x/ cj0 j .b
x / D 0; 8
2 f1; : : : ; mg ; 8j 2 J . / :
@Q
(4.39)
Combining (4.38) and (4.39), we obtain
2 3
@Pi Fk b
k
4Pi Fk b
i Ci0 b
i
k2I @qi
m
" #
X @1
Q
D Q.i /
cj0 . / j . / .b
x / j . / .b
x/ ; 8i 2 I; (4.40)
D1
@Q
where, for each

2 f1; : : : ; mg, j .
/ can be any element of J . / . Recall that for
every
2 f1; : : : ; mg, 1
is strictly decreasing in Q . Thus, (4.40) implies that
2 3
@Pi Fk b
k
4Pi Fk b
i
k2I @qi
X m

Ci0 b
i > Q.i / 0
c
j . /
j . / .b
x / ; 8i 2 I: (4.41)
D1
For each i 2 I , consider the maximization problem

m
X X
max Ui .b
x C z/ ci .b
xi C zi / bp C rj i bx j i C zj i (4.42)
D1 j 2J . /
subject to
m
X X
Uk .b
x C z/ck .b
x k C zk / bp C rjk b x jk C zjk b
vk ; 8k 2 I n fi g ;
D1 j 2J . /
X (4.43)

p C rjk b
b x jk C zjk cj j .b vj ; 8
2 f1; : : : ; mg 8j 2 J . / ;
x C z/ b
k2I
(4.44)
zjk bx jk ; 8 .j; k/ 2 J I; (4.45)
X X
zjk j b
x jk ; 8j 2 J; (4.46)
k2I k2I
p ; 8
2 f1; : : : ; mg ; 8 .j; k/ 2 J . / I;
rjk b (4.47)
p ; 8
2 f1; : : : ; mg ; 8 .j; k/ 2 J . / I
rjk pjmax b (4.48)
Thus, in the maximization problem (4.42)(4.48), we seek a vector of the

changes in the traded quantities given by vector b x and a vector of the changes in
the prices given by vector b p that maximize the profit of buyer i without decreasing
the profit of any other firm k 2 .J [ I / n fi g below b vk . (Note that r can specify
different changes in price b p for different producers in J . / and for different buy-
ers.) It maximizes a continuous function on a nonempty compact set, and hence has
a solution.
Let .b r/ be a solution of (4.42)(4.48) for firm i 2 I . We proceed in four steps.
z;b
Step 1. The changes in the prices in the analyzed market without any change in
the traded quantities cannot change the sum of the profits of the firms in I [ J .
Thus, they cannot increase the profit of buyer i without decreasing the profit of
any other firm. Suppose that (starting from b i ) buyer i infinitesimally increases
i . Further, suppose that he/she does so by proportionally increasing the purchased
quantity of each type
2 f1; : : : ; mg of good from every producer j 2 J . / with
b
x j i > 0 and increases his/her expenditure on the output of each j 2 J with b xj i > 0
by an amount equal to the increase in producer j s production cost (keeping the
quantities traded by and the expenditures of all other buyers unchanged). Using
(4.41), it increases i s profit without changing the profit of any firm belonging to
J .20 Taking into account (4.29) and part (ii) of Assumption 4.7, it does not decrease
the profit of any firm in I n fi g. Therefore,b z 0.
b
Step 2. Starting again from i and using (4.41), a decrease in i with a proportional
decrease in the purchased quantity of each type
2 f1; : : : ; mg of good from every
producer j 2 J . / with b x j i > 0, and a decrease in i s expenditure on the output
of each j 2 J with b x j i > 0 by an amount equal to the decrease in producer j s
production cost (again keeping the quantities traded by and the expenditure of all
other buyers unchanged) decreases i s profit.21 (Decreasing i s expenditure on some
20
We have

P
m P
d cj j bx b
xj i C
xOj i
b .i/
.i / i Q
Q

D1
b
j 2J .
/ Wx j i >0
d i
0 1
.i/
B Q
X m X C
D @ .i/ cj0 j b
x bxj i A
Q

D1
b
j 2J .
/ Wx j i >0
m
X n o
Q.i/

x j j 2 J .
/ ; b
max cj0 j b xj i > 0 :

D1
Recall that in (4.41), for each

2 f1; : : : ; mg, j .
/ can be any producer belonging to the set
J .
/ .
21
This argument is based on an infinitesimal change in i . Nevertheless, taking into account (4.30),
part (ii) of Assumption 4.9, parts (ii) and (iii) of Assumption 4.7, (4.28), and Assumption 4.10, the
difference between the right- and the left-hand side of (4.41) is nonincreasing in i . Thus, an
infinitesimal decrease in i does not decrease this difference. Therefore, we can view any decrease
good j by less than the decrease in firm j s production cost would further decrease
{Ks profit. Decreasing the expenditure by more than the decrease in firm j s pro-
duction cost would violate the constraint for firm j in (4.44). This conclusion does
not depend on proportionality of decrease in purchased quantities.22 ) Taking into
account (4.29) and part (ii) of Assumption 4.7, a decrease in i does not increase
the profit of any firm k 2 I n fi g. Therefore, .b
z;br/ withbzi < 0 andb zk D 0 for each
k 2 I n fi g decreases (in comparison with .0; 0/) the sum of the profits of all firms.23
This implies that it cannot increase or leave unchanged i s profit without decreas-
ing the profit of any other firm. Thus, it cannot solve the maximization problem
(4.42)(4.48).
Step 3. With respect to (4.29) and part (ii) of Assumption 4.7, a decrease in k for
any k 2 I n fi g does not increase the profit of firm i . Thus, using the results in Step
2, a decrease in k for more than one k 2 I without an increase in n for any
n 2 I decreases the sum of the profits of all firms. Therefore, .b r/ with b
z;b z < 0
cannot increase or leave unchanged i s profit without decreasing the profit of any
other firm. Thus, it cannot solve the maximization problem (4.42)(4.48).

Step 4. Suppose that B D .j; i / 2 J I j xj i > 0&b zj i < 0 and BC D
.j; i / 2 J I jbzj i > 0 . Since (as already shown in Step 1) .0; 0/ cannot
solve the maximization problem (4.42)(4.48) and .b r/, by assumption, solves it,
z;b
these changes increase the sum of the profits of all firms in J [I in comparison with
.b
x; b
p /. Suppose first that the changes in the inputs corresponding to B take place,
in i as a sequence of infinitesimal decreases, apply the argument in the text to each element of
this sequence, and reach the conclusion stated in the text. An analogous comment applies to the
other arguments in this section that are based on the signs of the derivatives.
22
Suppose that for
2 f1; : : : ; mg, producer j 2 J .
/ has share !j 2 .0; 1 in i s increased
purchase of type
of good. Let
X
Z D j b x bxj i b
x ki
k2J .
/ nfj g
Then, we have

dcj Z C !jb
i Q.i/

C 1 !j Q
.i/
x !j Q.i/
D cj0 j b
d i
and
X n o
x !j Q.i/
cj0 j b
Q.i/

x j j 2 J .
/ ; !j > 0 :
max cj0 j b
j 2J .
/ W!j >0
Thus, we can still use (4.41).

23
We have considered the specific changes in the prices in the analyzed market for which the
change in i s expenditure on good j 2 J equals the change in firm j s cost. Nevertheless, the
prices in the analyzed market do not affect the sum of the profits of all firms in J [I . An analogous
comment applies also to the following arguments in this section concerning the sum of the profits
of all firms in J [ I .
and then the changes in the inputs corresponding to BC take place. Further, suppose
that (for both groups), the changes in the quantities of the goods in J purchased
by the firms in I are accompanied by the changes in their expenditure on these
goods equal to the changes in the production cost of the firms belonging to J , and
that in both groups, the changes take place sequentially. (Thus, in each step of the
succession of changes, the change in the production cost of the affected producer
in J is well defined. Since the argument here concerns a change in the sum of the
profits of all firms, it does not depend on the order in which the changes within a
group take place.) The conclusion in Step 3 implies that the first group of changes
decreases the sum of the profits of all firms in comparison with .b x; b
p /. Thus, the
second group of changes increases the sum of the profits of all firms in comparison
with the result of the first group of changes. Moreover, the increase in the sum of the
profits of all firms in comparison with .b x; b
p / is greater when only the second group
of changes takes place than when both groups of changes take place. (Of course,
when only the second group of changes takes place, the changes in the expenditures
on the goods in J are equal to changes in the production costs of the firms in J
C C
stemming only from the latter changes.) Let and be the sum of the profits
of all firms after only the second group of changes takes place and after both groups
C C
of changes take place, respectively. Thus, > . As the prices in the analyzed
C
market do not affect the sum of the profits, is also the sum of the profits of all
firms generated by .b x Cb z; b
p Cb r/. For each k 2 J [ I , lete vk be the profit of firm k
generated by .b x Cb z; b
p Cb r/. Taking into account the constraints (4.43) and (4.44),
we havee vk bvk for each k 2 J [ I . We consider two distinguished cases.
First, there exists ` 2 J such that bx `k Cmax fb z`k ; 0g > 0 for each k 2 I . (That is,
after the second group of changes in the traded quantities alone takes place, producer
` trades with each buyer in I .) As such, we can find such changes in the prices in the
analyzed market after which the profit
C P of each firm k 2 .J [ I / n fi g equalsb vk and
the profit of buyer i equals k2.J [I /nfi g b vk . (We can do this in two rounds.
First, we change the prices in such a way that each j 2 J earns profit b vj .Then, if
some buyer k 2 I n fi g earns a profit exceeding b vk , we change the prices in a way
that increases his/her expenditure on good ` to the level that leaves him/her with
profit bvk and decreases by the same amount the expenditure of buyer i on good `.
We repeat the second step until we obtain the desired result.) Since
X X
C C
b
vk > e
vk D e
vi ; (4.49)
k2.J [I /nfi g k2.J [I /nfi g
we have a contradiction with the assumption that .b r/ solves the maximization

z;b
problem (4.42)(4.48).
Second, there exists no such ` 2 J . Take (arbitrary)
2 f1; : : : ; mg, n 2 J . / ,
and d 2 I with bx nd > 0. (Since n .b
x / > 0, such d exists.) For each k 2 I with
b
x nk Dbznk D 0, take
n o
k D min j 2 J . / W b
x jk > 0 :
Fix 2 .0; /. (See (4.34) for the fixing of .) Replace the zero delivery by producer
n to buyer k by , reduce the delivery by producer k to buyer k by , reduce the
delivery by producer n to buyer d by , and increase the delivery by producer k to
buyer d by . These changes leave the outputs and hence, the production costs
of the producers in J unchanged. Since Assumption 4.1 is satisfied in this section,
these changes leave the sum of the revenues of the firms in I , and the sum of their
costs expressed by the functional value of functions ck , k 2 I unchanged. Thus,
they leave the sum of the profits of all firms unchanged. This transforms the second
case to the first case, and we again obtain a contradiction with the assumption that
.b r / solves the maximization problem (4.42)(4.48). Therefore (taking also into
z;b
account Steps 13) we haveb z > 0.
Of course, (with respect to Assumption 4.6 and part (ii) of Assumption 4.7)b z>0
implies that fk .b x k Cb zk / fk .bx k / for each k 2 I . Moreover, the latter inequality
is strict for at least one k 2 I . Otherwise, we would have Uk .b x Cb z/ D Uk .b x/
and (using part (ii) of Assumption 4.8) ck .b x k Cbzk / ck .b x k / for each firm k 2 I
(with strict inequalityfor P
P k 2 I), but (using part (ii) of Assumption 2.1)
at least one
c
j 2J j j .b
x Cb z/ > j 2J j j .b
c x / . Thus, the sum of the profits of all firms
would decrease, contradicting the assumption that .b z;br/ solves the maximization
problem (4.42)(4.48).
Assume that the firms are using pure strategies in game ./ and V contains a
vector of the firms stage game profits v that weakly Pareto dominates b v, is strictly
Pareto efficient with respect to V C , and is the vector of firms equilibrium expected
average discounted profits in an SRPE. Let .b x C z; b
p CP r/ 2 X P generate P v. From
the fact that v weakly Pareto dominates b v (and hence, k2J [I vk > k2J [I b vk )
and the arguments in Steps 2 and 3 above, it follows that we cannot have z 0.
Then, the argument analogous to the one in Step 4 above shows that z > 0.24 This,
together with the fact that b x C z increases the sum of the profits of all firms in
comparison with b x , implies that fk .b x k C zk / fk .b x k / for each k 2 I with strict
inequality for at least one k 2 I . That is, at least one firm in I increases its output
and none decreases its output.
If the goods produced by the firms in I are consumer goods, in order to evaluate
the impact on consumer welfare, we have to also take into account the prices of the
goods produced by the firms in I . If no firm in I raises its price (as in the following
example), consumer welfare increases. If some firms in I raise their prices, we can
proceed in an analogous way as in Sect. 4.3.1.
The result in this section is driven by two differences in the firms motivation
between a non-collusive benchmark equilibrium and a collusive outcome. First, in
a Cournot equilibrium on the supply side of the analyzed market, the firms are dis-
couraged from increasing their output by its consequence a decrease in price. In a
24
In Step 4, we have shown that when only the increases in traded quantities take place, we can
increase the profit of buyer i (without violating any constraint in (4.43)(4.44)) by more than when
also the decreases in traded quantities take place. Here, we show that when only the increases in
traded quantities take place, we can give each firm a profit no lower than its profit when also the
decreases in traded quantities take place and we can give at least one firm a higher profit.
collusive outcome, the benefit of an increase in the output can be divided between
a producer and a buyer (with a positive impact on other buyers because the buyers
products are complements). Since both the firms in J and the firms in I have a mar-
ket power in the market for their products (unlike in Sect. 4.3.1 where the firms in J
are price takers), we have a case of double marginalization.25 Nevertheless, a rem-
edy, which is analyzed here, is a collusion, and not a vertical integration. Second,
in a Cournot oligopoly in the market for the buyers products (i.e., the products of
the firms in I ), an individual buyer does not take into account the positive effect
of his/her increased output on the revenues of other buyers. In a collusive outcome,
these effects are taken into account.
Now, we give an example of the situation analyzed above.

Example 4.3. We have J D f1; P2g, I D f3; 4g, Yj D 0; 50 and cj yj D 2yj C
15 for each j 2 J , ci .xi / D j 2J xj i C 1 for every i 2 I , and each buyer uses
the inputs purchased from the producers in J to produce a single good. The buyers
production functions have the form f3 .x3 / D min f0:5x13 ; x23 g and f4 .x4 / D
min fx14 ; 0:5x24 g. The inverse demand functions for their products have the form
P3 .q 3 ; q4 / D max f39 2q3 C q4 ; 0g (4.50)
and
P4 .q3 ; q4 / D max f39 C q3 2q4 ; 0g : (4.51)
Since each type of good is produced only by one producer, J .1/ D f1g,
.2/
J D f2g, and Y D Y . The buyers production and cost functions implythat
.3/
Q.3/ D .2; 1/, Q.4/ D .1; 2/, max 3 D 25, max
4 D 25, f 3 Q 1 ; Q2.3/ D
n o n o
.3/ .3/ .4/ .4/ .4/ .4/
min 0:5Q1 ; Q2 , f4 Q1 ; Q2 D min Q1 ; 0:5Q2 , and F i . i / D i
and Ci . i / D 3 i C 1 for each i 2 I .
For given prices p1 and p2 of the goods produced by firms 1 and 2, the Cournot
equilibrium in the market for the outputs of firms 3 and 426 is
3 2
3 .p1 ; p2 / D 12 p1 p2 (4.52)
5 5
and
2 3
4 .p1 ; p2 / D 12 p1 p2 : (4.53)
5 5
25
This problem was first pointed out by Cournot ([Cournot (1897)]; the original French publication
was in 1838). See [Spengler (1950)] for its analysis.
26
We compute the Cournot equilibrium using the buyers reaction functions obtained from the
solutions to the optimization problem (4.37) for i D 3; 4. The formulae given in the text are valid
for prices for which they give nonnegative outputs. This condition is satisfied in the non-collusive
benchmark equilibrium. Therefore, here, we do not deal with the prices for which the above formu-
lae are not valid. An analogous comment applies to the demand functions and the inverse demand
functions for the analyzed market given below.
Using (4.52), (4.53), Q.3/ , and Q.4/ , we obtain the demand functions for the
analyzed market
8 7
1 .p1 ; p2 / D 36 p1 p2 ; (4.54)
5 5
7 8
2 .p1 ; p2 / D 36 p1 p2 ; (4.55)
5 5
and from them, we have the inverse demand functions for the analyzed market
8 7
1
1 .Q1 ; Q2 / D 12 Q1 C Q2 ; (4.56)
3 3
7 8
1
2 .Q1 ; Q2 / D 12 C Q1 Q2 : (4.57)
3 3
Using (4.56) and (4.57), we compute the Cournot equilibrium in the analyzed
market: by 1 D 10=3 and b y 2 D 10=3. This gives the prices in the analyzed market
p 1 D 1
b 1 .10=3; 10=3/ D 98=9 and bp 2 D 1
2 .10=3; 10=3/ D 98=9; the outputs
of the buyers b 3 D 10=9 and b 4 D 10=9; the prices of the goods produced by
them P3 .10=9; 10=9/ D P4 .10=9; 10=9/ D 341=9; the traded quantities in the
analyzed market b x 13 D 20=9, b x 23 D 10=9, b
x 14 D 10=9, b x 24 D 20=9; and the
profitsbv1 D b
v2 D 14:63 and b v3 D bv4 D 1:4691:
The maximization of the sum of the profits of all firms gives the unique vector of
the traded quantities in the analyzed market:
max max max max
x max D x13 ; x23 ; x14 ; x24 D .30; 15; 15; 30/ : (4.58)
Thus, the buyers outputs are 3 D 15 and 4 D 15, the prices of the goods
produced by them are P3 .15; 15/ D P4 .15; 15/ D 24 < 341=9, the outputs of
the firms in J are 1 .x max / D 45 and 2 .x max / D 45, and the maximized sum of
the profits is D 418. Note that the buyers outputs are higher and the prices of the
goods produced by them are lower than in the non-collusive benchmark equilibrium.
We choose the following upper bounds on the prices in the analyzed mar-
ket: p1max D p2max D 671. These upper bounds satisfy both requirements, (2.6)
and (4.35).
Suppose that the firms agree on the following prices in the analyzed market:
p1 D p2 D 5. This gives profits v1 D v2 D 120 and v3 D v4 D 89. Thus,
v strictly Pareto dominates b v. Therefore, it is in the interest of all firms to move
from the non-collusive benchmark equilibrium to the collusive outcome with prices
p1 and p2 in the analyzed market. As already noted, such movement increases the
buyers outputs and decreases the prices of the goods produced by them. Therefore,
if the buyers products are consumer goods, it increases consumer welfare.
It can be shown that this example satisfies all assumptions of Proposition 3.4.
(Concerning requirements (3.33) and (3.34), we can take pji D 4 for each .j; i / 2
J I .) Thus, the described collusive outcome can be sustained in an SSPE of ./
for a sufficiently close to one.
References 83
References
Baumol, W. J.: On the Proper Cost Tests for Natural Monopoly in a Multiproduct Industry,
American Economic Review 67 (1977), 809-822.
Baumol, W.J.: Contestable Markets: An Uprising in the Theory of Industry Structure, American
Economic Review 72 (1982), 1-15.
Cournot, A.: Research into the Mathematical Principles of the Theory Wealth. Edited by N. Bacon.
New York: MacMillan, 1897.
Osborne, M.J. and A. Rubinstein: A Course in Game Theory. Cambridge, MA: MIT Press, 1994.
Spengler, J.J.: Vertical Integration and Antitrust Policy, Journal of Political Economy 58 (1950),
347-352.
Varian, H.R.: Microeconomic Analysis. Third Edition. New York, W.W. Norton, 1992.
Chapter 5
Afterword
In this book, we have shown that (for a discount factor close enough to one and
under certain conditions that we have identified in Chaps. 24), the firms on both
sides of a market can organize and sustain a collusive scheme with the following
properties.
First, all continuation equilibrium payoff vectors are strictly Pareto efficient.
Therefore, the grand coalition cannot abandon a punishment of a previous deviation
by a smaller coalition (i.e., weakly Pareto improve the vector of the continuation
payoffs of the members of a deviating coalition) without harming at least one
firm that did not deviate.1 Thus, a collusive scheme is an SRPE. Moreover, if the
requirements of Proposition 3.4 are satisfied or # .I / D 1 and the requirements of
Proposition 3.2 are satisfied, then for each proper coalition C of the firms in every
continuation equilibrium, for given continuation equilibrium strategies of the firms
outside C , the vector of the payoffs of the members of C is strictly Pareto efficient.
Thus, a collusive scheme is an SSPE.
Second, when the sum of the firms continuation equilibrium payoffs is maxi-
mized in an SRPE in some subgame, then in each period along the equilibrium
path in such a subgame the group of active producers of each type of good forms
a natural oligopoly (or a natural monopoly if there is only one active producer) for
the equilibrium output of that type of good. That is, the equilibrium output of each
type of good is produced with the lowest possible production cost. This result also
holds when the firms capable of producing a certain type of good have different
cost functions. This shows that a collusive scheme sustainable in an SRPE with the
1
This shows why an SRPE, which is based on the strict Pareto efficiency of the continuation equi-
librium payoff vectors, is preferable to various versions of renegotiation-proofness (e.g., [Farrell
& Maskin (1989)] and [Maskin & Tirole (1988)]), which are based on the weak Pareto efficiency
of the continuation equilibrium payoff vectors with respect to some set of allowed continuation
payoff vectors (e.g., in a weakly renegotiation-proof equilibrium of [Farrell & Maskin (1989)] with
respect to the set of all continuation equilibrium payoff vectors of an equilibrium that is weakly
renegotiation-proof or as in [Maskin & Tirole (1988)], with respect to the set of all continuation
payoff vectors that can be generated by Markov strategies.) In the latter case, an abandonment of
the punishment of a previous deviation can increase the continuation payoff of each deviator with-
out harming any player who did not deviate. In such a case, there need not be a player objecting to
the abandonment of the punishment.

86 5 Afterword
maximization of the firms continuation equilibrium profits is superior to the non-

collusive equilibria in the oligopolies containing firms with different cost functions
from the point of view of production efficiency.2 When the buyers are retailers and
the sum of the firms continuation equilibrium payoffs is maximized in an SRPE in
some subgame, then in each period along the equilibrium path in such a subgame
the group of active retailers forms a natural oligopsony (or a natural monopsony if
there is only one active retailer) for the equilibrium vector of the sales in the retail
market (which is also the equilibrium vector of the purchased quantities in the ana-
lyzed market). That is, the sum of the selling costs for the equilibrium vector of the
sales in the retail market is minimized. This result also holds when the buyers have
different cost functions (expressing their selling costs). We can combine this with
the result on a natural oligopoly (or a natural monopoly) on the demand side of the
analyzed market. Thus, the sum of the production and selling costs for the equilib-
rium vector of the sales in the retail market is minimized. For such an evaluation of
the efficiency of a market structure on both sides of the analyzed market, the expen-
diture on the goods sold in it is irrelevant the sum of the buyers expenditures
equals the sum of the producers revenues.
Third, we have identified two types of a market structure in the analyzed market
for which a non-collusive equilibrium is (under some additional conditions) inferior
from the point of view of consumer welfare to a collusive scheme sustainable in an
SRPE that weakly Pareto improves the vector of the firms profits. (This result does
not depend on the maximization of the sum of the firms continuation equilibrium
payoffs in an SRPE.) A monopsonist on the demand side of the analyzed market,
who chooses the purchased quantities, along with price taking behavior on the sup-
ply side is the first of them. In this case, we obtain the result that collusion increases
consumer welfare if the goods sold in the analyzed market are complements (or
independent) from the point of view of the monopsonists revenue and the marginal
cost of using any of them is nonincreasing in the used quantity of any other good
(see Assumption 4.5). Here, the result is driven by the fact that in a non-collusive
equilibrium, the monopsonist is discouraged from increasing the purchased quan-
tities of inputs (that would increase his/her sales to consumers) by the increases in
their prices that would increase his/her expenditure on them by an amount exceeding
the increases in the producers costs. Moreover, there is no input cannibalism an
increase in the use of one input does not decrease the marginal revenue from another
input (and it does not increase the marginal cost of using another input).3
2
Consider, for example, a Cournot oligopoly producing a single homogeneous good, with dif-
ferentiable inverse demand function at industry outputs giving a positive price and differentiable
convex cost functions. Take a Cournot equilibrium in it. For each firm with a positive equilibrium
output, its marginal cost equals the sum of the equilibrium price and the product of its output and
the derivative of the inverse demand function. Thus, the firms with different outputs have different
marginal costs. This implies that the sum of the costs of producing the equilibrium industry output
is not minimized.
3
Collusion can also increase consumer welfare when an increase in one input decreases the differ-
ence in the marginal revenue from and the marginal cost of using another input. This happens if the
latter effect is outweighed by the elimination of the increases in the prices of inputs (exceeding the
5 Afterword 87
A Cournot oligopoly on the supply side of the analyzed market coupled with
price taking behavior on its demand side is another market structure that we have
compared with a collusion. In this case, we obtain the result that collusion increases
consumer welfare if (besides some standard and technical conditions that we do
not repeat here) the buyers are themselves producers, each of them produces one
type of good using Leontieff technology, their outputs are complements, and each
buyers marginal cost of using any input is nondecreasing in the used quantity of any
other input (see Assumptions 4.64.9). Here, the result is driven by the fact that in a
non-collusive equilibrium, the producers of inputs are discouraged from increasing
their outputs by the decreases in their prices and buyers do not take into account the
positive impact of their increased outputs on the revenues of the other buyers.
The possible positive impact of collusion on consumer welfare described above
suggests that the antitrust policy in the US and the competition policy in the
European Union (and the antitrust policies in other countries)4 should not view
collusion that involves firms on both sides of a market as illegal. It should either
view it as legal (for reasons described below) or in each case, examine its impact on
consumer welfare.
The legislators who included the prohibition of cartel agreements5 restricting
competition into antitrust laws had obviously price cartels containing producers of
either identical products or close substitutes in mind. Moreover, in the case of the
Sherman Antitrust Act in the US, their thinking was influenced more by a fear of a
concentration of political power enabled by a concentration of economic power than
by a fear of the economic consequences of cartels.6 Nevertheless, currently antitrust
laws prohibit all cartel agreements that directly or indirectly set prices. Thus, the
prohibition applies also to cartels containing firms on both sides of a market and
even if the products sold to final consumers are complements.
Our analysis in Sect. 4.3 shows that under certain conditions, cartels containing
firms on both sides of a market besides being beneficial for participating firms
are beneficial for consumers. Such cartels should be legalized. Thus, the question
is whether or not current antitrust laws (on both sides of the Atlantic) make this
treatment possible.
The collusive schemes analyzed in this book represent the agreements about the
sequences of contracts (which specify prices and traded quantities) between trading
increases in the producers costs) caused by increased purchases of them. (Example 4.2 describes
such a situation.) Collusion can also be beneficial for consumers when there is more than one buyer
and the assumptions of Sect. 4.3.2 hold.
4
Unless we deal specifically with the competition policy or competition law of the European
Union, we use the term antitrust policy or antitrust law.
5
We use the term cartel agreement also for cartel agreements that are not legally binding and for
tacit collusion that has the same effects as a cartel agreement that is not legally binding. Antitrust
laws prohibit each of these three types of coordinated behavior if it restricts competition.
6
In 1890, Congressmen (or their advisers) could hardly have the knowledge of economics needed
for the evaluation of the economic consequences of cartels and other practices restricting compe-
tition. Alfred Marshalls Principles of Economics was first published in 1890 exactly the same
year, in which the Sherman Antitrust Act was passed.
88 5 Afterword
partners. They are not legally binding. They make contracts in each period (except
the first one) conditional on the history of contracts concluded up to that period.
Only the firms participating in them are affected by the fixing of prices in them.
These schemes do not fix the purchase or selling prices or any other trading condi-
tions for any third party (i.e., for any firm that does not participate in them). It is
true that an agreement on the quantities traded in the analyzed market affects the
prices and quantities sold in the market for the products sold by the buyers (i.e.,
by the firms in I ; this market is the retail market if the firms in I are retailers
and is the market for the goods produced by the firms in I if they are themselves
producers). Nevertheless, it affects the prices and quantities sold in the latter mar-
ket only through the fixing of the quantities of inputs used by the firms that sell
in it. Any contract between a seller and a buyer, who is not a final consumer of the
purchased goods, affects in the same way, the prices and quantities sold in the down-
stream market (i.e., in the market in which the buyer from the contract is a seller).
Despite this, the contracts are not viewed as an instrument of price fixing or limiting
production in downstream markets. Moreover, the collusive schemes analyzed in
the present book can be beneficial for the buyers in the downstream market. There-
fore, we think that they are not the cases of price fixing or limiting of production.
If they are implemented in the US, they are not a conspiracy in restraint of trade
or commerce among the several States, or with foreign nations. Thus, they do not
violate Sect. 1 of the Sherman Antitrust Act. Since they do not involve acquisition
by one corporation of the stock of another, they do not violate Sect. 7 of the Clayton
Antitrust Act. If they are implemented in the European Union, they do not have as
their object or effect the prevention or restraint or distortion of competition within
the Common Market. Thus, they do not violate [EU (2010a), Article 101(1)].
Moreover, collusive schemes that we have analyzed increase the output of at
least one good without decreasing the output of any other good. They are beneficial
for consumers. Thus, they contribute to improving the production and distribution
of goods.. Therefore, the European Commission should view them as compati-
ble with the Common Market also on these grounds, according to [EU (2010a),
Article 101(3)].
Allowing collusive schemes studied in the present book can yield an additional
benefit. Suppose that all firms involved in the collusive schemes beneficial for con-
sumers merge or through acquisitions create a business unit managed from one
center (create a concentration in the terminology of the European Unions com-
petition law; henceforth we use only the term merger). Such a new business unit
can achieve everything that can be achieved through analyzed collusive schemes.
(It can achieve the same even more easily because the decisions of its top manage-
ment are binding for the lower levels of management. Hence, there is no need to
use punishment schemes that are needed to sustain a collusion in an SRPE and an
SSPE.) Such a merger would have significant vertical components and if products
of at least some participating firms (namely, those on the demand side of the market
in question) are complements, it is likely that it would be cleared by the antitrust
5 Afterword 89
authorities on both sides of the Atlantic.7 If in the future, conditions change and
for some part of the new business unit (created by the merger), continued partici-
pation in it is no longer useful (i.e., participation in the business unit created by the
merger brings it lower profit than cooperation with some other firm), such a part
cannot secede. In the case of a collusive scheme, a participating firm can withdraw
from it whenever a withdrawal is useful for it.8 Thus, if a collusive scheme can be
legally implemented, then the firms would prefer its implementation to a merger.9
Further, suppose that in the future, an antitrust authority finds that the operation of
the business unit created by the merger violates the antitrust laws. The authorities
in the European Union cannot order (or seek through the court) a divestiture. In the
US, it is unlikely that the Antitrust Division of the Department of Justice will suc-
ceed in divestiture. In the case of a collusive scheme, if an antitrust authority finds
that the participating firms violate antitrust laws, it can prohibit the continuation of
their cooperation. Therefore, both from the point of view of participating firms and
from the point of view of antitrust authorities, collusive schemes are preferable to
mergers because the former can be undone much more easily.
As already noted, collusion between firms on both sides of a market should not
be viewed per se as a violation of antitrust laws. Antitrust authorities should adopt
one of the following approaches to it. First, they can view it as per se legal (for
reasons explained above) and intervene only when some specific action of partici-
pating firms is detrimental for consumers or violates the right of other firms to freely
choose their business partners (e.g., by exclusionary vertical contracts that cannot be
justified on efficiency grounds10). Second, they can introduce an approval process
for cartel agreements between firms on both sides of a market, analogous to the cur-
rent approval process for mergers, stipulating the right to revoke the approval if the
conditions change in the future. In the latter case, the approved cartel agreements
would be legally binding.
Our results (especially those on monopsonistic markets see Propositions 3.2
and 3.3 and Sect. 4.3.1) can shed light on an important real world economic prob-
lem: trading between (relatively small) farmers and other producers of foodstuffs
(or producers of other consumer goods) and chain-stores. The producers of food-
stuffs are weaker partners in this relationship. The chains-stores have considerable
7
The evaluation of the proposed mergers by the Antitrust Division of the US Department of Justice
is quite accommodating, at least since the period of the first Reagan administration. The European
Commission became more accommodating after critique of its handling of the proposed merger
between General Electric and Honeywell (which it prohibited in July 2001). For an analysis of
the impact of this case, see [Kolasky (2006)]. I am grateful to Tmea Antalicsov for drawing my
attention to the latter paper.
8
Recall that in our models, punishments have the form of a modified contract. Thus, they are
harmless for a firm that has found more attractive trading partners in another separated market.
9
[Chandler (1990), p. 424] points out that the legality of cartelization in Germany reduced the
pressure for industry-wide mergers.
10
See [Whinston (2006), chapter 4] for the economic evaluation of the various forms of
exclusionary vertical contracts.
90 5 Afterword
market power.11 Our results show that collusion can improve the situation of both
the producers of foodstuffs and chain-stores and increase consumer welfare.
References
Boel, M.F.: Food Prices in Europe: What the CAP Can and Cant Do, Konrad Adenauer
Stiftung Lecture, Brussels, 27 January 2009. http://europa.eu/rapid/pressReleasesAction.do?
reference=SPEECH/09/25&format=HTML&aged=0&language=EN&guiLanguage=en
Chandler, A.D. Jr.: Scale and Scope. The Dynamics of Industrial Capitalism. Cambridge, MA: The
Belknap Press of the Harvard University Press, 1990.
Consolidated Version of the Treaty on the Functioning of The European Union, Official Journal
of the European Union C83/49, 30 March 2010.
Farrell, J. and E. Maskin: Renegotiation in Repeated Games, Games and Economic Behavior 1
(1989), 327-360.
Kolasky, W.: GE/Honeywell: Narrowing, But Not Closing, the Gap, Antitrust Magazine, Spring
2006, 69-76.
Maskin, E. and J. Tirole: A Theory of Dynamic Oligopoly II: Price Competition, Kinked Demand
Curves, and Edgeworth Cycles, Econometrica 56 (1988), 571-599.
Whinston, M.D.: Lectures on Antitrust Economics. Cambridge, MA: MIT Press, 2006.
11
According to [Boel (2009)], the retail sector has the strength of a giant. Hence, in a model, its
market power can be approximated by a monopsony power.
Index
Antitrust law, see antitrust policy Hung, C., 5

Antitrust policy, vi, 5, 88, 89
cartel, see collusion
Celler Antimerger Act, 5 Kolasky, W., 89
Clayton Act, 88 Kreps, D.M., 6
collusion, 5, 67, 74, 81, 8690
merger policy, 5, 88, 89
Maskin, E., 4
Robinson-Patman Act, 5
Mertens, J.F., 1
Sherman Act, 87, 88
Monopsonist, see monopsony
Treaty on the Functioning of The European
Monopsonistic market, see monopsony
Union, 88
Monopsony, 6, 26, 29, 30, 3335, 53, 57, 62,
Aumann, R.J., 3, 15
63, 6569, 86, 89
natural monopsony, 53, 5861, 86
Baumol, W.J., 53, 54, 59
Bernheim, B.D., 4 Natural monopoly, 53, 5558, 85, 86
Boel, M.F., 90 Natural monopsony, 53, 5861, 86
Natural oligopoly, 5358, 85, 86
Natural oligopsony, 53, 5861, 86
Cartel, see collusion Norman, G., 5
Chain-stores, v, 6, 89
Chandler, A.D. Jr., 5
Collusion, 5, 61, 67, 74, 81, 8690 Peleg, B., 4
Competition policy, see antitrust policy Price taking, 6, 53, 86, 87
Complements, 63, 67, 72, 81, 8688
Consumer welfare, vi, 1, 53, 54, 61, 67, 68, 70,
Ray, D., 4
74, 80, 82, 86, 87, 90
Renegotiation-proof equilibrium, 4
Cooperative managerial capitalism, 5
internally consistent equilibrium, 4
Cournot oligopoly, 6, 53, 81, 87
strict renegotiation-proof equilibrium, see
Cournot, A., 81
SRPE
strong renegotiation-proof equilibrium, 4
strongly consistent eq., 4
Double marginalization, 81
weakly renegotiation-proof equilibrium, 4
Retail market, 1, 911, 58, 59, 61, 67, 68, 86
Rubinstein, A., 3
Farrell, J., 4
Scheinkman, J.A., 6
Greenhut, M.L., 5 Selten, R., 3

and Mathematical Systems 649, DOI 10.1007/978-3-642-19763-5,
92 Index
Spengler, J.J., 81 Strong Nash equilibrium, 3, 4

SRPE, 35, 17, 21, 53, 56, 61, 62, 66, 70, 80, Strong perfect equilibrium, 3
85, 86, 88
SSPE, 3, 5, 1719, 21, 26, 32, 33, 36, 38, 53,
57, 61, 82, 85, 88 Tirole, J., 4
Strict renegotiation-proof equilibrium, see
SRPE
Strict strong perfect equilibrium, see SSPE Whinston, M.D., 4, 89

(Lecture Notes in Economics and Mathematical Systems 649) Milan Horniacek (Auth.) - Cooperation and Efficiency in Markets-Springer-Verlag Berlin Heidelberg (2011)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Lecture Notes in Economics and Mathematical Systems 649) Milan Horniacek (Auth.) - Cooperation and Efficiency in Markets-Springer-Verlag Berlin Heidelberg (2011)

Uploaded by

Copyright:

Available Formats

Lecture Notes in Economics

and Mathematical Systems 649

For further volumes:

Cooperation and Efficiency

Library of Congress Control Number: 2011929346

c Springer-Verlag Berlin Heidelberg 2011

Cover design: eStudio Calamar S.L.

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

Bratislava, Milan Horniacek

3 Existence of an SRPE and an SSPE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 21

4 Efficiency of an SRPE and an SSPE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . 53

M. Horniacek, Cooperation and Efficiency in Markets, Lecture Notes in Economics 1

1.1 Choice of Solution Concepts

A subgame perfect equilibrium (first defined in [Selten (1965)] in German and

In many infinite horizon games of economic interest a strong perfect equilib-

1.2 Choice of Non-collusive Benchmarks

1.3 Mathematical Notations

For a set Z <n ; conZ denotes its convex hull.

2.1 Stage Game

M. Horniacek, Cooperation and Efficiency in Markets, Lecture Notes in Economics 9

Here, xj i is the quantity of good j 2 J purchased by buyer i 2 I . We will

For each i 2 I , Ui W X ! <C is buyer i s revenue function. It assigns to each

Assumption 2.3. For each i 2 I , Ui is (i) continuous and (ii) xi D 0 implies

Assumption 2.4. There exists x C 2 X such that

Since X .F / max is nonempty and compact for each

Clearly, taking into account (2.9)(2.12), we have

2.2 Repeated Game

The repeated game is a countable infinite repetition of G with discounting of future

every firm k 2 J [ I observes for every firm r 2 .J [ I / n fkg rs pure

We define function  W S ! <#.J [I / by  .s/ D .k .s//k2J [I . With this for-

s .C / of coalition C in subgame .h/ ./ to subgame .h;h0/ ./.

2.3 Solution Concepts

Definition 2.1. A strict renegotiation-proof equilibrium of  ./ is a strategy profile

(i) There do not exist a nonterminal history h 2 Hf , a firm k 2 J [ I , and its

Definition 2.2. A strict strong perfect equilibrium of  ./ is a strategy profile s 2

Thus, a strategy profile s is an SSPE if no coalition in no subgame can increase

Aumann, R.J.: Subjectivity and Correlation in Randomized Strategies, Journal of Mathematical

3.1 Existence of an SRPE

M. Horniacek, Cooperation and Efficiency in Markets, Lecture Notes in Economics 21

B. Description of strategy profile s . For each h 2 Hf n fg let L .h/ be the set

C. Strategy profile s is an SRPE of ./ for 2 .; 1/. Since ./ is continuous

exist k 2 J [ I and h 2 Hf such that k can increase his/her expected average

(The latter payoff is positive by (3.7).) A deviation gives j a nonpositive expected

Therefore, a deviation decreases j s expected average discounted profit in .h/ ./.

(The latter payoff is positive by (3.7).) A deviation gives him/her a nonpositive

Therefore, a deviation decreases i s expected average discounted profit in .h/ ./.

The strategy profile described in the proof of Proposition

3.2 Existence of an SSPE

In this section, we first deal with a monopsonistic market, where I D f# .J / C 1g.

Definition 3.1. A correlated strategy profile  2 is an SSE of G if

(ii) there does not exist  2 such that

Thus, a correlated strategy profile is an SSE of G if no coalition of players can,

Lemma 3.1. Assume that # .I / D 1, v 2 V C , and that there exist  2 , n 2

Proof. Consider (arbitrary) v 2 V C ,  2 , n 2 f1; : : : ; # .J [ I / C 1g, a.r/ 2

Take (arbitrary) x max P2 X max . Since

Take r 2 f1; : : : ; ng satisfying  .!/ D a.r/ . Suppose that all produc-

Lemma 3.2. Assume that # .I / D 1 and v 2 V C is the vector of the equilib-

Proof. Let v 2 V C be the equilibrium vector of the expected payoffs in an SSE

Proposition 3.2. Assume that #.I / D 1 and v 2 V C ; that there exists  2 ,

Proof. Consider (arbitrary) v 2 V C ,  2 , n 2 f1; : : : ; # .J [ I / C 1g, a.r/ 2

The following proposition shows that in each SSPE of ./ and in every

Proposition 3.3. Assume that # .I  / D 1. Let 2 .0; 1/, s 2 S be an SSPE of

We define function W S ! <#.J [I / by .s/ D .k .s//k2J [I . With this for-

s .C / of coalition C in subgame .h/ ./ to subgame .h;h0/ ./.

Definition 2.1. A strict renegotiation-proof equilibrium of ./ is a strategy profile

Definition 2.2. A strict strong perfect equilibrium of ./ is a strategy profile s 2

Definition 3.1. A correlated strategy profile 2 is an SSE of G if

(ii) there does not exist 2 such that

Lemma 3.1. Assume that # .I / D 1, v 2 V C , and that there exist 2 , n 2

Proof. Consider (arbitrary) v 2 V C , 2 , n 2 f1; : : : ; # .J [ I / C 1g, a.r/ 2

Take r 2 f1; : : : ; ng satisfying .!/ D a.r/ . Suppose that all produc-

Proposition 3.2. Assume that #.I / D 1 and v 2 V C ; that there exists 2 ,

Proof. Consider (arbitrary) v 2 V C , 2 , n 2 f1; : : : ; # .J [ I / C 1g, a.r/ 2

Proposition 3.3. Assume that # .I / D 1. Let 2 .0; 1/, s 2 S be an SSPE of

where P1 W X ! 0; 121 defined by

Proposition 3.4. Assume that there exists v 2 V C with. v 0, 2 , n 2

Proof. A. Preliminaries. Take (arbitrary) v 2 V C , 2 , n 2 f1; : : : ; #.J [ I / C

and set s .h/ D j D .h/.

3.3. We have J D f1; 2; 3g, I D f4; 5g ; Yj D 0; 20 for each j 2 J ,

max .52 2x14 / x14 22x14 70 subject to 0 x14 10 (3.60)

max U#.J /C1 .x/ c#.J /C1 .x/

Assumption 4.7. For each i 2 I , function Fi (i) is twice continuously 00