Document
Document
Document
Environmental Cooperation
NEW HORIZONS IN ENVIRONMENTAL ECONOMICS
General Editors: Wallace E. Oates, Professor of Economics, University of
Maryland, USA and Henk Folmer, Professor of General Economics, Wageningen
University and Professor of Environmental Economics, Tilburg University, The
Netherlands
This important series is designed to make a significant contribution to the develop-
ment of the principles and practices of environmental economics. It includes both
theoretical and empirical work. International in scope, it addresses issues of current
and future concern in both East and West and in developed and developing coun-
tries.
The main purpose of the series is to create a forum for the publication of high
quality work and to show how economic analysis can make a contribution to under-
standing and resolving the environmental problems confronting the world in the
twenty-first century.
Recent titles in the series include:
Designing International Environmental Agreements
Incentive Compatible Strategies for Cost-Effective Cooperation
Carsten Schmidt
Spatial Environmental and Resource Economics
The Selected Essays of Charles D. Kolstad
Charles D. Kolstad
Economic Theories of International Environmental Cooperation
Carsten Helm
Negotiating Environmental Quality
Policy Implementation in Germany and the United States
Markus A. Lehmann
Game Theory and International Environmental Cooperation
Michael Finus
Sustainable Small-scale Forestry
Socio-economic Analysis and Policy
Edited by S.R. Harrison, J.L. Herbohn and K.F. Herbohn
Environmental Economics and Public Policy
Selected Papers of Robert N. Stavins, 1988–1999
Robert N. Stavins
International Environmental Externalities and the Double Dividend
Sebastian Killinger
Global Emissions Trading
Key Issues for Industrialized Countries
Edited by Suzi Kerr
The Choice Modelling Approach to Environmental Valuation
Edited by Jeff Bennett and Russell Blamey
Uncertainty and the Environment
Implications for Decision Making and Environmental Policy
Richard A. Young
Game Theory and
International
Environmental
Cooperation
Michael Finus
University of Hagen, Germany
Edward Elgar
Cheltenham, UK · Northampton, MA, USA
© Michael Finus 2001
Published by
Edward Elgar Publishing Limited
Glensanda House
Montpellier Parade
Cheltenham
Glos GL50 1UA
UK
Printed and bound in Great Britain by MPG Books Ltd, Bodmin, Cornwall
To Kerstin
Contents
List of figures xi
List of tables xiii
Acknowledgments xiv
1 Introduction 1
vii
viii Contents
Appendices 317
I Chapter 3: appendix 317
II Chapter 4: appendix 319
III Chapter 5: appendix 324
x Contents
References 380
Index 405
Figures
3.1 Payoff space of the chicken game in Matrix 3.3 35
3.2 Convex payoff space of the matrix game in Matrix 3.7 36
4.1 Two-stage sequential move chicken game 45
4.2 Simultaneous move chicken game 47
6.1 Payoff space of the extended PD games III and IV 78
6.2 Renegotiation-proof equilibrium payoffs in a finitely
repeated chicken game 83
7.1 Payoff space of the extended PD game V 93
7.2 Payoff space of the PD game 95
8.1 Issue linkage of asymmetric PD games I and II 107
9.1 Possible curvatures of benefit functions 122
9.2 Nash equilibrium and parameter variations in the global
emission game 124
9.3 Reaction functions of slope less than –1 127
9.4 Positively sloped reaction functions 128
9.5 Non-intersecting reaction functions 129
9.6 Parallel reaction functions 130
9.7 Non-linear reaction functions 131
9.8 Nash equilibrium and social optimum in the global emission
game 140
9.9 Pareto frontier in the global emission game 145
10.1 Stackelberg equilibrium 153
10.2 Conjectural variation outcomes 158
10.3 Emission reduction offer curves in abatement space 163
10.4 Emission reduction offer curves in emission space 164
10.5 Sequential move matching game 169
12.1 Subgame-perfect, weakly and strongly renegotiation-proof
payoff space 198
12.2 Weakly renegotiation-proof payoff space of type A and B
games 199
12.3 Subgame-perfect, weakly and strongly renegotiation-proof
emission space 200
12.4 Weakly renegotiation-proof emission space for reciprocal
and restricted punishments 206
xi
xii List of figures
xiii
Acknowledgments
In writing this book I have benefitted from the help of many people. Most
of all, I am indebted to my mentor at the University of Hagen, Professor
Alfred Endres. First, I benefitted from many discussions on the subject and
received many comments that substantially improved the selection of
topics and the style and clarity of the exposition, and helped me to avoid
many errors. Second, I learnt how to write papers in economics by being
involved in many joint projects with him. Third, he provided me with an
ideal working environment, reducing my workload for the department to
the minimum, so that I found enough time to conduct my research and
write this book. Fourth, I received plenty of encouragement during times
in which I felt progress was slow.
I also feel obliged to my second mentor at the University of Hagen,
Professor Arnold, for carefully reading through the manuscript and provid-
ing me with many comments and hints that helped me to avoid some mistakes.
I also would like to thank my former mentor at the University of
Giessen, Professor Roland Herrmann, who was the first to stimulate my
interest in research during my studies of Agricultural Economics. He
encouraged me to orientate my research towards international standards.
Moreover, Bianca Rundshagen, my colleague and co-author of some
papers, spent much time in explaining some difficult math with a great
amount of patience. I enjoyed working with her very much. Some of the
research on which this book is based would, undoubtedly, not have been
possible without her. She also read parts of this book and forced me to be
more precise with respect to the technical exposition of the subject, helped
me to eliminate much nonsense and to avoid some murky prose.
I have also received many comments on preliminary work from scholars
at international conferences, which helped to improve the quality of my
research. Though I cannot name them all, I would like to particularly
mention Professor Henk Folmer (University of Wageningen and Tilburg
University), Professor Ayre Hillman (Bar-Ilan University), Professor
Michael Hoel (University of Oslo), Professor Rüdiger Pethig (University
of Siegen), Professor Sigve Tjøtta (University of Bergen), Professor Henry
Tulkens (University of Louvain, CORE), Professor Ekko van Ierland
(University of Wageningen) and Professor Heinz Welsch (University of
Oldenburg).
xiv
Acknowledgments xv
M. Finus
Hagen, March 2000
1. Introduction
International environmental problems have received increasing attention
from economists in recent years. Two basic strands of literature may be dis-
tinguished. The first strand estimates the costs and benefits of various
abatement targets under different cost allocation rules. It also discusses
institutional issues and the design of treaties with respect to their efficiency.
In particular the problem of global warming caused by the so-called green-
house gases (see, for example, Brunner 1991; Cansier 1991; Chapman and
Drennen 1990; Cline 1992a, b; Crosson 1989; Grubb 1989; IPCC 1996a, b;
Manne and Richels 1991; Michaelis 1992; Nitze 1990; Nordhaus
1991a, b, c; Schelling 1991; and Welsch 1995) and the ‘acid rain’ problem
due to sulfur and nitrogen oxides (for example, Crocker 1984; Førsund and
Naevdal 1994; Foster 1993; Newbery 1990; Tahvonen et al. 1993; and
Welsch 1990) have been studied in depth.
The second strand of literature has approached the problem of interna-
tional pollution control from a game theoretical perspective (for example,
Alho 1992; Andersson 1991; Buchholz and Konrad 1994; Chen 1997;
Endres and Finus 1998a; Heister 1998; Kölle 1995; Kuhl 1987; Müller-
Fürstenberger and Stephan 1997; Van der Ploeg and de Zeeuw 1992; and
Welsch 1993). The incentive scheme of countries which sign a treaty and
the stabilization of international environmental agreements (IEAs) are
typical issues analyzed by this literature. This book is in the tradition of this
second strand of literature.
Since, broadly speaking, game theory analyzes the interaction between
agents and formulates hypotheses about their behavior and the final
outcome in games, international environmental problems are particularly
suited to analysis by these methods. Global and transboundary emissions
exhibit a negative externality not only in the country of origin but in
other countries too. Hence, there is a high interdependence between
countries and strategic considerations enter the scene. Strategic aspects
are particularly important in international pollution control since there
is no ‘world government’ which could enforce IEAs. Therefore, the free-
rider problem is a distinguishing feature of international environmental
policy. Though countries are usually better off by coordinating their envi-
ronmental policy, cooperation is often difficult to achieve. Since each
country has only a marginal effect on aggregate emissions, it is always
1
2 Game theory and international environmental cooperation
all proofs is emphasized, those readers who are less interested in the tech-
nical part of a proof should be able to understand the central idea of all
results from reading the text. Thus, this book has the, surely, heroic aim of
being self-contained.
In what follows, Chapter 2 introduces some important terms of game
theory (Section 2.1) and some frequently used notations in this book
(Section 2.2). The second chapter also provides a taxonomy of game theory
(Section 2.3). In the course of the discussion of this taxonomy it will be
emphasized which parts of game theory are covered in this book.
Therefore, the outline of the book has been left to Section 2.4.
NOTES
1. For a complete classification of environmental problems, see Siebert (1985, 1992).
2. However, this does not imply that environmental damages are equally perceived and eval-
uated.
3. The details of this setting and the terms mentioned will be explained in Chapter 2.
4. In the environmental economics context, see, for instance, the volumes edited by Hanley
and Folmer (1998) and Pethig (1992).
5. Though some proofs may be very long and therefore may look tricky, basic algebra is
sufficient to follow all the proofs.
2. Important terms, notation and
classification of games
2.1 TERMS
The essential elements of a game are the players, actions, strategies, out-
comes, payoffs, equilibria, the information and the order of the game (see,
for example, Rasmusen 1989, pp. 23ff.). The players are the actors in the
game who take decisions. In the international environmental context the
players are countries or the political representatives of countries, such as
politicians or diplomats. Players can take actions, such as making catalytic
converters for automobiles mandatory or not.
In contrast, a strategy is a complete plan of action for each contingency
which might arise during the game. In a game comprising several stages, a
strategy specifies how a player reacts at each point in time to all possible
actions of fellow players. For instance, a participant to an IEA must specify
an answer if a signatory to an IEA complies with its obligations but also if
it violates the agreement.
A particular combination of actions (resulting from the play of some
strategy combination) leads to the outcome of the game. For instance, in the
catalytic converter example outcomes could be measured as the nitrogen
oxide concentration in the air, which depends on how many and which
governments introduce stricter car regulations. Alternatively, the outcome
could also be measured with respect to some other environmental index.
The choice will depend on the focus of the analysis (Rasmusen 1989, p. 25).
However, since game theory is mainly concerned with predicting which
of the possible outcomes will emerge in a game and which strategies will be
played in equilibrium, it is more important to have information on how
players evaluate these outcomes. Therefore, the outcomes must be trans-
formed into some form of utility. The utility derived from an outcome is
called the payoff to a player.
If all action combinations and their associated payoffs are known in a
game, the possible strategies for each player can be determined. If each
player chooses an equilibrium strategy, this strategy combination leads to
the outcome and the equilibrium of the game, which is sometimes also called
7
8 Game theory and international environmental cooperation
2.2 NOTATION
the book. The set of elements is denoted by a capital letter and the elements
themselves with the same lower-case letter. For indices we use lower-case
letters and the last element of an ordered set is denoted by the same capital
letter as the set itself. The only exception is the index i which runs from 1 to
N (and not from 1 to I). This is because the letter ‘I’ also denotes the set of
players and it has become a notational convention in game theory and
microeconomics to use N for the number of players.
In line with this convention, we denote a strategy by si r, r {1, ..., Ri},
where the first subscript refers to the player and the second to the strategy.
Si refers to country i’s strategy space and S is the strategy space of the entire
game where S S1 ...SN or S Si (see, for example, Friedman 1986,
pp. 23ff.; and Eichberger 1993, p. 64). The number of strategies may be
finite or infinite.
If each player plays a particular strategy s*i in equilibrium, we write s*
* * ). If there is a single equilibrium point in a game, we haveS* {s*}
(s 1, ..., sN
and if there is more than one equilibrium point this may be expressed by S*
{s*(1), s*(2), ...}. Generally, asterisks will be used throughout the book to
indicate an equilibrium. If the emphasis is on a particular concept, initials
will be used; for example, sN* for a ‘Nash equilibrium’.
The payoff to a player derived from some action combination is denoted
i l, l{1, ..., L}. More explicitly, we may write i l(a), where a denotes some
action combination. This stresses that a payoff function maps action com-
binations on payoffs. Since the actions which are chosen depend on the
strategies players pursue, we may also write il(s), with s denoting some
strategy combination.
The payoff set (or payoff space) of player i is i, and the set of all payoffs
. If the payoff space is infinite we write i [Li , U i ], where i denotes the
L
lower bound and i the upper bound of player i’s payoff space.
U
Character of Games
organization, which can enforce an agreement the game belongs to the cat-
egory of non-cooperative games.
It is important to note that the term ‘non-cooperative’ should not be mis-
interpreted as implying a general conflict between players and that of co-
operative games as an absence of conflict. For instance, most bargaining
problems in game theory, like the division of a cake between players, are
analyzed using concepts of bargaining theory, which belong (predomi-
nantly) to cooperative game theory, though an obvious conflict between
players exists.
Another instance of a non-cooperative game where there is no conflict is
a positive externality game. Imagine customers in a restaurant sitting on a
terrace, enjoying cows grazing in a meadow nearby. Hence, the production
of milk by the farmer exhibits a positive externality for the restaurant
owner (by attracting more customers). If the farmer stops producing milk,
the restaurant owner can negotiate with the farmer about transfer pay-
ments but cannot force him to continue production.
As we stated in the Introduction, since there is no agency at the global
level that is empowered to sanction the breach of a contract (which makes
cooperation in international pollution control so difficult), almost all
aspects of international environmental problems belong to the realm of
non-cooperative games (Congleton 1992; Heister 1997, p. 4). We depart
from this ‘rule’ only in Chapters 10 and 13 since some important contribu-
Terms, notation and classification of games 11
Cost–Benefit Structure
The second criterion in Table 2.1 classifies games as either constant sum or
non-constant sum games (Friedman 1986, pp. 30ff.). In constant sum games
players can only gain at the expense of other players and hence coopera-
tion cannot generate additional welfare. This is why they are sometimes
also called strictly competitive games. A typical example of a constant sum
game is matching pennies: player 1 gains if head–head or tail–tail appears,
otherwise player 2 gains. Since we are concerned with the formation and
12 Game theory and international environmental cooperation
Number of Players
{{1}, {2}, {3}}, {{1, 2}, {3}}, {{1, 3}, {2}}, {{1}, {2, 3}}, {{1, 2, 3}}.
The first possibility assumes that each player forms a coalition by him- or
herself. The next three possibilities are coalitions among two players.
Finally, we have the grand coalition, which is a coalition comprising all
players. Hence, it is easy to perceive that things get rather complicated if N
is large (for example, for N4 there are 15 possible permutations).
Therefore, some simplifying assumptions may be necessary to determine an
equilibrium in a coalition game. For instance, many papers on coalition for-
mation in international pollution control assume symmetric countries (see
Chapters 13 and 15). Moreover, other plausible assumptions, for example,
about the behavior of countries, can also reduce the complexity of a game,
making it possible to solve the game eventually (see Chapters 13–15).
Due to these complications we shall proceed stepwise. In Chapters 3–12
we either contemplate only two players or, in the case of N players, stra-
tegic aspects of the formation of sub-coalitions are discarded. Subse-
quently, in Chapters 13–15, the aspect of coalition formation is taken up
separately.
Strategy Space
one to illustrate many results and concepts of game theory with the help of
simple matrix games. This is why Chapters 3–8 exclusively focus on discrete
decisions, and only from Chapter 9 onward is the analysis extended to cover
continuous choices as well. However, both assumptions are less distinct
than one would assume. By increasing the number of actions, which also
raises the number of strategies in a game, a discrete strategy space trans-
forms gradually into a continuous strategy space. Moreover, when players
are allowed to randomize between strategies (mixing of strategies), that is,
playing several strategies with some probability, a discrete strategy space
turns into a continuous strategy space. In other words, the strategy space
can be convexified through mixing strategies (see Section 3.6).
Time Horizon
One can basically distinguish between static and dynamic games. Static
games are also called one-shot games (Chapters 3 and 9). Dynamic games
can further be divided into finite (Chapters 4, 6 and 10) and infinite
(Chapters 5, 7, 12 and 14) games. In finite games the game is played over
some (limited) time and the termination of the game is known with cer-
tainty. In contrast, in infinite games the game either lasts until perpetuity
or the end of the game is not known with certainty. There is also a partic-
ular type of game which does not belong to either of these categories.
Basically, these are static representations of dynamic games, but where a
time explicit story is missing. Time implicit models are treated in Chapters
10, 13 and 15.
The complexity of dynamic games is higher than that of static games.
This is immediately apparent when recalling the definition of a strategy.
The longer the time horizon, the more contingencies can arise during a
game for which a strategy must specify an action. Hence, it is sometimes
useful to approximate a situation as a one-shot game provided this
simplification ‘preserves’ the structure of the game. For instance, if the
investment in a new and cleaner power plant is associated with high sunk
costs, the investment decision might be modeled as a one-shot game. This
decision is not reviewed regularly by politicians and might be based on the
net present value of the payoff stream of this project.
In contrast, if an industrial country has to decide whether and how much
to contribute to an international environmental fund designed to foster
environmental projects in developing countries, the decision might be
better modeled as a dynamic game.7 Each year parliament debates the
budget for the forthcoming year. If this fund has been set up only for a
certain number of years (for example, because it was earmarked only for
the initial stage of an agreement), then the game is finite. However, if this
14 Game theory and international environmental cooperation
fund has been established without any further specification of its resolu-
tion, the game should be viewed as an infinite game.
In the context of dynamic games one can distinguish between two classes
of strategies: strategies that account for past actions, which are called closed
loop strategies (sometimes also called feedback strategies), and those that
ignore the past, which are called open loop strategies (Rasmusen 1989,
p. 100). Though open loop strategies are simpler to analyze, in most games
there is no reason why players should not use historical information.
Therefore, we do not restrict our attention to open loop strategies.
An immediate prerequisite for closed loop strategies is the assumption of
perfect recall of all past actions, that is, complete memory of the history of
the game (Brandenburger and Dekel 1989). This assumption seems to be
justified in the context of international environmental agreements because
all relevant data, as for instance historical emission levels, is statistically
reported.8, 9
Time Dimension
The time dimension refers to the feature whether time is counted in discrete
time intervals (t0, 1, 2, ..., T), also called periods, or whether a game is
viewed in continuous time. Discrete time intervals have two implications.
First, if the violation of a treaty is detected, punishment can take place only
one period later at the earliest and will therefore be delayed. Second, a
country which violates a treaty receives a transitory gain before being pun-
ished, which makes free-riding attractive. In contrast, in continuous time
actions can be taken at each instant (if no time lag is explicitly introduced
into the decision process) and hence free-riding is less of a problem. Since
political decisions usually take time and since we believe free-riding is a
feature by which IEAs are typically plagued, we assume in all time-explicit
models discrete time intervals. Only the time implicit models in Chapters
10 and 13 require a continuous time interpretation.
Time Structure
Information Requirement
Games in which all information is known to all players are called games of
complete information. If one piece of information is not available to at least
16 Game theory and international environmental cooperation
Sequence of Moves
Remark
From what has been said above it is clear that each criterion of Table 2.1
reflects a polar pair of assumptions characterizing a game. Most generally,
assumption (a) within each category simplifies the analysis, whereas
Terms, notation and classification of games 17
assumption (b) makes a game richer and more interesting, but also more
complicated. As a general rule, it seems sensible to follow the motto of ‘no
fat modeling’ – a phrase coined by Rasmusen (1989, pp. 14ff.). This implies
that simple models should be chosen to bring out the gist of the analyzed
problem. Models which are too complicated may distract attention from
the main focus of an analysis. For instance, if the role of time is the main
focus of an analysis, then it might be helpful to consider only two players
in a first step. However, if we are interested in the coalition formation of
countries, then we may start with a static or simple two-stage game model
and gradually extend the analysis to longer-lasting games.
Of course, there is also the danger of using too simple models which may
not capture important features driving a result in reality. For instance, if the
decision on whether to switch off and dismantle a nuclear power plant is
modeled as a static game with discrete strategy space, this may seem an
appropriate simplification. First of all, due to the high sunk costs of the
decision, a possible revision of a taken action can be ruled out and this is
captured by a static game. Second, a discrete strategy space also seems a
good approximation because the choice will be either to dismantle the
power plant or to operate it at a low, middle or high capacity. In contrast,
if countries negotiate an agreement on greenhouse gases this may be better
modeled as a dynamic and continuous strategy space game. Emission
reductions may range from 0 to 100 percent and the treaty might be in force
for a couple of years.
Unfortunately, sometimes assumptions cannot reflect what is the most
appropriate in the particular case, but have to be guided by the necessity
that dropping too many simplifying assumptions makes it impossible to
solve a game analytically, and one has to rely on simulations. Therefore, in
most parts of the book we follow a stepwise approach, altering only one
assumption at a time. However, though we have already ruled out some
assumptions with respect to the criteria displayed in Table 2.1, lack of space
forces us to confine the subsequent analysis further, considering only some
interesting combinations of assumptions in this book. The general struc-
ture of the book is briefly laid out below.
NOTES
1. In order to characterize an equilibrium, it is necessary to list the equilibrium strategies
of all players. Thus, the notion of an equilibrium in game theory is slightly different from
that in economics. For instance, whereas in economics the price resulting from the com-
petition between firms in a Cournot oligopoly is called an ‘equilibrium price’, this is an
outcome in game theory. From a game theoretical perspective, the strategies, which are
the chosen quantities by each firm in the simple static version of this game, would also
have to be listed for a full characterization of the equilibrium.
2. For details, see Section 2.3.
3. Generally, the set of players may also be infinite, see, for example, Fudenberg and Tirole
(1996) for examples. However, in this book no such instances will be encountered.
4. The terms action space and action set can be used synonymously. However, the former
term is used mostly in games with an infinite number of actions.
5. Many bargaining solutions belong to cooperative game theory, as for instance the Nash
bargaining solution or the Shapley value. In Chapter 11 we argue that those solutions
are based on some assumptions which are often violated in reality and therefore are not
treated in this book.
6. For a discussion of the problems associated with the aggregation of welfare functions,
see for instance Boadway and Bruce (1993); Just et al. (1982); and Sen (1984).
7. See, for instance, the Rio Declaration of 1992 and its successor protocols by which the
Global Environmental Facility (GEF) was set up. Industrial countries are supposed to
contribute to the GEF from which developing countries can receive financial support for
environmental projects (see, for example, Bergesen and Parmann 1997, pp. 90ff.; Hanley
et al. 1997, pp. 171ff.; Jordan and Werksman 1996, pp. 247ff.; Kummer 1994, p. 260; and
Sand 1994, pp. 98ff.). Contributions by industrial countries to a fund are also required
by the Convention on Biological Diversity (Beyerlin 1996, p. 617; and Gündling 1996,
pp. 806ff.) and the Montreal Protocol on the Protection of the Ozone Layer (DeSombre
and Kauffman 1996, pp. 89ff.; Ladenburger 1996, pp. 72ff.; and Sand 1996, p. 56). A
model analyzing the stability of such transfers is provided by Barrett (1994a).
8. For instance, statistical reports on historical emissions are issued by the Norwegian
Meteorological Institute in Oslo which provides emission data for the monitoring
program EMEP (Emission Monitoring and Evaluation Program) which is part of the
Convention on the Reduction of Long-range Transboundary Emissions in Europe
(Geneva 1979). In these reports data on SO2, NOx, and VOCs (volatile organic com-
pounds) have been reported since 1980.
Note that the fact of biased reporting by governments, which frequently occurs in
reality, is an issue of incomplete information but not of bounded recall.
9. The assumption of bounded recall is sometimes introduced into differential games (the
term is explained below) in order to simplify the analysis. A typical open loop strategy
assumed in this literature is a Markov strategy, where only the immediate previous
history is considered. See, for example, Dutta and Sundaram (1993); Hoel (1992b);
Mäler (1991, 1992); and Wirl (1994) in the environmental economics context.
10. Most game theorists follow this definition, implying that the finite repetition of a game
is not called a supergame. This is also how we shall use this term. For a different inter-
pretation, see, for instance, Friedman (1986, pp. 94ff.). He also calls finitely repeated
games supergames.
11. General references for dynamic optimization problems are Kamien and Schwartz (1991)
and Seierstad and Sydsater (1987). Differential games are treated in Basar and Olsder
20 Game theory and international environmental cooperation
(1982); Clemhout and Wan (1994); and Friedman (1994). In the environmental and
resource economic context they have been applied by Dockner and van Long (1993);
Hoel (1992b); Martin et al. (1993); and Tahvonen (1994).
12. This is not true for finite games, as we shall show in Chapters 4 and 6.
13. Exceptions include Chillemi (1996); Laffont (1993); and Steiner (1997).
14. A particular form of incomplete information is uncertainty of information (see, for
example, Rasmusen 1989, pp. 52ff.). Uncertainty refers to the particular feature whereby
after a player has moved, ‘nature’ makes a move which is either not completely observ-
able by players and/or cannot be influenced by them (for example, Holler and Illing 1993,
pp. 36ff.). An example is the global warming game where there is great uncertainty as to
how the climate will change in the future. To solve for such a game, assumptions with
respect to the risk attitude of agents (risk averse, risk neutral or risk loving) have to be
made. Though most books on game theory cover incomplete information in general,
they do not treat uncertainty of information since this warrants an analytical approach
in its own right. Exceptions include Chichilnisky et al. (1998) and Machina (1989). In
the context of international pollution control, see, for example, Endres and Ohl (1998a,
b); Mohr and Thomas (1998); and Na and Shin (1998).
3. Static games with discrete strategy
space
3.1 INTRODUCTION
21
22 Game theory and international environmental cooperation
a2 na2
3.2 1.4
a1
3.2 4.4
4.4 2
na1
1.4 2
nations, and (c) the payoffs received by each player for each possible
strategy combination in this game. For short, the normal form of a game
is given by (I, S, ).
Note that, due to the assumption of a one-shot game, actions and strate-
gies coincide in this game. It can easily be checked that in the PD game each
country has the dominant strategy ‘no abatement’; that is, regardless what
the neighboring country does, strategy nai delivers the highest payoff. If
country j plays aj, nai delivers a payoff of 4.4 whereas ai leads to a payoff of
only 3.2 to country i. If country j plays naj, country i nets 2 by playing nai
instead of 1.4 when playing ai. Hence, the unique equilibrium in this game
is SD* {(na1, na2)} where the superscript D stands for ‘dominant strategy
equilibrium’. The associated payoffs are printed in bold in Matrix 3.1. The
equilibrium of this game implies that cooperation fails and both countries
get stuck in the Pareto-inferior status quo. Hence, delegates of both coun-
tries either do not sign an agreement in the first place because they antici-
pate the instability of such a deal, or they sign an agreement but neither
country will comply.2
The predictive power of this kind of equilibrium may be regarded as
quite high. Each player has one strategy which leads under every contin-
gency to a higher payoff than any other strategy. That is, a player can deter-
mine his/her ‘best strategy’ without having to rely on any speculation about
the behavior of his/her opponents. Hence, this equilibrium is also immune
to any kind of strategic considerations of players and can therefore be
regarded as very robust.
Note that for the incentive structure in the PD game (and in all the games
discussed below) it is neither necessary for utility to be measured cardinally
(for example, a payoff of 3.2 gives 3.2/2 times more utility compared to the
status quo) nor that utility can be compared across countries (for example,
a payoff of 4.4 generates higher utility to country 1 than a payoff of 3.2 to
country 2). Also the assumption of symmetric payoffs is not necessary.
With reference to Matrix 3.2, all that is required for a game to qualify as a
(basic) PD game is ci
ai
di
bi iI.
However, to make the following analysis interesting from a policy point
24 Game theory and international environmental cooperation
of view, we assume that (1) utility can be aggregated, and that (2) a1 a2
b1 c2 and a1 a2
c1 b2 hold. The first assumption automatically implies
that a1 a2
d1 d2 (since ai
di iI) and hence together with the second
assumption ‘mutual cooperation’ is globally optimal.
a2 na2
a1 b1
a1
a2 c2
c1 d1
na1
b2 d2
To set the stage for a comparison of the outcome in the static PD game
with outcomes in other models and settings, we briefly review three main
assumptions which are responsible for the pessimistic result obtained
above.4
First, the discrete strategy space restricts the question of whether to
cooperate on environmental protection to a binary choice. In contrast, in
a continuous strategy space where a country can tune its decision more
finely, more optimistic results can be obtained (see Chapter 9). However,
if the policy options discussed on an international platform are clearly
distinct from each other, a discrete strategy space is the appropriate
setting.
Second, in a static setting no threats and rewards can be used to prevent
a country from seeking its immediate interests. Below, we shall show that
cooperation might be possible if a game lasts longer. However, as pointed
out in Section 2.3, a static setting is appropriate if an action is associated
with high sunk cost and is not reviewed at short intervals.
Third, the cost–benefit structure generates this Pareto-inferior result.
Apart from the free-rider incentive it does not pay a country to contribute
unilaterally to the international public good ‘environmental quality’. In
fact, a unilateral contribution is the worst outcome from the contributor’s
perspective.
Looking at the record of international environmental protection it is
evident that there has been successful cooperation in a few areas, though
for most international environmental problems either less has been
Static games with discrete strategy space 25
In chicken games the prospects for a cleaner environment are higher (see,
for example, Holler and Illing 1996, pp. 89ff.; Rasmusen 1995, pp. 72ff.).
Still, the free-rider incentive exists. However, the cost–benefit structure is
such that it pays a country to invest unilaterally, though it prefers its neigh-
bor also to contribute to a joint environmental policy. An example is shown
in Matrix 3.3.
a2 na2
4.6 2.2
a1
4.6 5.2
5.2 2
na1
2.2 2
and s*i Si. That is, in equilibrium every strategy is a best response to
the best strategies of the other players.
In the chicken game the strategy combinations (a1, na2) and (na1, a2) con-
stitute such mutual best responses to each other. That is, each player has no
incentive to deviate from his/her strategy, given that the other player does
not deviate. Hence,SN* {(a1, na2), (na1, a2)}. The payoffs in the two NE
are printed in bold in Matrix 3.3.6
There are basically two interpretations of a Nash equilibrium:
That is, country i invests if the probability that country j invests is less than
1/4. It does not invest if this probability is greater than 1/4, and it is
indifferent between both strategies if this probability is exactly 1/4. Hence,
if both countries invest with probability 1/4, both countries have no incen-
tive to deviate from their strategy and beliefs are mutually confirmed. Thus
p* (p*1 1/4, p*2 1/4) is the Nash equilibrium in mixed strategies. The
resulting payoff to each country is 2.8 which is more in line with the sym-
metric structure of the chicken game in Matrix 3.3 than those payoffs
derived from the pure Nash equilibria.8
Let us now define more formally what has been derived above:
Note that now in the general case pi denotes some probability distribution
of player i and not a probability that a particular strategy sir is played. From
the definition it is obvious that an NE in mixed strategy is a straightforward
extension of the definition of an NE in pure strategies. In fact, a pure strat-
egy is a special case of a mixed strategy which is played with probability 1.
Therefore, pure strategies are a subset of mixed strategies (Eichberger 1993,
pp. 20ff.).
28 Game theory and international environmental cooperation
Before dealing with some other games and other technical features of
mixed strategies in the subsequent sections, we shall pause here and briefly
discuss the pros and cons of mixed strategies.
Cons
Pros
In assurance games the payoff structure is such that it pays neither country
to invest unilaterally. A country prefers either the status quo or joint invest-
ment. The assurance payoff structure could be generated by economies of
scale in the development and production of new and cleaner power plants.
If the costs of R&D are high, an investment in the new technology might
30 Game theory and international environmental cooperation
a2 na2 a2 na2
9.5 0 9.5 5
a1 a1
9.5 1 9.5 8
1 2 8 2
na1 na1
0 2 5 2
With respect to the general payoff matrix (Matrix 3.2), the inequalities
ai
di , di
bi and di
ci iI define the basic payoff structure of an assu-
rance game. Though bi
ci has been assumed in Matrix 3.4, bi ci also pre-
serves the incentive structure of an assurance game. Obviously, mutual
investment is globally optimal if utility can be aggregated.
In this section we briefly discuss the extension of the previous models from
two countries to the general case of N countries. Analytically, this exten-
sion is straightforward. Conceptually, this is a little more difficult in the case
of a chicken and an assurance game. A simple way to display the incentive
structure in the symmetric N-country case is suggested by Matrix 3.6.
0 1 ... N* 1 N 1
N*1 N1
ai MB1i MCi MB1i MB2i MCi ... MB
j
j1 MC
i i MB
j
i
j1 MC
i
N*1 N1
nai 0 MB1i ...
j
MB1i MB
j
i
j
MBij stands for marginal benefits and MCi for marginal costs which
accrue to country i from the investment in the new technology and the
status quo payoff has been normalized to zero. The term ‘marginal’ is used
to stress that these benefits and costs occur additionally compared to the
status quo. For instance, MBi4 is the marginal benefit country i receives if
32 Game theory and international environmental cooperation
3.5.1 PD Games
The incentive structure of no-conflict games implies that the only NE,
which is at the same time an equilibrium in dominant strategies, has all
countries investing. Hence, we have to require MBij1
MCi i and j and
MBij
0 i I and 1 jN1. These conditions also ensure that co-
operation is attractive and full cooperation globally optimal.
In its ‘pure’ version the incentive structure of assurance games implies that
either all countries or no country invest. Moreover, joint investment is pre-
ferred among these two options. For such an incentive structure MBij1
MCi for all i I and j
N1, and MBij1
MCi and MBij
MCi for j
N1 and all iI is required. Since in an assurance game cooperation
among some countries never pays a country as long as not all countries
cooperate – even though it does not contribute anything to joint abatement
– MBij
0 for all 1 j
N1 must hold. That is, only if all countries
invest will marginal benefits exceed a threshold so that the investment is
beneficial to country i and globally optimal.
Static games with discrete strategy space 33
3.5.5 Discussion
From the discussion it is evident that with the help of the simple models
presented in this chapter an environmental agreement ranging from zero to
N participating countries can be depicted by varying the relation of margi-
nal benefits to marginal costs. For example, observing that a pollution
problem is not regulated within an IEA, one may suspect that the
benefit–cost structure is that of a PD game. Many authors argue that for
most international problems, in particular global ones, marginal costs
exceed marginal benefits on a country basis. This could explain why so far
not much success can be detected regarding the problem of global
warming. In contrast, if an effective IEA is in operation with some but not
all countries participating, a chicken or modified assurance game is a pos-
sible incentive pattern underlying the problem. Following the main opinion
in the literature, judging the Montreal Protocol as a relatively successful
treaty, the ozone problem could be such an instance.
However, though the models in Chapters 13–15 on coalition formation
confirm that the cost–benefit structure is a crucial variable for the success of
a treaty, for a final evaluation we first have to analyze whether modifications
of the assumptions made so far can also explain partial or full cooperation.
Since free-riding is a problem, particularly in PD and chicken games, the
analysis in Chapters 4–8 is confined to these two types of games.
π2
9
8
7
6
(2.2, 5.2)
5 (4.6, 4.6)
4 Π Pareto frontier
• πA
3
2 (5.2, 2.2)
(2, 2)
1
0 1 2 3 4 5 6 7 8 9 π1
π2
(1, 5)
5
3 πA = (3, 3)
2 Π
1 (5, 1)
0 1 2 3 4 5 π1
Figure 3.2 Convex payoff space of the matrix game in Matrix 3.7
in some way. Consider the simple example in Matrix 3.7 which is illustrated
in Figure 3.2. The example is chosen for illustrative purpose only and is not
economically motivated.16
This game has two NE in pure strategies, that is, SN* {(a11, a21) (a12,
a22)}, where the associated payoff tuples, that is, N*(1) (5, 1) and
N*(2) (1, 5), are printed in bold in Matrix 3.7. There is also NE in mixed
strategies, that is, pN* (p*i 1/6, pj* 5/6), with associated payoff tuple
N*(3) (5/6, 5/6).
Suppose we want to find an uncorrelated mixed strategy to generate the
payoff tuple A (3, 3) which is an element of the convex payoff space .
This implies that the following two equations must be satisfied:
5 0 1
a11 a11 0 a11 z1 z2
2
1 0
0 1 1
a12 a12 0 a12 z3 z4
2
0 5
Whereas it was the aim of the last section to demonstrate that in some
games correlated strategies are necessary to convexify the entire payoff
space, we are now concerned to show how correlated strategies can be used
to improve upon a non-cooperative status quo. By constructing an appro-
priate random device, global welfare can be raised in some games which
would not be possible by using pure strategy or an uncorrelated mixed NE
only. Before studying the details, we wish to clarify two issues first:
38 Game theory and international environmental cooperation
3.7.1 Motivation
3.7.2 Compliance
A basic prerequisite that players agree to make a third party responsible for
coordination is that the rules according to which these recommendations
are given are common knowledge. Moreover, due to our fundamental
assumption of a non-cooperative game, this coordinator has no enforce-
ment power. Hence, the recommendations must be self-enforcing and the
correlated strategies must constitute a Nash equilibrium in correlated strat-
egies. We define:17
z (s , s
u1
*
u i i
* )
i si s*i , si s*i and i I hold.
u1
3.7.3 Application
Proposition 3.1
In a two-player chicken game there is a Nash equilibrium in correlated
strategies with aggregate payoffs higher than in the pure and uncorre-
lated mixed strategy equilibria.
Proof: See Appendix I.
3.7.4 Discussion
NOTES
1. The game derives its original name from the following situation (see, for example, Luce
and Raiffa 1957, pp. 95ff.): Two suspects are arrested for some serious crime. The police
lack sufficient evidence to convict the suspects unless one confesses. Therefore, the police
confront the suspects with the following alternatives. If neither confesses, then both will
be convicted for some minor offenses and sentenced to a short period in jail, say one year.
If both confess, both will be sentenced to jail for five years. If one confesses and the other
does not, the one who confesses will be released and the other will be sentenced to jail
for ten years. If both prisoners were to agree not to confess, both would be released after
one year. However, such an agreement is not stable; both prisoners will confess and will
be sentenced to five years in jail.
2. If there existed an authority (in the sense of cooperative game theory) which was able to
enforce international treaties, then the problem could be solved trivially. Holler and Illing
(1993, pp. 23ff.) call this the Mafia solution to the PD game. In the original game (see
the previous note) this implies that both prisoners do not confess because of fear of
being killed by the Mafia members.
3. The following definitions assume the general case of possibly more than two players.
4. The degree of cooperation in experimental PD-type situations is reported in Axelrod
(1984); Feeny et al. (1990); Gardner and Ostrom (1991); and Ostrom et al. (1990, 1994).
5. A report of the US government identifies 170 multilateral IEAs (USITC 1991). Up until
now only political scientists have launched and conducted comprehensive empirical pro-
jects on the evaluation of the effectiveness of those IEAs. Due to different evaluation
methods the results are rather ambiguous; see, for example, Benedick and Pronove
(1992); Haas et al. (1993); Sand (1992); Victor et al. (1998); Werksman (1997); and Young
and von Moltke (1994).
6. From the definition of a Nash equilibrium it should be clear that a dominant strategy
equilibrium is always a Nash equilibrium too, that is, SD* SN*. (However, the opposite
does not hold.)
7. The chicken game derives its name from the following situation which has become
famous in the movie Rebel without a Cause (see Rasmusen 1995, p. 87). Two young rebels
race in their cars towards a cliff. The player who jumps out first is the chicken and the
other the winner. The game is constructed such that each wants to be the winner, but
prefers to be the chicken instead of being killed. Also, both players prefer to be embar-
rassed when they either both back out before the race or jump out of the car simultane-
ously. In the original game the strategies are continuous and therefore more complicated.
However, the discrete representation as given in Matrix 3.3 preserves the basic incentive
structure (ai would be the strategy ‘take no part in the race’ or ‘jump out first’; nai would
represent the strategy ‘jump out last’).
8. The expected payoff is computed from 0.25 · 0.25· 4.60.25· 0.75· 2.20.75· 0.25· 0.25·
0.250.75· 0.75· 2 2.8.
9. In the context of incomplete information there are more interpretations of mixed strat-
egies (see Gibbons 1997; and Harsanyi 1973).
10. Of course, the example requires expanding the analysis to N countries. However, this
extension is straightforward, as we argue in Section 3.5.
11. The term was coined by Schelling (1960). He used this term in games with several equi-
libria, where an equilibrium (or some equilibria) is (are) more convincing than others.
The criteria which define a focal point are rather vague and, strictly speaking, do not
follow from the game itself. They require some additional story, for example, psycholog-
ical or sociological motives and so on. In the present context it seems obvious that the
players play the Pareto-superior equilibrium because it is in both countries’ interest. For
a discussion of this point see, for example, Rasmusen (1989, pp. 36ff., pp. 228ff.).
12. There is a third Nash equilibrium in mixed strategies in which each country invests with
probability p*1 p*2 2/10.5. However, this would generate a lower payoff to each player
Static games with discrete strategy space 41
(i 1.8) than in either of the two pure strategy equilibria. Therefore, using the focal
point argument, this equilibrium can be discarded.
13. In Chapter 6 and the subsequent chapters we discuss the equilibrium concept of a strong
Nash equilibrium (Aumann 1959) which would pick sN*(1) as the only equilibrium in the
assurance game. Roughly speaking, in a static context this concept requires a Nash equi-
librium to be Pareto-efficient. In many games, however, this requirement implies that no
equilibrium exists at all, as would be the case in the PD game. In the chicken game all
pure uncorrelated strategy Nash equilibria are also strong Nash equilibria; this is not so,
however, with the mixed strategy equilibrium.
14 The term ‘assurance game’ is probably due to the fact that players like to ‘assure’ each
other of their interest in cooperation, so that they can coordinate on sN*(1) instead of
getting stuck in sN*(2).
15. Of course, alternatively, economies of scale could be expressed byMCi / N*
0 instead
of MBij1 / N*
0. This also applies to the subsequently discussed cases.
16. This game is basically a version of what is known in the literature as the ‘battle of the
sexes’; see, for example, Rasmusen 1989, p. 34.
17. As a matter of terminology, henceforth, when talking about a Nash equilibrium, we are
referring to a pure strategy Nash equilibrium. With mixed Nash equilibrium we shall
refer to an uncorrelated mixed Nash equilibrium and reserve the term correlated strat-
egy Nash equilibrium for a Nash equilibrium in which correlated mixed strategies are
played.
4. Finite dynamic games with discrete
strategy space: a first approach
4.1 INTRODUCTION
For the first question the punishment options in a game are important.
Obviously, the harsher the punishment, the higher is the potential of deter-
rence from cheating. However, if the player conducting the punishment
also suffers some loss because of the punishment, credibility becomes an
important issue. In the game theoretical literature the problem of credibil-
ity has attracted great attention and we shall deal with this issue through-
out this book, gradually strengthening the definition of credibility.
The second question has to be answered by comparing the sums of peri-
odically accruing payoffs. This is done by means of discounting with the
following implications: the lower the discount rate, also called time prefer-
ence rate, the lower are short-term gains from free-riding weighted against
the long-term gains from cooperation; also, the lower the discount rate, the
42
Finite dynamic games with discrete strategy space I 43
| stands for ‘provided that ...’ or ‘conditional on ...’. For instance, s21
(a2 |a1; na2 |na1) reads as ‘invest provided country 1 invests and do not invest
if country 1 does not invest’. For country 1 strategies and actions coincide
(as in the one-shot game) since it moves first; however, for country 2 actions
and strategies differ since it can condition its decision on country 1’s behav-
ior.
A routine check reveals that there are two (pure strategy) Nash equilib-
ria in this game: SN {sN(1) (s12, s22), sN(2) (s11, s24)} of which the payoffs
are printed in bold in Matrix 4.1. However, the question arises whether the
second equilibrium involves credible strategies. It implies that country 2
plays a strategy of ‘unconditional no investment’ and country 1 gives in to
this threat. That is, country 2 threatens to play na2 if country 1 chooses na1
even though its best reply is a2.
2 5.2 5.2 2
s12 na1
2 2.2 2.2 2
From the game tree in Figure 4.1, which emphasizes the sequence of
moves, the argument is even more transparent. Whereas a matrix is one way
to represent a game in normal form, a game tree is frequently used to rep-
resent a game in extensive form which is defined as follows (see, for example,
Eichberger 1993, pp. 14ff.; Gibbons 1992, pp. 115ff.):1
4 4.5, 4.6
a2
2
subgame 1
a1
na2
5 2.2, 5.2
1
6 5.2, 2.2
a2
na1
3 subgame 2
na2
7 2, 2
In Figure 4.1 the threat of country 2 entailed in the second Nash equi-
librium is indicated by the broken line whereas country 2’s best replies in
the second stage are represented by the dotted lines branching off from
nodes 2 and 3. Since (due to the assumption of complete information)
country 1 knows country 2’s strategies, it can solve the game for country 2
(in particular it knows that country 2 cannot do any better than to play a2
if it plays na1) and can use its first-mover advantage to enforce its preferred
equilibrium sN(1). In other words, the threat contained in sN(2) is disclosed
as an empty threat. Thus in the sequential chicken game, though there are
two Nash equilibria (NE), only the first equilibrium sN(1) is credible, that is,
it is a subgame-perfect equilibrium (henceforth abbreviated SPE), that is,
sSPE sN(1). Hence, in a two-stage sequential move chicken game, the
outcome (unilateral investment) does not differ from the one-shot chicken
game, though equilibrium strategies differ.
The concept of subgame-perfect equilibrium is due to Selten (1965) and
derives its name from the concept’s requirement that in equilibrium a strat-
egy must be a best response in every subgame of the game. Simply speaking,
a subgame is a part of a game which can be viewed as a game in its own
right. For an extensive form representation a subgame can be defined as
follows:
46 Game theory and international environmental cooperation
For the understanding of this definition two more definitions are necessary
(Rasmusen 1989, pp. 48ff.):
4 4.6, 4.6
a2
a1
na2
5 2.2, 5.2
1
a2 6 5.2, 2.2
na1
3
na2
7 2, 2
hold – using the general payoffs of Matrix 3.2 – this is an equilibrium strat-
egy. In fact, it is an SPE since the threat implies playing an NE in every
remaining subgame. The play of a stage game NE in the terminal rounds is
necessary for two reasons:
1. In the last round a stage game NE must be played in any case since
deviations cannot be punished.
2. The alternation between the two Pareto-undominated NE in at least
two terminal rounds is necessary because if countries played the same
equilibrium twice one (potential) deviator could not be punished.
the agreement on (a1, a2). In other words, there is no reason for country 1
to follow a contingent cooperative strategy and to sacrifice a free-rider gain
since such behavior is not rewarded. Consequently, the stage game equilib-
rium sN(1) will also be played in the second last round and, due to the sym-
metric structure of the game, in any previous round. Thus, the only SPE in
this game involves the repeated play of sN(1).
However, since sN(2) (s11, s24) is a stage game NE, cooperation can be
established as a Nash equilibrium by using strategy sN(1) against country 1
and strategy sN(2) against country 2 as a punishment in case it deviates, as
in the simultaneous move version.
The results may be summarized as follows:
Proposition 4.1
In a finitely repeated simultaneous move two-player chicken game which
is played for T periods, assuming payoff relations ci ai bi di iI
(see Matrix 3.2) and no discounting, cooperation can be supported as a
subgame-perfect equilibrium (in pure strategies) for the first T-1 rounds.
In a sequential move chicken game, cooperation can only be established
as a Nash equilibrium.
Proposition 4.2
In a finitely repeated PD game the only equilibrium of the whole game
is the T-fold repetition of the Nash equilibrium of the stage game.
Taken together, playing a chicken game for more than one period can
resolve the deadlock caused by the free-rider incentive. If the number of
50 Game theory and international environmental cooperation
stages is large enough (and time is not discounted), the fully cooperative
outcome can approximately be obtained. If moves occur sequentially co-
operation can only be established as an NE; if they are taken simultane-
ously cooperation can be supported as an SPE. In a PD game the only
equilibrium (SPE and NE) involves the repeated play of the stage game NE
and cooperation fails.
4.3.1 Preliminaries
From the chicken game example it appeared that results may depend cru-
cially on whether moves occur simultaneously or sequentially. This is true
whether the constituent game is played once or repeatedly. However, the lit-
erature on repeated games either implicitly or explicitly restricts attention
to simultaneous moves only. This restriction may be defended on two
grounds: first, from a conceptual point of view, including sequential moves
renders definitions more complicated; second, from a motivational point of
view, one may argue that in longer-lasting games players move simultane-
ously ‘on average’. In particular, if the problem itself does not suggest a
certain order of moves, simultaneous moves seem a ‘natural’ assumption
(see Chapter 2). Acknowledging these arguments, we shall continue to
stress the difference between simultaneous and sequential moves only in the
subsequent parts of this chapter, implying that the subsequent definitions
and theorems are more general than those typically found in the literature.
From Chapter 5 onwards, however, the term repeated games is exclusively
reserved for simultaneous move games.
From the previous section it appeared that a distinction can be made
between strategies played within a stage game and strategies which are
played in the overall game. From Chapter 3 it is known that by playing
mixed strategies any payoff tuple in the convex payoff space can be
obtained. This is also possible in dynamic games. However, in order to
avoid dealing with incomplete information games, the mixing device must
be assumed to be publicly observable. Moreover, now, there is an additional
possibility of generating convex payoff tuples if different stage game strat-
egies are played in various periods. This amounts to mixing strategies
between stages and not only within a stage. However, departing from the
assumption of stationary strategies implies that payoffs differ between
stages and simple strategy profiles in the sense of Abreu cannot be con-
structed when testing for an equilibrium (Abreu 1986, 1988). Therefore,
non-stationary strategies are not treated in the following.
Finite dynamic games with discrete strategy space I 51
Discounting9
There are three reason for assuming that agents value a payoff received today
higher than one at some future date: (a) agents are impatient; (b) future
payoffs are uncertain by their nature: there is always a risk that these payoffs
will never be realized for some reason; (c) payoffs received today can be
invested to yield interest: hence, delayed payoffs involve opportunity costs.
Discounting makes payoffs received at different stages of a game com-
parable. Since information about optimal strategies is required right from
the beginning of a game, it is advisable to discount all payoff streams to the
initial stage of a game. The net present value of a payoff stream received
over T periods at time t0 is computed by using the discount factor i, i
1/(1i ) and 0i 1, where the index refers to the player. i denotes the
discount rate or time preference rate. For example, a discount rate of 10
percent implies i 0.1 and i 0.909.
Omitting the subscript i (denoting the player) for convenience, we have
the following relations (see, for example, Eichberger 1993, p. 211; Holler
and Illing 1993, p. 140):
T1 T1
1 2 ...T1T t0
tt; 1 (1 ) if ...
t0
t T
1 T
(4.2)
Finite dynamic games with discrete strategy space I 53
T T
1 22 ...TT
t1
tt; 1 (1 ) if ...
t1
t T
1 T
(4.3)
1 T1 t 1 T1
¯ T1 · t; ·
if 1; if 1 ... T.
T t0 t
(4.4)
t0
t t0
In (4.2) the net present value at time t0 is computed when payoffs are
received at the beginning of each period, starting at t 0. Alternatively, it
is the net present value at time t1 if payoffs are received at the end of each
period and for the first time at t1. (T1 has to be replaced by T if players
receive payoffs in the last round.) In (4.3) the net present value of a payoff
stream at time t0 is computed, when payoffs are received at t 1 for the
first time, and then after each stage.
Sometimes it is convenient to express payoffs as average payoffs ¯
because this makes the payoffs received in a dynamic game comparable to
those derived in a one-shot game (see (4.4)). For large T, with T → at the
limit, the average payoff is defined as:
¯ (1)· if 1; if 1 and ... .
t0
t
t 1 T (4.5)
Since average payoffs may not be defined for 1 (see, for example,
Eichberger 1993, p. 211; Holler and Illing 1993, p. 140), we usually assume
to be strictly smaller than 1, though might be close to 1.
Normal form
The normal form of a repeated game, in which the constituent game
(I, , S) is repeated T times, may be written as T(I, , , , T), with I the
set of players, the set of payoffs, {1, ..., N} the set of history-depen-
dent strategies ( i i ), {1, ..., N} the set of discount factors, and T
the number of stages which may be finite, that is, TT, or infinite, that is,
T. Alternatively, the normal form of a dynamic game may be written
as T(I, , ) or T(I, , ), where is the set of discounted payoffs and
the set of average payoffs.
( (h
tg
i it i m), i(hm)) iI, m{1, ...,
}; g {1, ..., T} and i(hm)
t V * V V
Minimax payoff
If player j punishes player i, the minimax payoff to player i is defined as
follows:10
iM min max i(si , sj). (4.6)
sj si
That is, given player i chooses an optimal response to the punishment of
player j, s/he can guarantee him- or herself a payoff of iM. We denote the
minimax strategy of player j when punishing player i by mij and the best
reply of player i to the punishment by mii. Thus the superscript denotes the
player who is punished and the subscript a player’s strategy. We write
iM iM(mii, mij ) for short. However, sometimes it is necessary to distin-
guish between iM(i) and jM(i) where iM(i) iM and the latter payoff refers
to the punisher j’s payoff when minimaxing player i.
Maximin payoff
In contrast, the maximin payoff of player i, which is also called his/her
security level, is the lowest payoff a player can guarantee him- or herself in
a stage game. It is defined as:
a2 na2 p2 a2 na2 p2
4.4 2 0 4.4 2 0
na1 na1
1.4 2 0 1.4 2 0
0 0 0 0 0 1
p1 p1
1 0 0 0 0 1
Maximax payoff
The maximax payoff of a player i is defined as:
Deviation payoff
The deviation payoff, denoted iD, which may also be called the free-rider
payoff, is the maximum payoff a player can achieve in a stage game if s/he
deviates from an agreed strategy and the other player complies with the
agreement. Suppose si and sj denote some agreed strategy combination,
then we have:
With the definitions of the previous sections we are now prepared to derive
some theorems, some of which are a generalization of the results we
obtained in Section 4.2. Theorem 4.1 is useful in that it forms the basis for
many strategy profiles below:
Theorem 4.1
If a simultaneous move constituent game has a Nash equilibrium sN,
then the repeated play of sN in every stage game is a subgame-perfect
equilibrium of the repeated game T. That is, the strategy combination
(ht )(siN(ht1(ht )), si
N (ht1 (ht )), ..., sN (hT(ht ))) forms a subgame-
i
perfect equilibrium for any i [0, 1] and t {1, ..., T}. In a sequential
move game the repeated play of a stage game subgame-perfect equilib-
rium constitutes a subgame-perfect equilibrium of T.
Proof: Theorem 4.1 is rather simple and a proof obvious.11 Assume
simultaneous moves and that player i plays some other stage game strat-
egy si
instead of sNi, but all other players choose continuously siN . s
s N
i i
in at least one stage game implies i (si , si )i (si
, si ). Consequently
N N N
we have:
T T T1
tg
ti i(sN)i (s
i, si
N )
tg1
tii(sN) (s
, s
tg
t
i i i
N
i ) g{1, ...,T}.
(4.10)
Theorem 4.2
In a finitely repeated simultaneous move game T with a unique stage
game Nash equilibrium sN, the only subgame-perfect equilibrium of the
whole game is the strategy combination SPE(ht )( SPE t N
1 (h )(s1 ), ...,
(h )(sN )); that is, the stage game Nash equilibrium strategy is played
SPE t N
Proof: The proof is obvious (see, for example, Binmore 1990, pp. 353ff.;
Eichberger 1993, p. 224; Myerson 1991, pp. 309ff.). According to back-
ward induction in the last period T, subgame-perfection requires that a
player plays his/her best stage game response strategy. This is true
whether no deviation occurred in previous rounds or whether some pun-
ishment strategy is played as a response to defection. Thus, in period T
each player must play iSPE(hT)siN in a simultaneous move game or
iSPE(hT)siSPE in a sequential move game.12 Hence, in period T 1, no
other strategy than iN(hT1)siN ( iN(hT1) siSPE) can be played
because no adequate threat is available. Obviously, this holds also for any
period previous to T1.
With respect to the concept of NE, a similar theorem to Theorem 4.2 can
be established; however, the statement has to be slightly altered (see, for
example, Eichberger 1993, pp. 224ff.; Friedman 1986, pp. 95ff.):
Theorem 4.3
In a finitely repeated simultaneous (sequentially) move game T with
unique Nash equilibrium sN of the stage game and where
iN(sN ) iM holds for all iI (iN(sN ) iSC for the first mover and
iN(sN ) iM for the other players), the only Nash equilibrium of the
entire game is the strategy combination N( N 1 (s1 ), ..., N (sN )), that is,
N* N N*
Proof: The first part of Theorem 4.3 simply follows from the fact that in
the terminal round sN has to be played and if no harsher punishment is
available than the Nash payoff no other strategy combination can be
enforced in previous rounds. The proof of the second part of Theorem
4.3 is only briefly sketched since it is similar in spirit to the proof of
Theorem 4.4 below (see van Damme 1991, pp. 195ff.; Sabourian 1989,
pp. 75–6). First, one constructs a trigger strategy13 which calls for mini-
maxing a deviator until the end of the game. Second, one shows that for
discount factors close to 1 the payoff stream from cooperating for some
time t*, and then playing the stage game NE for the rest of the game (t*
1 until T) is higher than when deviating at any time tt0, t0 t*, and
then being minimaxed for the rest of the game. Third, one determines the
limit of the average payoff for large T if players follow the equilibrium
Finite dynamic games with discrete strategy space I 59
path which approaches the stage game payoff sustained in the finitely
repeated game for t* periods.
Theorem 4.2 and the first part of Theorem 4.3 cover the PD game in its
simultaneous and sequential move version. Since iN iM iSC in this
game no other strategy tuple than the stage game NE will be played.
Theorem 4.2 also covers the sequential move chicken game where there is
only one stage game SPE which has to be played throughout the game if
one requires strategies to be subgame-perfect. A typical example which
illustrates the second part of Theorem 4.3 is the extended PD game I in its
simultaneous move version where iN 2iM 1. To see this, assume that
the game is played three times and assume no discounting. In the first round
the parties agree to play (a1, a2), and in the second and third rounds to play
the stage game NE (na1, na2). The payoff stream of such an agreement is
3.2227.2. If a player deviates s/he will be minimaxed afterwards
which delivers the payoff stream 4.4116.4. Since 7.2 6.4, free-riding
can be deterred by such a strategy. Accordingly, by increasing the number
of repetitions, only in the last two rounds must the stage game NE be played
and hence for large T (T → ) the average equilibrium payoff converges to
the stage game payoff from mutual cooperation, that is, 3.2. With discount-
ing the same can be shown provided the discount factors are sufficiently
close to 1. Of course, the above strategy is only an NE of the entire game
and not subgame-perfect as follows from Theorem 4.2.
Though we cannot refer to an example to which the second part of
Theorem 4.3 applies in the case of sequential moves, it should be evident
that in repeated games payoff tuples which imply a payoff above the
minimax payoff to all players except the first mover, who may receive only
a payoff above his/her security level, can be sustained in the long run.
Moving first may in this case be a disadvantage since the other players
have a harsher punishment (in the sense of NE strategies) available to
them.
The next result is an extended version of a theorem derived by Friedman
(1985):
Theorem 4.4
Let T (I, , ) be a finitely repeated simultaneous move game with at
least two strictly Pareto-undominated stage game Nash equilibria, that
is, iN(1)(sN(1))i(2)(sN(2)) iI. Then for large T and i close to 1 any
average payoff vector * (*i , *i ) for which *i (s)iN(si ) iI holds,
where iN(si ) denotes the lowest Nash payoff to player i, can be obtained
in the T-fold repeated game by playing subgame-perfect strategies. In a
sequential move game at least two Pareto-undominated subgame-perfect
60 Game theory and international environmental cooperation
stage game equilibria must exist so that any payoff vector *i (s) iSPE(si )
can be obtained in the finitely repeated game.
holds. That is, the proposal is an equilibrium if players are not too impa-
tient. We denote min
i the ‘minimum discount factor requirement’ to estab-
lish cooperation. That is, i min i iI is a necessary and sufficient
condition to establish cooperation as SPE.
Increasing the number of stages at the end of the game in which a player
can be punished, min i decreases and cooperation becomes easier.14
Alternatively, looking at it from a different angle, if discount factors are
close to 1 and the number of stages T is large, then the number of co-
operative stages t* can be chosen rather large. At the limit for T → ,
t*/T approaches 1 from below. That is, the fraction of cooperative stages
compared to the total number of stages, , is large. Consequently, the
average payoff along the equilibrium path approaches the stage game co-
operative payoff *i 3.2.
The simultaneous move chicken game is another example covered by
Theorem 4.4. Here iN(1)(sN(1)) iN(2)(sN(2)) for player i and jN(1)
(sN(1)) N(2)
j (sN(2)) for player j. Though in Section 4.2 no discounting was
assumed, it should be obvious that the same result could be established with
discounting. The only problem in games in which a sequence of alternat-
ing stage game NE in the terminal rounds is required is that punishment
may be delayed. This leads to high discount factor requirements, though for
discount factors close to 1 Theorem 4.4 still holds at the limit, as demon-
strated in Appendix II.
Of course, as discovered before, the sequential move version of the
Finite dynamic games with discrete strategy space I 61
chicken game is not covered by Theorem 4.4 since there is only one stage
game SPE. To cover the NE strategy of alternating stage game NE in the
terminal rounds as laid out in Section 4.2, Theorem 4.4 must be weakened
in that Nash equilibrium strategies are also allowed to be played in the
entire game.
For completeness, note finally that there is an even more general result
than Theorem 4.4. Benoît and Krishna (1985) show that Theorem 4.4 does
not only hold for payoff vectors which give each player more than in his/her
worst stage NE but extends to all individually rational payoff vectors. Since
the proof requires far more complicated strategy profiles than those used
above, it is omitted here.
NOTES
1. According to Rasmusen (1989, p. 46), the only difference between an extensive form rep-
resentation and a game tree is that in (e) the word ‘payoffs’ is replaced by ‘outcomes’.
Thus, for our purposes, we can use both terms more or less interchangeably.
2. Some authors exclude the starting node in the definition above (see, for example,
Gibbons 1992, pp. 122ff.). However, this is only a matter of terminology. Definition 4.2
only implies that in a one-shot game an NE could also be called an SPE. Nevertheless,
we reserve the term SPE for an equilibrium in a dynamic game.
3. Of course, by the symmetry of a simultaneous move game, also country 2 could move
first.
4. If players are not completely informed about the history of the game this is referred to
as imperfect information. Consequently, a simultaneous move game is always a game of
imperfect information. Moreover, a game of incomplete information is automatically a
game of imperfect information, though the opposite is not implied.
5. In a simultaneous move repeated game, the terms ‘stages’ and ‘rounds’ are used synon-
ymously. A similar term is ‘period’. In sequential move games this is different. Each move
is one stage of the game. If after the last move in a constituent game (also called a stage
game), the same game is played anew, we refer to this as the second round or second
period in the game. Thus, in a sequential move game there are several stages within a con-
stituent game.
6. In a simultaneous move game V T and v1 by definition.
7 In the general case of mixed strategies, observability of past moves requires that the
random device of each player is common knowledge. This is the standard assumption in
the literature, though it lacks an intuitive explanation for why this should be the case.
Departing from this assumption, however, complicates the analysis tremendously.
8. For notational convenience, we refrain from indexing i. Of course, a player has as many
history-dependent strategies as there are possible histories in a game.
9. For an extensive discussion on discounting see, for example, Mohr (1995), Pearce et al.
(1989, ch. 6).
10. ‘Player j’ may comprise one but also more players.
11. Of course, a similar theorem can be established for Nash equilibrium strategies in a
dynamic game (see, for example, Eichberger 1993, p. 212, Theorem 8.1). However, since
every subgame-perfect strategy is also a Nash equilibrium, Theorem 4.1 is a stronger
result.
12. If this were not true, then for any history hT such that i (hT) si ( i (hT) siSPE), there
would be at least one player who could improve his/her payoff in period T without any
effect on the payoffs in previous periods. Therefore, we must have i(hT)si ( i (hT)
62 Game theory and international environmental cooperation
siSPE) for all players i, implying that the stage game strategy in period T will be chosen
history independent.
13. The trigger strategy is sometimes also called grim strategy (see, for example, Dasgupta
1990; Rasmusen 1989, p. 91). It refers to a strategy which calls for punishment until the
end of the game once a deviation occurs, regardless of what the deviator plays after-
wards.
14. For example, adding a fourth stage to the game, it is easily checked that min
i 0.61.
5. Infinite dynamic games with discrete
strategy space: a first approach
5.1 INTRODUCTION
63
64 Game theory and international environmental cooperation
discount rate i , also the probability that the game continues, p, 0p1, so
that i p/(1i ). Thus, a low probability that the game continues shows up
in a small discount factor. Since for i 1 the average payoff may not be
defined, we assume in what follows discount factors strictly smaller than 1,
though they may nevertheless be very close to 1.1 To compute the net present
value of an infinite payoff stream, we can use formulae (4.2)–(4.4), replac-
ing T by (Gibbons 1992, pp. 90, 97). For convenience, the formulae are
reproduced below (subscript i, denoting the player, has been omitted):
1 2 ... ; 1 if ...
t0
t
t 1 2 (5.1)
1 22 ...T ; 1 if ...
t1
t
t 1 2 (5.2)
¯ (1 ) · if
1; ¯ if ....
t0
t
t
1 2 (5.3)
sij if in any ht(0) ... ht(t 1)(sj, s j); ht(0) ... ht(t 1) (sjj, s j
j )
si in t0
si if in any ht(0) ... ht(t 1)(si, s i) t 0 ... t 1 (5.4)
*i(s) N(sN)
iD (si , s i ) i i ⇔ *i (1 i )iD iiN
1 i 1 i
Di *i
⇔ i min
i iI (5.5)
Di Ni
must hold. The first inequality is expressed as net present values, the second
inequality in terms of average payoffs. This is to stress the equivalence of
both approaches. The first inequality implies that the net present value from
complying (LHS term of the inequality sign) must be at least as high as
when taking a free-ride (first term on the RHS of the inequality sign) and
then being punished for the rest of the game (second term on the RHS of
the inequality sign).
Thus, for given discount factors i, and min i i
1, i i must hold
* N
for all i, so that stability can be guaranteed using the trigger strategy above.
Since in the basic PD game playing the stage game NE for the rest of the
game is the harshest punishment players can inflict on a free-rider, that is,
iN iM, there is no other strategy which could deliver a lower min i .
Therefore, i mini iI may be regarded as a necessary and sufficient con-
dition for stability.
For instance, if we assume payoffs of the PD game as given in Matrix 3.1,
and that mutual investment shall be sustained, then iD 4.4, *i 3.2,
iN 2 and, consequently, min i 0.5. Of course, other payoff vectors (for
example, resulting from playing mixed strategies within a stage or from
alternating between pure strategies in different stages) could also be sus-
tained as long as they give each player at least an average payoff of 2.
From (5.5) it is evident that the larger the free-rider incentive iD *i ,
the more difficult it is to ensure stability. In contrast, a harsh punishment
in the form of a low value of iN is conducive to cooperation (min i will be
lower).
The ‘actual’ discount factor of players itself has also a great influence on
the stability of an agreement. In a politico-economic context, one would
expect that, if politicians seek short-term success, the value of i would be
high and therefore stability may be jeopardized. Though it is common prac-
tice to assume that the discount factor remains constant over time, an
assumption we shall follow throughout this book, it is nevertheless conceiv-
able that short-term success is particularly important before elections
(Hahn 1989).2 Hence, long-term commitments may be particularly jeop-
ardized during election campaigns.
Moreover, though one may think of p as an objective probability, it may
also be a subjective estimation by agents about the uncertainty of future
events. In the latter case, p could be indexed as well and pi may reflect a
bundle of factors, for example, the general risk attitude of politicians; the
66 Game theory and international environmental cooperation
i min
i is possible. Since the cooperative strategy vector s delivers *i in
each period, the average payoff is *i (see (5.3)).
Proof: The proof of Folk Theorem II is very much in the spirit of the
proof of Folk Theorem I. It also uses a trigger strategy. Along the equi-
librium path each player plays the stage game strategy si which generates
the payoff vector *i (si, s i )iM(mii, m i i ) i I. Once a player i deviates
can arise in games of the second type where the punisher receives less than
his/her minimax payoff. Then, if a player deviates from conducting the pun-
ishment, the threat to punish him/her for some time might not be sufficient
to deter deviation. Due to discounting, punishment payoffs in the future
become less of a deterrent and the punishment hierarchy may stop working
at some higher order. Put differently, with discounting, arbitrarily long
punishments are not arbitrarily severe since far-off punishments are rela-
tively unimportant (see Fudenberg and Maskin 1986 for details).
It took exactly ten years for Fudenberg and Maskin (F/M) to extend
Folk Theorem III in 1986 to games with discounting. The proof is partic-
ularly straightforward in the two-player case. In the general case of N
players, slightly more complicated strategy profiles have to be used and
some restrictions on the payoff structure have to be imposed. In the follow-
ing we proceed stepwise: first, we prove F/M’s folk theorem for two players
(Folk Theorem IV) and then, second, we cover the general case of N players
in Folk Theorem V.
m ij if in any ht(t tP)...ht(t 1)(si, sj), or ht(t tP)...ht(t 1)(m ij, s
j )
i
si if in any ht(t tP)...ht(t 1)(si, sj), or ht(t tP)...ht(t 1)(m ij, mij)
(5.7)
where ht(t tP ) ... ht(t 1) means all tP periods previous to period t, tP1
tP2 tP. The strategies imply that the game may be divided into a cooper-
ative and a punishment phase. In the case of compliance, the cooperative
phase continues and players choose strategy si. The minimax strategy m ij
is played if there is either a deviation in the cooperative phase of the
game or a deviation in the punishment phase. Thus, the strategy implies
that players ‘agree’ to minimax each other mutually for tP periods in the
case of a deviation where Gi (m ij, mij)Mi (mii, mij) iI holds. Thus, the
severity and length of the punishment is independent of the degree of
deviation and of the player who deviates. That is, the strategies are simple
in the sense of Abreu. For an SPE three basic conditions have to be
satisfied which are reflected in (5.8)–(5.10).
70 Game theory and international environmental cooperation
Now we have to show that for sufficiently high values of the discount
factor, that is, 1 i min
i , the strategy i always satisfies conditions (5.8)
and (5.9), as long as *i iM iI. This is done as follows.
First, substitute (5.10) into (5.8) and rearrange terms to have:
Mi Gi Mi Gi
itP ⇒ C : 1i (5.11)
*i Gi 1 i
*i Gi
where, due to *i iM, the RHS term of each inequality is smaller than
1 and C1 is a necessary condition for an SPE since i
1, tP 1, and there-
fore i it.
From (5.9) it follows that iP should be as small as possible which, con-
sidering (5.10), is equivalent to minimizing itP. Thus, if we choose the
punishment time tP such that:
Infinite dynamic games with discrete strategy space I 71
Mi Gi
tP t*
log
*i Gi
logi
then iPiM and (5.9) becomes binding.7 Upon substitution, (5.8) reads:
Di *i
*i (1 i )iD iiM ⇒ C2: i i2 (5.12)
Di M
i
where i2 1 due to *i iM. Taken together, C1 and C2 are necessary con-
ditions for the discount factor so that the strategy in (5.7) is an SPE.
Hence, these conditions provide the starting values of min i since
min
i max(i , i ) holds. The actual i
1 2 min is determined as follows.
is, i is too big, consider raising tP. Thus, eventually, (5.8) will be
P
value in (5.8) gives 3.2 3.3375, which is obviously not true. Therefore, we
choose i min
i 0.375 so that (5.8) is satisfied. A routine check reveals that
choosing alternatively tP 2 leads tomin i 0.56 and this alternative can
therefore be discarded.
From (5.8) the following corollary is readily proved:
Corollary 5.1
For i →0 the payoff space which can be sustained by subgame-perfect
strategies converges to the Nash equilibrium payoff space of the stage
game , that is, SPE (N(1), ...).
72 Game theory and international environmental cooperation
Proof: For i → 0, (5.8) becomes *i (si, sj ) iD (si, (sj ), sj ), which can only
be true if *i (si, sj ) iN (siN , sjN). N(1), ... refers to the general case that
there might be more than one NE.
Let us now consider the case of more than two players. The proof of Folk
Theorem IV was based on the punishment strategy of mutual minimaxing.
With this strategy we killed two birds with one stone: on the one hand, the
deviator was punished; on the other hand, the punisher had an incentive to
conduct the punishment. For both players the incentive to go along with
the punishment is the prospect of returning to cooperation as soon as pos-
sible. Now, in the case of more than two players, mutual minimaxing might
not be possible any more. That is, some punishers might receive more than
their minimax payoff during the punishment phase if they deviate from
punishing. To render this a non-beneficial option, a three-phase strategy is
needed which may be summarized as follows.
Players cooperate in phase 1 if no player deviates. If a player deviates,
phase 2 starts in which the deviator is punished by being minimaxed for
some time. The deviator must be punished long enough for the gains from
the deviation to be wiped out. If a player deviates in phase 2, this phase is
started anew. If there is no deviation in phase 2, phase 3 starts in which the
punishers are rewarded who went through with the punishment. In this
phase the punishers of phase 2 receive a payoff which by construction
exceeds their minimax payoff.
This three-phase strategy allows us to state the following Folk
Theorem V, which is proved in Appendix III:
5.3 DISCUSSION
Basically, there are two approaches to tackling this problem. First, one
thinks of whether it is possible to develop the definition of an equilibrium
concept further, for example, by defining more restrictively what is meant
by a credible threat and punishment strategy. This is the route we follow in
Chapters 6 and 7. Second, a bargaining stage is added to the game in which
the negotiation of players to settle for an equilibrium is modeled.
Moreover, in richer games the problem itself may suggest some restrictions
on the set of possible equilibria. For instance, in a second-best world insti-
tutional restriction may exclude some equilibria. We shall encounter such
examples in Chapters 11–14.
We wish to end this chapter with two remarks. First, as noted already in
Chapter 4, when mixed strategies are played, it has to be assumed that the
mixing device is known to all players, otherwise incomplete information
must be modeled explicitly, which is more complicated. Second, in the case
of two players the mutual minimaxing strategy, though it is a feasible
subgame-perfect strategy from a game theoretical point of view, seems a
rather curious kind of punishment when thinking of applications.
Basically, it implies that if one player free-rides and the other player does
not conduct the agreed punishment, s/he in turn is punished by the free-
rider. It seems rather paradoxical to expect a player to punish someone if
s/he does not get punished him- or herself. Therefore, the three-phase strat-
egy in the context of more than two players appears to be more convincing.
Players agree on a penalty code according to which each party is required
to participate in the punishment. Since enforcement itself is a public good
(since punishment is costly), potential punishers have an incentive to ensure
that all parties fulfill their punishment obligations. This is done by estab-
lishing a two-stage hierarchy of punishments.
However, the difficulties in international politics of reaching stable agree-
ments suggests that it is not always that easy to establish an incentive-com-
patible punishment code. The failure of many trade embargoes may serve
as an example. The available second-order punishments are obviously often
not severe enough to overcome the incentive of some potential punishers
to trade with the country on which a boycott is imposed and therefore a
successful punishment code (including first-order punishments) cannot be
established. In particular, receiving only slightly more than the minimax
payoff may not be a strong enough incentive for players to penalize a free-
rider. This concern is taken up in the next two chapters.
74 Game theory and international environmental cooperation
NOTES
1. This assumption does not affect the generality of the subsequent proofs. As will become
apparent, if a payoff can be backed by equilibrium strategies for discount factors close to
1, this will be even more true for a discount factor equal to 1.
2. This idea is elaborated in various models of political business cycles. For a survey see, for
example, Willet (1988) and the literature cited therein.
3. A survey of the issues treated in this section can be found in Pearce (1992) and Sabourian
(1989).
4. For an explanation of the term ‘folk theorem’, see below.
5. Of course, though mii is a best reply to m i
i , and hence a punished player will ‘comply with
his/her punishment obligations’, the punishers could do better than playing m i i and hence
6.1 INTRODUCTION
In Section 4.4 we demonstrated that in games with two or more stage game
Nash equilibria (NE) all payoff vectors can be sustained by subgame-
perfect equilibrium strategies which give each player more than in his/her
worst NE provided discount factors are close to 1. Such an abundance of
equilibria was also found in supergames where even weaker conditions
must be satisfied to derive folk theorem type of results. Thus, although we
strengthened the equilibrium concept for dynamic games by requiring
strategies not only to be an NE but also to be a subgame-perfect equilib-
rium (SPE), the set of equilibrium payoffs remains large.
A concept which is capable of reducing (though not eliminating) this lack
of predictability in repeated games is the concept of renegotiation-proofness.
Though there emerged many versions of this concept, in the context of
finite games the interpretation seems not very controversial (Benoît and
Krishna 1993; Bergin and MacLeod 1993; Bernheim et al. 1987; Fudenberg
and Tirole 1996, pp. 174ff.). The subsequent discussion is based on Benoît
and Krishna’s definition which is restricted to two-player games. In this
case their concept coincides with Bernheim et al.’s definition of coalition-
proof equilibria, which we discuss in Chapter 15.
In the above-cited literature it is argued that threats which imply a lower
payoff to deviators and punishers alike will be subject to renegotiations and
therefore lose their credibility. If defection occurs, it is in the interest of all
players to treat bygones as bygones, and punishment will not be carried out.
Three examples will illustrate the idea. In each of them no discounting is
assumed and only pure stage game strategies are considered.
Suppose that the extended PD game II in Matrix 4.3 is played three times
and assume no discounting. Both countries agree to cooperate in the first
round by playing (a1, a2) and subsequently playing the good equilibrium
75
76 Game theory and international environmental cooperation
(na1, na2) twice. Defection in the first round will be punished by playing the
bad equilibrium ( p1, p2) until the end of the game. Since 3.222 4.4
11 this strategy is subgame-perfect. However, suppose a country deviates
in the first round; then, before period 2, the deviator will suggest that the
punisher forget about the past and play the good NE in the last two rounds.
Obviously, since there is no reward phase, it is in the punisher’s interest to
accept the proposal. Since a ‘potential’ free-rider knows this before round
1, s/he will not be afraid of the punishment and will take a free-ride.
Therefore, the only renegotiation-proof equilibrium, henceforth abbrevi-
ated RPE, in this game will be the thrice-repeated play of the good stage
game NE (na1, na2). (The bad NE (p1, p2) is no RPE because it is Pareto-
inferior to (na1, na2).)
Since this result holds also if the game is repeated more than three
periods, it also holds for the limited case of T→ . Hence, the set of average
equilibrium payoffs shrinks from IR SPE (which includes all payoff
tuples which give each player at least a payoff of 1 in the example) to the
single payoff tuple RPE ¯ RPE (2, 2) if we require equilibrium strategies
to be renegotiation-proof and not only subgame-perfect. Hence, RPE
SPE or in the more general case ,RPE ,SPE where i 1 and T is not
necessarily large. Though in the following we concentrate mainly on equi-
librium payoffs, it should always be kept in mind that the above relation has
its analogy with respect to strategies, that is, RPE SPE.
From the example it is evident that though renegotiation-proofness
requires equilibrium strategies to be subgame-perfect and to be Pareto-
efficient, renegotiation-proof equilibria, also known as Pareto-perfect equi-
libria may not be efficient with respect to the set of all SPE (Fudenberg and
Tirole 1996, p. 177).
Next consider the extended PD game III in Matrix 6.1 where the associated
payoff space is displayed in Figure 6.1. This game has three-stage game NE.
Here pij denotes a punishment strategy of player i to punish player j. The
best reply to this has player j choose pjj. For this game it is easily checked
that the minimax payoffs in this game are iM 2
i I which correspond
to the worst NE payoff to each player. Hence, SPE {i SPE i 2
6 1 0 0
a1
6 7 0 0
7 3 0 0
na1
1 3 0 0
0 0 0 4
p21
0 0 0 2
0 0 2 0
p11
0 0 4 0
6 1 0 0
a1
6 7 0 0
7 0.5 0 0
na1
1 0.5 3 0
0 3 0 4
p21
0 0 0 2
0 0 2 0
p11
0 0 4 0
78 Game theory and international environmental cooperation
π2
7
5 ΠIR
Pareto frontier
4
0 1 2 3 4 5 6 7 π1
SPE(T 2): (4, 8), (8, 4), (5, 7), (7, 5), (6, 6), (9, 9)
where payoffs are summed over the two stages and SPE stands for the set
of (total) SPE payoffs. The first two tuples result from payoffs (2, 4), (2, 4)
or (4, 2), (4, 2); the second two from (3, 3), (2, 4) and (3, 3), (4, 2); the fifth
from (3, 3), (3, 3); and the sixth from (6, 6), (3, 3). Whereas the first five
payoffs result from the play of stage game NE, and therefore the underly-
ing strategies are automatically subgame-perfect, the last payoff vectors can
be obtained by using the threat strategy to play the NE ( pii, pji ) against devi-
ator i. Since a deviator can net a gain of 1 and loses 1 through punishment,
deviation does not pay (assuming no discounting).
Requiring the SPE to be efficient, leads to:
is the number of rounds at the beginning of the game in which one of the
stage game NE is played, followed by [T/m] times in which the sequence of
length m is played.1 Therefore, for large T, the average payoff to a player in
the extended PD game IV approaches 4. Compared to the extended PD
game III, the lack of a symmetric Nash equilibrium causes the equilibrium
average payoff to be lower in the extended PD game IV because the
sequence length has to be extended to three periods.
Set:
T:1
X(T) ,SPE(T),
,RPE(T)Eff(,SPE(T)).
Finite dynamic games with discrete strategy space II 81
For T1:
T:T1
In words, the process works as follows. First, determine the set of SPE
payoffs in the last round. Second, select the efficient ones among them.
Third, increase the number of rounds by 1 as long as T T* and find for
this game those SPE payoffs, given the continuation RPE payoffs in the
rounds succeeding the first round. Choose the efficient SPE payoffs deter-
mined for T. From then onwards the procedure is repeated. The definition
also covers the general case of discounting, examples of which will be pro-
vided below.
We now summarize and generalize the results of the PD game examples
in the introductory section. PD game II is covered by Theorem 6.1, PD
games III and IV by Corollary 6.1 below.
Theorem 6.1
In a finitely repeated game T, where there is either a strictly Pareto-dom-
inant stage game Nash equilibrium or only a single stage game Nash
equilibrium, the only renegotiation-proof equilibrium is the T-fold rep-
etition of the Pareto-dominant Nash equilibrium or the single Nash
equilibrium respectively.
Corollary 6.1
In a finitely repeated game T at least two weakly undominated stage
game Nash equilibria must exist to establish a stage game strategy com-
bination different from the Nash equilibrium.
(2 2) where only player 1 can gain from deviating. In this case player 1
gains 1 in the first period, but if in the second period the second equilib-
rium is played s/he receives only 2 instead of 4. Since player 2 is indifferent
between both NE, free-riding can be deterred efficiently as long as 1 1/2
holds.
In the PD game examples discussed above, the set of average RPE
payoffs contained only a single element (assuming large T and no discount-
ing) which did not lie on the Pareto frontier. This is a first category of games
identified by Benoît and Krishna (1993). However, there are also games of
a second category in which the (closed) set of RPE payoffs is a subset of the
points lying on the Pareto frontier. Benoît and Krishna show that all games
belong either to the first or to the second category.
Theorem 6.2
In finitely repeated games T with no discounting and large T, the average
equilibrium payoff tuple is either unique and lies in the interior of the
payoff space or the set of average payoff tuples is a closed and connected
set lying on the Pareto frontier.
Since the proof is quite involved and rather lengthy, we skip it here, but
provide an example of the second category of games below. Assume that
cooperation will be established in a chicken game where both countries
invest in abatement technology (action ai ) and receive a payoff of ai (see
Finite dynamic games with discrete strategy space II 83
π2
πA
c2
A
a2
Pareto frontier
B
C
b2
πB
d2
d1 b1 a1 c1 π1
Matrix 3.2). Recall that there are two strictly undominated (pure strategy)
Nash equilibria in this game. Like in the extended PD game IV, these equi-
libria have to be played in turn in the last two periods. However, in the
chicken game, the Nash equilibria are Pareto-efficient and can therefore be
played at any time and for any duration. Moreover, the payoff in the Nash
equilibria is for one player superior to the cooperative payoff (ci ai ) and
may therefore be used as punishment. Hence, in contrast to extended PD
games III and IV, no sequence has to be played involving playing a co-
operative strategy in one period followed by playing Nash equilibria sub-
sequently. Consequently, total equilibrium payoffs are given by (T
2)· ai
ci bi and (T
2)·aj bj cj with average payoffs ¯ RPE
i ai
iI if T is
large.
An immediate implication of the arguments presented above is that to
play: (a) always one of the two NE; (b) a combination of (a1, a2) and one
NE, or (c) a combination of NE, are also an RPE as long as average payoffs
lie on the Pareto frontier. Playing only one Nash equilibrium throughout
the game leads to an average payoff tuple at the boundary of the Pareto
frontier (either A or B in Figure 6.2). A combination of (a1, a2) and one
Nash equilibrium implies a payoff tuple on the segment A or B of the
Pareto frontier. A combination of the two NE may, depending on the
specific payoffs, lead to average payoff tuples lying on the Pareto frontier.
84 Game theory and international environmental cooperation
Theorem 6.3
In finitely repeated chicken games the set of renegotiation-proof average
payoffs comprises all points on the Pareto frontier provided discount
factors are close to 1.
Proof: The proof basically follows from the discussion above. It remains
to stress the validity in the case of discounting. First, deviation does not
pay if an NE is played in some stage. Second, if (ai, aj ) is played, then
deviation may occur. The weakest threat is if the equilibrium play involves
playing a Nash equilibrium for t* periods in which the potential deviator
receives only a payoff of bi subsequent to the deviation. This implies that
punishment is delayed by at least t* periods. After these t* periods either
the last two terminal rounds have been reached in which an alternation
between the two NE would be played in equilibrium, or the game does
not terminate after the t* periods and equilibrium play involves resum-
ing the play of ai/aj. In the former case the ‘worst’ case implies that the
deviator suffers a loss of ci
bi from the punishment only in the last
round. In the latter case, the lower bound of the loss due to punishment
is ai
bi in period t* 1. Thus, deviation does not occur if either
i [(ci
ai )/(ci
bi )](1/(t*2)) min
i or i [(ci
ai )/(ai
bi )](1/(t*1)) min
i
hold. Both inequalities can generally be satisfied provided discount
factors are sufficiently large.
It should be evident that, in the case of the extended PD games III and
IV, RPE can be derived in the presence of discounting. Generally, however,
it should be noted that, though discounting does not change the main
results for considerations at the limit (sufficiently high discount factors), the
discount factor may very well affect equilibrium play in a game and hence
the average payoffs obtainable in a game. This will be briefly demonstrated
for the extended PD game III.
Finite dynamic games with discrete strategy space II 85
(6.2) can only be satisfied if i 1/2. Suppose this to be the case, then it is
straightforward to show that mmin/i 0. That is, the lower the discount
factor, the longer will be the sequence m in which the symmetric NE must
be played m
1 times. Since in this game the cooperative payoff strictly
Pareto-dominates the Nash equilibrium payoffs, this implies that with
decreasing i the average payoff to each player decreases. For i →1/2
and large T, we have ¯ i,RPE →3 whereas for i 1 we found ¯ i,RPE 4.5
above.
Summarizing, the concept of renegotiation-proofness reduces the
number of equilibria and average payoffs sustainable in repeated games
compared to subgame-perfection. Renegotiation-proofness defines cred-
ible threat strategies more narrowly. It requires that during the punish-
ment (that is, off the equilibrium path) the punisher receives at least the
same continuation payoff as along the equilibrium path. To sustain a
cooperative outcome for some time, a game must have at least two undom-
inated stage game NE. Whereas this is true in the chicken game and the
extended PD games III and IV, this does not hold in the ordinary PD
game and the extended PD game II. In the case of no discounting, the
equilibrium average payoff either reduces to a single point below the
Pareto frontier or comprises a closed set of average payoffs on the Pareto
frontier.
Set:
T: 1
P()Eff((T)),
,SSPE(T)SPE(P()).
For T1:
T:T1
P() Eff((T)),
Theorem 6.4
In the finitely repeated PD game and its extended versions II, III and IV
no strongly perfect equilibrium exists.
Theorem 6.5
In the finitely repeated chicken game the set of strongly perfect equilib-
rium average payoffs comprises all points on the Pareto frontier provided
discount factors are close to 1.
Proof: Follows from SSPE Eff(RPE) where the RPE have been derived
above.
6.4 DISCUSSION
NOTES
1. The bracket indicates an integer value.
2. An extension to mixed strategies can be found in Benoît and Krishna (1993, s. 6).
3. If this were the case, this would have to be modeled explicitly in the utility functions of the
players.
4. In other games the RPE average payoffs are a true subset of the Pareto frontier. See Benoît
and Krishna (1993).
5. For games with more than two players, SSPE requires more than subgame-perfection and
Pareto-efficiency of all continuation payoffs. See Chapter 15. The basic idea of this equi-
librium concept has been formulated by Aumann (1959) for static games. A definition for
dynamic games may be found in Rubinstein (1980).
7. Infinite dynamic games with discrete
strategy space: a second approach
89
90 Game theory and international environmental cooperation
sij if in any ht(ttjP) . . ht(t1)(si, sj), or ht(ttjP) . . ht(t1)(sij, sj)
i
si if in any ht(ttjP) . . ht(t1)(si, sj), or ht(ttjP) . . ht(t1) (sij, sij).
(7.1)
Here payoffs are expressed as average payoffs of the infinite game (dis-
counted payoffs are multiplied by (1 i )) assuming i 1 (see Section 5.1).
Inequality (7.2) states that the discounted payoff from cheating, iD (best
Infinite dynamic games with discrete strategy space II 91
Corollary 7.1
For i → 0 i the set of weakly renegotiation-proof equilibria converges
to the set of stage game Nash equilibria.
Proof: Since (7.2)–(7.6) are continuous in i, the set of equilibria satisfy-
ing these WRPE conditions for i → 0 iI converges to the set for
i 0 i I which contains only stage game Nash equilibria. For i →0
iI, (7.4) becomes iD *i iI. Hence, si si (sj ) and sj sj (si ) from
which sN (siN, sjN) follows.
Thus, Corollary 7.1 is analogous to Corollary 5.1 in Section 5.2, where this
result has been shown to apply to subgame-perfect strategies.
To see what changes are needed if equilibrium strategies are required to
be weakly renegotiation-proof instead of subgame-perfect, we consider
some examples below. We start by assuming i → 1 i I and hence condi-
tions (7.5) and (7.7) apply (Sub-section 7.1.2). Subsequently, we consider
discount factors strictly smaller than 1 and therefore conditions (7.3), (7.4)
and (7.5) must be satisfied (Sub-section 7.1.3).
Extended PD game V
Consider the extended PD game V in Matrix 7.1, the payoff space of which
has been drawn in Figure 7.1. This PD game version is basically the same
as the extended PD game I (Matrix 4.2), except that now the punishment
Infinite dynamic games with discrete strategy space II 93
π2
5
πW1
4 C1
A
3
B C2
X
2 π W2
C
1
0 1 2 3 4 5 π1
*2 R 1 1 1 1
2 (p1, p2, q1, q2). (7.9)
Since we know that any payoff vector * (*1, *2) can be generated by
mixed strategies, the probabilities on the LHS of both inequalities have
been omitted as arguments. The probabilities on the RHS refer to the strat-
egies in Matrix 7.1. Again, superscripts refer to the player who gets pun-
ished.8
94 Game theory and international environmental cooperation
a2 na2 p2
(q1) (q2) (1 q1 q2)
3.2 1.4 0
a1( p1)
3.2 4.4 0
4.4 2 0
na1(p2)
1.4 2 0
0 0 0
p1 (1 p1 p2)
0 0 0
*1 4.4q11
2q12 (7.10)
*2 3.2q11
4.4q12. (7.11)
2*1
C2: *2 . (7.13)
4.4
Both conditions are drawn in Figure 7.1. All feasible payoff tuples which lie
to the right of C1 and to the left of C2 can therefore be backed by WRPE
strategies. Therefore,
WRPE corresponds to the area denoted B in Figure
7.1 which is a (true) subset of
SPE.
Ordinary PD game
Let us now turn to the ordinary PD game and check for the WRPE payoff
space. Denote p1 the probability that a1 is played and q1 the probability that
a2 is played. Assume payoffs as given in Matrix 3.1, the payoff space of
which is drawn in Figure 7.2. The entire payoff space,
, comprises
IR
and, additionally, the areas A and B (including the hatched triangles).
π2
6 C2
5 πW1
4 A πC
3
L1 C1
2 πW2
B
L2
1
0 1 2 3 4 5 6 π1
*1 2
2.4q11 (7.14)
Since the minimax payoff in this game is 2, we assume the first inequality
to be binding and solve for q11. We get q11 (*1 2)/2.4 and substitute this
into (7.15) which delivers:
*1 *
C1: *2 5.4 ; C2: *1 5.4 2 (7.16)
2 2
where the second WRPE condition follows from symmetry.
Both conditions are drawn in Figure 7.2, from which it is evident that
both conditions are non-binding. All payoff tuples in
IR are below the C1
and C2 line and hence
SPE
WRPE. Thus, the PD game is an example
where the WRPE concept does not reduce the set of equilibria payoffs com-
pared to the SPE concept.
Theorem 7.2
In an infinitely repeated (ordinary) PD game with stage game payoffs as
given in Matrix 3.2 and payoff structure ci ai di bi, a1
a2 b1
c2,
a1
a2 b2
c1 and a1
a2 d1
d2,
SPE
WRPE for i →1 i I.
Proof: See Appendix IV.2 for a general proof.
The reason for this result can easily be seen in Figure 7.2. For each payoff
tuple in
IR there is a payoff tuple in
which can be used to punish devi-
ation and which gives the punisher more than in the cooperative phase. This
is particularly important for tuples on the border of
IR which comprises
lines L1 and L2 and the part of the Pareto frontier between W1 and W2.
Take for instance point W1 (1W1, 2W1) which is defined as that payoff
tuple which delivers the lowest weakly renegotiation-proof payoff on the
Pareto frontier to player 1. Accordingly, the free-rider incentive for player 1
is high and a severe threat is needed to deter deviation. In the case of *i
W1 one may choose (1.4, 4.4) as a punishment tuple or any payoff tuple
lying in the hatched area which is located at the upper left part of area A in
Figure 7.2. More generally, in a PD game for any payoff tuple *
IR
there are strategy combinations available for which iC(sii , sji )*i (si , sj ) and
jR(sii , sji )*i (si , sj ) hold – the conditions of Theorem 7.1.9
In contrast, in the game depicted in Figure 7.1,
IR and therefore it
must be that payoff tuples along the border of
IR cannot be backed by
WRPE strategies. For instance, consider point X. Punishing player 1
requires a punishment tuple in a north-west direction, which is not feasible
because this would lie outside
.
Chicken game
Drawing
and
IR for a chicken game would also show that we can find
for all *
IR strategies such that iC(s i )*i (s) and jR(s i )*i (s) hold.
Infinite dynamic games with discrete strategy space II 97
Theorem 7.3
In an infinitely repeated chicken game with stage game payoffs as given
in Matrix 3.2 and payoff structure ci ai bi di, a1
a2 b1
c2, a1
a2 b2
c1 and a1
a2 d1
d2,
SPE
WRPE for i →1 i I.
Proof: Follows along exactly the same lines as the proof of Theorem 7.2
which is provided in Appendix IV.2. Note that in the chicken game pun-
ishment is particularly easy because both stage game Nash equilibria are
Pareto-efficient and satisfy the inequalities above. Therefore they can be
used as punishments.
Extended PD game V
Assume that the payoff tuple * (3.2, 3.2) shall be backed by WRPE strat-
egies (see Matrix 7.1). For a start, consider only pure strategies and hence
iD 4.4, R(i) (1.4, 4.4) (the superscript i refers to the player who gets
punished and where R(i) (iR(i), jR(i)) and iC 2. To determine min
i in
this symmetric game, we substitute (7.6) into (7.3) and assume for a start
punishment time to be given by tiP 1. Then, we have:
(1 i )1.4
i3.22⇔ i(1) 1/3 (7.17)
Though any further increase of the punishment time would reduce i(2), it
increases i(1) above 0.5774 and would therefore deliver a higher min i .
Consequently, min
i (WRPE)0.5774.
Next consider what will change if we require only SPE strategies to be
played in this game. For this, we substitute (5.10) into (5.9), noting that
iG iM 0, and find:
which is trivially satisfied for any punishment time and discount factor.
Hence, we may choose tP and iP iM 0. Inserting this into (5.8)
reveals i 0.27 min
i (SPE), which is obviously lower than the discount
factor requirement for WRPE strategies. The reason is simple: for a WRPE
less severe punishment strategies are available in this game because it must
be ensured that the punishment payoff exceeds the cooperative payoff of the
punisher.
Now allow for mixed strategies so that all payoff tuples in
may be
obtained. For min
i (SPE) nothing changes because we have already used the
harshest possible punishment. However, for min i (WRPE) there will be a
modification.
From (7.4) it is evident that to determine the lowest min i (WRPE) we
should look for a harsh punishment. However, the punishment is restricted
by conditions (7.3) and (7.5). For the latter condition we know from the
previous section that if, say, player 1 is punished, p11 1 is the ‘best’ repen-
tance strategy. Since, due to (7.5), *2 3.2P2 must hold, we choose q11
0, q12 0.72 and 1q11 q12 0.28 so that (7.5) becomes binding and
P2 3.2. From q12 0.72, C 1 , 1.45 follows (if q1 1) which, due to (7.3),
constitutes a lower bound for the punishment. If player 1 complies with the
punishment s/he receives a stage game payoff of P1 1.018 during the tPi
periods.
Computing as above i(1) and i(2) reveals that min
i 0.447 for t i 2. Thus,
P
by enlarging the set of strategies to include mixed strategies reduces the dis-
count factor requirement from min i (WRPE)0.577 to i (WRPE)
min
0.477. Nevertheless, also for mixed strategies the discount factor require-
ments for WRPE strategies are higher than for SPE strategies.
Ordinary PD game
Next, consider the minimum discount factor requirements in the ordinary
PD game in Matrix 3.1 and assume * (3.2, 3.2). From Section 5.1,
min
i (SPE)0.5 is known. To determine i (WRPE) in the case of pure
min
strategies is an easy task because the relevant payoffs are the same as in the
extended PD game V, namely iD 4.4, jC 2 and R(i) (1.4, 4.4). Hence,
min
i (WRPE)0.577, which is greater than i (SPE)0.5.
min
Theorem 7.4
In an infinitely repeated two-player (ordinary) PD game with stage game
payoffs as given in Matrix 3.2, ci ai di bi iI, a1
a2 b1
c2, a1
a2 b2
c1 and a1
a2 d1
d2, mini (WRPE) i
min (SPE) iI and
holds.
* IR
Infinite dynamic games with discrete strategy space II 99
For the example in Matrix 3.1, the details are provided in Appendix IV.3.
Chicken game
For the chicken game a similar result as for the PD game holds.
Theorem 7.5
In an infinitely repeated two-player chicken game with stage game
payoffs as given in Matrix 3.2 and payoff structure ci ai bi di iI,
a1
a2 b1
c2, a1
a2 b2
c1 and a1
a2 d1
d2, min
i (WRPE) i
min
In the previous section it became apparent that the WRPE are not neces-
sarily Pareto-efficient. On the one hand, the average payoff tuples of a
WRPE may not be Pareto-efficient. On the other hand, the continuation
payoffs induced by punishment may not be Pareto-efficient. A WRPE only
requires that no continuation payoff of an equilibrium strategy is Pareto-
dominated by a payoff of another subgame of this strategy. Put differently,
100 Game theory and international environmental cooperation
Figures 7.1 and 7.2 illustrate Theorem 7.6. All points between W1 and W2
are Pareto-efficient WRPE. Whereas in Figure 7.2 these efficient WRPE
payoff tuples comprise all individually rational payoff tuples on the Pareto
frontier, in Figure 7.1 the efficient WRPE payoff tuples are a (true) subset
of P(
IR). Hence, we may write
WRPE P(
IR) P(
WRPE) orP(
WRPE)
SRPE.
From Figures 7.1 and 7.2 the next theorem is intuitively appealing:
Theorem 7.7 states a result which is similar to the findings for finite games.
There we found that for i 1 i I the renegotiation-proof payoff space
for large T is either a single point below the Pareto frontier or a closed
Infinite dynamic games with discrete strategy space II 101
subset on the Pareto frontier. Now for infinite games only the second pos-
sibility is relevant.
A corollary of Theorem 7.7 is that for each payoff tuple * P(
WRPE)
there exists a strategy tuple such that iC(s i )Wi and j (s )i j
i P i Wi *
Condition iC W i i is a rather mild requirement and is almost always
i Wj
NOTES
1. Sub-section 7.1.1 draws on Endres and Finus (1998a) and Finus and Rundshagen
(1998b).
2. The origin of the idea can be traced back to the independent work of Farrell (1983) and
Bernheim and Ray (1985). For similar concepts, see Abreu et al. (1993); Asheim (1991);
Bergin and MacLeod (1993); Bernheim and Ray (1989); and Ray (1994). For more intui-
tive explanations of the gist of the concept, see Fudenberg and Tirole (1996, pp. 174ff.);
and Mohr (1988, pp. 551ff.).
3. Farrell and Maskin (1989a) assume the deviation payoff iD to be the maximax payoff of
country i, iU max (sj ) max (si )i (si , sj ) which constitutes an upper bound of the devi-
ation payoff (see Section 4.3). Though this simplifies matters and is of course valid for
considerations at the limit, it introduces a bias when computing the minimum discount
factor requirements ( min i ) of a particular cooperative strategy combination. Since we
shall be concerned below with computing min i for different ‘cooperative’ payoff tuples,
we are already using iD instead of iU at this introductory stage.
4. Thus (7.3) follows from iP (1 i)iC
iP after rearranging terms.
5. Condition (7.5) is equivalent to *j jP because of jP (1 j tPi )jR(sii, sji )
jtPi *j (si, sj).
6. For a derivation, see Appendix IV.1.
7. Note that Farrell and Maskin’s proof also includes showing that the two strategy com-
binations s and si are sufficient to generate any payoff tuple in
IR (see Farrell and
Maskin 1989a, pp. 332ff.). Since this part of the proof is not important for the central
idea of the Theorem 7.1, we skip it.
8. Probabilities p11 and p12 in (7.9) imply that player 1 accepts the punishment, probabilities
p1 and p2 in (7.8) refer to the defection strategy of player 1.
9. It should be kept in mind that
SPE
WRPE does not imply SPE WRPE. For instance,
the trigger strategy which calls for an infinite punishment by reverting to the stage game
NE in case of defection is not a WRPE because (for any *j jN) the punisher receives
less than in the cooperative phase. This violates condition (7.5).
10. Van Damme (1989) proves Theorem 7.4 using a particular example. However, acknowl-
edging that any more general proof would require a great amount of notation and would
be rather messy in the context of mixed strategies, we refrain from giving one.
11. For more than two-player games, SSPE requires more than subgame-perfection and
Pareto-efficiency of all continuation payoffs. See Chapter 15.
12. Farrell and Maskin (1989a), Theorem 4, specify sufficient conditions for the existence of
an SSPE.
8. Issue linkage
8.1 INTRODUCTION
103
104 Game theory and international environmental cooperation
1. Each party tries to hide its ‘true’ preferences for strategic reasons. In
particular, upstream/upwind governments will exaggerate their abate-
ment costs in order to extract high compensation payments from
downstream/downwind governments (Mäler 1990).
2. There seems to be a widespread consensus in international politics that
the polluter-pays principle instead of the victim-pays principle should
be applied to tackle international environmental problems (United
Nations 1972). There is the fear that if the victim-pays principle were
applied, the polluters would reduce their efforts at a preventative envi-
ronmental policy.
3. There is the danger that governments which pay compensation pay-
ments may be judged weak bargaining partners, which diminishes their
bargaining power with respect to other issues (Mäler 1990).
4. Following Heister (1997, pp. 247ff.) transfers themselves may be inter-
preted as prisoners’ dilemma games or as games with a similar incen-
tive structure. For both countries, the payer and the payee, there exists
Issue linkage 105
a free-rider incentive. On the one hand, the payer prefers the payee to
increase its abatement efforts without paying the transfer. On the other
hand, the payee would like to receive the transfer without increasing its
abatement obligations in turn. Thus if transfers are not accompanied
by appropriate punishments they constitute no equilibrium strategies
and cannot be used to stabilize an abatement game.3
a2 na2
3 1
a1
1 2
4 0
na1
3 0
a2 na2
1 3
a1
3 4
2 0
na1
1 0
Issue linkage 107
π2
6
5
4
3
2
1
–4 –3 –2 –1 1 2 3 4 5 6 π1
–1
–2
–3
–4
2 0 2 4
a11/a12
2 3 5 6
5 3 1 1
a11/na12
2 1 1 2
3 1 1 3
na11/a12
0 1 3 4
6 4 2 0
na11/na12
4 3 1 0
which are the sums of minimax payoffs to the players in the isolated games
have been normalized to zero.
Matrix 8.4 Payoffs in the isolated PD game
a2k na2k
a1k b1k
a1k
a2k c2k
c1k 0
na1k
b2k 0
1. cik aik;
2. 0bik;
3. a1k a2k b1k c2k, a1k a2k c1k b2k and a1k a2k 0 iI and k.
110 Game theory and international environmental cooperation
An implication of (3) is that aik 0 must hold for at least one country and
from (1) and (2) 2i1cik 2i1aik 2i 1bik k follows. Now we claim the
following:
Proposition 8.1
In an infinitely repeated PD game with payoff relations as defined above,
the possibility of issue linkage either improves upon the chances that
mutual cooperation can be sustained as a strongly renegotiation-proof
equilibrium or leaves them unchanged.
Proof: In Theorem 8.1 cooperation implies the strategy combination
(a1k, a2k) with associated payoff tuple (a1k, a2k). In order to sustain co-
operation as an SPE in an isolated game k, a necessary and sufficient con-
dition is:
aik (1 i )cik
0 iI and k{1, 2} (8.1)
k1
aik (1 i) c
k1
ik
0 iI (8.2)
It should be evident that Theorem 8.1 can easily be extended to cover the
case where more than two PD games are linked. Linking issues either
improves the chances for cooperation or leaves them unchanged. Note also
that it is not necessary that in each game the strategy tuple (a1k, a2k) is
played as a cooperative strategy. It is also conceivable that other strategy
combinations are played as long as they deliver individually rational payoff
tuples in the linked game. However, in this case Theorem 8.1 can only be
stated with respect to SPE strategies because it is not possible to tell at a
general level whether such a payoff tuple lies within the ‘imaginary PD
payoff space’ of the linked game.
Proposition 8.2
In an infinitely repeated PD game the possibility of issue linkage
improves the chances that cooperation can be sustained as a subgame-
perfect equilibrium or leaves them unchanged.
Issue linkage 111
Proof: Let ik(sk) 0 denote the payoff to player i from some coopera-
tive strategy sk in game k, and Dik(sik(sjk), sjk) the deviation payoff in the
stage game (ik N(s N) 0). Then cooperation can be sustained in an iso-
k
lated game provided:
ik(sk) (1 i )D
ik(sik(sjk), sjk)
0 iI and kK (8.3)
k1
ik(sk) (1 i) (s (s ), s )
0 iI
k1
D
ik ik jk jk (8.4)
where the number of games is K and (8.4) is again less restrictive than
(8.3).
Conditions (8.5) and (8.6) stress the main reason why issue linkage has a
positive effect on the possibilities for cooperation in infinite games: it is
simply because the payoff space is enlarged. Of course, by varying some of
the conditions above, results in the spirit of Theorems 8.1 and 8.2 may also
be derived for other classes of games.
Whereas in the context of infinite games the main focus of issue linkage was
on the enlargement of the payoff space, in finite games the focus is on the
number and the nature of stage game NE. Of course, also in finite games
issue linkage may lead to more symmetric payoffs so that cooperation
becomes attractive to all participants. However, since in finitely repeated
games special requirements with respect to the stage game NE must be
satisfied to establish cooperation, it seems more interesting to analyze how
issue linkage affects these requirements. Recall that for SPE strategies two
stage game NE – a good and bad NE – are sufficient to sustain a coopera-
tive solution for sufficiently large discount factors whereas for renegotia-
tion-proof strategies at least two Pareto-undominated stage game NE are
required. Consequently, the main concern is to answer the question of how
many stage game equilibria result from the linkage of two games.
First, note that the stage game NE of the linked game are the strategy
combinations of the NE of the isolated games. This is so because it can
never be a best reply in a linked game to play an NE strategy, say, in game
I but not in game II. Hence, if games I and II each have a single stage game
NE, the linked game will also have only a single NE. Thus, a necessary
112 Game theory and international environmental cooperation
8.4.1 Introduction
a22 na22
Definition 8.1
Issues are substitutes in agents’ objective functions if 2ui /xi1xi20 and
they are complements if 2ui /xi1xi2 0 holds.
That is, if issues are substitutes, additional utility derived from increased
physical payoffs in game I will be lower if the physical payoffs of game II
are already high than if they are low. By symmetry, the same holds for
game II. In contrast, if issues are complements, marginal utility will be
higher in game I if the payoffs in game II are also high. If 2ui /xi1xi2 0 we
are led back to the case we discussed in previous sections.
Basically, there are two main cases to be considered. First, linking a new
issue to an existing issue: that is, issue 1 already exists and has already been
negotiated. After the negotiation the second issue emerges. The new issue
Issue linkage 115
may or may not be linked to the first issue. Second, ‘linking two existing
issues’: that is, governments first seek cooperation on each issue separately
and then consider linking both issues in negotiations.
In both cases four subcases have to be distinguished.
Proposition 8.3
If two infinitely repeated (ordinary) prisoners’ dilemma games are linked
to each other, mutual cooperation becomes easier to sustain as a strongly
renegotiation-proof equilibrium than in the isolated games, provided
issues are substitutes in governments’ objective functions. The opposite
holds if issues are complements in governments’ objective functions.
Roughly speaking, the reason for this result is the following. Once a
country deviates in the interconnected game it is punished with respect to
both issues. If issues are substitutes, the loss through the punishment is par-
ticularly severe as the marginal utility from cooperation is high at low
payoff levels. Moreover, if countries are cooperating and receiving a rela-
tively high level of utility, free-riding pays less than at lower levels of utility.
Thus, though in the linked game a country deviates with respect to both
issues and is punished by terminating cooperation on both issues, the latter
effect is stronger than the former, implying additional enforcement power
in the linked game. By symmetry, exactly the opposite holds if issues are
complements.
Consequently, if issues are complements an obvious countervailing
measure would be to separate issues. This could be done if a government
delegates decision-making power on one or both issues to separate inde-
pendent agencies. For instance, in Germany monetary policy is indepen-
dently conducted by the central bank, economic competition is enforced by
an antitrust agency, and economic and fiscal policy is conducted by the
government itself. Of course, a basic prerequisite for such an ‘isolation
strategy’ to be successful is that the delegation contract must be based
on a long-term relationship between the government and the agencies;
otherwise, delegation is not credible and can always be reversed. In the
116 Game theory and international environmental cooperation
Proposition 8.4
In an infinitely repeated PD game, if issues are substitutes in govern-
ments’ objective function long-term delegation of decision-making
power on one or two policy issues to independent agencies with the same
objective function as the government will make it more difficult to
sustain cooperation. If issues are complements, delegation will make
cooperation easier.
Proof: Is obvious and therefore omitted.
Proposition 8.5
Suppose that the relation between two governments is described by an
infinitely repeated PD game. Then if a government decides to delegate
decision-making power to an agency and has the option to choose
between several agencies, then it should select that agency with the most
concave utility function to sustain international cooperation.
Proof: Assume material payoffs of the General Payoff Matrix 3.2. Then
cooperation can be sustained in the infinitely repeated game k pro-
vided:
Issue linkage 117
cik aik
i
min (8.5)
i
cik dik
holds. For any monotone transformation this condition reads:
i(cik) i(aik)
i
imin# (8.6)
i(cik) i(dik)
where i is agency i’s utility function. This transformation will ease co-
operation if min
i
imin#. Using (8.5) and (8.6) this implies:
i(aik) i(dik) i(cik) i(dik)
NOTES
1. Exceptions include the North Pacific Seal Treaty signed in 1957 which requires the USA
and the former USSR to pay Canada and Japan 15 percent of their annual harvest of pelts.
Another instance is the 1972 agreement on the reduction of the salinity of the Rhine, where
the Netherlands agreed to compensate France for 35 percent of its costs.
2. The Rio Declaration comprises mainly the Climate Framework Convention in which states
declare their concern about the global warming problem. Therefore, it may be viewed
mainly as a declaration of good will rather than an actual IEA.
3. Basically, this implies that the abatement game is extended to include transfer strategies,
where each strategy set may be interpreted as belonging to a separate game. Heister (1997)
calls the combinations of several games, like the combination of abatement and transfer
games, hypergames. According to this definition the issue linkage games considered below,
where two no-transfer games are linked, could also be termed hypergames.
4. Other interesting examples where issue linkage has played some role may be found in
Ragland et al. (1996).
5. Note that it is important that the cooperative payoff tuple lies on the boundary of this
imaginary payoff space. Other individually rational payoff tuples of the linked game may
lie outside this payoff space and hence Theorems 7.2 and 7.4 cannot be applied.
6. There are three cases to consider: (1) Inequality (8.1) is satisfied for both isolated games,
consequently, inequality (8.2) will hold as well; issue linkage does not improve upon the
chances for cooperation but also does no harm. (2) Inequality (8.1) does not hold for both
isolated games, consequently, inequality (8.2) is not satisfied either; again, issue linkage has
no effect. (3) Inequality (8.1) is satisfied for one issue, but not for the other; this leads to
two subcases: (3a) inequality (8.2) is also not satisfied. In this case cooperation is not pos-
sible in the linked game but countries can cooperate on that issue for which (8.1) holds; (3b)
118 Game theory and international environmental cooperation
inequality (8.2) is satisfied and cooperation can be sustained with respect to both issues.
Due to this last subcase Theorem 8.1 holds.
7. In the general case, this is an implication of 2i1cik 2i1aik 2i1bik k as derived above.
8. We assume that each game possesses at least one NE.
9. Static games with continuous
strategy space: global emission game
9.1 INTRODUCTION
Up to now we have assumed that players have discrete action sets so that
the normal form of the game could conveniently be displayed in a matrix.
Though for many policy problems modeling decisions as a discrete choice
seems adequate, other situations may be better modeled as a continuous
choice problem such as the amount of emission reduction in a global
policy game, for example, greenhouse gases (see also the discussion in
Section 2.3).
A continuous strategy set allows finer tuning of actions and reactions
and therefore leads to some interesting results which are absent in discrete
policy games. This is true at least as long as mixed strategies are ruled out
for discrete policy games. Though we dealt with mixed strategies in Chapter
3 and also mentioned some instances in which one can expect players to use
mixed strategies, they were introduced mainly for technical reasons; that is,
mixed strategies were required in the discrete strategy context to capture the
entire feasible payoff space when deriving folk theorem type results. In a
continuous strategy setting it suffices to consider only pure strategies. This
is true at least as long as games with a convex payoff space are considered.
Since all games in the remainder of this book satisfy this condition, we no
longer have to bother about mixed strategies.
The following analysis is based on a rather simple emission model. In
particular, the payoff or net benefit functions contain emissions as the only
argument. Thus, the aspect of issue linkage as well as agents’ choice
between environmental quality and other goods is not considered. The
reasons are the following:
119
120 Game theory and international environmental cooperation
That is, the strategic interplay of players with respect to international envi-
ronmental issues would remain unchanged. Although, of course, such an
extended framework might be more convincing from an economic point of
view, it would only complicate the analysis.1 Only if ‘joint production’ were
considered, could the results of the analysis change. Joint production
implies that the consumption or production of private goods is directly
related to environmental quality. That is, a (representative) consumer’s
choice concerns not only how much money to allocate to the purchase of
private goods versus the public good ‘environment’, but the consumption
of private goods affects environmental quality and vice versa. If joint prod-
ucts are substitutes (the consumption of the private good causes environ-
mental damage which, ceteris paribus, decreases the demand for the private
good) the main conclusions of the simple model continue to hold. Only if
joint products are complements (for example, an increase in carbon emis-
sions, which increases the demand for refrigerators, which in turn creates
more global warming and so on), this may partially lead to qualitatively
different results (see Sandler 1996). Though such a complication could be
handled in a static framework, treatment in a dynamic context would cer-
tainly be beyond the scope of this book.2
Since ‘several-goods models’ are typically set up as maximization prob-
lems subject to budget constraints, the influence of a change in income and
a redistribution of income on equilibrium emissions can be analyzed.
Whereas the first aspect can easily be integrated in the simple model by
interpreting an increase in income as equivalent to a reduction in abatement
costs, transfers can only be analyzed by assuming that the tastes of players
(or the willingness to pay for environmental quality) are unaffected by these
transfers. Though this is certainly a disadvantage of the model used here,
working with a simple model at the present stage allows us to relate the
setting of this chapter to the more advanced game theoretical later chap-
ters which require, by their nature, such a simplification anyway.
Another simplification worth mentioning is that the subsequent analysis
is restricted to a global emission game. An extension to cover transboundary
pollutants – though it is conceptually straightforward – would require us to
deal with additional variables, namely transportation coefficients reflecting
different spillover patterns between countries. Since most qualitative results
remain valid for transboundary pollutants also, this extension is not con-
sidered in what follows. Extension may be found, for example, in Kuhl
(1987) and Nentjes (1994).
As in the previous chapters, we take a stepwise approach. We start the
analysis in this chapter in a static framework, which we shall continue to
assume in parts of Chapters 10 and 11. We also integrate two-stage games
into the analysis in Chapters 10 and 11. Finally, we extend the analysis to
Static games with continuous strategy space 121
Let the payoff function (net benefit function) of country i, i , comprise the
benefits from emissions, i (ei ), and the damages caused by global emis-
sions, i (ei ). In particular assume:
A1: i
0 and i
0 ei
0, i (0)0; i0⇒ii ei
0
i
0 0
ei
ei0, i 0 ei ei0, i
0 ei
0, i (0) 0. (9.2)
The assumptions regarding the first derivative of the benefit functions imply
that benefits increase in emissions up to a level ei0. If emissions are further
122 Game theory and international environmental cooperation
1990; Endres 1994; Turner et al. 1994) and acknowledges the fact that
raising emission removal levels requires successively sophisticated and
costly abatement technologies.
With respect to the damage function assumptions, A1 describe the fact
that damage increases in emissions at an increasing rate. The last property
reflects the diminishing self-purification of environmental systems at higher
rates of contamination. Of course, there might be an upper bound of
aggregate emissions above which environmental systems collapse. In the
following, however, it is assumed that such an upper bound lies well above
eimax and therefore its implications for equilibrium considerations can be
discarded.
Another interpretation of i0 follows if the evaluation of environ-
mental damage is interpreted as society’s willingness to pay for emission
reductions. Then at higher levels of aggregate emissions this willingness to
pay is higher than at low levels and decreases more than proportionally if
environmental quality improves due to lower emissions.4, 5
9.2.2 Example
1 c N 2
i bi dei ei2 i ei ; bi 0, ci 0, d0 i and j I (9.3)
2 2 j1
The derivation of the Nash equilibrium proceeds in two steps. In the first
step the best reply of each country i for given strategies of the other players
is established. In the second step that strategy tuple for which no country
likes to modify its choice given the optimal strategies of the other countries
124 Game theory and international environmental cooperation
9.3.1 Example
The best response or reaction functions for the example are derived by differ-
entiating (9.3) with respect to ei , setting the first derivative equal to zero and
solving for ei (assuming N2, b1 b2 b, c1 c2 c):
bd cej d ej
ri:ei (ej ) ⇔Ei ; b/c and b c. (9.4)
bc 1
It will turn out that most of the later results can be expressed in terms of ,
which may be interpreted as a parameter describing the benefit–cost ratio
from emissions. From (9.4) it is evident that the best reply of country i
depends on the strategy of country j and hence no equilibrium in dominant
strategies exists. Country i’s optimal emission level is negatively related to
emissions in country j due to the global character of the negative external-
ity. The best reply functions of countries 1 and 2, r1 and r2, are drawn in
Figure 9.2.
Both best reply functions start from ei (bd )/(b c) where ej 0. The
e2
bd
—c r (1)
r1 1
r (2)
2 eo
d
e N(1)
bd
——–
b+c
r (1)
2
e2(2) eN
e2(1)
r (2)
1
r2
e1(2) e1(1) bd d bd e1
——– —
b+c c
optimal choice of ei becomes smaller at higher levels of ej. For ej (bd )/c,
ei 0. One possibility, to rule out negative emissions as a best reply, that is,
ri (ej )
0 (since without further assumptions it is not clear whether
eimax ei0 d is greater or smaller than (bd )/c), is to assume b
c or
1, as
is done in (9.4). Alternatively, one could define piecewise reaction func-
tions:
d ej
ei ej
d
ri (ej ): 1 . (9.5)
ei 0 ej d
This would imply that reaction functions do not stop at bd/c but continue
along the e1 and e2 axes in Figure 9.2 (not drawn).
From Figure 9.2 it appears that the NE emission tuple, eN, is determined
as the intersection of the best reply functions r1 and r2. Due to the assump-
tion of complete information, each player can form expectations about the
best reply of the other player and s/he knows that the only stable strategy
tuple is eN. To see this, consider that country 1 chooses e1(1) instead of e1N.
The best reply would have country 2 choosing emission level e2(1). This in
turn would motivate country 1 to correct its previous choice, e1(1), to e1(2).
Continuing with this kind of reasoning one derives at e2(2), and, finally, as
with a cobweb, the NE is reached. This method neatly stresses the logic
behind the NE as the result of the convergent expectations of rational
players (see Section 3.2).6
Mathematically, the NE is derived by substituting the reaction functions
mutually in each other and solving for the remaining variable. The follow-
ing is obtained:
bd d
eiN ⇔eiN (9.6)
b 2c 2
which upon substitution in the net benefit functions (9.3) delivers:
b3d 2
iN . (9.7)
2(b 2c)2
It can easily be seen from (9.6) that as long as environmental damage is
not neglected, that is, c0 in the example, eiN ei0 d iI. In particular,
from (9.4) (or (9.5)) it is evident that even if country j chose ej 0, the
maximum emission level of country i would only be ei bd/(bc)d.
With the help of (9.4) and (9.6) the implications of a variation of
the benefit–cost ratio on equilibrium emissions can be analyzed.
Differentiating NE emissions with respect to , we find eiN/0. That is,
the higher the opportunity cost of abatement compared to environmental
126 Game theory and international environmental cooperation
damage, the higher will be equilibrium emissions. In this case reaction func-
tions move outward, as indicated by the arrows in Figure 9.2, with the ‘new’
reaction functions ri(1) and the ‘new’ Nash equilibrium eN(1). Thereby, the
movement of the reaction functions may be broken down into two parts.
First, the starting point of the reaction function (ei bd/(bc), ej 0) of
country 1 (country 2) moves to the right (up) on the abscissa (ordinate).7
Second, the slope of the reaction function becomes less steep since environ-
mental damages are now valued at less than the opportunity costs of abate-
ment. This follows from:
What has been derived for the example can be shown for the general case
too. The first-order conditions (FOC) in the NE (assuming an interior solu-
tion8) are given by:
e2
e N(3)
2
r2
e A2
eN(1)
r1
e1
e1A e1N(2)
slopes of the reaction functions are less than 1 in absolute terms since i0
and are negative due to the assumption of the concavity of the net benefit
functions, that is, ii0 and i0. The greater i (the less negative
or smaller in absolute value), the more negatively sloped are the reaction
functions. At the limit, for i close to zero, reaction functions have a slope
of 1. The same is true for large values of i, that is, i→.
Recall in the example that ib and ic so that an increase of b/c
implies now in the general case that i/i is increasing.
There are six interesting cases to note with respect to the properties of
the reaction functions, four of which depart from assumptions A1.
First, allowing for i0 (contrary to A1) but still assuming ii0
(according to A1), the slope of the reaction functions would still be nega-
tive, though smaller than 1 (greater than 1 in absolute terms). In
Figure 9.3 such an example is shown where both countries’ reaction func-
tions have a slope less than 1. In this example there are three NE, two at
the boundaries, that is, eN(2) and eN(3), and one in the interior of the strat-
egy space, that is, eN(1). As above, the equilibrium eN(1) is determined as the
intersection of both best reply functions. To see that there are two more
equilibria, consider that country 1 chooses emission level e1A to which
country 2’s best reply is e2 0. Given e2 0, country 1 will respond by
choosing e1N(2) which results in the equilibrium eN(2). By symmetry, equilib-
rium eN(3) is derived where country 1 emits nothing. Of course, this requires
128 Game theory and international environmental cooperation
e2
r2 eN
r1
e1
reaction functions to be piecewisely defined (see (9.5)) and that eimax eiA
holds, otherwise eN(1) would be the only NE.
From the example one would expect that the question of whether a game
has a unique equilibrium or multiple equilibria would be closely related to
the slope of the reaction functions. This is indeed the case and this issue will
be taken up in more detail in Section 9.3.
Note that an immediate implication of the arguments above is that the
equilibrium conditions (9.9) hold only for an interior NE. In Figure 9.3,
though (9.9) holds for country 1 at eN(2), 2 2 0 must be true for
country 2. An ‘adjustment’ to 2 2 0 would imply negative emissions,
which are ruled out by the definition of the strategy space, that is, ei
0
iI. By the same token, at eN(3 ) 2 2 0 but 1 1 0.11
Second, from (9.10) it appears that one possibility to generate positively
sloped reaction functions would be to assume i 0 and i i0 (con-
trary to A1). However, i i 0 implies a convex payoff function for
which no interior maximum exists. In particular, as will be set out in
Section 9.3, the sufficient conditions for the existence of a Nash equilibrium
would not be met. Therefore, this possibility is discarded.
Another theoretical possibility to obtain positively sloped reaction func-
tions would be to assume i 0 (according to A1) but i 0 (contrary to
A1). In particular, one would need ii. This assumption would be in
line with a concave payoff function, though, of course, in the present setting
one has to think hard to find a case where i 0 would hold true.12
Nevertheless, this case is illustrated in Figure 9.4, since we refer to it later,
in Section 10.3. As it is drawn, country 1’s reaction function has a slope of
greater than 1 and that of country 2 has a slope of less than 1.
Third, best reply functions may not intersect for any emission tuple in the
Static games with continuous strategy space 129
e2
eA
2
eB2 r1
r2
e N1 e1
positive quadrant. Such an example is shown in Figure 9.5 (see Hoel 1991)
where reaction functions are drawn as if they were defined piecewise and
where eN is the NE tuple. To see this, pick eA (e1 0, e2 e2A) as the ‘initial
situation’ which is also the ‘origin’ of country 1’s reaction function. A best
reply to e1 0 has country 2 choosing eB2. This move and the subsequent
adjustment reactions are indicated by arrows. It is apparent that finally eN
(e1N, e2N 0) is reached which, in contrast to Figure 9.3, is a unique equilib-
rium.
Fourth, consider the possibility of linear damage cost functions, which
implies i0 (contrary to A1) and hence r i 0. Consequently, reaction
functions are orthogonal to each other as, for instance, r1(2) and r2(2) in
Figure 9.2. That is, countries have dominant strategies. Though linear
damage cost functions are sometimes chosen for mathematical simplicity
(see Bauer 1992; Hoel 1992a; Mäler 1994), they may be criticized on two
grounds: (a) they ignore the ecological relations mentioned at the beginning
of this section and may only be justified in the range of low emission levels;
(b) a linear specification does not depict a typical feature of global pollu-
tion control, namely that of the interaction and dependency of countries.
Fifth, another interesting case occurs if i 0 (contrary to A1) because
then r i 1. That is, reaction functions run parallel to each other, as shown
in Figure 9.6. In Figure 9.6(a), country 1’s reaction function lies to the left
of country 2’s reaction function13 and therefore in equilibrium country 1
emits nothing. (The arguments to derive eN are basically the same as those
used to derive the equilibrium in Figure 9.5.)
In Figure 9.6(b), net benefit functions of all countries are assumed iden-
tical (and i0). Then ri rj and there are multiple NE which comprise all
emission tuples on these reaction functions.
130 Game theory and international environmental cooperation
(a) (b)
e2 e2
eN
r2
eN
r1 r1 = r2
}
e1 e1
Sixth, though all reaction functions have been drawn as straight lines for
simplicity in the previous figures, this is not true in general. Whenever
second-order derivatives are not constant, best reply functions are not linear.
Then the curvature of the reaction functions depends on third-order deriv-
atives. Basically, reaction functions may be concave (Figure 9.7a), convex
(Figure 9.7b) or may contain some segments which are convex and some
which are concave (Figure 9.7c).14
In order to analyze the curvature of these reaction functions, the second-
order derivatives of these functions have to be determined. This is demon-
strated with respect to country 1’s reaction function assuming a
two-dimensional emission space. Defining convexity or concavity with
respect to the origin of the e1 e2 space, country 1’s slope of the reaction
function may be written (based on (9.10)) as:
1 e2 1 1
(9.11)
r 1 e1 1
from which follows the second-order derivative (see Appendix VI.1 for
details):
1 2e2 2 1 2 2e
2 1 1 1
⇒ 22 () 0 if 112 121 () 0.
r1 e 1 ( 1 ) 3 e 1
(9.12)
eB e1 eB e1 eB e1
From the discussion of the slope of the reaction functions and of the
point(s) where they intersect the question arises when we can expect a
unique Nash equilibrium. As preliminary information one may find out
first whether an equilibrium exists at all. The sufficient conditions for the
existence of a Nash equilibrium are readily checked in the present context
(Friedman 1986, p. 23):
132 Game theory and international environmental cooperation
The first property is satisfied. The strategy space is compact, that is, Ei [0,
eimax] and the strategy space of player i is convex because any strategy lies
within this interval. For instance, any strategy tuple for which Ei [0, d]
iI holds in Figure 9.2 can be connected by a straight line which is entirely
contained in the rectangle defined by the emission tuples (0, d ), e0 (d,
d )and (d, 0) and the origin (0, 0).
The second property also holds since in (9.1) a continuous payoff func-
tion for each player is defined. Payoffs are bounded since an upper and
lower bound for the strategy space has been defined (see also Section 9.4).
The third property is implied by A1 but also, as we know from the previ-
ous section, by i 0 as long as i i 0 holds and by upward-sloping
reaction functions requiring i 0 and ii.
Though the sufficient conditions for the existence of a Nash equilibrium
are rather general, this is not true for the uniqueness of a Nash equilibrium.
In fact, one may need more than one theorem to show uniqueness, depend-
ing on the particular circumstances. We shall introduce two of them.
If best response functions are twice differentiable and continuous, then
uniqueness can be checked by the following theorem (see also Friedman
1986, pp. 44ff. for a proof):
Since i /sj 0 are the implicit reaction functions, the Jacobian matrix
takes the following form:
21 21
…
s 1
2 s1sN
J ……………… . (9.14)
2N 2N
… 2
s 21 sN
Static games with continuous strategy space 133
If we try to establish that the inequality sign in (9.16) holds generally in the
emission game by applying assumptions A1 we shall be disappointed. At
such a general level the opposite sign in (9.16) cannot be ruled out a priori.
Only by introducing restrictive assumptions, for example, 12, will the
above inequality sign generally hold. Thus, for instance, in the case of
orthogonal reaction functions uniqueness of the Nash equilibrium would
follow due to i0 iI.
This lack of predictability suggests the need to search for other theorems
to establish a unique Nash equilibrium for the general payoff functions
(9.1) and assumptions A1. Before doing so we apply Theorem 9.1 to
example (9.3). In order to discuss some general features, it will prove helpful
to depart from the previous assumption of symmetric countries. Moreover,
it will turn out to be convenient to rearrange the implicit reaction functions
slightly to have:
id 1 b
i /ei bidbiei ci (ei ej )
0 ⇒ ei ej
0; i i (9.17)
1 i 1 i ci
from which matrix JA can be computed:
1 1
2
JA JJT 1 1 1 2 (9.18)
1 1
2
1 1 1 2
1 1 2
20 and 4 0. (9.19)
(1 1) (1 2)
Since 1/(1i )1, where the LHS term is the slope of a country’s reac-
tion function (see (9.8)), (9.19) is always satisfied. From (9.19) it is also
134 Game theory and international environmental cooperation
evident that it is not necessary that the slopes of both reaction functions are
smaller than 1 but it suffices if their sum is less than 2. Figure 9.4 represents
such a case (though reaction functions are positively sloped!) where the
slope of country 1’s reaction function is greater than 1 and that of
country 2’s reaction function less than 1; however, the sum of both slopes
is less than 2 and therefore there is a unique Nash equilibrium.
Thus in the case of linear reaction functions, Theorem 9.1 may be spec-
ified as follows:
Corollary 9.1
In a strictly smooth two-player game (N, S, ) with linear reaction func-
tions and where the sufficient conditions for the existence of a Nash equi-
librium hold, there is a unique Nash equilibrium provided /r1/s2
r2/s1/ 2.
Proof: Is obvious and therefore omitted.
Note that Corollary 9.1 also covers the case where the reaction functions
have opposite signs. In order to make progress in establishing uniqueness
of the equilibrium for the general payoff functions in (9.1), some definitions
on the way to another uniqueness theorem are needed (see, for example,
Friedman 1986, pp. 42ff.):
Thus, roughly, a contraction leaves the images of two points closer than the
original points. If f(x) is a differentiable function (which is not necessary
for a contraction and Theorem 9.2 below), then a contraction implies
fi (x)/xj
for each component fi (x) of f(x)( f1(x), ..., fN(x)).
s1*(1) and s1*(2) and thus s1*(1) r(s1*(1)) and s2*(2) r(s2*(2)). Then d(r(s1*(1)),
r(s1*(2)))d(s1*(1), s1*(2)), but r being a contraction means that d(r(s1*(1)),
r(s1*(2)))
d(s1*(1), s1*(2)) for 0
1. This requires, however, s1*(1) s1*(2)
which implies that the equilibrium is unique.
The upper bound of the payoff space in this game is the maximax payoff,
iU. Since i /ej 0 is true, the maximax payoff is determined by assum-
ing that all countries emit nothing and country i maximizes its payoff, that
is, maxi (ei , 0), which delivers iU i (ei (0), 0).
For the example it is known that ei bd/(bc) is country i’s best reply to
ej 0. Substituting these emissions in (9.3) gives:
b2d 2
iU . (9.20)
2(b c)
From i /ej 0 it follows that the lowest payoff to a country is obtained if
all other countries choose their highest emission level eimax. If i (ei ,
eimax)/ei 0 for small values of ei , country i’s lowest payoff is given by iL
max). If the first derivative is negative for all e 0, the lowest payoff is
(0, ei i
i (eimax, ei
L max).
In the example ejmax ej0 d and since i (ei , d )/ei 0 if b c, which
we assume to hold, iL (0, d ). For the function in (9.3) we get:
cd 2
iL . (9.21)
2
To punish country i, all countries will choose their maximum emission level
ejmax to which country i chooses its best reply, that is, iM (ei (ei
max, e max).19
i
Obviously, the assumption regarding the upper bound of the strategy space
Static games with continuous strategy space 137
if one is more specific regarding the curvature of the benefit curve, then
some candidates for eimax may suggest themselves.
In the example ejmax ej0 holds by assumption and country j minimaxes
country i by choosing ejM(i) ej0 d to which country i reacts by choosing
its best reply ei (d )eiM(i ) d(bc)/(bc) according to (9.4). Con-
sequently, we have:
bd 2(b 3c) M(i ) bd 2(b c)2
iM(i) ; j . (9.22)
2(b c) 2(b c)2
With this information we can now write {i
iL iI} and
IR {i
iM iI}. The normal form of this game is given by (N,
S, ) where N{1, ..., N, {, S{S1, ..., SN}, Si [0, eimax], {1, ..., N}
and i [iL, iU]. In the following it is assumed that all relevant emissions
are not restricted by the definition of the strategy space, that is, eimax eiN,
eiM(i ), eiS
0 where socially optimal emission levels, eiS, are derived in the
next section.
For the example Si Ei [0, d ], N2 and the lower and upper bounds
of the payoff set of player i are given by (9.20) and (9.21) respectively.
The social optimum in the global emission game follows from maximizing
global net benefits according to:
CBA: (2) i (eiS) k(eiS 0) j(0) k(eiS 0) if eiS 0 ejS 0
(3) i (0) k(0) if eiS 0 i, j, k
(9.24)
138 Game theory and international environmental cooperation
where the superscript S stands for social optimum.21 That is, a country con-
siders not only the damage caused by its emission in its own country, as in
the non-cooperative Nash equilibrium, but also those in the neighboring
country. Since (9.24) states the optimality conditions of a cost–benefit
analysis, these conditions are abbreviated CBA. The first condition in (9.24)
assumes an interior solution. That is, emissions in all countries are positive
in the social optimum. In this case social optimality implies that marginal
opportunity costs of abatement are equal across all countries. The second
condition takes care of the possibility that for some countries (the j coun-
tries) emissions might be zero in the social optimum. This could be the case
if marginal opportunity costs of abatement are very asymmetric in various
countries. The third condition reflects the possibility that damages are so
high that no emissions are advisable in the social optimum.
Comparing aggregate emissions in the social optimum with those in the
Nash equilibrium we find:
Proposition 9.1
Let the global emission game be described by the payoff functions in (9.1)
and assumption A1 in (9.2), then eiN eiS.
Proof: This assertion is readily proved by contradiction. Compare
the FOC (9.24) with (9.9), where the latter assumes an interior Nash
equilibrium, and assume that ekN ekS would be true. Then,
(1) i (ekN ) k(ekS) i, j and k. Consequently, from (9.9) and the first
constellation in (9.24), we must have (2) i (eiN ) i (eiS) i I, which is
only possible provided eiN eiS i I, and hence ekN ekS, which
contradicts the initial assumption. For the second constellation in (9.24),
again, assume ekN ekS and (1) above follows. Then for a country i for
which eiS 0 holds, this implies i (eiN ) i (ekN ) k(ekS) i (eiS)
which is only possible if eiN eiS for all i countries which, again, contra-
dicts the initial assumption. For the third constellation ekN ekS is
obvious because eiN
0 i I by assumption. The proof in the case of a
corner solution in the Nash equilibrium proceeds exactly along the same
lines and is therefore omitted here.
Note that though ekN ekS always holds, this should not be mistaken to
imply eiN eiS i I. For instance, assume that abatement cost functions
are equal across all countries, that is, i j i and j I, but that environ-
mental damages are perceived differently. Consequently, countries with
high damages will have low emissions in the Nash equilibrium and vice
versa. In the social optimum, where, due to i j, eiS ejS holds and where
the countries with low damage also have to consider high damage in the
Static games with continuous strategy space 139
neighboring countries, the high damage countries may very well have to
increase emissions compared to the Nash equilibrium.22
Moreover, note that iS iN and even iS iM may be possible for some
countries, though iS iN and iS iM always hold. For instance, if
a country has low opportunity costs of abatement compared to other coun-
tries and perceives damages as less severe than its neighbors, a globally
rational solution may imply that the individual rationality constraint of a
country is violated.23 Since this result is important for subsequent chapters
it is summarized in the following proposition:
Proposition 9.2
Let the global emission game be described by the payoff functions in (9.1)
and assumption A1 in (9.2), then iN iS and iM iS is possible if
eimax
eiN, though iM(i) iN iS.
Proof: The first part of the assertion has already been proved by example.
The second part follows trivially from:
max i
max 1 . . .max N
e1, …, eN e 1 e N
9.6.2 Example
For the example socially optimal emission levels are derived by differentiat-
ing i with respect to e1 and e2, setting both derivatives equal to zero and
solving for the respective variable. Thus, one derives a sort of ‘cooperative
reaction functions’ (that is, eiS(ejS) or in implicit form i j 0) which
have to be mutually inserted in each other in order to solve for socially
optimal emission levels:
bd d
eiS ⇔eiS . (9.25)
b 4c 4
Inserting these emission levels into the net benefit functions in (9.3) gives:
b2d 2
iS . (9.26)
2(b 4c)
A routine check confirms eiS eiN and iS iN i I due to the assumption
of symmetric countries.
140 Game theory and international environmental cooperation
Figure 9.8 Nash equilibrium and social optimum in the global emission
game
From the previous chapters we know that in a static context no other emis-
sion tuple than the (unique) Nash equilibrium is stable. If, for instance,
countries agreed to reduce emissions to eS, then each country would have
an incentive to deviate to eiD ei (ejS) which is indicated by the arrows in
Figure 9.8. More generally, for any emission tuple – which is not necessar-
ily socially optimal – for which i (ei ) i (ej ) holds, a country has an
incentive to increase emissions. Thus, even if a country has to increase
142 Game theory and international environmental cooperation
Note: a Based on payoff functions (9.3), N 2 and symmetric countries, that is, b b1 b2,
c c1 c2. In particular b 5, c1, d 10 are assumed.
In Figure 9.8 the Pareto frontier has also been drawn, on which by defini-
tion the socially optimal emission tuple must lie. The Pareto frontier repre-
sents all those emission tuples for which it is not possible to increase the
payoff to one country without reducing the payoff of any other country.
That is, for a given level of welfare of country j (or countries i in the N
2 case), one determines that emission tuple which maximizes country i’s
welfare. Thus, in contrast to the social optimum which implies an uncon-
strained optimization (with respect to welfare, see (9.23)), finding the Pareto
optima in this game requires a constraint optimization procedure.
Constructing the Lagrangian L for this problem, restricting the number
144 Game theory and international environmental cooperation
of countries to two for simplicity,25 and taking the relevant derivatives, the
Kuhn–Tucker conditions can be derived:
L L
1 ( 2 2)
0, e2
0, e2 0,
e2 e2
L L
2 2 ¯ 2
0,
0, 0
from which the following FOC along the Pareto frontier can be derived: 26,27
1 1 1 1 2
(a) e1 0, e2 0, 0: ⇔ 1 ,
2 2 2 1 2 2
1 1
(b) e1 0, e2 0, 0: ,
2
(9.30)
1
(c) e1 0, e2 0, 0: ,
2 2
(d) 0: i(i) 1 1 0, 2 2
0 ⇒ e1 0, e2 0,
(ii) 1 1
0, 2 2
0 ⇒ e1 0, e2 0,
(e) →: i(i) 1 1
0, 2 2 0 ⇒ e1 0, e2 0,
(ii) 1 1
0, 2 2
0 ⇒e1 0, e2 0.
Condition (a) holds in Figure 9.8 along the entire segment of the Pareto
frontier between eX and eY (where the emission tuples themselves are
excluded). Note that the LHS term of the equality to the right of the arrow
in condition (a) corresponds to the slope of the indifference curve of
country 1 and that the RHS term is identical to the slope of country 2’s
indifference curve (see (9.27) and (9.28)). Hence that part of the Pareto
frontier connects all those emission tuples where the slopes of countries’
indifference curves are identical.28 Since iso-net benefit curves are strictly
concave (convex), this implies tangency of these curves along the Pareto
frontier. This is also evident from Figure 9.8, where it is shown for indiffer-
ence curves I1M, I1(1) and I2M, I2(1) respectively.
Rearranging condition (a) in (9.30) we get:
1 2 2 1 1 2. (9.31)
Static games with continuous strategy space 145
π2
300
150 πs
100 πN
πM
50
Note: Payoff functions (9.3) apply, assuming b 5, c1 and d10.
Comparing conditions (b) and (c) in (9.29) with those in the social
optimum as given in (9.24) for a boundary solution involving eiS 0 for
some i and ejS 0 for some j, it is, again, evident that aggregate welfare will
generally be lower than in the social optimum as long as 1.
Conditions (d)(i) and (e)(i) correspond to emission tuples eA and eB
respectively in Figure 9.8. At eA country 1’s payoff is maximized, which
requires e2 0 due to 1/e2 0. Thus, eA gives country 1’s maximax
payoff, 1U(eA). By the same token, eB represents the emission tuple gener-
ating country 2’s maximax payoff. Hence, eA and eB imply that one country
emits nothing and the other chooses its best reply which correspond to the
origins of the reaction functions. eA and eB constitute the boundaries of the
Pareto frontier in Figure 9.8 and correspond to A and B respectively in
Figure 9.9.
Conditions (d)(ii) and (e)(ii) are not reflected in Figure 9.8. These con-
ditions would imply that marginal damages are already so high at low emis-
sions compared to marginal benefits as to call for no emissions at all. For
the example in (9.3) we would need bd
c to generate this result.29
Turning now again to the relations in the payoff space (see Figure 9.9) it
is evident that payoff tuples C and D are at the edges of the boundary of
the payoff space where one country chooses ei 0 and the other ej d. This
implies that at these points one country derives its lowest payoff iL. Of
course, these points are not Pareto-efficient and therefore do not belong to
the Pareto frontier. The whole payoff space of the emission game, , is more
lightly shaded than the individually rational payoff space, IR.
The concavity of the Pareto frontier (in the payoff space!) follows from
the envelope theorem. To see this, note that for the Lagrangian L1(e1,
e2)(2(e1, e2) ¯ 2) in (9.29) the envelope function may be written as
ev1() where indicates the amount by which the constraint 2(E1,
E2)
¯ 2) must be relaxed to obtain one unit more of 1. Thus the Pareto
frontier can be interpreted as the function ev. Since it is known that if 1(e1,
e2) and 2(e1, e2) are strictly concave, the envelope function is also strictly
concave (see Birchenhall and Grout 1984, pp. 176ff.; Varian 1984,
pp. 322ff.), we only have to establish these properties for these functions.
This, however, has already been done in note 3 based on A1.30,31
NOTES
1. More sophisticated models may be found, for instance, in Arnold (1984); Kuhl (1987);
Pethig (1982); and Sandler (1996). However, the main results of these works can also be
accommodated in the subsequent simple model. This will also be apparent in the core-
coalition models in Section 13.3, where the original setting is more sophisticated but
where all conclusions remain valid in the ‘simple’ framework as well.
Static games with continuous strategy space 147
2. To the best of our knowledge, a several-goods model, let alone a joint-production model,
still awaits game theoretical treatment in a dynamic context.
3. This condition follows from differentiating the net benefit function twice with respect to
the own strategy ei , that is, 2i /ei2 0. In case one requires strict concavity of the net
benefit function in the two-dimensional emission space, i0 must be assumed since the
Hessian determinant is only negatively definite if i i0 and (ii)(i)
(i)2 0⇔ ii0 i I hold (see, for example, Chiang 1984, pp. 307ff.). In the N-
dimensional emission space (N2) only concavity but no strict concavity holds since
there are many ej–ej combinations for a given level of ei which deliver the same net
benefit to country i. Recall, in a global emission game, only the sum of emissions in other
countries is relevant to country i, not its composition.
4. For the evaluation and measurement of environmental damage, see Endres and Holm-
Müller (1998); Finus (1992a, b).
5. An alternative way of modeling a global emission game would assume the payoff func-
tion to be given by i i (ri )i (rj ) where i is an abatement cost function, i a
damage cost function and ri emission reduction in country i. Thus the optimization
problem involves minimizing costs. Alternatively, one could assume i i (rj )i (ri )
where i are benefits from emission reduction (for example, Stähler 1998a). Assuming
the appropriate first- and second-order conditions would deliver the same quantitative
results obtained below.
6. Of course, this illustration is only an additional device to make the NE plausible and is,
strictly speaking, only valid for dynamic games because an adjustment process requires
at least two periods by definition.
7. This follows from ei bd/(bc)d/(1) and ei /d/(1)2 0.
8. The case of corner equilibria will be taken up below.
9. This is an application of the implicit function rule. See, for example, Chiang (1984,
pp. 204ff.).
10. Put differently, since ei /ek 1, ei /ek ei /ei.
11. Setting up the maximization task subject to ei
0 and deriving the Kuhn–Tucker condi-
tions would deliver the above FOC.
12. A simple way of modeling positively sloped reaction functions would be to
assume i b(dei 12 ei2)c(aej 12 ej2) instead of (9.3). If bc, then ii0 and
1r i 0, which delivers an interior unique Nash equilibrium.
13. This may be due to the higher marginal opportunity costs of abatement and/or the lower
marginal damage costs in country 1 compared to country 2.
14. As it is drawn, all examples in Figure 9.7 assume an interior Nash equilibrium.
15. For example, the benefit function could be given by the logarithmic function
i bln(1(eiI ei )/eiI ) bei where b is some positive scaling parameter, eiI denotes some
initial emission level, and hence the term in brackets denotes the fraction of emission
reduction. It is easily checked that i 0, i 0 and i 0 ei 0.
16. An example of a damage cost function featuring these properties is i c(ej )3.
17. Unfortunately, there is not much more one can say about when to expect the first, second
or a mix of both constellations. For the functions as specified in notes 15 and 16, i2/i
b/(2eiI ), and hence this term is constant, whereas i2/i 6c(e1 e2 )2. Thus, as long
as b is not too big, i2/i i2/i implies a convex reaction function.
18. Hence, the predictability of the equilibrium eN(1) may be regarded as rather low consid-
ering the possibility that players may make small errors (‘they tremble’) when choosing
their best reply. Any small error would upset this equilibrium. An equilibrium which is
immune against such errors is called a trembling-hand perfect equilibrium (Selten 1975),
which is a refinement of the NE concept.
19. It is easily checked that minimax and maximin payoff coincide in this game.
20. Thus, (9.24) reflects the relevant Kuhn–Tucker conditions of maximization problem
(9.23) where the Lagrangian multipliers have been omitted for convenience.
21. Constructing the Hessian determinant for this maximization problem and using assump-
tions A1 confirms that the aggregate net benefit function is strictly concave and hence
eS (e1S, ..., eSN) constitutes a unique global maximum.
148 Game theory and international environmental cooperation
22. A simple way of seeing this would be to assume cost coefficients c1 c2 for country 1 and
2 instead of c1 c2 c in (9.3). Assume, for instance, b 10, c1 9, c2 1 and d 10.
Then, e1N 1, e2N 9, e1S e2S 3.3, which demonstrates the assertion that eiN eiS is pos-
sible, though eiN eiS holds.
23. For instance, consider the example in the previous note for which we find 1N 355,
2N 445, 1S 77.7, and 2S 255.5, which stresses that iN iS is possible, though
iN iS is true. Computing the minimax payoffs delivers 1M 447.4 and
2M 318.2, which demonstrates that even iM iS may be true, though iM iS
holds.
24. In the emission game minimaxing another country leads to a higher payoff to a country
than if it is minimaxed itself, that is, iM(i ) iM(j) whereas in Matrix 4.2 iM(i) iM(j) has
been assumed. However, the relation of iM(i) and iM(j) is not important for the classifi-
cation of an extended PD game.
25. The extension to N > 2 countries should be apparent but is omitted here because the sub-
sequent derivations are mainly conducted to support the discussion of Figure 9.8.
26. An alternative way to derive the Pareto optimality conditions is to maximize the
weighted welfare function 1 (1)2 for different values of , 0
1, subject to
the constraints of the strategy space, that is, ei
0 i I. See Friedman (1986, p. 14).
27. The second-order sufficient conditions for a maximum hold since (1) the objective func-
tion 1 1 is twice differentiable and concave in the non-negative orthonant; (2) the
‘constrain function’ 2 2 ¯ 2 is twice differentiable and concave; and (3) the FOC for
each emission tuple in the emission space can be satisfied (see, for example, Chiang 1984,
pp. 738ff.). The concavity of these functions has been established in note 3.
28. The similarity to the Edgworth box should be obvious.
29. From the discussion it should be evident that, due to the restriction of the lower bound
of the strategy space to ei
0, the Pareto frontier cannot be simply derived from the
Lagrangian conditions, which would only deliver the FOC in the segment eX to eY in
Figure 9.8, but would miss those along the segments eA to eX and eB to eY.
30. By analogy with the arguments presented in note 3, strict concavity of the Pareto fron-
tier holds only in the 1 2 space. With respect to the N-dimensional payoff space (N
2) only concavity of the Pareto frontier holds.
31. If instead of the Lagrangian the weighted welfare function 1 (1)2 is used to
derive the Pareto frontier (see note 26), strict concavity follows from the simple fact that
the sum of two strictly concave functions is a strictly concave function (see Chiang 1984,
pp. 342ff.).
10. Finite dynamic games with
continuous strategy space and static
representations of dynamic games
10.1 INTRODUCTION
From the previous chapter it became evident that in a static setting there is
underprovision of the public good ‘environmental quality’. In the Nash
equilibrium global emissions are too high from a global point of view. Thus
one may wonder whether more positive results could be obtained in a
dynamic, though finite time horizon. This question will be analyzed within
two approaches.
The first approach remains in the tradition of repeated games, as encoun-
tered in previous chapters. More precisely, we proceed as in Chapter 4: first
the equilibrium (or the equilibria) of the constituent game is determined; and
second, one investigates whether the finitely repeated play of the stage game
leads to more optimistic results. Again, a simultaneous and a sequential
move version of the constituent game can be distinguished. In the former
case the analysis is simple. If the constituent game is the global emission
game described in Chapter 9, where there is a unique Nash equilibrium (NE)
due to assumption A1, the repeated play of this stage game NE is the only
equilibrium in a finite time horizon. This is an immediate implication of
Theorem 4.2. In the case of sequential moves, it is shown in Section 10.2 that
there is a unique subgame-perfect equilibrium (SPE) of the emission stage
game and hence, again, by Theorem 4.2, the finitely repeated play of this
stage game equilibrium is the only equilibrium of the overall game. Thus, the
outcome in the finitely repeated emission game does not differ from that if
the game is played only once. Hence, the main point to be clarified in the sub-
sequent analysis is whether the outcome in the simultaneous or in the sequen-
tial move game is preferable from a global point of view. In Section 10.2 this
is done for a filterable externality (as assumed in Chapter 9) and for the case
of a transferable externality in Section 10.3. Crudely speaking, the term
transferable externality refers to a ‘policy of high stacks’ where emissions are
not truly reduced but merely transferred to third parties.
149
150 Game theory and international environmental cooperation
From this it may appear that non-zero conjectures are more convincing
in the context of infinite dynamic games and incomplete information where
‘players learn gradually about their economic environment’ (Cornes and
Sandler 1985a, p. 125) and the conjectural equilibrium may be interpreted
as a long-term steady state (Sabourian 1989, pp. 86ff.). However, this inter-
pretation is also not very convincing for two reasons: (a) an explicit learn-
ing story is missing in these models; (b) all models assume complete
information. If they assumed incomplete information, then a natural way
of modeling this would be to assume that players assign a probability dis-
tribution over different possible strategies of their fellows when choosing
their best reply. The probabilities would be revised in the course of the game
if new information became available to players. However, as will become
evident, all models are deterministic in that players expect a very specific
reaction from their fellow players throughout the game, which is not revised
(though new information becomes available).
Consider a stage game where the payoff functions of countries are given by
(9.1). Furthermore, suppose that assumption A1 holds and assume without
loss of generality only two countries for simplicity. Let country i move in
the first stage and country j in the second stage. Since country i knows
country j’s best reply of the second stage, that is, ej (ei ), it can incorporate
this knowledge in its objective function when choosing its equilibrium
emissions, eiST, in the first stage (see (10.1)). Given these emissions, country
j chooses ej ej (eiST) in the second stage. Borrowing a term of oligopoly
theory, player i may be called the Stackelberg leader, player j the
Stackelberg follower and the SPE of this game a Stackelberg equilibrium,
which also explains the superscript ST.2 Thus, the objective functions of the
leader and follower read:
i i (ei )i (ei ej (ei )) and j j (ej )j (ei ej ). (10.1)
follow.4 From (10.2) it is apparent that for the follower the maximization
problem does not change compared to the assumption of simultaneous
moves. The FOC of the leader, however, now contains an additional term
(see (9.9)). In the conjectural variation context, ki could be interpreted as
the conjecture of the leader about the reaction of the follower.5 In the
present context, this parameter is nothing else than the slope of country j’s
reaction function. Due to A1, 1
ki
0 holds. The implications in the
Stackelberg equilibrium may be summarized as follows:6
Proposition 10.1
In the Stackelberg (stage game) equilibrium of the global emission game
with payoff functions (9.1) the Stackelberg leader i increases emissions
and the follower j reduces emissions compared to the Nash equilibrium,
assuming conditions A1 to hold. Global emissions will exceed those in
the Nash equilibrium of the simultaneous move game. Global welfare
will be lower than in the Nash equilibrium if the follower has the same
or higher marginal abatement costs than the leader in the Nash equilib-
rium. The leader gains and the follower loses from the strategic move.
Proof: See Appendix VII.2.
ei
rj
e ST
eA
I ST
j
eN
INj
ri
ej
Shogren and Crocker (1991) and Shogren et al. (1992) point out that some-
times countries or regions reduce their emissions at the cost of higher emis-
sions in other countries and regions. For instance, this could happen if a
government does not pursue an environmental policy which either fosters
preventative measures or ‘truly’ reduces emissions based, for instance, on
an end-of-pipe technology, but induces firms to build high stacks. Such
‘abatement measures’ do not reduce externalities but merely transfer them
to third parties. Following the terminology of Shogren and Crocker this
type of externalities are called transferable externalities in contrast to filter-
able externalities which we have considered so far. Of course, a transfer
strategy can only work in the case of impure public pollution, such as acid
rain, since for pure global pollutants only total emissions are relevant to
damage in a country and not their regional distribution. Hence, in this
section (and only in this section) we depart from our previous assumption
of a global pollutant.
A simple explanation for a ‘transfer policy’ could be that the costs of
transferring externalities is cheaper than truly curbing emissions. This case
causes some fundamental changes compared to the previous analysis.
Without loss of generality, we again restrict the analysis to two countries.
With respect to A1 the assumptions regarding the benefit functions do
not have to be changed. The same holds true for the effect of own emissions
on damage costs. That is, i / ei 0 and 2i / ei2 0 are assumed. Now,
however, we have j / ei
0, 2j / ei2 0 and 2i / ei ej
0. That is, reduc-
ing emissions in country i increases damage in country j at an increasing
rate. At low levels of foreign emission reductions, marginal damage caused
by own emissions is lower than at higher levels. These modifications lead to
the following results which are clearly distinct from those obtained in
Chapter 9 and Section 10.2.9
Proposition 10.2
In a transferable externality game with payoff function i i (ei )
i (ei ,ej ), where the benefit function has the properties enumerated in
156 Game theory and international environmental cooperation
1. The above result relies on a given technology and does not take into
consideration the possibility of a switch to another abatement policy
which might be globally more efficient. In other words, the transferable
externality technology is viewed as the only option to obtain a socially
optimal solution.
2. Whether aggregate emissions in the Stackelberg equilibrium exceed or
fall short of socially optimal emissions depends on the specific func-
tions. As in the case of filterable externalities, emissions in the
Stackelberg equilibrium are higher than those in the Nash equilib-
rium. A general comparison of global welfare in the NE with that in
the Stackelberg equilibrium is not possible in the case of asymmetric
countries for the same reason as laid out in Section 10.2. Thus, again,
from an ecological perspective whether the game is played once or
repeatedly, simultaneous moves are superior to sequential moves.
With respect to global welfare such a clear-cut conclusion is not pos-
sible.
3. It is interesting to note that in the case of transferable externalities
reaction functions are upward-sloping in emission space, as illustrated
in Figure 9.4 in the previous chapter. To ensure a unique Nash equilib-
rium, i2/ ei2 i / ei ej has been assumed in Proposition 10.2,
which is a sufficient condition so that the slopes of the reaction func-
tions have a slope less than 1 and Theorem 9.2 can be applied. The
details are laid out in Appendix VII.3.
The assumption that players hold conjectures about how other players
react to their strategies may be written in its most general form as (see
Cornes and Sandler 1983, 1984b):
e i
f i
C (, e , e )
i i (10.3)
ei
where the responsiveness by others is a function of a bunch of factors
which are summarized in the parameter (or vector of parameters) and
the level of the strategy of players i and i. To demonstrate how the non-
zero conjectural approach develops its full potential in capturing the inter-
action of agents, it is helpful to adopt an example by Cornes and Sandler
(1983) to the present context. The example assumes:
ei
e
i (10.4)
ei e i
where the relation in brackets expresses the conjecture that low emissions
in country i will induce other countries to reduce their emissions as well. In
particular, large emitters have a greater influence on others to follow suit
than if a country’s share of total emissions is small.10 The parameter
may
be interpreted as the elasticity of the conjectured response with respect to the
relative importance of a country’s emission level.11
Assuming a symmetric game allows us to study the effect of the number
of countries on the outcome.12 Since for symmetric countries ei
(N 1)· ei , (10.4) becomes:
e i
1
. (10.5)
ei (N 1)
That is, the expected conjectured reaction diminishes with the number of
countries. In other words, the greater the number of countries involved in
the pollution problem, the less is the emission reduction of a single country
noticed and the less is the free-riding incentive mitigated. The first-order
conditions implied by (10.5) are (assuming an interior solution):13
i i
1
. (10.6)
i (N 1)
158 Game theory and international environmental cooperation
That is, in the optimum the slope of a country’s indifference curve (LHS
term in (10.6); see also (9.27)) is equal to the conjecture (RHS term in
(10.6)). Since in (10.6) a positive conjecture is assumed, the optimum is
located on the ascending part of a country’s indifference curve.
10.4.2 Illustration
Figure 10.2 illustrates the effect of the elasticity parameter
on the con-
jecture of countries and global emissions. On the abscissa, emissions of a
representative country i, ei , are measured, along the ordinate those of the
other N 1 countries, ei. The figure is based on the payoff functions in
(9.3) and assumes the parameter values of Chapter 9 (without restricting N
to two countries).
e– i
23 N = 5, slope 4
22 X4 N = 4, slope 3
21 θ=1 N = 3, slope 2
eN(4)
20
19 θ=0
18 1/4 X3
C4
17
16 eN(3)
15
1/3
14 C3
D4
13 X2
eN(2)
12 N = 2, slope 1
D3
11
1/2
10 C2
riC
9 D2
8
eS(2) eN(1) X1
7 eS(4) eS(3)
6 θ = –1
5 θ = –2 eS(1) = A1
A2 = C1 = D1
4
–A3
3A ri
4
2
1
0
1 2 3 4 5 6 7 8 9 10 11 12 ei
Note: Payoff functions (9.3), b5, c1 and d10 are assumed.
ai )) i, j; ij and ai Ai , that is, the rationality requirement holds
at every status quo and not only in equilibrium, then the RCVE would
even resemble a subgame-perfect equilibrium in a ‘conventional’
dynamic game. Hence, by imposing the rationality requirement on a
conjectural equilibrium the time-implicit story of the conjectural vari-
ation setting is placed in a consistent framework.
With the help of Definition 10.1, we can now give an answer to the ques-
tion whether a conjectural variation outcome is stable as stated above:
Proposition 10.3
In a global emission game there is only a non-zero conjectural variation
equilibrium with negative conjectures. Global emissions will be higher
than in the Nash equilibrium. In a symmetric game, country-specific and
global welfare in the conjectural variation equilibrium are lower than in
the Nash equilibrium.
Proof: The first assertion is proved by applying the definition of an RCVE.
In the present context, country i’s best reply in the conjectural variation
outcome is defined by its reaction function, riC, and hence part (2) of
Definition 10.1 requires (see also Cornes and Sandler 1983, 1984b):17
i (1 ki )
riC ki (10.7)
i i ki (1 ki )
to hold where for simplicity we assume that ei / ei ki does not depend
on the level of ei and ei (as in the example (10.5)) and is therefore con-
stant, that is, f iC(, ei , ei )ki. In other words, we linearize the slope of
the reaction function around the conjectured equilibrium. Since riC 0
for ki 0, equality (10.7) cannot hold for positive conjectures. Only for
negative conjectures can the equality be satisfied.
The second assertion of eiC eiN can be shown by comparing the FOC
in the conjectured equilibrium for ki
0 (see note 17) and those in the NE
(see (9.9)) using a similar argument in the spirit of the proof of Proposition
10.1. The third assertion follows from strictly concave payoff functions,
i / ei
0 and i / ei
0 at emissions above eN and the fact that for sym-
metric games eiC eiN eiS and eiC eiN eiS i I.
Thus, in the context of global pollutants the expectation that positive con-
jectures lead to more optimistic results is disappointed as long as consis-
tency of conjectures is required in a ‘non-Nash-behavior equilibrium’ (see
Cornes and Sandler 1985a; Sudgen 1984, p. 773). In fact, non-Nash
behavior makes it more difficult to explain voluntary contributions to the
162 Game theory and international environmental cooperation
and the implicit offer curves of countries follow from differentiating (10.8)
with respect to ri:23
(1 1)e1I 2 e1I
, . (10.9)
1e 2I (2 2 )e 2I
For an example (10.9) can be solved explicitly.
10.5.2 Illustration
The example illustrated in Figure 10.3 assumes the net benefit functions of
(9.3), N2, b1 b2 10, c1 5, c2 1 and d10. In the figure use has been
made of the fact that global reduction r, that is, rr1 r2, implies r
(eiI eiA)/eiI or rr1(e1I ·e2I )/(e1I e2I ). Country i’s offer may thus be
µ
1.2 r
N(2) r 0(2) r N(1) r 0(1)
1.1
1.0
0.9
0.8
0.7
0.6
0.5 µ0*
0.4 m *2
0.3 µ N*
0.2 m *1
0.1
0
10 20 30 40 50 60 70 80 90 100 %r
Note: Offer curves are based on payoff functions (9.3), assuming b1 b2 10, c1 5, c2 1,
d10.
Figure 10.3 Emission reduction offer curves in abatement space
in the auction equilibrium are higher than in the initial situation. This
becomes apparent from Figure 10.3, assuming eiI eiN, where the auction
equilibrium emission tuple lies within the lens formed by indifference
curves I1N and I2N.
The main result illustrated in Figure 10.3 may be summarized in the follow-
ing proposition:
Proposition 10.4
Let eA be the equilibrium emission tuple of an auction of emission reduc-
tions as described above and let eiI denote initial emission levels, then eA
constitutes a Pareto optimum. In equilibrium, emissions in each country
and global emission are lower than in the initial situation, that is, eiA
eiI
and eiA
eiI and individual and aggregate welfare are higher than in
the status quo, that is, i (eA)i (eI) iI, and i (eA)i (eI) pro-
vided eiI eiN 0 iI. Each country receives a payoff in excess of its
minimax payoff, i (eA)iM(eiM(i ), ejM(i)) i I.
Proof: The first part of Proposition 10.3 follows simply by comparing
(10.9) with (9.31). The second part is proved in Appendix VII.5.
10.6.1 Introduction
induction. That is, first, equilibrium flat actions are determined, assuming
the matching rates to be given. The solution may be denoted by r̄ * (r̄ *1(M),
..., r̄ *N(M)) where M is an (N )(N1) matrix of matching rates where in
each row the matching rates of a country vis-à-vis the other N1 countries
is listed and r̄ * is a vector of flat contributions of dimension N.27 Hence,
r̄ i*(M) are the best replies of the second stage.
Second, the solution of the second stage is inserted into the payoff func-
tions and one solves for the optimal set of matching rates, M*. Substituting
M* into the best replies determined above, r̄ i*(M), delivers the ‘subgame-
perfect equilibrium’ of the overall game which is given by the two strategy
sets r̄ * and M*. Together with (10.10) overall equilibrium reduction levels
ri* and r* can be computed.
Once the best replies have been determined, it suffices to check stability
of a strategy combination by focusing on M* solely. Stability is then defined
in the standard fashion of a Nash equilibrium in which no player has an
incentive to change its matching rates. That is, i (r(M*))i (r(M*i , mij ))
i and j I, ij must hold where M*i denotes the matrix of equilibrium
matching rates except that one element is different, that is, mij m*ij.
Of course, other specifications than the linear relation as given in (10.10)
may also seem plausible but are far more difficult to work with. For most
games of economic interest a linear relation between activity levels of
different agents can hardly be solved by simulations, let alone analytically.
Therefore, Guttman and Schnytzer devote most of their attention to show
that, for a broad class of positive and negative externality games, socially
optimal matching rates constitute a subgame-perfect equilibrium if moves
occur simultaneously. However, it appears to us that this is trivially satis-
fied in almost all games by the construction of the equilibrium concept. For
sequential move games, the equilibrium concept exhibits even more severe
shortcomings. In particular, it seems difficult to establish the existence of
an equilibrium without imposing some restriction on the strategy space.
by eI eN, rN(1) is country 1’s offer curve and the resulting emission tuple is
e(m*1) as shown in Figure 10.4. By the same token, one can determine
government 1’s matching rate m*2 which it offers to country 2 provided it has
the first move. This would deliver point e(m*2) in Figure 10.4. The associated
matching rates m*1 and m*2 are also depicted in Figure 10.3.
To highlight the main forces at work, parts of Figure 10.4 have been
reproduced in Figure 10.5 on a greater scale. In order to keep the relations
e2
10
I2(m*1)
9 eN
IN1
I1(m*1)
8 e(m*1)
r e(m *1 )(2) I N2
A
7 e(m 2*(m*1))
1 2 3 4 e1
in Figure 10.5 tractable, only country 1’s offer curve, rN(1), is shown. Now it
is evident how the offer m*1 and therefore e(m*1) is determined. Country 2
chooses m*1 such that e(m*1) is located on the highest indifference curve of
country 2, I2(m*1), which is tangential to country 1’s offer curve rN(1).
Country 1’s welfare level at e(m*1) is represented by indifference curve I1(m*1)
and that of country 2 by I2(m*1). Since both indifference curves lie within
the lens formed by the Nash equilibrium indifference curves, I1N and I2N,
both countries have obtained a higher welfare level than in the initial situ-
ation. It is also apparent that the indifference curves I1(m*1) and I2(m*1)
themselves form a lens which indicates that there is room for further mutu-
ally beneficial offers.
In the next step, based on e(m*1), country 1 can offer country 2 that it will
match an additional reduction by m*2. To determine m*2, country 1 chooses
170 Game theory and international environmental cooperation
that point on country 2’s offer curve which delivers country 1 the highest
obtainable welfare. This is e(m*2(m*1)) where indifference curve I1(m*2(m*1)) is
tangential to the offer curve re(m*1)(2). Again, if one were to draw country 2’s
indifference curve (not shown in Figure 10.5), it would turn out that a
further Pareto improvement is possible. Hence, country 2 has an incentive
to commit itself to a matching rate based on e(m*2(m*1)).
It seems intuitively plausible that by extrapolating this process both coun-
tries will finally reach the Pareto frontier. An exact statement is difficult
since, when determining matching rate offers at some stage, the first-order
conditions deliver more than one solution (in the example, four). Though
some of these solutions can be discarded since they define a minimum for
the proposing country, negative matching rates may well be in the interest
of a country.28 Hence, if negative matching rates are not ruled out a priori,
then there might be cyclical offers, and it is not clear whether a final equi-
librium will be reached.
However, if negative matching rates are deleted from the solution set, the
Pareto frontier will be reached. By the definition of emission tuples along
the Pareto frontier, there is no positive matching rate which could improve
welfare of both countries and no further offer will be accepted. Thus, in this
narrow sense (ignoring negative matching rates) the matching equilibrium
would be stable though this procedure stretches the equilibrium concept to
the limit (and hence renders it almost meaningless). From the discussion it
is also evident that for a complete determination of the equilibrium the
sequence of moves (matching path) and the initial stage must be specified ex
ante. Both aspects of the practical determination of an equilibrium (restric-
tion of strategy space and specification of the details of the game) seem to
be unnoticed by Guttman and Schnytzer.
In the case of simultaneous moves equilibrium determination may also be
difficult. Therefore, Guttman and Schnytzer (1992) focus on socially
optimal matching rates and show that for the case of two players these
matching rates may constitute an equilibrium.29 The intuition is obvious in
the case of symmetric countries where m*1 m*2 1. If a country were to
reduce its matching rate, then emissions in both countries would increase,
which implies a welfare loss both to the deviator and to the country stick-
ing to its commitment alike. By the same token, an increase in the match-
ing rate would imply that the opportunity costs of abatement increase more
than damage decreases. Thus, Guttman and Schnytzer’s finding that
socially optimal matching rate equilibria may exist in positive or negative
externality games is not particularly surprising.
A more general problem with the strategic matching approach (which the
authors claim to be capable of solving the deadlock inherent in externality
games while staying within the realm of non-cooperative game theory) is
Finite dynamic games with continuous strategy space 171
that its ‘positive’ result crucially depends on the assumption that matching
commitments are fulfilled. For instance, if in the above-mentioned case of
symmetric countries matching rates are kept at the socially optimal equilib-
rium rates but countries are allowed to alter their implied reduction level,
which seems a natural conjecture within a non-cooperative setting, such an
equilibrium is no longer stable. It is hard to see how such an equilibrium
can be established without some external enforcement power. Of course, by
the construction of the game any deviation is instantaneously matched by
countermeasures which, in a ‘true’ dynamic setting, would be called pun-
ishments. Due to immediate reactions, the deviator cannot obtain any tran-
sitory gains as this is possible in a discrete time setting. Hence, the free-rider
problem is trivially solved.30
In a strict sense, Guttman and Schnytzer’s logic implies in an ordinary
static prisoners’ dilemma game that each country affirms the fellow player
that it will match his/her action ‘cooperation’ by ‘cooperation’, and refrains
from taking a free-ride because then both would end up in the non-
cooperative and Pareto-inferior Nash equilibrium. Since we did not accept
this logic previously, the question arises why we should accept it now,
though, admittedly, the result comes about in a neatly disguised term called
strategic matching.
Thus, we are led back to the question raised in the Introduction to this
chapter: Why not model a dynamic process as a dynamic game? It seems to
us that one does not gain any new insight by using such an approach com-
pared to the traditional models discussed at some length in Chapters 4–8 in
the context of discrete strategies (though Guttman and Schnytzer 1992,
pp. 74ff., claim exactly this to be the case). In particular, in contrast to the
previous models of this chapter, one can hardly argue that the advantage
of the strategic matching approach over the ‘conventional’ dynamic
approach lies in its greater simplicity.
What seems more convincing to us is to interpret strategic matching as a
negotiation process leading parties to sign an IEA. Like a minimum partic-
ipation clause which is part of many treaties (that is, the treaty only
becomes binding if a minimum of the signatories have deposited their sig-
nature with the treaty secretariat; see Black et al. 1992), matching rates may
be interpreted as the minimum commitment of reduction obligations by
each party. Compliance would then not be defined in absolute terms but
towards each party via matching rates. However, the treaty would have to
be enforced by the conventional type of threat and punishment strategies
within an infinitely repeated game framework.
172 Game theory and international environmental cooperation
Thus, no member of some group is expected and actually will not contrib-
ute more than others in that group, but also not less. Hence, there is no
problem of free-riding within a group.
Sudgen sets out to analyze the case of the symmetric preferences of
players and then discusses the more general case of asymmetric preferences.
Since the symmetric case is sufficient to demonstrate our main concern with
this theory, we restrict our attention to this case.
In the global emission context a contribution may be interpreted as emis-
sion reduction ri , where ei (1ri )·eiI and rri as in the previous section.
In the symmetric country case, the equilibrium contribution is r* N·ri*.
Let riL be that reduction level where i (riL, ri 0)0 and i i (rj )
i (ri ) be a strictly concave function in ri. Now Sudgen (1984, p. 778) argues
that for the equilibrium contribution rL r* rS must hold, where the sub-
scripts L and S stand for lower bound and social optimum respectively.
rL
r* is necessary to protect the self-interest of a country if it has no expec-
tations that others reciprocate. r*
rS must hold because otherwise coun-
tries contribute more than they are obliged to do. In sum, he postulates
that: (a) r* must be somewhere between rL and rS; (b) there is more than one
equilibrium; (c) one equilibrium is socially optimal; and (d) all other equi-
libria imply underprovision of the public good.
Though these conclusions are correct, at least if one chooses the ‘right’
interpretation of the equilibrium path, they are not terribly exciting.
Following Sudgen’s arguments, and taking a sequential point of view, one
should expect r* rN since i (riL, ri 0)i (riL, ri
N 0). This implies that
Finite dynamic games with continuous strategy space 173
in a first step each country chooses at least riL riN which, according to the
theory, it cannot adjust downward. Then, however, it is not ensured that
riL
riS holds, as Sudgen claims. That is, the overprovision of the public
good is possible. If this sequential view is given up in favor of a simultane-
ous move game, then of course riL riN by definition.
Sudgen notes that any contribution above rL faces the assurance problem
where each country likes to contribute if others will follow suit. Thus, it
seems that Sudgen has the second interpretation in mind. If so, then his
theory does not lead to more ‘positive’ results. If, however, the first inter-
pretation was correct, then riL riS may lead to a welfare loss if the overpro-
vision is pronounced enough. Moreover, the principle of reciprocity implies
that players accept a minimum of obligations, which also requires players
to be guided by some moral motives. Admittedly, the moral requirements
are not as strong as assuming unconditional commitment which Sudgen
(rightly) criticizes, but, as we think, his requirements are also not convinc-
ing in a game theoretical context.
NOTES
1. Of course, from previous chapters it is known that an extension to an infinite dynamic
time horizon is capable of solving this putative deficiency. For the emission game this
will be demonstrated in Chapters 12 and 14.
2. In oligopoly literature, assuming competition in quantities, if players move sequentially,
this is referred to as Stackelberg competition (von Stackelberg 1934). If all players
choose their quantities simultaneously, this is referred to as Cournot competition
(Cournot 1938); see also Koutsoyiannis (1991) and Malinvaud (1985). Hence, in some
environmental economics literature the NE is sometimes referred to as the
Cournot–Nash equilibrium (for example, Endres 1993, 1997). The similarity to the emis-
sion model should be obvious: output is equivalent to emissions; the output of player i
exhibits a negative externality on player j via a price drop. All players would be better off
by forming an output cartel and by reducing supply.
3. To keep the discussion brief we consider only an interior Stackelberg equilibrium in the
following. That is, we assume that the equilibrium is not restricted by the boundaries of
the strategy space. In particular, for the Stackelberg leader eiST
eimax and for the fol-
lower ej(eiST) 0 are presumed to hold. Moreover, we assume that the second-order con-
ditions in the Stackelberg equilibrium hold (which is not automatically ensured). Then
it follows immediately that the Stackelberg equilibrium is unique. See Appendix VII.1
for details.
4. Recall i / ei i / ej i / e due to the assumption of a pure public bad type of pol-
lutant.
5. ki may also be interpreted as a weighting factor of marginal damages.
6. A similar result can be obtained for transboundary pollutants as well.
7. For the example in (9.2), assuming bi bj b and ci cj c, we find eiST [d(b2 bc c2)]/
(b2 3bc c2), ejST [d(b2 bc c2)]/(b2 3bc c2) and ekST [2d(b c)]/(b2 3bc c2).
Comparing these emissions with those in the Nash equilibrium, as given in (9.6), con-
firms ekST ekN . Due to the symmetry of this example 1(e1N )2(e2N ) and so one can
immediately conclude that global welfare is lower in the Stackelberg equilibrium than in
the Nash equilibrium.
174 Game theory and international environmental cooperation
8. See the general discussion in Section 2.3 and, in the context of the chicken game, Section
4.3.
9. i / ei i / ej is assumed in Proposition 10.2 to guarantee an interior Stackelberg
equilibrium. Again, the second-order conditions in the Stackelberg equilibrium are
assumed to hold, which may not generally be true. See Appendix VII.3 for details.
10. Of course, one could also specify the share in brackets as ei /ei and could derive the
same qualitative results as presented below. However, in the later computations it turns
out that the specification in (10.4) is more convenient to work with.
11. Let ei / ei ki as in Section 10.2, then:
dki e /e
· i i .
d(ei /ei ) ki
12. The term outcome instead of equilibrium is used intentionally, since so far it has not
been checked whether the conjectured outcome is stable. We shall turn to this issue below.
13. This follows from i i i · ( ei / ei )0, substitution of (10.5), and rearrangement
of terms.
14. For
1, ei / ei N1 and the first-order condition in the conjectured outcome
(see previous note) becomes i i i ·(N 1)0 ⇒ i i 0.
15. This follows from the facts that (1) 0
ei / ei
N 1 implies eiS
eiC
eiN i I in the
symmetric country case; (2) eiS i I maximizes country specific and global net benefits;
and (3) net benefit functions are strictly concave in emissions by A1. For ei / ei well
above N1 (below
1 in the example) welfare in the conjectured outcome may be
below that in the Nash equilibrium since opportunity costs of abatement are far above
optimal levels.
16. This is the second point of the claimed shortcomings of conjectural variation models
which are listed in the Introduction to this chapter.
17. This follows from totally differentiating the first-order condition i i (1ki )0 in the
conjectured optimum.
18. Cornes and Sandler (1985a, pp. 128ff.), claim that more positive results could be
obtained in the case of impure public bads. To see this, assume for instance a transboun-
dary pollutant and only two countries. Let payoff functions be given by
i i (ei )i (aiiei aij ej ) where 0aii 1 and 0aij 1 are transportation coefficients,
aii indicating that portion of emissions from country i which remains in this country and
aij that portion of emissions from country j which is transported to country i. Then (10.7)
reads [i(aiiaij aij2 ki )]/[ii(a2ii aiiaijki )]ki which can be satisfied for positive con-
jectures provided aiiaij aij2 ki
0 ⇔aii /aij
ki (see also Cornes and Sandler 1984a).
c(1 k)
19. For the example employed in Figure 10.3, riC (k kj k due to symme-
b c(1 k) i
try) and the rationality requirement implies ri k to which the solutions are
C
k (2cb 4bc b2 )/2c.For the parameter constellation of Figure 10.2 it turns out
that k0.146 is the only stable conjecture. The second solution leads to emissions
outside the domain of the strategy space.
20. When deriving a negative conjecture equilibrium one has to keep in mind that neither
the first- nor the second-order conditions for an interior solution may generally hold.
This is demonstrated in Appendix VII.4. See also Cornes and Sandler (1984b, p. 375) on
this point.
21. Note that the conjectural variation approach allows us to link the Nash equilibrium, the
social optimum and the Stackelberg equilibrium to each other in a symmetric global
emission game in a simple manner (Varian 1984, Section 2.10). We have: Nash equilib-
rium: ki 0 i I; social optimum: ki N1 i I and Stackelberg equilibrium: ki
0, kj rj where the FOC for an interior solution reads i i (1 ki )0.
22. The basic idea of this section goes back to Nentjes (1994).
23. We show below in the proof of Proposition 10.3 that the SOC are always satisfied. See
Appendix VII.5.
24. Aggregate emissions implied by a particular offer can simply be measured by drawing a
Finite dynamic games with continuous strategy space 175
45° line through an emission tuple (not shown in Figure 10.4). The point where this line
cuts the ordinate or the abscissa gives aggregate emissions.
25. From country 1’s perspective m*2 would be the optimal exchange rate; from country
2’s perspective this is m*1.
26. The reason is that from (10.9) such general conclusions cannot be drawn. Even for the
specific functions (9.3), the equilibrium price and resulting emissions cannot be obtained
analytically any more (except for symmetric countries) but only by simulations.
27. Of course, M* could also be defined as an NN matrix where the diagonal elements are
1.
28. This is also noted by Guttman (1987, pp. 9ff.). See the arguments when comparing
(A10.6) with (A10.7) in Appendix VII.5 in deducing whether ri , as the solution to the
FOC, will be positive.
29. This equilibrium is characterized by matching rates m*ji [ j/ ri]/[ j / rj]. If actor i
maximizes i (ri , rj ) with respect tor̄ i , noting ri r̄ i mijr̄ j and rj r̄ j mjir̄ i , the FOC
read: i / r̄ i ( i / ri )( ri / r̄ i )( i / rj )( rj / r̄ i )0, from which i / r̄ i ( i / ri )
( i / rj )· mji 0 follows. Substitution of m*ji from above delivers ( i / ri ) ( j / ri )0,
which are the FOC in the social optimum.
30. This weakness is even more striking since the authors do not impose any rationality con-
straint of the sort described in the conjectural variation context. See Definition 10.1.
31. Sudgen’s paper sets out to criticize Margolis’s ‘Theory of Unselfish Behavior’ (Margolis
1982). Basically this ‘theory’ assumes that an agent is not only concerned about his/her
own utility but also about that of the fellow players. Essentially, this theory postulates
altruistic or any kind of moral motivation of individuals when contributing to the pro-
vision of a public good. Like Sudgen, we feel that this approach is too simple to explain
voluntary contributions. If players were altruistic, free-riding in international coopera-
tion would hardly be a problem worth analyzing.
11. Bargaining over a uniform
emission reduction quota and a
uniform emission tax
11.1 INTRODUCTION
176
Uniform emission reduction quota and emission tax 177
role in international politics has not yet been answered, to our knowledge.
Nevertheless, it has been suggested that uniform solutions are apparently
perceived to be ‘fair’ and therefore find relatively easy acceptance. Uniform
solutions thus constitute some kind of a focal point in the sense of
Schelling (1960).3 Moreover, to agree on differentiated solutions may take
time and is therefore associated with higher transaction costs (Endres
1996a). However, in the sequel we do not model transaction costs or
psychological motives. These arguments just serve to motivate the focus of
this chapter on ‘uniform solutions’.
The lowest common denominator decision rule, henceforth abbreviated to
LCD decision rule, can frequently be observed in international politics
(Endres 1996a, b; Unerdal 1998a, p. 6, and 1998b, p. 109). Typical exam-
ples comprise the voting procedure within the European Union or within
the UNO on particular issues (Lenschow 1996). The reason for this deci-
sion rule in international politics is simple: due to the lack of an enforce-
ment authority at the global level, the accession to IEAs and the agreement
on abatement targets have to be voluntary. Moreover, if the LCD decision
is applied an agreement can be reached with only limited information
requirements (see Section 11.3). Applying the LCD decision rule also gives
negotiators no incentive to misrepresent their preferences strategically (see
Section 11.4).
From an economist’s point of view it is surprising that uniform reduc-
tion quotas, which belong to the group of command and control instru-
ments, are part of so many IEAs since they are generally inefficient. For a
given agreed reduction level, marginal abatement costs will typically differ
between countries. Therefore, economists have persistently argued that
market-based instruments, such as effluent charges or emission permits, are
better suited to achieve abatement targets cost-efficiently. However, this
advice has not yet fallen on fertile soil. This chapter and the subsequent
chapters present some evidence why this might be the case.
To achieve this task efficiently, we select only one market-based instru-
ment for comparison, namely a uniform emission tax, and contrast this
instrument with the quota. As we shall set out in Section 11.2, in the context
of global pollutants, which we assume throughout this chapter, the assump-
tion of a uniform application of a tax in all countries is in fact a necessary
condition for cost-efficiency. As will become apparent below, a uniform tax
also implies emission reductions at a fixed exchange rate, however, one
which differs from that of a uniform quota. Similar to the quota regime, we
consider a bargaining game in which countries settle for the lowest bid, that
is, the lowest tax proposal.
The subsequent analysis of the quota and tax bargaining equilibrium
focuses on two politically interesting questions:
178 Game theory and international environmental cooperation
1. Why are abatement targets within most IEAs rather low, though
cost–benefit considerations would suggest lower emissions?
2. Why have effluent charges (as one representative of market-based
instruments) not been applied in any IEA so far whereas quotas have
found widespread application?
CEA: ei ek, ej 0⇒i (ek)j (0)
ei ej ek 0⇒i (0) and j (0) (11.2)
ti (eTi ) if ti (0)
max i (ei ) tei ⇒ eTi 0 if t i (0)
ei
eTi eimax if t 0. (11.3)
Shapley value (see, for example, Barrett 1992b, 1997b; Botteon and Carraro
1997).5 These concepts are well founded in the game theoretical literature,
possess some interesting axiomatic properties and also satisfy some norma-
tive criteria. Among other factors, this explains their frequent applications
in the environmental economics literature.6
A prerequisite of most cooperative bargaining concepts is that there is
some cooperative spirit when it comes to choosing an abatement target and
allocating abatement burdens. Additionally, a cooperative spirit is also
required when the agreement is implemented since there is no suprana-
tional authority which could enforce compliance by all parties. Moreover,
these concepts rest on the assumptions of complete information and unlim-
ited transfers. Typically, they also determine a solution in terms of payoffs
and abstract from the problem of how this is institutionally translated.
That is, the choice of the policy instrument is not considered in these con-
cepts. Since IEAs typically specify an abatement target below a Pareto-
efficient level and employ inefficient instruments, this seems to suggest that
some important features of international politics are not captured by those
concepts.
Acknowledging these deficiencies, we propose to consider a two-stage
game in which some of the above-mentioned assumptions are relaxed or
changed. In the first stage – the bargaining stage – the two countries nego-
tiate and agree either on the level of the emission quota or the emission tax.
In the second stage – the implementation stage – the agreement is enforced.
With respect to the particular assumptions the following modifications
apply:
max i (eiI (1 ri ))
i (eI(1 ri )). (11.5)
ri
That is, the exchange rate of the auction market is set to 1, that is, 1 (see
(10.9)). According to the LCD decision rule, the agreed reduction level,
denoted rA, is equal to r1 if r1 r2 and is equal to r2 if r1 r2 where ri denotes
a country’s proposal. With respect to Figure 10.3 depicting the offer curves
in the auction market, this implies that one country (in the example
182 Game theory and international environmental cooperation
country 2) suggests an emission reduction below and one country (in the
example country 1) above r(N*) (r(0*)) at 1 if eiI eiN (eiI ei0).
We call the country which makes the lowest bid the bottleneck country
(which is in the following denoted country i). Further, as mentioned above,
we assume that the status quo is represented by the NE, eiI eiN
i I.12
Under the tax regime a country’s proposal ti follows from:
where, again, according to the LCD decision rule, we have tA t1 if t1 t2,
tA t2 if t1 t2 and tA t1 t2 if both proposals are identical.
The relation ei (ti ) and ej (ti ) is known from (11.3). Since ei /ti 1/i
(see Appendix VIII.1), the exchange rate defined in terms of emissions is
given by ei /ej j /i. For quadratic benefit functions this exchange rate
is constant; for benefit functions of higher order the exchange rate is a
function of emissions itself. Thus, each participant with information on
ei /ti and ej /ti can compute an optimal uniform tax from its perspec-
tive.
The solution to (11.5) is unique since payoff functions are strictly concave
with respect to a reduction level ri. This has been shown in the proof of
Proposition 10.4 (see Appendix VII.5) where strict concavity has been
established for any positive exchange rate (and hence the case of 1 is
covered as well). It turns out that a sufficient condition to establish strict
concavity of the payoff function in the tax rate is to require that second-
order effects dominate third-order effects (see for details Appendix
VIII.1).13 Therefore, we assume:
Proposition 11.1
Let ki {ri, ti} denote the bargaining proposals of country i derived from
(11.5) and (11.6) under the assumption of partial information where
each country knows that an agreement according to the LCD bargain-
ing rule implies kA min [ki, kj ]. Further, let ki ki and kj kj denote
some arbitrary proposals of countries i and j and assume properties A1
Uniform emission reduction quota and emission tax 183
and A2 to hold, then the lowest bid game has an equilibrium in dominant
strategies, that is, i (ki, kj )i (ki , kj )
iI.
Proof: Let ki solve the problem max i (ei, (kA), ej (kA)), kA min[ki, kj ]
ki
and pick arbitrarily kj Kj.14 Then if ki kj , this implies i (ki, kj )i (ki ,
kj ) since ki solves the maximization problem stated above. However, if
ki kj , then if country i proposes ki kj instead, then bid ki kj leaves i
unchanged, and a bid ki kj reduces i by strict concavity.
Proposition 11.1 implies that, assuming partial information, the LCD rule
possesses the property that ‘truth-telling’ is a dominant strategy. Hence, it
does not pay a country to make a strategically manipulated bid instead of
its ‘true’ bid in order to bias the final agreement in its favor. This may reduce
the transaction costs of bargaining. Other agreement procedures, such as
agreeing on the arithmetic mean of the proposals or the auctioning equi-
librium of Section 10.3, are generally more vulnerable to strategic behavior
and may constitute neither an equilibrium in dominant strategies nor a
Nash equilibrium. Countries may have an incentive to misrepresent their
preferences in a first step, but also may have an incentive to alter their pro-
posals once offers are disclosed. The result may serve as an additional
explanation for the popularity of the (inefficient) LCD rule in international
politics.
The index Ij simply reflects a country’s best reply as defined in (9.9) and
where in the (interior) NE Ij 0
jI holds. In both cases, whether a
country adjusts or free-rides, we assume that a country’s action is guided
by its best reply function. In the case of an adjustment, equilibrium emis-
sions are given by eik:ei (kA), e jk,a:ej (ei (kA)) ej (kA) and ekk,a:eik ejk,a
where the superscript a stands for adjustment. If no adjustment takes place
and countries comply with their obligations, equilibrium emissions imme-
diately follow from the agreement on kA. That is, eik:ei(kA) and ejk:
ej (kA). Free-riding would imply (out-of-equilibrium) emissions eiD:
ei (ej (kA)) ei (kA) and eDj :ej (ei (kA)) ej(kA) respectively where the super-
script D stands for deviation.
Therefore, as a matter of terminology, we distinguish between a bargain-
ing outcome and an equilibrium. Outcome refers to the agreement on kA;
equilibrium emissions or tax and quota equilibrium refer to emissions after
possible adjustments have taken place.
With these definitions we can now state the following result:
Proposition 11.2
Let the bargaining outcome be determined by the LCD decision rule
under partial information and allow countries to overfulfill but not to
violate the terms of the agreement, then there is always a free-rider incen-
tive under a quota agreement, that is, Ij 0
j I and rA [0, 1], but
never an incentive to overfulfill the terms of the agreement. Under the
tax regime the non-bottleneck country may have an incentive to unilat-
erally reduce emissions below what is implied by the agreement.
Proof: Under the quota regime any agreement implies rA 0. This
follows from Appendix VII.5, where it has been shown for the auction
market that offers will be positive for any exchange 0 and under the
quota regime 1 holds. Since an (interior) NE would imply rA 0 and
Ij 0 (recall ejIejN
jI), Ij j (ejN(1 rA))
j (ekN(1 rA)) 0 for any
rA 0 must therefore hold.
Uniform emission reduction quota and emission tax 185
1 c 1 c
1 b de1 e12 (e1 e2 )2, 2 b de2 e22 (e1 e2 )2 (11.8)
2 2 2 2
and hence the tax proposals of countries 1 and 2 are:
2bcd( 1) 2bcd( 1)
t1 ; t2 , (11.9)
b2 2c 2c c
b 2c 2c c
The results obtained in Section 11.4 for the bargaining proposals together
with the investigation of possible incentives to overfulfill an agreement in
Section 11.5 allow us to derive some general results:
Proposition 11.3
Let the bargaining outcome be determined according to the LCD deci-
sion rule under partial information and allow countries to overfulfill but
not to violate the terms of the agreement, then global emissions under
the quota and tax regimes are always lower than in the Nash equilibrium,
that is, eik eiN, k{Q, T}. Under the quota regime global emissions
may be below socially optimal levels, that is, eiQ eiS, whereas under
the tax regime global emissions are at least as high as in the social
optimum, that is, eTi eiS.
Proof: See Appendix VIII.3.
Proposition 11.4
In the uniform quota and uniform tax bargaining equilibrium aggregate
welfare is always higher than in the Nash equilibrium but generally lower
than in the social optimum. Under the quota regime equal proposals are
a necessary condition to obtain socially optimal welfare levels whereas
under the tax regime it is a sufficient condition.
Proof: See Appendix VIII.4.
Taken together, the results suggest that environmental quality and global
welfare improve under both regimes compared to the NE, though global
emissions are usually above and global welfare below the social optimum
due to institutional restrictions.
From an individual government’s perspective, however, what matters is
not global gains to be realized in an agreement but the welfare implications
for its ‘own’ country.
Proposition 11.5
Under a quota agreement each country gains compared to the Nash
equilibrium, that is, iQ iN
iI, regardless of which country deter-
mines the terms of the agreement. Under a tax agreement the non-bottle-
neck country always gains from an agreement, that is, jT jN; the
bottleneck country may gain or lose compared to the Nash equilibrium,
that is, Ti iN, though it receives at least an individual rational
payoff, that is, Ti iM. If the non-bottleneck country j has an incentive
to reduce emissions unilaterally below what is implied by the agreement
on tA, that is, Ij (tA)0, then Ti (eT,a)iN(eN ) for the bottleneck country
i.
Proof: See Appendix VIII.5.
Thus, even though countries may exhibit very asymmetric interests, both
countries gain from a quota agreement in any case. This may not be true
under a tax agreement. At first glance, the fact that the bottleneck country
could lose under a tax regime may seem at odds with the rules of the game:
that the bottleneck country determines the terms of the agreement.
188 Game theory and international environmental cooperation
Corollary 11.1
For the functions in (11.8), assumption A3 in (VIII.4), Appendix VIII.2,
partial information and the LCD decision rule, 1 implies r1 r2
rA so that ekQ ekS and Q i i hold. If 1, then r1 r2 r
S A
from which ek ek (r )ek and k k (r )k follow. If
Q Q * S Q Q * S
, then r1 r2, implying ekQ ekQ(r*), ekS and kQ kQ(r*)kS.
is a necessary and sufficient condition for a Pareto-efficient bar-
gaining equilibrium under the quota regime.
Under the tax regime 1/ implies t1 t2 rA from which ekT ekS
and kT kS follow. If 1/, then t1 t2, implying ekT ekS and
kT kS. 1/ is a necessary and sufficient condition for a Pareto-
efficient and socially optimal equilibrium under the tax regime.
For all parameter constellations Qi i and i i
i I.
N T M
Proof: Most parts of the proof are contained in the general proofs of
Propositions 11.3–11.5 (see Appendices VIII.3, VIII.4 and VIII.5). The
statement with respect to Pareto-efficiency follows from the auction
market in Section 10.4, where it has been shown that at the equilibrium
exchange rate (which implies in the present context equal proposals) the
emission tuple is an element of the Pareto frontier. Since under the tax
regime no adjustment takes place for 1/ (bargaining outcome
equals tax equilibrium), Pareto-efficiency follows.
regime may be lower or higher than in the tax regime. Of course, whenever
eiQ eTi holds, iQ Ti must be true because then aggregate emissions
under a tax regime come closer to those in the social optimum and abate-
ment is conducted cost-efficiently. However, if eiQ eTi this general con-
clusion cannot be drawn any more because this may imply iQ Ti . In
such cases, though reduction is conducted cost-efficiently under a tax
regime (CEA condition (11.2) holds), aggregate emissions under a quota
regime come closer to what is required from cost–benefit considerations
(see CBA condition (9.24)). If eiQ eTi , then the cost-efficiency effect
(CEA effect) may be overcompensated by the cost–benefit effect (CBA
effect) which renders global welfare higher under a quota regime than
under a tax regime. The reason is that in these cases the bottleneck country
finds the conditions under the quota regime more in line with its interests
and therefore accepts higher abatement targets than under a tax regime.15
Proposition 11.6
Assuming that an agreement is reached according to the LCD decision
rule, global welfare might be higher and global emissions lower in the
quota than in the tax equilibrium.
Proof: The proof follows by example, such as using net benefit functions
in (11.8) and Table VIII.1 in Appendix VIII.2.
To say a little more about the conditions when eiQ eTi or iQ Ti
hold, we consider two cases. Case 1 assumes for the example in (11.8) equal
damage in both countries but allows for different benefit functions, that is,
1 and 1. Case 2 considers equal benefits but different damages, that
is, 1 and 1.
For both examples we find eiQ eTi and iQ Ti if b/c is large.
The opposite holds for small values of .16 That is, if abatement is relatively
expensive compared to perceived environmental damage the quota regime
performs better than the tax regime. Of course, the term ‘relatively expen-
sive’ is unspecific and difficult to relate to real world problems. Therefore,
conclusions can be only indicative and must be drawn with caution. With
these restrictions in mind, however, the literature suggests that a problem
where abatement costs are perceived to be rather high compared to possible
damage costs is the greenhouse effect. In contrast, the depletion of the
ozone layer could belong to the second category of low values because
abatement costs are rather low compared to damage (Barrett 1991b, 1994b;
Nordhaus 1993).
Taken together, the results suggest that in a second-best world with insti-
tutional restrictions a quota may perform better with respect to ecological
190 Game theory and international environmental cooperation
and welfare criteria than a tax regime. In these cases the bottleneck country
finds the conditions under the quota regime more in line with its interests
and therefore accepts higher abatement targets than under a tax regime. Of
course, we have to remind ourselves that this result hinges on the assump-
tion that bargaining equilibria are stable, which is still to be analyzed in
Chapters 12 and 14.
So far we have assumed that the bottleneck country does not consider the
possible adjustment of the neighboring country when putting forward its
proposal. As long as damage in the neighboring country is not known
(assumption of partial information) the adjustment reaction cannot be
computed. However, if we assume a game of complete information, then
overfulfillment of a country might be anticipated and used strategically by
its neighbor. Of course, a quota regime is immune to strategic offers
because there is no proposal ri 0 for which Ij (eQ)0 is true. Under a tax
regime, however, a country performs:
Proposition 11.7
A strategic proposal under a tax regime yields higher global emissions
than without strategic considerations and compared to the stage game
Nash equilibrium. Global welfare will be lower than without strategic
considerations. The non-bottleneck country loses from the strategic pro-
posal and its welfare is below that in the Nash equilibrium; however, it is
above the minimax payoff. The bottleneck country will gain from the
strategic bid and its welfare will be above that in the Nash equilibrium.
Under a quota regime strategic proposals will never occur.
Proof: See Appendix VIII.6.
11.8 SUMMARY
1. uniform solutions;
2. no transfers; and
3. agreement on the lowest common denominator offer.
The results may serve as a first indication why within many IEAs reduction
targets are specified as uniform emission reduction quotas though more
efficient instruments are available. In a next step the stability aspects of the
quota and tax equilibrium have to be analyzed; this is done in Chapters 12
and 14.
NOTES
1. Uniform emission quotas are also part of the amendments signed in London (1990) and
in Copenhagen (1992).
2. Also other IEAs which are not concerned with emission reductions often specify
uniform duties for signatories. Examples include the 1972 London Convention on the
Prevention of Marine Pollution by Dumping from Ships and Aircraft, the 1974 Helsinki
Convention on the Protection of the Marine Environment of the Baltic Sea Area
(HELCOM), and the 1973 Washington Convention on International Trade in
Endangered Species of Wild Fauna and Flora. See also Barrett (1992d) and Hoel (1991,
1992a) on ‘uniform regulations’.
3. The focal point as a device to select a particular equilibrium among a larger set has
already been discussed in the context of the assurance game. See Section 3.4.
4. A uniform tax does not imply that taxes are increased either by the same absolute () or
by a relative () amount compared to the Nash equilibrium. Let taxes which bring about
a Nash equilibrium emission allocation in countries 1 and 2 be denoted by t1N and t2N
(i (eiN ) tiN), then, t1N t2N and t1N(1 )t2N(1) will generally be true.
5. Since we do not use these concepts in the following we do not give a detailed description.
Roughly speaking, the Nash bargaining solution allocates the gains from cooperation in
proportion to the threat points of players (Nash 1950b). A frequently made assumption
is that the threat points are the Nash equilibria or the minimax values of the one-shot
game. See, for example, Binmore et al. (1986) and Owen (1982). The Shapley value dis-
tributes the gains from cooperation according to the average marginal contribution of a
country to the overall welfare gain (see, for example, Moulin 1988; Schotter and
Schwödiauer 1980). All concepts assume that the cooperating players maximize the
aggregate welfare of the participants.
6. Bargaining models in the context of international pollution control are treated, for
instance, in Barrett (1992a, d); Compte and Jehiel (1997); Eyckmans (1997); Kuhl (1987);
Myerson (1997); Richer and Stranlund (1997); Rotillon and Tazdait (1996). General ref-
erences on bargaining models are Binmore and Dasgupta (1987); Canning (1989);
Osborne and Rubinstein (1990); Roth (1979, 1985); Rubinstein (1987); and Stähler
(1998b).
7. To render the following analysis interesting, we assume an interior Nash equilibrium
(which is unique by A1) to represent the status quo so that there is an incentive for both
countries to reduce emissions within an agreement. Since eiN eimax
i I some environ-
mental policy must already be in place at the national level in the status quo.
8. This procedure seems to be suggestive since it turns out in Chapter 12 that, though the
equilibrium set in a supergame framework may be substantially reduced by integrating
Uniform emission reduction quota and emission tax 193
some ‘real world restrictions’ into the analysis, nevertheless in a global emission game a
unique equilibrium can hardly be expected.
9. Within the Convention on Long-range Transboundary Air Pollution for the agreements
on sulfur dioxide (SO2), (Helsinki 1985, Oslo 1994), nitrogen oxide (NOx) (Sofia 1988)
and VOC (volatile organic compounds) (Geneva 1991) emissions are gathered within
EMEP (Environmental Monitoring and Evaluation Program) which was signed in 1984.
In Article 4.1 the UN Framework Convention on Climate Change also calls on all parties
to publish emission data which are accessible on the internet (http://www.unfccc.de). Of
course, in some cases the reporting is incomplete (Benedick and Pronove 1992).
10. In a strict sense, we just assume that information on neighbors’ damage costs are not
used when deriving a proposal. We do not model incomplete information in a game theo-
retical sense, that is, putting priors on countries’ expectations about their neighbors’
damage costs. Since it will turn out that introducing the more restrictive assumption of
‘complete information’ may lead parties to put forward biased proposals, this possibil-
ity is also considered in Section 11.7.
11. This problem of insufficient information about damage costs has led in the national
context to Baumol and Oates’s standard-price approach as a pragmatic modification of
a first-best Pigouvian tax (Baumol and Oates 1971). The problems associated with the
measurement of damage are discussed in Endres and Holm-Müller (1998); Garrod and
Willis (1999); Johansson (1990); Willis et al. (1999).
12. The possibility of eiI eimax is discussed in Endres and Finus (1999).
13. For an analogous assumption, see Marino (1988).
14. Thus, strictly speaking, in (11.5) and (11.6) the agreement procedure kA min[ki, kj ]
should have already been incorporated for a complete description of the maximization
problems. However, as Proposition 11.1 demonstrates, this would not change equilib-
rium behavior.
15. This result is confirmed by using other initial emission levels as the starting point and
other functional forms of the net benefit functions. In particular, it also holds for the
functions assumed in Chapter 14 on coalition formation.
16. A detailed derivation is available upon request.
17. It is also possible that both countries receive more than in a strategic equilibrium. For
instance, if countries are symmetric and settle for a socially optimal tax, a strategic pro-
posal would be to the disadvantage of both.
12. Infinite dynamic games with
continuous strategy space
12.1 INTRODUCTION
194
Infinite dynamic games with continuous strategy space 195
since any payoff tuple has only to be measured against the benchmark
minimax payoff. This allows us to draw rather general conclusions with
respect to the stability of the quota and tax equilibrium, the auction equi-
librium and the social optimum:
Proposition 12.1
Let the stage game of the global emission game be described by the
general payoff functions (9.1), assumptions A1 hold, and the strategy
space be given by Ei [0, eimax], eiN eimax iI, then the quota, tax and
auction equilibrium (as derived in Chapters 10 and 11) can be sustained
by subgame-perfect strategies provided discount factors of all players are
sufficiently close to 1. Under the tax regime this holds regardless of
whether adjustment is anticipated.
Proof: Follows from Propositions 10.4, 11.5, 11.7 and applying Theorem
5.4.
Preliminaries
As we know from Chapter 7, there is no simple benchmark to check for a
WRPE or an SRPE. Only for particular payoff functions can the renegotiation-
proof payoff space be determined. Therefore, the following analysis is based on
the familiar payoff functions in Chapter 11, equation (11.8). Weakly renegoti-
ation-proof conditions C1 and C2 in the payoff space are derived, proceeding
as in Chapter 7. Additionally, these WRPE conditions are illustrated in the
two-dimensional strategy space. It will be investigated whether the WRPE con-
ditions are binding compared to the SPE conditions (as in the extended PD
game V, Chapter 7) or not (as in the ordinary PD game, Chapter 7).
Assume for a start that country 1 is the potential defector; then, using
payoff functions (11.8), these conditions can be written as:
1 c
*1 b(de1(e12) e1(e12)2) (e1(e12)
e12)2 (12.3)
2 2
1 c
*2 b(de12 e122) (e11
e12)2 (12.4)
2 2
where e1(e12) is country 1’s best deviation according to its reaction function.
For (11.8) e1(e12)(bd ce12)/(b
c) which may be substituted into (12.3).
Then:
b(bd 2 ce122 2cde12
*1 . (12.5)
2(b
c)
Since
R
2 /
e1 0, we set e1 0 in (12.3) to have:
1 1
1 c
*2 b de12 e122 e212. (12.6)
2 2
That is, (12.4) is ‘most easily satisfied’ for e11 0 which allows us to deter-
mine the outer boundaries of the WRPE payoff space.3 This assumption
basically implies that country 1 accepts the harshest possible punishment
in the game. Modification of this assumption will be treated in Sub-
section 12.2.4.
Solving (12.4) for e12 gives:
db d 2b 22 2*2(b
c) db
d 2b 22 2*2(b
c)
e12 b
c
,
b
c
.
(12.7)
First note that the root is always positive if *2 U
2 which obviously holds
by the definition of the upper bound of the payoff space.4 Second, note that
e12 should be chosen as big as possible because
1C/
e12 0. However, e12 is
restricted by the definition of the strategy space to e12 [0, d ] (see A3 in
(VIII.4) in Appendix VIII.2). Whether the RHS expression in (12.7) or d is
larger depends on the specific parameter values. Thus, we take e12 to be:
db
d 2b 22 2*2(b c)
e12 min d,
b
c
. (12.8)
Assume, first, that the second term in (12.8) is larger than the first term and
hence e12 d. Substituting this into (12.5) gives the first condition for stabil-
ity, L1:
Infinite dynamic games with continuous strategy space 197
bd 2(b 3c)
L1:*1 ⇔*1 M
1 . (12.9)
2(b
c)
Of course, L1 is nothing other than requiring payoffs to be above the
minimax payoff which is the condition for subgame-perfection.5
Now consider the second possibility, that the second term in (12.8) is
smaller than d. Inserting the second expression in (12.8) into (12.5) gives:
c(bd
A)2 2cd(bd
A)
b
C1:*1 bd 2
2(b
c) (b
c)2 (b
c)
c(bd
B)2 2cd(bd
B)
b
C2:*2 bd 2
2(b
c) (b
c)2 (b
c)
π2
70 πW1
60 A
C1
50
40
πS
30 Pareto frontier
20
πN
L1
10 B
0
–10 πW2
–20 C2
–30 L2
πM
–40
–30 –20 –10 0 10 20 30 40 50 60 70 π1
Note: Payoff functions (11.8) apply; b 2, c 1, d 10 and 1 are assumed.
is a true subset of the set of SPE payoffs, that is, WRPE SPE. However,
it is easily checked that for other parameter values this must not be true.
That is, the C1 curve may run above and the C2 curve below the boundar-
ies defined by L1, L2 and the Pareto frontier. For instance, increasing the
parameter value of b from b 2 to, say, b 5 and keeping constant the other
parameter values assumed in Figure 12.1 would be such a case.
All payoff tuples lying on the Pareto frontier between W 1 and W 2 in
Figure 12.1 are SRPE by Theorem 7.8. Depending on the parameter values,
the set of SRPE payoffs may or may not be a true subset of the Pareto fron-
tier, that is, SRPE P(IR) or SRPE P(IR).
(a) π2 (b) π2
C2
L1 L
C1 C1
πB πB
πM L2 π1 πM L2 π1
own minimax payoff (jM(i ) jM( j)), whereas in a type B game s/he may
receive more (jM(i ) , jM( j )). Thus, in a game of type A punishment is
always costly, but in a game of type B this depends on the specific circum-
stances (in the emission game on the particular parameter values; see also
the discussion of Folk Theorem III, Section 5.2).
Proposition 12.1
In an infinitely repeated two-player game of type A, that is, in a game in
which the renegotiation-proof conditions C1 and C2 are more restrictive
than the minimax conditions with respect to the entire individually ratio-
nal payoff space, minimaxing a fellow player implies that the punisher
will receive less or at best his/her own minimax payoff.
Proof: Assume a type A game and a payoff tuple B in the payoff space
for which B1 M 1 and 2 2 is true (see Figure 12.2a). In order for
B M
C1
8
6 eW1
A IM
2
eN
4 eS
eM(2)
2 B C2
Pareto frontier
eW 2
0
2 4 6 8 10 e1
Note: Payoff functions (11.8) apply; b 2, c 1, d 10 and 1 are assumed.
Figure 12.3 Subgame-perfect, weakly and strongly renegotiation-proof
emission space
Infinite dynamic games with continuous strategy space 201
Stability of the social optimum, the quota, tax and auction equilibrium
The example displayed in Figures 12.1 and 12.3 assumes symmetric param-
eter values (that is, 1), implying that the quota and the tax bargain-
ing equilibrium as well as the auction equilibrium coincide with the social
optimum, that is, S Q T A and eS eQ eT eA. Hence, trivially,
for this symmetric case all payoff tuples are weakly and strongly renegotia-
tion-proof. For the more general case, considering also the possibility of
asymmetric payoff, we find:
Proposition 12.2
Let the stage game of the infinitely repeated emission game be described
by payoff functions (11.8), let assumptions A3 in (VIII.4) hold and let
i →1 iI, then the quota and tax equilibrium are weakly renegotia-
tion-proof. Under the quota regime and under the tax regime
1/ (implying equal proposals) are necessary and sufficient conditions
that the bargaining equilibria are strongly renegotiation-proof. The
auction equilibrium, where the exchange rate is defined in terms of emis-
sion reductions from the stage game Nash equilibrium, that is, eiI eiN
i I, is always a strongly renegotiation-proof equilibrium.
Proof: The first part of Proposition 12.1 is proved by computing net ben-
efits in the quota and tax equilibrium for all possible parameter constel-
lations using emissions from Table VIII.1, Appendix VIII.2. For each
parameter range it has to be checked whether C1, C2, L1 and L2 derived
above hold, using assumptions A3. Since the proof requires a substantial
amount of computation, it is not reproduced here; however, it may be
obtained upon request.
The second part of Proposition 12.2 simply follows from the fact that
the quota and tax equilibria must lie on the Pareto frontier to qualify as
potential SRPE candidates. This is the case for and 1/
respectively (see Corollary 11.1). Since the above-mentioned proof holds
for all parameter constellations, it therefore also covers and
1/.
The third part of the statement is an implication of Proposition
12.4 below, which claims that the auction equilibrium is a strongly
202 Game theory and international environmental cooperation
indicates the punished player.6 Then, inserting e12 into the best reply of
country 1, e1(e12), allows us to derive 1C. For the functions in (11.8) we have:7
b2d 22 b2d 2(b22
2bc 3bc2 2c2
c22)
R(1) , 1C .
2
2(b
c) 2(b
c)(b
c)2
(12.13)
Infinite dynamic games with continuous strategy space 203
where the respective payoffs are those given in (12.13) and (12.14) and
where condition S3 ensures that the payoff tuple in the cooperative phase is
Pareto-efficient. That is, all payoff tuples which simultaneously satisfy con-
ditions S1 and S2 are SSPE. Of course, in the present game the upper bound
is not a binding constraint, so one may write only S1: 1C *1 and
S2: 2C *2 for short.
In Figures 12.1 and 12.3 all points between points A and B satisfy the
SSPE conditions. From the graphs it is evident that in the global emission
game the set of SSPE is a true subset of the SRPE. Recall that for the exam-
ples in Chapter 7 this was different, namely the set of SSPE payoff tuples
coincided with the SRPE payoff set. This finding for the emission game
holds not only for the specific parameter values underlying Figures 12.1
and 12.3 but is in general true for the functions in (11.8).
Proposition 12.3
In the infinitely repeated global emission game where the stage game is
defined by the functions in (11.8), and conditions A3 in (VIII.4),
Appendix VIII.2, hold, SRPE SSPE.
Proof: The proof is obvious. Comparing the punishment emission levels,
e12, in (12.8) with e12 as given in note 7 we find that the latter is smaller than
the former. Recalling that
1C/
e12 0 and that 1C is country 1’s lower
bound for stable equilibria, then the claim is proved from country 1’s per-
spective. A similar procedure would show that the same holds from
country 2’s perspective.
Proposition 12.4
Let the stage game payoff function of the infinitely repeated emission
game be given by (11.8), assumptions A3 in (VIII.4) hold and let i →1
204 Game theory and international environmental cooperation
i I, then equal proposals are a necessary and sufficient condition that
the bargaining equilibria under the quota and tax regimes are strongly
subgame-perfect equilibria. The auction equilibrium, where the
exchange rate is based on emission reductions from the stage game Nash
equilibrium, is always a strongly subgame-perfect equilibrium.
Proof: The first part of the statement is proved by computing net bene-
fits under the quota and tax regimes, using emission levels of
Table VIII.1, Appendix VIII.2. Then it is shown that iU iQ iC and
iU Ti iC iI{1, 2} holds for (quota regime) and 1/
(tax regime) which are the conditions for Pareto efficiency (and which
imply equal proposals by both countries; see corollary 11.1) by using A3
in (VIII.4), Appendix VIII.2, and where iC and iU are those bounds
given in (12.13) and (12.14).
The second part of the statement is proved by noting that for the
payoff functions in (11.8) 1C 1N and 2C 2N hold for all parameter
values and that the auction equilibrium constitutes a Pareto optimum for
which iA iN iI is true provided eI eN holds by Proposition 10.4.
Proposition 12.5
Restricting the strategy space of the stage game defined by the payoff
functions in (11.8) and assumptions A3 in (VIII.4) to Ei [0, eiN] iI,
the quota equilibrium can always be sustained by weakly renegotiation-
proof strategies in an infinitely repeated game provided i →1 iI,
whereas the tax equilibrium and the social optimum may fail to be stable.
If post-adjustment takes place after countries have agreed on a tax rate
according to the LCD decision rule, the tax equilibrium is not stable,
regardless of whether a country makes a strategic proposal.
Proof: The first part of the proposition with respect to the quota regime
is proved in the spirit of the proof of Proposition 12.2. The second part
of the proposition follows from Propositions 9.2, 11.5 and 11.7 and the
necessary condition i* iN as noted above. 9
(a) e2 (b) e2
10 I M
1
10 IM
1
r1 r1
8 B1 A1 8 B1 A1
6 A2 IM 6 A2 IM
eN 2 eN 2
e B
4 eB r2 4 B r2
2 eS
e
S
2 B2 2
B2 B2
B1 B1
0 0
2 4 6 8 10 e1 2 4 6 8 10 e1
Note: Payoff functions (11.8) apply; b 2, c 1, d 10, 1 are assumed. The left
graph assumes 1 and 0.2, the right graph 0.8 and 0.05.
Figure 12.4 Weakly renegotiation-proof emission space for reciprocal and
restricted punishments
Infinite dynamic games with continuous strategy space 207
Rule 2 The second restriction reflects the assumption that the punishment
cannot exceed a certain level of the cooperative emission level where 0
is a parameter. For instance, 0.1 implies that emissions during the pun-
ishment cannot be increased by more than 10 percent.
Results Apart from the same (and symmetric) parameters of the previous
figures (b 2, c 1, d10, 1), 1 and 0.2 have been assumed
in Figure 12.4(a) and 0.8 and 0.05 in Figure 12.4(b). Condition (2)
is reflected in Figure 12.4 by curve A1 for country 1 and by curve A2 for
country 2 where, as pointed out above, A1 is the envelope of inequalities
(12.1) and (12.2) if country 1 is the potential defector and A2 if this is
country 2. Hence, emission tuples in the interior of A1 and A2 satisfy con-
dition (2) in (12.16).
From the figures it is evident that only emissions tuples which are in the
208 Game theory and international environmental cooperation
e2
eM(1) IM
1
10
C1
8 C (1)
1
C (2)
1
6
A IM
2
eN
A(1)
4 S C (1) C (2)
A(2) e C2 2 2 eM(2)
B(2) B(1)
2 B
Pareto frontier
0
2 4 6 8 10 e1
Note: Payoff functions (11.8) apply; b2, c 1, d 10 and 1 are assumed.
From Chapter 7 it is known that for discount factors smaller than 1 three
inequalities must be satisfied simultaneously, which are reproduced in the
context of the global emission game:
iI where:
P P
iP (1 ti i )iR(eii, eji )
ti i i*(ei, ej ). (12.20)
tiP*
log
*
i
R
i
approx.: tiP*
log .
*
i
R
i
(12.22)
log i log i
iD i*
(1 i)iD
iiC i* ⇔ CiB: i if i* iC. (12.24)
iD iC
Hence, for a given ei (eii, eji ):
follows. In a second step we vary ei (eii, eji ) to minimize max {CiA, CiB}
subject to the condition that the punisher should not be worse off than in
the cooperative phase ((12.19)). For a given equilibrium emission tuple
e(ei, ej ), the minimum discount factor requirement of a country i is there-
fore given by:
Technically, for a given eii the largest eij is chosen which still satisfies (12.19)
(recall eji d ); this is substituted into CiA and CiB and the maximum of both
is minimized. From Sub-section 12.2.2 we know that eji is given by (12.8)
where i 1 and j2. As argued there, such an eji satisfying (12.19) (which is
equivalent to (12.2)) exists for any payoff of country j in the domain of
rational payoffs, that is, j* [jM, jU].
Observing (12.23) and (12.24), it is clear that, for i close to 1, CiA CiB.
Since we chose eii 0 (harshest possible punishment) in Sub-section 12.2.2,
this implies that we minimized CiA, and therefore automatically CiB as well.
Hence, C1 and C2 in Figures 12.1 and 12.3 are CiA and CjA (and/or CiB and
CjB) drawn for i 1. Since we defined i to be strictly less than 1, all emis-
sion tuples lying on the boundary of C1 or C2 do not belong to the renego-
tiation-proof emission space. Hence, also all boundary emission or payoff
tuples which have been mentioned in the context of SRPE (for example,
W1 and W2 in Figure 12.1 and eW1 and eW2 in Figure 12.3) are excluded.
Now, for a discount factor smaller than 1, eii 0 is not always optimal
when solving (12.26). Instead, it might be conducive to the stability of an
agreement (lower discount factor requirement) to be not so harsh on
country i in the punishment phase (eii 0), but to choose a longer punish-
ment duration. For instance, if i is small and iC iR, then tiP 1 and
eii 0 could already be too long to be acceptable to the defector. In other
words, there is an optimal mix of punishment duration and punishment
level.
212 Game theory and international environmental cooperation
δ min,S
2 0.05 δ min,Q
1
δ min,T
2
0 0
23 5 10 20 50 γ 23 5 10 20 50 γ
Note: Payoff functions (11.8) and 1, 5 are assumed. The -axis is logarithmic scaled.
where iST stands for Stackelberg payoff. The details of the derivation of
CiC are laid out in Appendix IX.1. As for i → 1, one must ensure that for
Infinite dynamic games with continuous strategy space 215
C B2
8 8
C D1 A C D1 A
6 I ST
1
= C C1 6 I ST
1
= C C1
I N1
I N1
4 I N2 B 4 B
C B2 I N2
eS eS
C D2 C B2
2 C D2 2
I ST = C C2
I ST
2
= C C2 IM
2
C B2 2
IM
2
0 0
2 4 6 8 10 e1 C B1 2 4 6 8 10 e1
Note: Payoff functions (11.8) apply; b2, c 1, d 10 and 1 are assumed.
the punishment eji d holds.16 Hence, from (12.18) a third condition has to
be derived:
iD i*
CiD:i (12.29)
iD iM
which takes into account eji d.17
Thus for tiP , CiB, CiC and CiD iI must be simultaneously satisfied.
These requirements are shown in Figure 12.7 for two examples. Note that
the assumption of symmetric discount factors in the examples is chosen
just for convenience; however, it is not important for the following argu-
ments. Let us first concentrate on Figure 12.7(a) where there are ten curves.
The minimax and Nash indifference curves of countries 1 and 2 are indi-
cated by IM M N N
1 , I2 , I1 and I2 and have already been discussed in Chapter 9.
ST ST
I1 and I1 are Stackelberg indifference curves, tracing out all emission
tuples that give a country the same payoff as if it were a Stackelberg leader
(see Section 10.2). These two indifference curves represent conditions CC 1
and CC 2 as given in (12.28).
Thus, all emission tuples lying to the right of the CB1 curve, above the CC
1
curve and below the C1D curve are WRPE from country 1’s perspective. By
the same token, all emission tuples lying above the C2B curve, below the C1C
curve and to the left of the C2D curve are WRPE from country 2’s perspec-
tive. (The CiB curve stops at points A and B because there the CiC conditions
216 Game theory and international environmental cooperation
become binding.). Taken together, the gray area represents the WRPE
emissions space for tiP .
It is evident that only emission tuples which generate relatively low
payoffs to both countries are WRPE. On the one hand, for tiP the
problem is that if the punishment emission level of the defector, eii, is too
small, his/her payoff is so low that s/he is not prepared to accept the pun-
ishment, and in particular such a long punishment. On the other hand, for
high levels of eii it might no longer be possible to satisfy (12.19).
Turning now to Figure 12.7(b), it is obvious the more that actors dis-
count time, for example, i 0.5 instead of i 0.9, the smaller becomes the
WRPE emission space. This is a similar result as stated in Corollary 7.1,
where, we may recall, for i → 0 the WRPE emission space shrinks to the
NE of the stage game. Put differently, the more that actors discount time,
the more difficult it becomes to stabilize agreements, deviating substantially
from the stage game NE emission tuple.
From Figures 12.7(a) and (b) it is evident that the equilibrium emission
tuple of the social optimum and the two bargaining regimes (which all co-
incide due to the assumption of 1) are not renegotiation-proof for
tiP (eS lies outside the gray area). It can be shown that this finding carries
over as a general result:
Proposition 12.6
Let the stage game of the infinitely repeated emission game be described
by (11.8) and conditions A3 in (VIII.4) hold, then irrespective of the
parameter values, the punishment level and the discount factors of
players, the social optimum, the quota and the tax equilibrium cannot be
backed by renegotiation-proof trigger strategies.
Proof: Compute net benefits for the emission levels as given in
Table VIII.1 in Appendix VIII.2 and show that for any parameter con-
stellation at least for one country ik iST, k{S, Q, T} holds and hence
condition CiC (see (12.28)) is violated. 18
The result nicely stresses the difference between SPE and WRPE. For
instance, the quota equilibrium can always be backed as an SPE by using a
trigger strategy which reverts to the stage game NE in case of defection,
since iQ iN iI. For the tax equilibrium and the social optimum this
may also be possible; however, this depends on the specific parameter
values (since iS iN and iT iN may hold). However, Proposition 12.6
clearly denies this possibility for renegotiation-proof trigger strategies.
Infinite dynamic games with continuous strategy space 217
NOTES
1. Parts of this chapter draw on Endres and Finus (1998a) and Finus and Rundshagen
(1998b).
2. Applications of the WRPE and SRPE concept, assuming i → 1 i I, may be found in
the context of a global emission game with N players in Barrett (1994a, b) and in the
context of a Cournot duopoly in Driffill and Schultz (1995) and Schultz (1994).
3. In other words, we search for the lowest WRPE payoff which is sustainable in the emis-
sion game.
d 2b22
4. 2U which follows from max 2(e1 0, e2) as explained in Section 9.5.
2(b
c) e2
5. Note that condition (12.6) is satisfied for e12 d. This follows from
22R/
e122 0 (2R is
the RHS term in (12.6)) and the fact that e12 d is bigger than the LHS expression and
smaller than the RHS expression in (12.7) by assumption.
6. This amounts to maximizing 1(e11, e12)
(1 )2(e11, e12) for 0. See Chapter 9, note
26.
7. The associated punishment emissions are:
bd bd
e11 0, e12 , e2 and e12 0.
b
c 1 b
c
218 Game theory and international environmental cooperation
8. Barrett (1994a) determines the parameter ranges for his models which allows one to sta-
bilize a socially optimal IEA. That is, he computes iU, iC and iS for symmetric games
and then determines the parameter values for which iU iS iC i I is possible.
9. Though in the auction equilibrium iA iN i I holds if eI eN, this does not suffice
to establish stability in the sense of WRPE since i* iN is only a necessary WRPE con-
dition but not a sufficient one. Additionally, one has to test whether conditions C1 and
C2 can be satisfied subject to the restriction ei [0, eiN] i I. For this payoffs iA must
be computed which, as pointed out above, is only possible for specific parameter values
but not at a general level. Of course, iA iN i I is sufficient to qualify the auction
equilibrium as an SPE.
10. For a similar interpretation in the context of a global emission game using the stability
concept of the core, see Sub-section 13.3.4.
11. There is little scope for legal punishments in common international law. There are only
two ways in which the injured party can legally punish the breach of a bilateral treaty.
On the one hand, the injured party can suspend or terminate the violated treaty provi-
sions, however, only to the extent to which they are violated by the other party (Simma
1970, pp. 20ff.). On the other hand, the injured party can take reprisal measures against
the violator of the treaty. These measures may consist of reciprocal violations of the vio-
lated treaty provisions. However, reprisals must also be in proportion to the delict
according to the generally recognized principle of proportionality in common interna-
tional law (Kelsen and Tucker 1967, pp. 20ff.). It is worth mentioning that leading schol-
ars of international law think that neither of the described reactions to a breach of a
treaty constitutes legal options for parties of a multilateral treaty, that is N 2 (Heister
1997, pp. 91, 135). Only if punishment rules are an explicit part of a multilateral envi-
ronmental treaty are punishments covered by international law (Sand 1992, pp. 14ff.).
12. Of course, one could argue that ei (ej ) ei 0 should not be punished at all because a
country would abate more than required. However, agreeing on a high emission level ei
(ei eiN ) implies also a sub-optimal choice from both countries’ point of view and
should therefore not be regarded as stable per se.
13. Drawing the curves C1(1), C2(1), C1(2) and C2(2) in the payoff space would reveal exactly the
same pattern. That is, these curves lie more inward than the curves C1 and C2 in
Figure 12.1.
14. All parameters have been chosen to ensure that assumptions A3 in (VIII.4), Appendix
VIII.2 hold. This explains, for instance, why the smallest value in Figure 12.6 is 1.6.
15. The interested reader may consult Finus and Rundshagen (1998b), where the discussion
is continued based on the present setting.
16. Recall, for i →1 i I the restriction eji d delivered conditions L1 and L2 in the payoff
space and the minimax indifference curves in the emission space.
17. Thus, CiD is basically CiB if eji d has been substituted into iC(ei (eji ), eji ).
18. A detailed proof is available upon request. With respect to the auction equilibrium, a
statement is not possible for the reasons given in note 9.
13. Coalition models: a first approach
13.1 INTRODUCTION
219
220 Game theory and international environmental cooperation
but not the subsequent moves (chain reactions) which may be triggered.
Additionally, these models consider only one coalition (signatories), the
members of which coordinate their strategies and assume that all other
countries (non-signatories) behave as singletons (see Carraro and
Siniscalco 1998). The concept of the core (Section 13.3) is defined either in
a static world or in a dynamic setting. However, in the latter case only
Markov (history-independent) strategies are considered. Moreover, the
core abstracts from the individual strategies of players and reduces deci-
sions to the payoff dimension (characteristic function). The supergame
coalition model (Chapter 14), although it takes up some of the concerns
raised with respect to the models presented in this chapter, also restricts the
coalition formation process to signatories and non-signatories. Moreover,
as with the previous models, the choice of abatement targets and transfer
schemes within the coalition is basically exogenous to the model, though a
large number of alternative abatement targets and transfer schemes are
considered and their selection is thoroughly justified.
Due to these conceptual shortcomings, we shall investigate in Chapter 15
whether new developments in the game theoretical literature on coalition
formation are suitable to be integrated into the analysis of the genesis of
IEAs. From the discussion it will be apparent that there is still much work
to be done in order to construct models which are capable of explaining the
entire coalition formation process endogenously.
To the best of our knowledge, all theoretical coalition models on the for-
mation of IEAs assume an externality of the pure public bad type. This is
because for transboundary environmental problems, different spillovers
cause an additional asymmetry between countries, the strategic implica-
tions of which would have to be considered.1 Hence in the tradition of pre-
vious chapters, we again restrict attention to global environmental
problems.
13.2.1 Preliminaries
The conjectural variation coalition models first made their appearance in the
analysis of oligopolies (see, for example, d’Aspremont and Gabszewicz
1986; and d’Aspremont et al. 1983). In the context of the formation of
IEAs, their origin can be traced back to Barrett (1991c, 1992e) and Carraro
and Siniscalco (1991). Typical representatives of these models are Barrett
(1994b); Bauer (1992); Carraro and Siniscalco (1991); and Hoel (1992a).
These models are typically set up as stage games comprising two or three
Coalition models I 221
In the third stage the allocation of the welfare gains among the coalition
members is decided. If countries are symmetric, this stage seems superflu-
ous since, intuitively, one would expect that signatories would agree on a
symmetric abatement allocation which implies equal payoffs and therefore
that no transfer scheme is needed.2 Probably, any morally motivated
welfare allocation rule or a focus point type of argument would not alter
this result. Hence, it is also not surprising that all the well-known bargain-
ing concepts of cooperative game theory would not call for a reallocation
of payoffs via transfers.
However, if countries are asymmetric the choice of the transfer scheme
is less obvious. In particular, stability of a coalition may depend on the
sharing rule of the gains of the coalition. Typical assumptions are: (a) no
transfers; (b) Nash bargaining solution; and (c) Shapley value, where the
latter two are welfare allocation rules of cooperative game theory.3
Assumption (a), again, implies that the third stage is redundant and each
signatory receives those payoffs which directly follow from the emission and
participation strategies. That is, iJ( p*, e*) where p* ( j1*, ..., j N*
* , nj *
N*1, ...,
nj N ), e (e , e ) and where the asterisk denotes an equilibrium strategy
* * J* NJ*
combination. Only for assumptions (b) and (c) is the third stage important
where the result of the first two stages may be altered. In this case iJ( p*, e*,
t*) where t* denotes a matrix of transfers.
In the literature there are basically two assumptions regarding the
sequence of moves in the first two stages: (a) players choose their strategies
in both stages simultaneously (Carraro and Siniscalco 1991; Bauer 1992;
Hoel 1992a); (b) players choose their participation strategy in stage one
222 Game theory and international environmental cooperation
e
1 c N
Type 1: i bi dei ei2 i k (13.1)
2 N k1
224 Game theory and international environmental cooperation
ci N 2
Type 2: i biei ek (13.2)
2N k1
1 c N 2
Type 3: i bi dei ei2 i ek . (13.3)
2 2N k1
k e k I1
e N S I
I1: ; 0 for (13.1)–(13.3) and 1 0 for (13.1) and (13.3).
e k
S N
(13.4)
Coalition models I 225
With respect to the parameter
b/c the relations hold for payoff functions
of types 1 and 3; for type 2 the index I1 does not contain b and c at all.
I1/
0 implies that abatement is particularly attractive from a global
point of view if environmental damage is high compared to the opportunity
costs of abatement. A similar index can also be defined in terms of payoffs:9
kS N
k I2 I
I2: ; 0 for (13.1)–(13.3) and 2 0 for (13.1) and (13.3).
kS N
(13.5)
1 2 c
max N* b deiJ eiJ (N* ·eiJ ejNJ) (13.6)
e iJ 2 N
which is easily solved since countries have a dominant strategy (towards
non-signatories):10
bdN N *·c
eiJ*(N*) . (13.7)
Nb
Thus, the more signatories, the higher are the abatement duties (the lower
are emissions) of a coalition member. In contrast, non-signatories, which
are supposed to behave as singletons, perform:
bde
1 NJ2 c
j (eiJ ejNJ ejNJ
max NJ e (13.8)
e NJ
j
2 j N
which delivers:
bdN c
ejNJ* . (13.9)
Nb
It is obvious that non-signatories’ emission levels are independent of the
number of signatories and the emission level of signatories. That is, each
non-signatory has a dominant strategy and ejNJ* ejN.
226 Game theory and international environmental cooperation
* I I I I
0, 3 0, 3 0, 4 0, 4 0,
N N
N
b
* N*/N,
, ek* eiJ ejNJ ekN, k* iJ jNJ kN,
c
(13.12)
e*k e kS kS *k
I3: , I : .
e kS 4
kS
That is, the greater the number of countries involved in the externality
problem, the smaller will be the percentage of signatories. Though the
coalition improves upon the status quo, whenever cooperation would be
needed most from a global point of view (large N and small
; see (13.4)
and (13.5)), equilibrium emissions and welfare differ markedly from the
social optimum.
For the second type of payoff function it is easy to show that, due to the
slope of 1, no coalition exists. Any emission reduction by the coalition is
offset by non-signatories. Hence, the profitability of a coalition always fails
to hold and the equilibrium of the game is the NE. Accordingly, I1 I3 and
I2 I4 and the comparative static results in (13.12) carry over to this type of
payoff function as well (except that I3 /
and I4 /
are zero).
For the third type of payoff function the equilibrium number of signato-
ries is determined in the same way as was laid out for the first type of payoff
Coalition models I 227
function. However, since the damage cost functions are quadratic, no dom-
inant strategy exists, which renders computation more cumbersome. It
turns out that the equilibrium coalition size is N* 2 and that the profit-
ability constraint C1 may fail to hold, depending on the parameter values.
The relations in (13.12) also hold for this type of payoff function.
Carraro and Siniscalco (1992, 1993) frequently stress that the reason for
the greater coalition size in the case of payoff functions of type 1 compared
to that of types 2 and 3 is orthogonal reaction functions. They point out
that in the latter two cases there is a higher free-rider incentive on the side
of the non-signatories which expand their emissions compared to the Nash
equilibrium. Hence, the abatement efforts of the signatories are completely
(type 2) or partially (type 3) compensated. In contrast, payoff functions of
type 1 imply that non-signatories choose their emissions irrespective of the
abatement efforts of the signatories. In other words, there is no leakage
effect.12 Though this argument is valid to explain the differences in the equi-
librium coalition size for the different payoff functions under the Nash–
Cournot assumption, it cannot be applied to the Stackelberg assumption.
1 2 c NJ
max b dejNJ ejNJ (e e
NJ e J)2 (13.13)
NJ
ej 2 2N j j i
where eNJj denotes emissions of all non-signatories except j, and eiJ are
aggregate emissions of signatories. Using * N*/N, ejNJ (1*)· N·ejNJ
and ejNJ eNJj ejNJ, the FOC deliver the ‘aggregate’ reaction function of
non-signatories:
(bdN ce Ji )(1 *)
ejNJ (13.14)
b c(1 *)
which exhibits the familiar property of a negative slope of less than 1 in
absolute terms. Thus, abatement efforts by signatories will be partially
offset by an expansion of emissions by non-signatories.
Signatories, maximizing the welfare of the coalition, perform:
1 2 c
max * ·N· b deiJ eiJ (* ·N ·eiJ ejNJ (eiJ))2 (13.15)
J
ei 2 2N
where the information about ejNJ (eiJ) in (13.14) is used. After some basic
manipulations, the FOC deliver:
N*d(
2 2
2
* 1 2* *2
N*
N*2)
ẽ iJ .
2 2
2
* 1 2* *2
N*2
(13.16)
Substituting (13.16) into (13.14), one derives non-signatories’ emission
levels:
N(1 *)d(
2 2
*
N*2
* *2)
ẽ jNJ . (13.17)
2 2
2
* 1 2* *2
N*2
A routine check reveals that neither (13.16) nor (13.17) need necessarily be
positive. In other words, there may be boundary solutions for which either
Coalition models I 229
if ẽ jNJ 0
0
eiJ* ẽ iJ if ẽ jNJ [0,
dN] (13.18)
dN if ẽ jNJ
dN
0 if ẽ iJ*
dN
–20
–30
30
20
10
πiJ(n)
n*
0
1 2 3 = N* 4 5 n
–10
–20
–30
equilibrium coalition size is not unique and N* {n*L, n*U}. However, since
N* n*U delivers a higher payoff to signatories and non-signatories than
N* n*L one would expect that in this case N* n*U.16 One can show that
n* n*L N*
n*U so that n* is a lower bound for N*. Whereas it is easy to
compute n*, n*L and n*U N* can only be determined for a particular param-
eter set or for large N, that is, for N→ at the limit.
Irrespective of the number of countries, the following general conclu-
sions can be drawn:
high participation rate in the Montreal Protocol, one cannot expect that
much has been achieved from a global point of view compared to the full
cooperative outcome (social optimum).
As mentioned above, in the case of small N the relations in (13.21) can
no longer be established analytically. However, a large set of simulations
suggest that (13.21) also holds in the case of small N.
In this section we state the results obtained so far in the form of proposi-
tions and briefly comment on them.
Proposition 13.1
Under the Nash–Cournot and the Stackelberg assumptions the equilib-
rium number of signatories is N* 2 or N* 3 for payoff functions of
type 1 ((13.1)). No stable coalition exists for type 2 ((13.2)) under the
Nash–Cournot assumption and the grand coalition forms under the
Stackelberg assumption, that is, N* N. For the payoff function of type 3
((13.3)) the equilibrium number of signatories is N* 2 under the
Nash–Cournot assumption and N* [2, N ] under the Stackelberg
assumption.
Proposition 13.2
Let * N*/N be the degree of participation in an IEA,
b/c the
benefit–cost ratio from emissions and I3:(ei* eiS)/eiS and
I4:(iS i*)/iS indices measuring the degree of externality in the
coalition equilibrium where ei* denotes aggregate emissions and i*
aggregate payoffs of signatories and non-signatories in equilibrium, then
for the payoff functions (13.1)–(13.3), */N 0, I3/N 0, I4/N 0,
I3/
0 and I4/
0 hold under the Nash–Cournot assumption,
except I3/
0 and I4/
0 for payoff function (13.2). For payoff
functions (13.1) and (13.3) these relations also hold under the
Stackelberg assumption. For payoff function (13.2) the degree of exter-
nality is zero.
Proposition 13.3
For the payoff function of type 3 ((13.3)), assuming Stackelberg leader-
ship of signatories, the percentage of participation in the stable coalition,
*, is a declining function of the benefit–cost ratio from emissions,
,
that is, */
0.
Coalition models I 233
Propositions 13.1 and 13.3 suggest that assuming quadratic benefit and
damage cost functions (satisfying assumptions A1 in (9.2)) and signatories
behaving as Stackelberg leaders allows us to model a broad spectrum of
IEAs of different coalition sizes.19 Another advantage of Barrett’s model is
that it allows us to relate the equilibrium number of signatories to the
benefit–cost ratio from emissions and the number of countries suffering
from the externality problem. Hence, the explanatory power of this model
version is higher than that of the other versions discussed above. Of course,
the assumption of Stackelberg leadership on the side of the signatories is a
rather special assumption but, as argued above, there are arguments which
can be put forward in defense.
To this end it is also evident that Carraro and Siniscalco’s claim that
orthogonal reaction functions would be conducive to cooperation applies
only to their model version (Nash–Cournot assumption). As demon-
strated, negatively sloped reaction functions are in fact a prerequisite for
Stackelberg leadership on the side of the signatories to be used strategically
so that a higher participation in an IEA than N* 3 may be obtained. In
fact, for a payoff function of type 2 with slope of 1, N* N. This result
also contrasts with that obtained in Chapters 10 and 11. In the coalition
context, strategic Stackelberg leadership can be conducive to cooperation,
increasing global welfare and reducing aggregate emissions compared to
the NE. This is so since signatories as Stackelberg leaders do not reduce
emissions in a ‘naive’ way but are aware of the leakage effects caused by
non-signatories.
It is interesting that for all model versions – except for the Stackelberg
assumption and payoff functions of type 2 – the same qualitative results
could be derived, namely that whenever cooperation is needed most from a
global point of view a coalition achieves only little (Proposition 13.2). In
the case of Barrett’s model, this part of the analysis had to rely partially on
simulations (if N is small), though for the other model versions this was
shown to hold in general. It turns out that this result carries over to the
dynamic coalition model in Chapter 14. Roughly speaking, this is due to
the fact that whenever the difference between NE and social optimum is
large, the free-rider incentive is also very strong and hence a stable coalition
can only achieve a moderate abatement target.
at best the same size of the coalition. For instance, recall that N* 2 or
N* 3 for the payoff function of type 1. Suppose that N* 3 and that one
country, say 1, receives more than an equal share of the gains from co-
operation, which of course implies that another country, say 2, must receive
less of the pie. Since, for N* 3, a signatory is indifferent between remain-
ing in the coalition or leaving it, country 2 would leave the coalition if such
a reallocation took place.
It also seems obvious that if countries are symmetric they choose that
abatement level which maximizes joint welfare. Any other target would reduce
welfare to a signatory and cannot secure a higher participation in an IEA.
In the case of asymmetric countries the implications of the welfare allo-
cation rule and the choice of the abatement target(s) on the equilibrium
coalition is less obvious. That is, both issues, which may be summarized as
the design of an IEA, affect the participation rate in an IEA and the com-
position of its members and hence, ultimately, affect the success of an
agreement. Of course, if unlimited transfers are possible, the highest par-
ticipation and global welfare is obtained if signatories maximize the coali-
tion’s welfare and hence only the relation between the welfare allocation
rule and the success of an agreement must be investigated. A similar sim-
plifying assumption (for which a justification was given in Chapter 8) is to
rule out any transfers, so that only the relation between abatement target(s)
and the success of an IEA has to be analyzed.
So far the question of how the design of an agreement affects its success
has received only limited attention. On the one hand, the importance of this
issue has not been recognized in the literature to date; on the other hand, a
systematic analysis of this issue is difficult. The main obstacle to obtaining
more general results is, apart from the numerous possible designs of IEAs
which one would have to investigate, that results crucially depend on the
type and degree of asymmetry between countries.
In the following we briefly summarize the findings of four papers. It is
important to stress that in these papers transfers are restricted to taking
place within the group of signatories. The possibility that signatories ‘bribe’
non-signatories via transfers to participate in their agreement or the pos-
sibility of transfers from non-signatories to signatories to induce higher
abatement efforts of signatories (‘abatement leasing’) are neglected. This
issue will be taken up in Sub-section 13.2.6.
reductions, the second term (quadratic) costs from abatement. Thus, payoff
functions (13.22) are of type 1, that is, reaction functions are orthogonal.
Consequently, a distinction between the Nash–Cournot and the
Stackelberg assumptions is not necessary.
The idea behind the specification in (13.22) is that the index i represents
differences in the size of countries. Accordingly, large countries benefit pro-
portionally more from emission reductions since a larger area is affected by
environmental damages. There are economies of scale from abatement so
that it is less costly for larger countries to reduce emissions than for smaller
countries. Taken together, the function in (13.22) implies that a country that
is double the size of a smaller country can reduce twice as much emissions
at the same marginal costs as the smaller country.
Bauer finds that only coalitions of two countries of similar size are
stable; that is, countries 1 and 2, 3 and 4, or 2 and 3, 4 and 5 and so on may
form stable coalitions. Countries of different size with rank differences
larger than 1, for example, countries 1 and 3, will never form stable coali-
tions. It is particularly interesting that in Bauer’s model several coalitions
may coexist at the same time. However, coalitions comprising more than
two countries are only stable if coalitions of two countries act as single
countries. This is, of course, an unrealistic assumption which is not in line
with the notion of self-enforceability.
What is crucial in driving Bauer’s result is the assumption that signato-
ries choose that reduction level which maximizes aggregate welfare and that
transfers within the coalition are ruled out. Thus, Bauer does not pursue
the interesting question of whether there is a transfer scheme which could
balance the asymmetries between countries so as to induce a higher partic-
ipation rate. She also does not investigate whether the choice of another
abatement target would lead to more optimistic results (given the assump-
tion of no transfers). Since the abatement target which maximizes joint
welfare imposes a particular strain on the smaller partner in the coalition,
one should expect that either a lower abatement target or an asymmetric
abatement allocation would achieve more.
Botteon and Carraro (1997) use estimates from Musgrave (1994) of five
world regions for abatement and damage cost functions of type 1 and
conduct a stability analysis allowing for transfers. Among signatories payoffs
are distributed either according to the Shapley value or the Nash bargaining
solution.20 They confirm the result of the symmetric country case, namely
that the maximum number of signatories is N* 3. Moreover, they demon-
strate that there is not only one single set of potential signatories but that
various combinations of players could form a stable coalition. They find that
membership in the coalition depends on the burden-sharing rule.
For their data set the Shapley value leads to a coalition generating a
higher global welfare than a coalition in which the Nash bargaining solu-
tion is applied. This finding supports our conjecture mentioned above, that
the design of an agreement has an impact on global efficiency.
Barrett (1997b) proceeds in two steps. In the first step he considers payoff
functions of type 1, in the second step those of type 3 where he defines these
functions in reduction space. For both types he considers two model ver-
sions, one without and one with side-payments. He assumes that signatories
maximize the aggregate payoff to the coalition. Barrett allows for different
marginal benefit and damage cost parameters between countries (though the
functional form of the net benefit functions is the same for all countries) and
thus the possible degree of asymmetry is relatively large.
Payoff Functions of Type 1: Barrett shows that without transfers the
maximum number of signatories is three, but also that no coalition may
exist at all, depending on the degree of asymmetry. With transfers, employ-
ing the Shapley value, he finds that the minimum number of signatories is
two, but the maximum number does not exceed N* 3. Though the results
claim no general relevance, it seems very likely that the finding of the sym-
metric country case of a maximum coalition of three countries carries over
to heterogeneous countries.
Payoff Functions of Type 3: Barrett assumes that countries belong to
one of two groups of countries with homogeneous group characteristics.
For both assumptions – no transfers and transfers – he basically finds his
results of the symmetric country case confirmed (by using simulations).
That is, N* [2, N ], and whenever cooperation would be needed most, a
coalition achieves only little.
Coalition models I 237
In this section we want to discuss briefly some model extensions and aspects
which have not been covered yet.
Proposition 13.4
If countries are symmetric it is not possible to enlarge an internally and
externally stable coalition by self-financed transfers.
Proof: First note that self-financed transfers refer to transfers which are
paid out of the welfare gain that signatories obtain by forming a larger
coalition compared to the status quo. For signatories it is only attractive
to pay transfers to a non-signatory to induce it to join if the additional
gain is larger than the transfer. Furthermore, a non-signatory only joins
the coalition if the transfer which it receives, t, is larger than the welfare
loss it occurs from becoming a signatory. If we denote the status quo
coalition size by N*, then the enlarged coalition size is N* 1 and the two
conditions taken together are:
However, from the definition of stability and the assumption that a trans-
fer is needed to induce a non-signatory to join the coalition (the inequal-
ity sign of the external stability condition C3 is strict!) it follows that
jNJ(N*)jJ(N* 1) and hence compensating a non-signatory implies
an overall gain of [iJ(N* 1)iJ(N*)]t/N* to a signatory. However,
238 Game theory and international environmental cooperation
Fairness
There are two papers which have taken another route to generate more pos-
itive results with respect to coalition formation which are very much in the
tradition of Sudgen’s theory of reciprocity and Margolis’s unselfish behav-
ior (see Section 10.7). Jeppesen and Andersen (1998) introduce ‘non-
material’ payoffs in Barrett’s model which are derived from fairness. They
argue that free-riding would be seen as unfair behavior by governments.
Therefore, governments derive some payoff from the membership as such.
Similarly, in Hoel and Schneider’s (1997) model governments receive dis-
utility from breaking an agreement (loss of reputation). Not surprisingly,
both papers find that such moral aspects increase the number of participat-
ing countries in an IEA. However, it is evident that such arguments are not
Coalition models I 239
Economies of scale
Heal (1994) views coalition formation as being similar to an assurance
game (see Sections 3.4 and 3.6) where cooperation pays once enough
players accede to an IEA. He argues that there must be a minimum
number of countries in order for a coalition to be profitable and to come
into force. He calls this a minimum critical coalition. Most of Heal’s anal-
ysis is concerned with the derivation of the conditions of a minimum crit-
ical coalition. However, basically, transforming his approach into the
context of the previous setting, it turns out that his definition of a
minimum critical coalition is nothing other than the definition of the
profitability of a coalition. Since it is known from above that profitability
is only a necessary but by no means a sufficient condition for the stability
of a coalition, Heal misses an important point. Hence, the really interest-
ing aspect of Heal’s analysis lies somewhere else: it is the idea that there
might be economies of scale in forming a coalition. For instance, sharing
the cost of R&D in developing new abatement technology may reduce
costs substantially so as to induce more countries to join an IEA. Thus,
integrating this idea into the conjectural variation models above would
lead to more optimistic results.
Since the conjectural variation models are technically rather simple, they
are ideally suited to analyzing the effects of linking an environmental agree-
ment with another agreement. Since the instability of an IEA is caused by
the non-excludability of the public good environment, the second agreement
must deal with the provision of an excludable good, as for instance a club
good (Carraro 1997).22 For instance, Carraro and Siniscalco (1997),
Botteon and Carraro (1998) and Katsoulacos (1997) consider the link
between an IEA and cooperation on R&D. Signatories of an IEA share the
costs of R&D and can exploit more-efficient technologies. Hence, the unit
production costs of the signatories decrease through cooperation and they
obtain a competitive advantage over non-signatories. Thus, if a signatory
thinks about leaving the coalition it has to expect not only that abatement
240 Game theory and international environmental cooperation
of the remaining signatories will be lower after they have reoptimized their
strategies, but also that it will have higher production costs. Thus, the forces
inducing a country to remain in the coalition are stronger through issue
linkage and therefore, not surprisingly, these papers find (using simula-
tions) that the equilibrium number of signatories may be greater in the
linked agreement than in the pure IEA.
Since Carraro and Siniscalco (1997) assume that there are decreasing
economies of scale in cooperation on R&D, the equilibrium number of sig-
natories may fall short of the grand coalition. Moreover, as the authors
assume an oligopolistic market structure of the Cournot type, the gains
from cooperation on R&D are limited. Since lower unit production costs
imply a higher output by signatories and hence a lower market price, it
might be in the interest of signatories not to have too many coalition
members. In fact, for some parameters of the market demand and produc-
tion function, it may happen that the linked agreement comprises fewer sig-
natories than the original IEA (Carraro and Siniscalco 1997). Thus, issue
linkage may also destabilize an IEA; this would be even more pronounced
if account were taken of transaction costs from coordinating two agree-
ments.
In a similar spirit Barrett (1997c) sets up a model in which an IEA is
linked to a trade agreement. Trade of signatories with outsiders is com-
pletely banned, but outsiders may trade among themselves. Barrett shows
(using simulations) that if minimum participation is ensured in the linked
agreement, the agreement may be capable of securing a large number of sig-
natories. An example of such a link is the Montreal Protocol on the deple-
tion of the ozone layer where trade with non-signatories is partially
restricted (Barratt-Brown 1991; Benedick 1991; Blackhurst and Subra-
manian 1992).
The problem with all the issue linkage models mentioned above is that
they make rather special assumptions regarding the market structure. In
particular, Barrett’s (positive) result rests crucially on the assumption of an
oligopolistic market structure. Assuming instead perfect competition, a
‘trade cartel’ would be of no interest to signatories. Moreover, the assump-
tion in Barrett’s model that imports of non-signatories to signatories’
markets are completely cut off seems also a very restrictive assumption and
is not compatible with the rules of GATT/WTO. Probably, it would be
more realistic to assume only a tariff imposed on non-signatories (see, for
example, Finus and Rundshagen 1999). In reality it is also hardly conceiv-
able that all countries belonging to a large trade agreement would automat-
ically be required to become members of an IEA. However, admittedly,
allowing for mixed membership creates a lot of analytical problems.
Finally, we should like to point out that Barrett implicitly assumes fixed
Coalition models I 241
13.2.8 Discussion
alternative coalition could form. Thus the myopic view where a player con-
siders only the immediate reaction to his/her deviation is given up in favor
of a farsighted view in which all subsequent actions of players are
accounted for (Carraro and Moriconi 1997).
In the context of symmetric countries this definition implies that the
former equilibrium N* according to Definitions 13.1 and 13.2 is no longer
an equilibrium. Since it is always beneficial to be a non-signatory if a coali-
tion exists, a signatory has an incentive to leave the coalition, knowing that
it will be replaced by a former non-signatory.23 Hence, the number of sig-
natories would permanently remain at N*; however, C2 does not hold and
membership permanently circulates. Consequently, in a strict sense, the
former equilibrium N* is no equilibrium any more (since C2 and C3 are not
simultaneously satisfied), though it may be called a ‘steady state’. Thus, also
the threat to resolve a coalition of N* once a signatory leaves is an empty
threat since by the definition of a profitable coalition there is an incentive
to form a coalition.24 Only in the context of heterogeneous countries may
a stable coalition at N* exist. If the new coalition reoptimizes its strategies
to a lower abatement level after a signatory has left, then the defector may
lose. This is because if countries are heterogeneous the abatement target of
the coalition depends on the membership and some countries may have a
lower payoff as a non-signatory than as a signatory.
However, even in the case of symmetric countries an equilibrium coali-
tion may exist, though not at (the former) N*. We denote this ‘new’ equilib-
rium by N*C. To see this, consider Matrix 13.1 which assumes payoff
function of type 3 and Nash–Cournot behavior between signatories and
non-signatories. The rows are ordered according to the number of signato-
ries. The first entries are the payoffs of non-signatories, the subsequent
entries those of signatories. From previous sections it is known that N* 2.
As argued above, N* 2 is only a steady state according to the modified
stability definition since a signatory leaving the coalition knows that it will
be replaced by a non-signatory and hence can enjoy a payoff of iNJ 0.19.
Now consider a coalition of four countries. If a signatory leaves, it must
conjecture that in the new coalition of three countries, again, a country
leaves. This leads to the steady-state coalition of two countries. Comparing
the payoff as a signatory in a coalition of four countries (iJ 0.43) with
that as a non-signatory, if a coalition of two countries forms (iNJ 0.19)
does not make it attractive to leave the coalition. Hence, according to the
modified stability definition above, the coalition of four countries is an
equilibrium coalition, that is, N*C 4. Consequently, the grand coalition
cannot be stable since it pays a country to be a non-signatory in a coali-
tion of four countries instead of being a signatory in the grand coalition
(iJ 1.14 2.49iNJ ).
Coalition models I 243
Number of
Payoffs
signatories
1 (singletons) 0.9 0.9 0.9 0.9 0.9
N* 2 0.19 0.19 0.19 0.84 0.84
3 1.5 1.5 0.3 0.3 0.3
N*C 4 2.49 0.43 0.43 0.43 0.43
5 1.14 1.14 1.14 1.14 1.14
Note: aPayoff function (13.3), b 2, c1, d2, N 5 has been assumed.
smaller coalition size than in the isolated IEA would have to be reconsid-
ered.
Third, under the Stackelberg assumption it is assumed that a signatory
which possesses superior information as a Stackelberg leader maximizes its
payoff according to Nash–Cournot once it leaves the coalition. This
assumption obviously requires some form of irrationality by players and is
difficult to justify. It is this very assumption which leads, for instance, to the
grand coalition for a payoff function of type 2 under the Stackelberg
assumption. Assuming instead for the preliminary coalition of N* N that
a signatory is as clever when leaving the coalition as when remaining in the
coalition, no stable coalition would exist.
Though the concept of the core belongs to cooperative game theory, it is dis-
cussed here since it represents an important branch of the literature on
coalition formation. As laid out in Section 2.3, a typical feature of cooper-
ative game theory is the assumption that legally binding contracts can be
signed. Consequently, this concept seems inadequate in the context of
international pollution control. However, rejecting this concept on this
ground would be too simple. In fact, a closer look at the concept reveals
that some form of implicit punishment is indeed part of the definition of
the core.
There are basically two definitions of the core, one assuming no transfers,
the other assuming unlimited transfers among the coalition members. Since
the latter assumption has exclusively been made in all papers on interna-
tional pollution control until now, we shall restrict our attention to this
assumption as well (though we have raised objections to this assumption in
Chapter 8). Before defining the core, we need two definitions (Luce and
Raiffa 1957; Moulin 1995; Ordeshook 1992):
w(IJ)max
eJ
iIJ
i
where eJ is the emission vector of the coalition and i i (ei )i (ek )
the payoff function of a player.
246 Game theory and international environmental cooperation
iIJ
i w(I )
* J IJ I.
From Definition 13.5 it is evident that the core depends on the worth of
the coalition, w(IJ), which in turn depends on the behavior of the players
outside the coalition.27 A frequently made assumption is that the non-
coalition members put the coalition members down to their minimax or
maximin payoffs. This leads to the and characteristic function:
w(IJ ) implies that the non-coalition members move first and choose their
highest emission levels, ejmax, to which the coalition members react by
choosing their best replies, ei(ejmax). w(IJ ) implies that coalition members
move first and choose their best replies, assuming that the non-members
will punish them as hard as possible. In the global emission game context
both characteristic functions are identical, that is, w(IJ ) w(IJ ) (since in
both cases ej ejmax j INJ ).28 Of course, if IJ I, w({I})w({I})
iS.
In the international environmental context the assumption about the
behavior of non-members in the and characteristic functions seems not
very realistic. It implies that the coalition members assume that players
outside the coalition will carry out punishments which are injurious to their
own welfare. Therefore, Chander and Tulkens (1995, 1997) propose the
assumption that non-coalition members choose their strategy according to
Nash behavior. If the coalition maximizes the aggregate welfare of its
Coalition models I 247
1. ẽ J maximizes for e
iIJ
i
NJ ẽ NJ;
13.3.2 Results
Proposition 13.5
Let the payoff function from global emissions be given by
i i (ei )i (ek ), assumptions A1 in (9.2) hold and let eS (e1S, ..., eNS )
be the (unique) socially optimal emission vector. Then the imputation
* (1*, ..., N*) defined by i* iS(eS)ti* where ti* is given by
(13.27) lies in the core provided one of the following assumptions is true:
(a) damage functions are linear; (b) IJ I, IJ2, kIJ k(eS) i (eN )
i IJ; and (c) countries are symmetric.
Proof: See Appendix X.2.
13.3.3 Discussion
1. In many papers, for example, Chander and Tulkens (1992, 1997), the
authors use a far more complicated model than the simple global emis-
sion model that we introduced in Chapter 9. Since all main results
of Chander and Tulkens, which have been summarized in Prop-
osition 13.5, can also be established (without qualifications) in the
simple framework, the motto of Rasmusen of ‘no fat modeling’ has
been violated by these scholars.37
2. Though this has not been explicitly mentioned above, the core concept
assumes as the conjectural variation models only two groups of coun-
tries: those which form a coalition and those outside the coalition
which either play as singletons (
core) or jointly form a ‘punishment
coalition’ ( and core).
3. By its very nature, the core is a static equilibrium concept. However,
from the definition of the characteristic function it is clear that some
behavioral assumptions are made as to how players react if a country
250 Game theory and international environmental cooperation
leaves the coalition. Thus, the assumptions of the core are very much
in the tradition of the conjectural variation models which may be inter-
preted as static representations of dynamic games (see Tulkens 1998).
Instant reactions are assumed which do not allow a defector to net any
transitory gain. Thus, the same critique as was raised with respect to
conjectural variation models applies here (see Chapter 10 and Sub-
section 13.2.8), namely that the core does not adequately account for the
free-rider incentive in international pollution control.
In view of this, defending the core concept by criticizing concepts
like the strong Nash equilibrium (see Chapters 6, 7 and 12) or the coali-
tion-proof Nash equilibrium (see Chapter 15), that they would not
account for reactions by players to a deviation, seems a rather weak
argument of Chander and Tulkens (1997, s. 3). Their critique is based
on the fact that, in a static context, these equilibrium concepts deter-
mine (in the tradition of an NE) a best reply, given the best reply of the
other players.38 However, with a dynamic story in mind when analyz-
ing a coalition formation process, it seems natural to evaluate the
dynamic and not the static version of these alternative equilibrium con-
cepts. A casual look at the definitions of the dynamic versions of these
concepts (or recalling the definition of a strongly subgame-perfect
equilibrium or renegotiation-proof equilibrium in a finite (Chapter 6)
or an infinite (Chapters 7 and 12) time horizon in a two-player game)
would immediately make clear that the critique is false. Moreover, after
all, a consistent assumption in a one-shot game implies no reaction by
the very meaning of this term (see the discussion in Section 10.1).
It is also not correct to claim that the core has an advantage over the
conjectural variation models because the latter models would not
account for reactions after a deviation of a (original) coalition member
(Tulkens 1998, s. 4). As has been demonstrated above, conjectural vari-
ation models assume that if a signatory leaves the coalition, the
reduced coalition reoptimizes its strategy.
4. If the core is given a static interpretation, then it is immediately evident
that it is not an equilibrium in a non-cooperative game theoretical
sense. If the core is given a dynamic interpretation, then the and core
versions imply a non-credible threat. Neither in a finite nor in an infinite
time horizon is minimaxing a player for the rest of the game a
subgame-perfect trigger strategy. In contrast, the threat to play an NE
once a player defects as assumed in the
core concept is a subgame-
perfect equilibrium in a supergame framework.39 However, such a
threat would neither be a weakly or strongly renegotiation-proof nor a
strongly perfect equilibrium.
5. It seems not very plausible that according to the
-characteristic function
Coalition models I 251
then the social optimum would hardly lie in the core, which is particu-
larly true for the
core. Then smaller coalitions and abatement targets
below socially optimal ones may only lie in the core. A reason why such
a modification has not been considered so far is the more surprising
since no plausible explanation is given as to how such a transfer scheme
can be realistically enforced. The assumption of Chander and Tulkens
that such a transfer scheme is simply monitored and administered by
an international agency seems rather naive.
In some papers the core concept has been integrated into an explicit
dynamic story. The basic idea is to test whether it is possible for a coalition
to reduce emissions gradually from NE to socially optimal levels (Tulkens
1979). Along such an emission path the local optima have to be in the core
in order to view such a path as stable. The motivation for the assumption
of a dynamic emission path is twofold: first, it is hardly realistic to assume
that players move at once from the non-cooperative equilibrium to the full
cooperative equilibrium; second, governments may only possess informa-
tion about the damage and abatement cost functions around the present
state but not over the full range of these functions. Thus, governments
adjust their emissions only gradually towards the final state by using local
information.
Whereas in Chander and Tulkens (1991, 1992) such a ‘time iterative
process’ is analyzed in continuous time, in Germain et al. (1995, 1996),
Kaitala et al. (1995) and Germain et al. (1998) this is done in a discrete
time setting. Whereas some of the papers (such as Chander and Tulkens
1991, 1992, and Kaitala et al. 1995) assume a notion of the stability
concept slightly different from the previously discussed core concepts,
Germain et al. (1996, 1998) base their stability analysis on the
core
concept. All papers assume a flow pollutant, except the last-mentioned
paper.42 Since all previous models in this book have assumed a flow pollu-
tant and since we have identified the
core concept as the most suitable
among the core concepts, the following discussion is based on Germain et
al. (1995, 1996).
The basic idea of all the papers is not much different from that in the
static context. One checks whether an imputation of the grand coalition lies
at each time t1, 2, ..., T in the
core, assuming a locally optimal emis-
sion vector and some transfer scheme. After T iterations the final optimal
state will have been reached. This state is the global optimum, or in the
former terminology, the social optimum. The transfer scheme assumed to
Coalition models I 253
achieve this is very much in the spirit of that used in the static context above
(see (13.27)).
More concretely, assume payoff at time t to be given by:
It can now be shown that max i,t subject to constraint (13.29) leads to
e*t
the global optimum after a finite number of iterations where e*t denotes the
locally optimal emission vector of the grand coalition at time t, that is,
e*tT eS (see Germain et al. 1995, Theorem 1).
The transfer scheme assumed during the iterative process is given by:
t*i,t [i (eiN )i (e*i,t )][i (eiN )i (e*i,t )] i [gj (e*j,t )gj (ejN )]
(13.30)
where:
Proposition 13.6
Let the payoff function from global emissions at time t be given by
(13.28) where the benefit function is strictly concave and damages are
linear and let e*t (e*1,t, ..., eN,t
* ) be the (unique) locally optimal emission
vector. Assuming that countries have local information about payoffs for
all emissions as defined by (13.29) and that they gradually approach the
global optimum via local optima in several stages, then the imputation
t* (*
1,t , ..., N,t ) defined by i,t i,t (et )t i,t where t i,t is given by
* * * * * *
We would like to finish this section with two remarks. First, the assump-
tion of a time-iterative process in which countries gradually approach an
254 Game theory and international environmental cooperation
optimal state is a major step in adapting models to reality (see also Sub-
section 12.2.4). Many international environmental agreements have seen an
evolution as described by this process (Caldwell 1984). Commonly, to begin
with, a formal IEA is only a framework convention, stating good inten-
tions. Then moderate abatement obligations are specified in a next step;
these are successively revised and increased over time in successive proto-
cols. Examples include the Montreal Protocol on the reduction of the
depletion of the ozone layer and the Helsinki Protocol to reduce sulfur.
Second, though the adjustment of emissions is modeled in a dynamic
framework, the threat and punishment strategies do not change compared
to the static version of the core concept (see Appendix X.3). From a game
theoretical point of view this is disturbing. As demonstrated for repeated
games, it should be expected that the time horizon would have an influence
on players’ strategies. Basically, the core concept as applied in the above-
mentioned papers implies that players follow a strategy which neither is
based on the history of the game nor does it take the future into account:
an assumption which can hardly be rationalized.
NOTES
1. An empirical paper which analyzes the stability of a grand coalition set up to fight acid
rain in Europe is Finus and Tjøtta (1998). Since their setting resembles that of
Chapter 14, their work will be discussed there.
2. This issue will be considered more thoroughly in Sub-section 13.2.5.
3. See the brief discussion of these concepts in Chapter 11, Section 11.3.
4. The definition implies that the status quo is given by the Nash equilibrium where no
coalition has formed. The term coalition is reserved for the cooperation of at least two
countries.
5. The ‘’ instead of the strict inequality sign is used for two reasons. First, it seems equally
likely that a signatory remains in a coalition or leaves it if nothing is gained from leaving.
Also, a non-signatory may remain outside or may join the coalition if this does not affect
his/her payoff. Second, the strict inequality sign would imply that for the net benefit func-
tion of type 1 below no coalition equilibrium exists. (Though Carraro and Siniscalco
1993, 1998 use the strict inequality sign in their definition of stability, they report on
small coalitions obtained in various model versions based on payoff functions of type 1.)
6. The advantage of this relative index is twofold. First, the index is independent of the unit
of measurement. Second, the signs of the derivatives hold for all three payoff functions
and for the Cournot and Stackelberg assumptions. Moreover, they would also hold if the
functions were slightly modified, and therefore rather general conclusions are possible
using this relative index. In contrast, Barrett (1994b) uses the absolute index and derives
results which only hold for specification (13.3) but would not hold if damages were
assumed to be given by i fi (eiJ ejNJ ) instead of i (1/N )fi (eiJ ejNJ ).
7. In Endres (1997) the parameter c is interpreted as the environmental awareness of the
society of a country. He analyzes the effect of a change of this variable on the degree of
externality in a static game with two countries. In Endres and Finus (1998c) the analysis
is extended to N countries and a dynamic framework. Particular attention is given to the
effect of a change in the environmental consciousness of society on the success and
stability of an IEA.
Coalition models I 255
8. We assume that the appropriate non-negativity constraints are invoked on the parame-
ters such that eiN 0 and eiS0 i I hold.
9. iS 0 i I holds for all payoff functions.
10. We assume in the following that the appropriate constraints on the parameters apply so
that positive emissions are ensured.
11. A signatory in a coalition of N* 3 is indifferent about leaving the coalition, implying
N* 2, and receiving a non-signatory payoff. Hence, N* 3 and N* 2 seem equally
likely, though Carraro and Siniscalco (1991) claim that there is a unique equilibrium of
N* 3.
12. Leakage effects have been studied in the energy market by Bohm (1993); Felder and
Rutherford (1993); and Golombek et al. (1995). Energy conservation by ‘green’ countries
leads to a reduction in demand for crude oil. Consequently, energy prices drop. This in
turn triggers higher demand by less environmentally conscious countries and renders the
environmental policy of green countries less effective. See also Barrett (1992c); Bohm
and Larsen (1993); and Hoel (1994) on leakage effects and countervailing measures.
13. In a symmetric Nash equilibrium eiN b/c and iN (b2(2 N ))/2c. As long as N* N,
the best strategy of signatories as Stackelberg leaders is to expand emissions up to
eiJ* bN/c so that non-signatories emit nothing. Then, a signatory receives
iJ* (b2N(2 N*))/2cN* and a non-signatory iNJ* Nb2/2c. It is easily confirmed that
C3 0 for any N* N, and C2 0 for N* N. (External stability is automatically ensured
for N* N by definition.)
14. This result contrasts to Barrett’s (1994b) finding of no stable coalition for this payoff
function.
15. Differentiating a signatory’s payoff function with respect to eiJ, assuming ejNJ 0, gives
iJ/eiJ bd beiJ cN2eiJ. Substituting bd/c for eiJ (which follows from eiJ
dN )
we get iJ/eiJ bdb2d/c Nbd 0 where the sign follows from 1/n. Since
2iJ/eiJ2 0, it follows that iJ/eiJ 0 eiJ bd/c which implies by symmetry
eiJ
dN.
16. This claim and all subsequent assertions in this section are proved in Appendix X.1.
17. Barrett (1994a, b) does not consider the profitability constraint when determining the
coalition equilibrium.
18. Only a framework convention has been signed stating ‘good intentions’ at zero costs,
which explains why it has been signed by many countries.
19. Of course, a shortcoming is that the failure of some IEAs to come into force cannot be
explained by this model version since N* [2, N ]. Then, payoff functions of the second
type and Nash–Cournot behavior could be a possible explanation. See also the model in
Chapter 14 which does not suffer from this deficiency and where N* [0, N ].
20. Both concepts imply that aggregate welfare of the coalition is maximized.
21. Xepapadeas (1997, p. 205) tries to justify such assumptions of commitment. He writes:
‘The internal stability requirements . . . can, however, be regarded as too restrictive in
reality.’ However, we feel that such arguments are misguided. Once a stability concept
has been chosen, one should stick to the implied results and not reinterpret them because
they are ‘too strong’ compared to the gray tones of reality.
22. On the characteristics of private, club and public goods, see Arnold (1992); Cornes and
Sandler (1986).
23. Therefore, Carraro and Siniscalco (1992, 1993) frequently stress that the underlying
incentive structure of international environmental problems is not that of a PD game
but rather of a chicken game, where each country would like to be a non-signatory but
would join the coalition if it did not form (see Section 3.6). However, as shown above for
a continuous strategy space, in a consistent conjectural variation framework no coalition
exists for symmetric countries and in the traditional stage game framework the incentive
structure resembles that of a PD game more than of a chicken game. Only for the
‘hybrid’ Definition 13.2 of a coalition equilibrium is their assertion true. See also the dis-
cussion in Sub-section 9.7.2 on the incentive structure in a global emission game.
24. This issue is taken up again in the context of the core (see Section 13.3) where exactly
this kind of inconsistency occurs.
256 Game theory and international environmental cooperation
25. As Keohane (1995, p. 217) puts it: ‘compliance is not very adequate. I believe that every
study that has looked hard to compliance has concluded . . . that compliance is spotty.’
In their prominent empirical study on compliance of IEAs, Brown Weiss and Jacobson
(1997, pp. 87ff.) find severe instances of non-compliance for all IEAs covered by their
study. Sand (1997, p. 25) reports that no fewer than 300 infractions of the CITES treaty
have been revealed per year. Also the whaling convention has been frequently breached
by all important parties to the treaty (Heister 1997, p. 68). Moreover, many IEAs also
have a very poor compliance record with respect to reporting requirements (GAO 1992,
pp. 3ff.; Sand 1996, p. 55; Bothe 1996, pp. 22ff.). Since official compliance-monitoring in
almost all IEAs relies exclusively on self-reporting by the states, some suspicion with
respect to good official compliance records of some IEAs seems also to be justified
(Ausubel and Victor 1992, pp. 23ff.).
26. See, for example, Finus and Tjøtta (1998) and Murdoch and Sandler (1997). The exten-
sive IIASA study on the effectiveness of IEAs draws a similar conclusion: see Victor et
al. (1998, pp. 661ff.).
27. Due to the assumption of transfers, only payoff vectors based on the socially optimal
emission vector (imputations) can qualify as potential candidates to lie in the core.
(Otherwise a Pareto improvement for all coalition members would be possible.) For the
assumption of no transfers this would be different.
28. More generally, the and characteristic functions are equivalent for games with trans-
ferable utility (Ordeshook 1986, p. 331). For the definition of minimax and maximin
payoffs, see Sub-section 4.3.2.
29. Chander and Tulkens define the PANE with respect to IJ I. However, there is no reason
why this definition should not also comprise the grand coalition, that is, IJ I.
30. See Tulkens (1998) on the similarity of both approaches.
31. w({i}) may also be called the disagreement point. This is the payoff a player can expect to
receive when players cannot agree on a grand coalition.
32. Note that the definition of the characteristic function is puzzling since it implies that if
a player leaves the grand coalition s/he forms a ‘one-player coalition’ and the former
coalition members become non-members. Whereas in the conjectural variation models
the coalition members are the ‘good guys’ and the non-members the ‘bad guys’, this is
reversed in the core. In other words, a country which does object to the terms of a treaty
proposes an alternative coalition (Tulkens 1998).
33. Shapley (1971) showed that the core of convex games is non-empty. A game is called
convex if its characteristic function is convex. (Since the characteristic function already
includes the behavioral assumption of the countries outside the coalition, the result is
true for any core concept.) This function is convex if w(I1J )w(I2J )
w(I1J I2J ) w(I1J I2J )
I1J , I2J I ⇔ w(I1J {i}) w(I1J )
w(I2J {i}) w(I2J ) i I2J , I1J I2J hold. That is, there
is a kind of increasing marginal utility for coalition membership. It is easily checked that
the latter condition does not necessarily hold in the global environmental context. For
instance, consider N 4, then w({1, 2, 3}) w({2, 3, 4})
w({1, 2, 3, 4}) w({2, 3}) must
hold. However, for a payoff function of type 3 in (13.3), b 0.2 and c1 and the
char-
acteristic function, this condition is not satisfied.
34. An intuitive reason for this result is the observation that w({i})
w
({i}) holds.
35. As a matter of notation: IJ I implies that IJ is a true sub-coalition of I.
36. It is surprising that Chander and Tulkens (1995, p. 18) claim that this assumption would
hold for a broad class of quadratic utility functions since a counter-example is readily
constructed. For instance, this assumption fails for payoff functions (14.1) in Chapter 14,
assuming N 3, 1
4 (which takes condition NNC1 in (XI.3), Appendix XI.1 into
account) and if countries 2 and 3 form a coalition, that is, IJ (2, 3).
37. Of course, in some papers a transboundary pollution problem and not a global environ-
mental problem has been assumed. However, this does not affect the main conclusions.
38. A definition in an N-country context is given in Chapter 15.
39. The underlying assumption is the payoff function i i (ei )i (ek ) as described in
(9.1) which satisfies assumptions A1 in (9.2) and hence by Theorem 9.2 there is a unique
Coalition models I 257
NE. Therefore no SPE except playing the stage game NE repeatedly exists in a finite
setting by Theorem 4.2.
40. The behavior of non-coalition members according to the and characteristic func-
tions has already been criticized in Sub-section 13.3.1.
41. See the justification for the LCD decision rule as assumed in the bargaining process in
Chapter 11.
42. Germain et al. (1998) may be regarded as the most advanced of these papers. Whereas
Chander and Tulkens (1991, 1992) are purely theoretical papers, the others apply their
framework to an acid rain game between seven northern European regions.
14. Coalition models: a second
approach
14.1 INTRODUCTION
258
Coalition models II 259
made. The results if all aspects of the formation process are considered
simultaneously are reported in Sub-section 14.4.4.
The analysis of the first (grand coalition) and second step (sub-coalition)
presents ample evidence to answer the two questions raised in the Introduc-
tion of Chapter 11 and one more:
1. Why are abatement targets within most IEAs rather low, even though
cost–benefit considerations would suggest higher targets?
2. Why do only sub-groups of countries sign an IEA though more coun-
tries suffer from an environmental externality?
3. Why have effluent charges (as one representative of market-based
instruments) not been applied in any IEA so far whereas quotas have
found widespread application?
14.2 PRELIMINARIES
where b and d are benefit parameters and c is a cost parameter. This implies
equal benefits from emissions in the N countries but different damages
which are (discrete) uniformly distributed. Countries with a higher index i
suffer more from pollution than do those with a lower index number.3
Though equal benefits may be seen as a restrictive assumption in (14.1) this
specification allows us to identify the interests of the N countries uniquely.
Countries with a higher index have a greater interest in emission reduction
than those with a lower index. This is evident from computing emissions in
the static Nash equilibrium (NE) and the social optimum and where
eiN/i 0, (iS iN )/i0 and iS iN , , 0 hold.4 That is, countries
with a higher index choose lower emissions in the non-cooperative equilib-
rium than those with a lower index. The high index countries also benefit
Coalition models II 261
more from a socially optimal solution than low index countries due to their
higher damage. In fact, low index countries may receive a lower payoff in
the social optimum than in the NE (recall Proposition 9.2).
For specification (14.1) a familiar result from Chapter 13 also applies:
e N
k e k
S Sk N
k I1 I I I
I1: , I : , 0, 1 0, 2 0, 2 0. (14.2)
e k
S 2
kS N N
That is, the externality is distinct if many countries suffer from the global
pollutant and if environmental damage is very high compared to the
opportunity costs of abatement.
For payoff functions (14.1) it is straightforward to derive a globally
optimal uniform tax, t*, or expressed as relative emission reductions from
the NE, rT* rkT* where eiT* (1riT*)·eiN. As mentioned above, since a
uniform tax is efficient in the context of global emissions, t* tS and riT* riS
where rS rkS, eiS (1riS)·eiN and tS denotes a socially optimal or
Pigouvian tax. Note that for (14.1) eiS ejS
i and j I. By the same token,
a globally optimal uniform emission reduction quota, rQ*, can be derived
where eiQ* (1rQ*)·eiN. Since eiN ejN and hence eiQ* ejQ*
i and j I,
such a solution is not cost-efficient if benefit functions are identical in all
countries as assumed in (14.1).
It is also straightforward to derive proposals ti under a uniform tax
regime and ri under a uniform emission reduction quota regime. For (14.1)
it turns out that t1 t2 ... tN1 tN and r1 r2 ... rN1 rN. That is,
under both regimes country 1 is the bottleneck which determines the terms
of the agreement if the LCD decision rule is applied. This is not surprising
since, given equal opportunity cost of abatement, country 1 evaluates envi-
ronmental damages lower than its N1 neighbors and is therefore inter-
ested in the lowest emission reduction.
As mentioned in the Introduction, t(N1)/2 t* and r(N1)/2 rQ* for (14.1).
This implies that if in the subsequent analysis a globally optimal tax or
emission reduction quota is not stable, this is also true for the median
country proposal.
For the different emission vectors mentioned in the previous sections stabil-
ity according to the WRPE concept can easily be checked. All relevant back-
ground information has already been derived in Chapters 7 and 12 so we can
immediately proceed to the results. The only difference is that now the
262 Game theory and international environmental cooperation
WRPE inequality system may be very large if N is large. This is evident when
recalling that in the case of discount factors close to one N2 (see (12.1) and
(12.2)) and in the case of discount factors smaller than N ·(N1) (see
(12.17)–(12.19)) inequalities have to be satisfied simultaneously. Thus, in
contrast to Chapter 12 where a stability check could be conducted analyti-
cally for → 1, in the case of more than two countries the analysis has to rely
on simulations. However, this procedure seems to be justified because all
simulation results presented below are quite robust and show a clear pattern.
Moreover, due to the sheer size of the inequality system an algorithm is
needed to check for stability. The details are provided in Appendix XI.2 for
the case of → 1 and in Appendix XI.3 for the case of 1.
As in the case of two countries, only deviations by one country are con-
sidered. In the context of a global emission game this assumption seems to
be justified (see also Sub-section 13.2.1). On the one hand, a country plan-
ning to take a free-ride has no incentive to do so jointly with other coun-
tries because the transitory gain is larger if other countries comply. On the
other hand, on the side of the punishers there is no incentive to form a
coalition when breaching the punishment obligations. If during the punish-
ment phase a country among the punishers were to emit more (or less) than
requested to punish the defector, it would not ask other countries to follow
suit. The punishment constitutes a public good to control an agreement
and therefore free-riding is most attractive to a country if conducted only
by itself. Therefore, if a punisher does not meet its obligations, it is treated
as a free-rider and will be punished. Of course, by the construction of the
WRPE concept, such a punishment is regarded as unnecessary since the
punishers receive more during the punishment phase than during the co-
operative phase.
In Table 14.1 the results of the stability analysis are reported. To evaluate
the different emission vectors a welfare indicator is provided which meas-
ures the ‘degree of optimality’. The indicator is defined as follows:
ki Ni
DOk · 100 (14.3)
Si Ni
where k stands for the policy regime and the particular abatement target,
for example, kT1 or h Qmax. Put differently, DOk indicates how close a
particular emission vector comes to the social optimum. By definition, the
degree of optimality in the social optimum is 100 percent and since riT* riS,
DOT* DOS 100 percent holds. From Table 14.1 four main results may
be derived:
Coalition models II 263
The results summarized in 1 and 2 confirm a result established for the con-
jectural variation models in Chapter 13: namely, that whenever cooperation
is needed most from a global point of view, a coalition achieves only little.
In particular, for global environmental problems where N is large, a stable
grand coalition seems very unlikely even if politicians were almost perfectly
patient. But even if a grand coalition were stable, reduction targets agreed
upon would have to be very low. This is a result which is confirmed by the
historical evidence of IEAs signed so far.
The results summarized in 3 and 4 can be taken as a first indication that
in a second-best world5 the choice of a quota regime to tackle global emis-
sions might be rational. Though governments cannot be expected to be
concerned about global efficiency, it is evident that the typical efficiency
argument in favor of the tax regime no longer holds in a second-best envi-
ronment. From an institutional economics point of view this may explain
why the quota and not the tax has emerged from the evolution of institu-
tions.
Table 14.1 Stability analysis of the grand coalition for → 1a
Quota Tax
rQ* rQ1 rQmax DOQ1 DOQmax rT* rT1 rTmax DOT1 DOTmax
N (%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
5 10 48.40 19.60 42.00 64.0 97.4 48.00 13.30 23.30 49.6 74.4
5 30 26.80 10.10 26.80 60.3 98.9 26.70 5.70 19.30 39.5 92.6
5 50 18.50 6.70 18.50 58.6 98.8 18.50 3.60 16.70 36.7 99.1
5 100 10.50 3.70 10.50 56.9 98.7 10.40 1.90 10.40 34.2 100.0
5 500 2.30 0.80 2.30 55.3 98.7 2.30 0.40 2.30 32.1 100.0
5 1000 1.18 0.40 1.18 55.0 98.6 1.18 0.20 1.18 31.8 100.0
5 2000 0.60 0.20 0.60 54.9 98.6 0.60 0.10 0.60 31.7 100.0
264
5 3000 0.40 0.13 0.40 54.9 98.6 0.40 0.07 0.40 31.6 100.0
10 10 76.50 27.60 36.00 59.0 71.9 76.20 22.50 3.90 51.4 11.9
10 30 58.40 17.80 24.00 51.6 65.2 58.20 11.30 10.50 35.5 33.3
10 50 47.20 12.80 19.00 46.9 64.2 47.10 7.50 10.50 29.8 40.1
10 100 32.00 7.50 14.00 41.4 68.2 31.90 4.10 9.50 24.4 50.8
10 500 8.90 1.70 6.00 35.0 89.0 8.90 0.90 5.50 19.1 85.5
10 1000 4.69 0.88 4.20 34.0 98.6 4.69 0.45 3.73 18.4 95.8
10 2000 2.40 0.45 2.40 33.5 99.7 2.41 0.22 2.39 18.0 100.0
10 3000 1.60 0.30 1.60 33.3 99.7 1.62 0.15 1.62 17.9 100.0
20 10 90.80 31.10 31.00 56.8 56.6 90.70 31.70 31.20 58.4 77.4
20 30 83.20 25.90 22.00 52.6 45.9 83.10 19.00 3.50 40.9 7.9
20 50 76.80 20.70 18.00 46.7 41.4 76.70 13.60 1.50 32.5 4.3
20 100 64.40 13.50 13.00 37.6 36.3 64.40 7.90 4.50 23.3 13.8
20 500 28.10 3.50 6.00 23.5 38.1 28.10 1.80 4.40 12.7 28.9
20 1000 16.49 1.83 4.50 20.9 47.1 16.49 0.93 3.54 11.1 38.4
20 2000 9.03 0.93 3.20 19.6 58.3 9.03 0.47 2.72 10.2 51.3
20 3000 6.22 0.63 2.50 19.1 64.2 6.21 0.31 2.24 9.9 59.2
50 30 95.80 32.30 18.00 56.1 34.0 95.70 30.60 43.10 54.0 109.0
50 50 94.30 30.10 16.00 53.7 31.0 94.30 24.50 23.50 45.4 55.5
50 100 90.90 23.80 12.00 45.5 24.7 90.90 16.30 8.80 32.8 20.1
50 500 70.40 8.20 6.00 21.8 16.3 70.40 4.50 1.40 12.3 4.0
50 1000 54.93 4.45 4.30 15.6 15.4 54.92 2.33 1.99 8.3 7.2
50 2000 38.15 2.33 3.10 11.9 15.6 38.15 1.20 1.92 5.0 9.8
265
50 3000 29.23 1.58 2.50 10.5 16.4 29.23 0.80 1.76 5.4 11.7
100 50 98.00 32.90 14.00 55.8 26.5 98.00 33.00 64.82 56.1 175.0
100 100 97.10 30.40 11.00 52.9 21.4 97.10 24.80 30.60 44.6 72.8
100 500 90.10 14.10 5.00 28.8 12.9 90.10 8.30 3.30 17.5 7.5
100 1000 82.64 8.24 4.30 18.9 10.1 82.64 4.50 0.43 10.6 1.0
100 2000 70.92 4.50 3.10 12.3 8.6 70.91 2.36 0.69 6.6 2.0
100 3000 62.11 3.09 2.50 9.7 7.9 62.11 1.60 0.93 5.1 3.0
Note: a rQ* globally optimal emission reduction quota; rQ1 emission reduction quota according to the LCD decision rule; rQmax maximum uniform
emission reduction quota which can be backed as a WRPE; DOQmax degree of optimality of rQmax; DOQ1 degree of optimality of rQ1; rT* rS
socially optimal emission reduction; rT1 emission reduction under the tax regime according to the LCD decision rule; rTmax maximum emission
reduction under the tax regime which can be backed as a WRPE; DOTmax degree of optimality of rTmax; DOT1 degree of optimality of rT1; italic
figures indicate instability in the sense of WRPE.
266 Game theory and international environmental cooperation
which a country can be deterred from taking a free-ride by the other N1
countries. Since it turns out that country 1 has always the lowest maximum
discount rate requirement, attention has been restricted to imax in Table 14.2
exclusively. In other words, imax
i (where i is the actual discount rate) is
a necessary and sufficient condition for country 1 (i1) not to take a free-
ride (provided all other countries comply with the agreement). However, it
is a sufficient condition for all other countries, i 1.
The following analysis is restricted to the abatement target resulting
under the LCD decision rule. Two main results can be derived from
Table 14.2:
N 10 Quota 1.8 2.1 2.0 1.7 1.5 1.5 1.5 1.4
Tax — — 1.0 1.2 1.1 1.1 1.1 1.1
14.3.4 Extensions
From the previous section it appeared that for global externalities a grand
coalition is unlikely to be a WRPE if the number of countries is large.
Considering that this result has been obtained assuming ‘optimal’ punish-
ment profiles, it is clear that if punishment profiles are restricted for some
reason, stability of a grand coalition would be even more unlikely. In par-
ticular, in the case of many countries, the optimal coordination of punish-
ment may be difficult and associated with high transaction cost. Therefore,
in reality one should expect that simple punishment profiles according to
rules of thumb will be applied to allow immediate and deterrent punish-
ment. The effect of such simple punishment profiles on the discount rate
requirement is investigated in Finus and Rundshagen (1998a). It is not sur-
prising that the authors find that the discount factor requirements are
higher for restricted than for unrestricted punishment profiles (that is, rimax
(unrestricted) rimax (restricted); see Chapter 12). What is more interesting
is their finding that, particularly for the critical parameter values (N large
and small), the discount factor requirements are greatly increased if
simple punishment profiles are applied. That is, countries face a dilemma:
on the one hand, particularly if many countries participate in an IEA,
simple strategy profiles will be employed to make coordination easier; on
the other hand, in this case the discount rate requirements are already very
high assuming optimal punishment strategy profiles. Moreover, Finus and
268 Game theory and international environmental cooperation
would not be possible. This stresses that the WRPE concept is a useful
extension of the SPE concept in order to explain politically interesting and
relevant phenomena.
This last assertion is supported by an empirical study by Finus and
Tjøtta (1998). The authors analyze the stability of a grand coalition which
intends to reduce acid rain in Europe. They show that a socially optimal
agreement would not be a WRPE. The reason is twofold. First, due to the
wind patterns in Europe emission transportation coefficients (see Chapter
10, note 18) exhibit a large asymmetry. Second, environmental preferences
and abatement costs vary substantially between countries. Both reasons
imply that the gains from a socially optimal agreement are very unevenly
distributed among signatories. Therefore, the authors go on to determine
the maximum backable emission reduction (such as rQmax in Table 14.1) by
WRPE strategies. They find that the maximum emission reduction is only
marginally above that in the NE. This result helps to explain why emission
reductions as laid down in the Oslo agreement signed in 1994 are not a
‘great leap forward’ and basically freeze emissions at the NE level of the
year 2000.
So far it has been assumed that if countries decide jointly to improve envi-
ronmental quality they seek to form a coalition of all countries. It became
apparent that there are many obstacles which render the probability of a
stable grand coalition rather low. This result is perfectly in line with exist-
ing environmental conventions in which only some of N countries partici-
pate if they come into force at all. For example, the Convention for the
Regulation of Whaling has been signed by 36 countries and the Antarctic
Treaty by 26 countries, even though all states should have an interest in the
problem.
Basically, environmentally conscious countries face a dilemma. On the
one hand, they like to see many countries participating in an IEA to improve
environmental quality and to share the abatement burden with them. On the
other hand, if the LCD decision rule is applied, the abatement target which
can be established within a grand coalition might be very low. Moreover, it
has been demonstrated that less environmentally concerned countries are
also the critical ones regarding stability. Therefore, one would suspect that
those countries which have a high interest in an international environmen-
tal policy seek partners but only among those with similar interests. This
270 Game theory and international environmental cooperation
procedure seems even more likely when taking into account that coordina-
tion in international politics is associated with transaction costs.7
As mentioned in the Introduction to Chapter 13, the number of possible
sub-coalitions is very large.8 To narrow down the analysis, it is assumed that
countries with higher environmental damages will form a coalition among
themselves (see also Botteon and Carraro 1997; and Hoel 1992a).
Therefore, in the context of the present model the set of non-signatories is
given by INJ {1, 2, ..., N0}, and the set of signatories by IJ {N0 1, ...,
N }. Within a potential coalition three issues have to be decided simultane-
ously: abatement target, instrumental choice and coalition size.
Abatement target
It is assumed among signatories that the LCD decision rule is applied to
determine the abatement target. As in the case of a grand coalition, it turns
out that for a given coalition size it is the country with the smallest index
which makes the lowest proposal. Therefore, it is this country’s government
which determines the terms of the agreement (see Appendix XI.1). For
instance, if N10 and IJ {7, 8, 9, 10}, then country 7 is the bottleneck
country and its proposal is laid down in the treaty.
For the non-signatories it is assumed that they choose their initial non-
cooperative emission level, eiN. Since non-signatories have an incentive to
increase their emissions above eiN as a result of the higher abatement efforts
of signatories, it has to be checked whether signatories can control the ‘free-
riding’ of non-signatories through suitable punishments.
From this it follows that the stability of an IEA comprises two compo-
nents:
Instrumental choice
Potential signatories choose the instrument employed in the IEA according
to the majority decision rule (MDR). For instance, if N 10, IJ {7, 8, 9,
10} and preferences are given by 7: Q, 8: Q, 9: Q and 10: T, this coalition of
Coalition models II 271
1 1125
Enlargem.
yes:no 1:0 2:0 3:0 4:0 5:0 6:0 3:4 2:6 0:9 —
Countries in favor
of a quota 9 8 7–8 6–7 5–7 4–7 3–10 2–10 1–10
Countries in favor
of a tax 10 9–10 9–10 8–10 8–10 8–10
Instrument chosen T T T T T Q Q Q Q
Note: a The example assumes N10, b 30, c1 and d10.
274 Game theory and international environmental cooperation
From the table it is evident that in the quota regime all potential signa-
tories favor an enlargement of the coalition until a coalition size of eight
countries is reached, including countries 3–10. If eight countries participate
in the IEA, only three countries are in favor of an enlargement. Therefore,
according to the UDR, the equilibrium coalition size comprises eight coun-
tries (shaded column).
Up to a coalition size of five countries the reduction level proposed by
country N0 1, rN0 1, increases. In this range ambitious abatement targets
are proposed by country N0 1 because the marginal benefits from a joint
emission reduction are high. Then, for a coalition size above five countries,
the proposal of country N0 1 decreases. Nevertheless, all potential signa-
tories benefit from an enlargement until a coalition of eight countries has
formed. In this range, the gains to the potential signatories from an addi-
tional contributor are stronger than the (negative) effect of a lower joint
abatement target. If more than eight countries were to join, then the reduc-
tion target would become too low for countries with high environmental
damage and a majority of countries would be against an enlargement.
In contrast, in the tax regime the equilibrium coalition size is not eight
but seven countries. In this example country 4’s tax proposal, t4 165.5
cannot be backed by WRPE strategies. Therefore, it is assumed that the
maximum stable tax rate, tmax 165 will be applied in the agreement. Since
countries 5–10 prefer a coalition of seven countries (which then includes
country 4) to a coalition of only six countries, though only the low tax rate
tmax is applied country 4 is asked to join.
For other parameter constellations Table 14.4 summarizes the relevant
information which will be discussed in Sub-section 14.4.4.
So far the policy regime has been assumed to be exogenously given and the
equilibrium coalition size has been derived for each regime. Now, the choice
of the policy regime within the negotiation process is also considered. This
choice will affect the equilibrium coalition size and the abatement target
agreed upon in an IEA as well.
Generally, countries may hold different views regarding the instrument
which is applied in an IEA. There is a tendency for countries with low index
numbers to prefer the emission quota because for a given reduction level
they have to carry a lower reduction burden than under the tax regime.12 In
contrast, countries with high index numbers prefer a tax because this leads
to an emission allocation which favors them.
However, since the instrumental issue is linked to the other two issues,
countries with high index numbers might also favor a quota if this implies
Coalition models II 275
14.4.4 Results
In Table 14.4, the equilibrium number of signatories (if all three issues are
decided simultaneously) has been marked bold in the column headed
‘Coalition size’. The equilibrium instrumental choice resulting from the
coalition formation process is listed in the last column.
From this table five main results can be derived:13
1. The relative coalition size (* N*/N ) is not very sensitive to changes
in the parameter values, but decreases in N and increases in . This
implies that the number of signatories is lower for the ‘critical’ param-
eter values (N large and small).
2. The degree of optimality of the sub-coalition, DOSC, decreases in .
3. The degree of optimality of the sub-coalition, DOSC, is higher than
that of the grand coalition, DOGC, if the LCD decision rule is applied.
The difference is particularly distinct for large N.
4. For the critical parameter values (N large and small) the degree of
optimality of the sub-coalition, DOSC, is higher than that of the grand
coalition which chooses the maximal backable emission reduction
target, DOGCmax (for example, N20 and 100).
5. In most cases the quota regime emerges from the decision process. Only
if N is small may a tax regime be chosen among signatories (for
example, N5 and 10).
The first result confirms the findings of the conjectural variation models
of Chapter 13 (see Propositions 13.2 and 13.3). The second, third and
fourth results underline the rationale behind the formation of sub-coali-
tions. Even though the signatories are not concerned with global welfare
but only with welfare in their countries, agreements based on compromises
(LCD decision rule) might be more efficient if they were signed among a
small group of relatively homogeneous countries instead of if a grand
coalition is formed. In the light of possible high transaction costs, this argu-
ment receives even more momentum. In particular, for the critical param-
eter range of low values of and large N the formation of a sub-coalition
seems particularly attractive. The rationale for the fifth result is the follow-
ing: according to (14.1) a low value of N implies a low variance of damage
between countries and therefore also a low variance of payoffs. Therefore,
the allocation of payoffs from a joint abatement policy is also relatively
even under the tax regime for small N. Consequently, a majority of coun-
tries votes for a tax in order to capture the efficiency gains associated with
this instrument.
However for large N, the interests of countries are heterogeneous, which
Coalition models II 279
1. Abatement targets of most IEAs are rather low since otherwise the free-
rider incentive cannot be controlled by the signatories. It was shown
that a grand coalition is very unlikely if signatories aim at a globally
optimal abatement target or if the median country proposal of a
uniform emission tax or a uniform emission reduction quota is applied
within the agreement. However, even if more moderate abatement
targets are laid down in the agreement, for example, the smallest tax or
quota proposal among the N countries (LCD decision rule), stability
is still a problem whenever an environmental problem is of a real global
character (N large) and/or environmental damages are high compared
to abatement costs (low ).
2. Due to the above-mentioned stability problems, it is more likely that
only some environmentally conscious countries would form a sub-
coalition. For the initiator countries a smaller coalition may have an
advantage in that more demanding abatement targets can be realized.
Particularly for the critical parameter values (large N and small ) it
was shown that a sub-coalition may be more efficient than a grand
coalition.
3. The analyses of the grand and the sub-coalitions suggested many
explanations as to why the quota may be so popular in international
280 Game theory and international environmental cooperation
Fourth, the present model derives endogenously the choice of the policy
instrument to achieve an abatement target. The model allows us to explain
why a uniform emission quota has found such widespread application in
many IEAs. In contrast, the conjectural variation and core models have
nothing to say on this issue.
Nevertheless, the present model is far from perfect with respect to the
theoretical requirements of an ideal coalition model as laid out in the
Introduction of Chapter 13. First, in the tradition of all models assuming
heterogeneous countries (except Bauer 1992), the analysis had to rely on
simulations. However, this disadvantage was compensated by considering
a large set of parameter values and various decision rules about how the
abatement targets (seven different emission vectors), the coalition size
(grand and sub-coalition) and the instrumental choice (UDR and MDR)
are agreed upon in the coalition, which delivered very robust results.
Next, in the tradition of the conjectural variation models, only the exis-
tence of one but not of several sub-coalitions was considered.
Finally, the punishment options were restricted to the emission space. A
more realistic assumption would be to extend the punishment strategy
space to other policy fields in the tradition of issue linkage models.
However, in the case of many signatories the question arises whether simul-
taneous membership of all signatories in two or more agreements is a realis-
tic assumption (see Sub-section 13.2.7). Moreover, governments may also
hesitate to punish a country using other policy issues since this may jeop-
ardize the possibilities for cooperation on those issues too. Taking these
qualifications into account the assumptions of the present model seem less
restrictive and, in fact, may reflect the limited punishment options in inter-
national relations quite well. This is even more true when recalling that the
present model allows us to take account of various limitations of punish-
ment profiles (see Chapter 12, in particular Sub-section 12.2.4, and Sub-
section 14.3.4 of this chapter).
NOTES
1. This chapter draws heavily on Finus and Rundshagen (1998a).
2. See Chapter 8 for a justification of this assumption.
3. See Hoel (1992a) for a similar specification of the payoff function. Note that all qualita-
tive results obtained below also hold if damages are not divided by N. Thus, 1/N is only
a scaling factor. See Finus and Rundshagen (1997).
4. All results mentioned in this section are derived in Appendix XI.1.
5. The term second-best world with respect to result 4 is used because in a first-best world
the stability problems of an agreement are usually neglected.
6. This definition implies that the probability that the game continues is assumed to be 1
(see Section 5.1). Note that i → 1 is equivalent to i →0.
7. In the following, transaction costs are not modeled since their effects are rather obvious
282 Game theory and international environmental cooperation
15.1 INTRODUCTION
283
284 Game theory and international environmental cooperation
strategy against other coalitions (second stage). The idea corresponds to the
Nash–Cournot assumption of the conjectural variation models encoun-
tered in Chapter 13. How these models differ is in the first stage, in which
players decide on their membership. This issue will be discussed below.
In a non-cooperative game theoretical context a compact device to
analyze the coalition formation process is the equilibrium valuation, also
called a per-membership partition function. It is defined as follows (see also
Bloch 1997):2
First, note that the definition assumes away one of the original problems
of endogenous coalition formation, namely how the gains from coopera-
tion are divided among coalition members. That is, the equilibrium valua-
tion assumes a fixed sharing rule. Though for symmetric countries the
assumption of an equal sharing rule seems quite plausible (see the
Introduction to Chapter 13), such an assumption (like any other assump-
tion) is exogenous to the model.3 However, since almost all general results
which have been obtained so far by applying ‘new’ concepts to the problem
of coalition formation rely on the assumptions of ‘ex ante symmetric
players’4 and an equal sharing rule, we shall make these assumptions for
simplicity in this chapter, too.5
Second, note that a similar critique can be raised with respect to the
assumption of the choice of the strategies within the coalition. Again, for
symmetric countries it is plausible that the coalition members will max-
imize joint welfare (see the discussion in Chapter 13). It is also plausible
that a coalition will play a Nash strategy toward outsiders since this implies
that a Nash equilibrium is played in the final stage of the finite game which
is a necessary prerequisite for a strategy to be subgame-perfect. However,
as pointed out in Chapter 13, this implies either that players are assumed
to abide by the rules of the coalition or that the instant reactions of players
Coalition models III 285
C1: *i (ci , c) *i (ci , c) where ci c, c and c\ci can be derived from c\ci
by merging coalitions in c\ci.
C3: *i (ci, c)*i (ci\{k}, c) where cc\{ci , cj}{cj {k}, ci\{k}}
and cj ci 2.
C4: *k(ci , c) *k(cj {k}, c) where cc\{ci , cj}{cj {k}, ci\{k}}
and cj ci 2.
Proposition 15.1
The symmetric global emission game with payoff functions (13.1) in
which the coalition maximizes the aggregate payoff of its members and
288 Game theory and international environmental cooperation
Proposition 15.2
In the symmetric global emission game with payoff functions (13.1) in
which the coalition maximizes the aggregate payoff of its members and
plays a non-cooperative strategy toward outsiders (Nash–Cournot
assumption) the largest coalition in a coalition structure which is stand-
alone stable comprises no more than three coalition members.
Proof: Is straightforward and therefore omitted.
From the definition it follows that the payoff vector of an SNE must be
Pareto-efficient with respect to the entire payoff space since otherwise a sub-
group of countries or the coalition as a whole would jointly alter their strat-
egies. From this it is immediately evident that no SNE exists in a standard
type of global emission game. This is so since the Nash equilibrium (for
example, payoff function of types 1 and 3) or Nash equilibria for example,
payoff function of type 2 in the case of symmetric countries) are inefficient
in positive externality games (as defined in this chapter).
As has been argued already in the two-player context (see Chapters 6 and
7), requiring equilibrium strategies to be Pareto-efficient with respect to the
entire strategy space may be an unnecessarily restrictive condition. In the
N-player context, the SNE concept is fraught with an additional weakness.
Deviation by a sub-group of countries is deemed to be feasible even though
the deviation itself may be subject to further deviations which may lead to
an outcome which is Pareto-dominated by the original agreement. It is this
very restrictive definition of a stable equilibrium that is responsible for no
SNE existing in many games.
The CPNE concept takes up this concern and rules out such non-credible
deviations. It allows an equilibrium to be challenged only by self-enforcing
deviations. That is, efficiency is only required with respect to the set of self-
enforcing agreements.12
The basic idea of a CPNE may be illustrated with the following example
(Bernheim et al. 1987). Suppose there are four players in a room who agree
on some strategy vector. Then one player, say 4, leaves the room and hence
cannot change his/her strategy any more. Given player 4’s strategy, the three
remaining players may now think about altering their strategies. Again, one
player leaves the room after the agreement and cannot change his/her strat-
egy any more. The process continues until one player is left in the room. The
original agreement among the four players is a CPNE provided that, at any
stage of the process, no sub-group of players left in the room wishes to
290 Game theory and international environmental cooperation
Thus every sub-group of players plays a CPNE against the other players.
From the definition it is evident that every SNE is a CPNE, too; however,
the opposite does not hold. Hence, SSNE SCPNE SN.13,14
It is important to note that the CPNE imposes not only a restriction on
the deviation strategies (which must be self-enforcing) but also on the set of
players who deviate. Only among the players who originally deviated are
further deviations considered. That is, if players 1 and 2 deviate, only these
players may further deviate, but not players 1 and 3 for instance. Though
this a conceptual weakness of a CPNE, it reduces the number of deviations
to consider and therefore eases computations.
In a (static) global emission game the NE is also a CPNE. By the defini-
tion of an NE, no single country wants to deviate. Moreover, any sub-group
of players which jointly deviated would again be subject to a further devi-
ation. This follows simply from the fact that any such deviation implies an
emission tuple which does not simultaneously lie on the reaction function
of all deviators, implying an incentive to further deviations by at least one
player.
To highlight the difference between the CPNE and the SNE concept con-
sider the following simple matrix game. Each of three countries has two
actions ai and nai where country 1 chooses rows, country 2 columns and
country 3 matrices. In this game countries 1 and 2 may be small countries
whose interests more or less coincide, whereas country 3 is a big country.
Coalition models III 291
a3 na3
a2 na2 a2 na2
a1 3, 3, 9 0, 1, 4 a1 5, 5, 10 0, 1, 2
na1 1, 0, 4 4, 4, 6 na1 1, 0, 2 6, 6, 0
From the payoff structure it is evident that i (a1, a2, na3) i (na1, na2, a3)
i I and therefore one should expect at first thought that countries would
play (a1, a2, na3). However, this NE is upset if countries 1 and 2 coordinate
their strategies and jointly deviate to (na1, na2). It is evident from
Matrix 15.1 that no further deviations will occur from (na1, na2, na3) and
hence this deviation is a credible challenge to the NE (a1, a2, na3).16
Consequently, (a1, a2, na3) is not a coalition-proof equilibrium. In contrast,
the (Pareto-inferior) NE (na1, na2, a3) is coalition-proof since no group of
two players has an incentive to deviate jointly. Also the group of all coun-
tries will not deviate, though it has an immediate incentive to move to the
Pareto-superior NE, since this deviation would be subject to a further devi-
ation to (na1, na2, na3). Since the payoff vector resulting from the play of
(na1, na2, na3) does not Pareto-dominate the NE (na1, na2, a3), the deviation
of the group as a whole is not credible in the sense of a CPNE.
Note that in this game no SNE exists. Only the Pareto-efficient NE (a1,
a2, na3) qualifies as a potential candidate; this, however, is subject to a joint
deviation by players 1 and 2, as argued above.
We saw above that the two-stage coalition formation game may be ana-
lyzed in a reduced (one-stage) form (since payoffs are received at the end of
the second stage and equilibrium strategies of the second stage are fixed
and known in the first stage). Thus, for our purposes, the definition of an
292 Game theory and international environmental cooperation
N = 5: {{1, 2, 3}, {4, 5}}, {{1, 2}, {3, 4}, {5}} N = 5: no equilibrium
Core stability N = 3: {{1, 2, 3}}, {{1, 2}, {3}}
N = 4: no equilibrium
N = 5: no equilibrium
stability N = 3: {{1, 2, 3}}, {{1, 2}, {3}}
N = 4: {{1, 2, 3, 4}}, {{1, 2, 3} {4}}
N = 5: {{1, 2, 3, 4, 5}}, {{1, 2, 3, 4}, {5}}
stability N = 3: {{1, 2, 3}}, {{1, 2}, {3}}
N = 4: {{1, 2, 3, 4}}, {{1, 2, 3} {4}}
N = 5: {{1, 2, 3, 4, 5}}, {{1, 2, 3, 4}, {5}}
Farsighted equilibrium N = 3: {{1, 2, 3}}, {{1, 2}, {3}}
N = 4: {{1, 2, 3}, {4}}, {{1, 2}, {3, 4}}
N = 5: {{1, 2, 3, 4, 5}}
Note: a Assumption: payoff function (13.1) and symmetric countries.
294 Game theory and international environmental cooperation
Proposition 15.2
In the symmetric global emission game with payoff functions (13.1) and
where the coalition maximizes the aggregate payoff of its members and
plays a non-cooperative strategy toward outsiders (Nash–Cournot
assumption) a CPNE coalition structure implies for any N5 (a) [N/3]
coalitions of size 3 if R0; (b) [N/3] coalitions of size 3 and one coali-
tion of size 2 if R2; or (c) [N/3]1 coalitions of size 3 and two coali-
tions of size 2 if R1 where R Nk·3, k[N/3]. This equilibrium is
unique. There is no SNE equilibrium coalition structure for N4.
Proof: Since the proof is straightforward using results of the discussion
above, conditions C1–C4, Proposition 15.2 and results provided in
Appendix XII.2, it is omitted here.
games
In games all players announce which coalition they want to belong to,
including the list of all participants of this coalition (Hart and Kurz 1983).
In order for a coalition to form, the unanimous agreement of all players is
required. More precisely, in the game the message space of country i is a
set of coalitions to which it wants to belong Mi {Ci N, i Ci , ci Ci}. A
coalition ci forms if and only if all members i of ci have announced mi ci.
This definition implies that, whenever a member deviates and leaves the
coalition, the whole coalition breaks apart. Due to this assumption, the set
of NE is very large (for example, for N5 it comprises all permutations, see
Appendix XII.4). For instance, if each country announces a singleton coali-
tion, this announcement is immune to a deviation. This is so since an NE
considers only single deviations, and forming any other coalitions requires
a change of at least two messages. Thus, in games the equilibrium
296 Game theory and international environmental cooperation
games
In
games this last-mentioned concern is taken up (Hart and Kurz 1983).
As in games, all members announce a coalition to which they want to
belong and a list of coalition members. In contrast to games, however, in
games the participation of all listed members is not needed for a coali-
tion to form. Thus, as in the open-membership game, the message space is
used as a coordination device (Bloch 1997, pp. 322ff.). However, member-
ship is exclusive because only coalition structures which have been
announced can eventually form. The definition implies that, whenever a
member leaves the coalition, it assumes that all other coalition members
remain together (as in the conjectural variation models in Section 13.2).
For instance, consider N4. Now, in contrast to the open-membership
game, {{1, 2, 3},{4}} is an NE-coalition structure. Though any of the coali-
tion members in the larger coalition would like to join the singleton coali-
tion, this is not in 4’s interest. Since unanimous agreement is required by
the exclusive membership rule, and 4 will never propose {{1, 2},{3, 4}},
{{1, 2, 3}, {4}} is stable.
From Table 15.1 it is evident that according to the CPNE concept more
than one equilibrium coalition structure exists. (See Appendix XII.5 for a
derivation of the equilibria.) For N 4 the grand coalition is no CPNE and
the coalition structure comprises several smaller coalitions. As in the open-
membership game, the reason is that any CPNE in a
game must be an NE
and an NE must be stand-alone stable. Thus, no coalition structure will be
an equilibrium involving coalitions of size greater than 3.
Obviously, the coalition structure in the
game allows for finer equilib-
rium coalition structures than the open-membership game. The reason is
the following. Consider – with a slight abuse of notation – the coalition
structure c{2, 2, ..., 1} where the numbers indicate the size of each coali-
tion. Then any of the coalitions of size 2 would like to merge with the sin-
gleton coalition. Since, however, the singleton member is indifferent to such
a move, the coalition structure above is a CPNE in the
emission game. In
contrast, in the open-membership game such a merger could be enforced.
Coalition models III 297
Again, from ci3 it follows that for N5 no SNE exists, confirming
our reservation against the SNE concept.
Thus, taken together, the coalition formation process in
games seems
quite plausible. Coalitions are not resolved once a country deviates and
exclusive membership seems a plausible assumption in the global emission
context.20
From the definitions it is evident that, in general, deviations occur more fre-
quently under core stability, than under stability and under stability
more often than under stability. Consequently, if CC, C and C denote
the set of core-, - and -stable coalition structures, CC C C holds
(Kurz 1988). In fact, it is easily checked for the global emission game
example that C C holds.
As it is evident from Table 15.1, a core-stable coalition structure for N4
does not exist. (See Appendix XII.6 for a derivation of the equilibria.) This
is not surprising since, given the optimistic conjectures about external
players, there is always a coalition structure for which it is beneficial for a
group of players to deviate. As under the SNE concept no restriction is
imposed on such a deviation to be immune to further deviations.
Under and stability either the grand coalition or one large coalition
and a singleton coalition is stable. The reason is that any deviation triggers
further deviations, leading to small partitions which are Pareto-dominated.
Again, no restriction is imposed on such a deviation to be immune to
further deviations, which allows us to stabilize large coalitions.
Thus, an obvious weakness of all three stability concepts is the ad hoc
assumption regarding the behavior of external players. Neither the opti-
mistic assumption under core stability nor the pessimistic assumption
under and stability is usually a best reply for external players. Thus,
these behavioral assumptions are rather unrealistic and confirm our reser-
vations against the core concept, as has been raised already in
Section 13.3.
It is important to note that at each stage of the deviation process, all players
who currently deviate must prefer the new coalition structure. It does not
suffice if the original members would like a further deviation, but not all
the members of the present coalition of which the original deviators are
members.
As laid out in Appendix XII.7, to determine the largest consistent set,
first, a list of all possible permutations is needed. Second, a preference rela-
tion between different coalition structures must be established. Third, a
coalition structure which is not obviously Pareto-dominated by another
coalition structure is picked and any possible deviation and ‘chain devia-
tions’ are tested for.22
Obviously, the farsighted coalitional stability concept introduces some
interesting strategic aspects into the coalition formation process which are
absent from the concepts discussed so far. For instance, consider N5, for
which the grand coalition is stable (see Table 15.1). Since the grand coali-
tion is not stand-alone stable, a country, say 5, has an incentive to leave the
grand coalition, that is, {{1, 2, 3, 4, 5}}⇒{{1, 2, 3, 4},{5}}, which is subject
to a further deviation, say by country 4, that is, {{1, 2, 3, 4},{5}}⇒{{1, 2,
3},{4},{5}}. Since {{1, 2, 3},{4, 5}} is strictly preferred by all players to
{{1, 2, 3},{4},{5}}, this would be, potentially, the final coalition structure.
However, since {{1, 2, 3, 4, 5}} strictly Pareto-dominates {{1, 2, 3},{4, 5}},
this chain deviation is not valid and therefore the grand coalition is consis-
tent.
In contrast, for N4, the grand coalition is not consistent since {{1, 2,
3, 4}}⇒{{1, 2, 3},{4}} and {{1, 2, 3},{4}} is not dominated by any other
coalition structure. {{1, 2},{3, 4}} is also stable. Though {{1, 2},{3, 4}} is
directly dominated by {{1, 2, 3, 4}},{{1, 2, 3, 4}} is dominated by {{1, 2,
3},{4}} and hence for {{1, 2},{3, 4}} to be not consistent it must be indi-
rectly dominated by {{1, 2, 3},{4}}. This is, however, not the case since not
all countries (here it is country 3) forming a grand coalition from {{1, 2},{3,
4}} benefit from coalition structure {{1, 2, 3},{4}}.
Note that {{1, 2, 3},{4}} is only consistent by the definition of a strict
Pareto-dominance. Since country 3 would prefer to join coalition {4} one
could think of the following scenario: {{1, 2, 3},{4}}⇒{{1, 2},{3},
{4}}⇒{{1,2},{3,4}}, where the final coalition structure is stable according
to the discussion above. That is, country 3 leaves the coalition to become a
300 Game theory and international environmental cooperation
singleton and thereby creates an incentive for country 4 to accept its mem-
bership. Since the first move leaves the payoff to country 3 unchanged
(though the second move implies an increase in its payoff ), this chain devi-
ation is ruled out by Chwe’s definition.
On the one hand, requiring only weak dominance would produce a
unique equilibrium in the example, and therefore the above definition seems
unnecessarily restrictive. On the other hand, this definition avoids cyclical
deviations in other examples and may therefore be justified.
Taken together, the farsighted coalitional stability concept appears the
most advanced among the simultaneous move concepts. It takes the strate-
gic considerations of players fully into account. It therefore satisfies the
condition of a rational conjectural variation equilibrium at every state (see
Chapter 10, Definition 10.1), and may therefore be regarded as a convinc-
ing static representation of dynamic games. It is a further development of
the farsighted concept mentioned in Sub-section 13.2.8 in that the coexis-
tence of several coalitions is in addition considered. The strict Pareto-dom-
inance at each stage of a deviation process implies de facto exclusive
membership, for which we have indicated some sympathy already. In con-
trast to the and
games, the FCS neither assumes in an ad hoc fashion
that the coalition breaks apart nor that the remaining coalition structure is
preserved following a deviation but that the external players play a best-
reply strategy. Thus, reactions to a deviation are consistently modeled.
Therefore, it seems promising to apply the FCS concept to a more general
model of coalition formation (such as payoff function (13.3) and heteroge-
neous countries and so on) in future research.23
Ray and Vohra (1997) motivate their equilibrium binding agreement (EBA)
with the help of the following story behind a coalition formation process.
Initially the grand coalition gathers. Then some leading perpetrators may
propose a different coalition structure if this is in their interest. The perpe-
trators split up to form a coalition, say ci , by themselves. In a next step,
either the coalition ci or the coalition C\ci may be subject to further devia-
tions. Those countries which initiate further deviations are called secondary
perpetrators. The process of disintegration continues until a coalition
structure has been reached where no country wishes to split up into finer
partitions. From this brief introduction it is already apparent that the term
equilibrium binding agreement is a misnomer: it implies that the EBA
Coalition models III 301
Due to the recursive definition and no further selection criteria, the set
of EBA is very large. For instance, for N4 {{1},{2},{3},{4}} is stable by
definition. {{1, 2},{3},{4}} is stable since it Pareto-dominates the finer par-
tition {{1},{2},{3},{4}} which is stable. {{1},{2, 3, 4}} is stable since it
Pareto-dominates {{1},{2},{3},{4}} and so on. By continuing this recur-
sive process, it is evident that the set of EBA is quite large. Thus, it seems
sensible to introduce a selection device to reduce the number of equilibria.
An obvious criterion follows by recalling the story at the beginning of this
sub-section to illustrate the EBA concept. By assuming that the coalition
formation process evolves from the grand coalition, fine partitions are
Pareto-dominated by coarser ones.25 Therefore, in Table 15.2 only
CEBA* Eff(CEBA) are displayed. Though the original concept of Ray and
Vohra does not have this selection device, we take the illustrative story as
302 Game theory and international environmental cooperation
an important and distinct feature of the formation process and have there-
fore placed this concept in the context of sequential move coalition models.
From Table 15.2 it appears that rather large coalitions can be supported:
either the grand coalition or a coalition comprising N1 countries and a
singleton coalition. (See Appendix XII.8 for a derivation of the equilibria.)
In particular for large N large coalitions are supported by the unfavorable
prospect of further deviations following a deviation. Thus, the set of effi-
cient EBA contains large coalitions if coalition structures of intermediate
coalition sizes are not stable.
To summarize, the EBA concept is convincing regarding the behavior of
players following a deviation. The assumption that only players belonging
to the same coalition can jointly deviate seems rather restrictive as in the
CPNE concept. However, in one respect the EBA concept is even more
restrictive than the CPNE concept in that only finer but not coarser parti-
tions can be formed by perpetrators. This is an obvious disadvantage.
The original idea of the sequential move unanimity game (SMUG) goes
back to Bloch (1995, 1996). The game is in the spirit of Rubinstein’s (1982)
Coalition models III 303
remaining players will form? The answer to the latter question will of
course affect the proposal at the initial stage. Thus, an initiator must solve
the entire game backward for all players to find his/her best strategy.
We define now more formally the SMUG and the equilibrium concept.
For this we need the following definitions. Some of these definitions have
already been encountered above but are listed here for convenience.
Let the set of players be denoted by I, i I. A coalition ci is a non-empty
sub-set of players. A coalition c is a partition on the set I and the set of all
coalitions is denoted C. For any sub-set of players K of I, the set of parti-
tions on K is denoted by CK with typical element cK. *i (cj , c) is the payoff
to player i in coalition j and coalition structure c.
Bloch (1996) has shown that for symmetric valuations a simple finite pro-
cedure can be used to determine the equilibrium/equilibria in the game. The
algorithm works as follows. The first player proposes an integer c1 [1, N ]
which indicates the coalition size. The player c11 proposes a coalition of
size c2[1, N c1]. This process continues until c1c2... cMN.
From Table 15.2 it appears that the SMUG according to Bloch does not
produce a unique equilibrium in the emission game. The reason is that in
this symmetric game players face an indifference between two strategies (see
Appendix XII.9) at some stages. Therefore, two equilibria can be sup-
ported.
It is also apparent from the example that rather large coalitions can be
supported. However, for a larger number of countries this may be different.
In the extension considered below a more exact statement is possible.
Ray and Vohra (1999) have generalized the sequential formation game of
Bloch. In particular, in contrast to the previous concepts which assume a
fixed sharing rule, they have endogenized the sharing rule. Since their exten-
sion is quite involved, we shall not pursue this issue further and consider
only a slight modification of Ray and Vohra. They assume that in the case
of indifference between two strategies a country selects the largest coalition
size. In the emission game this assumption seems plausible since all other
players (except the player who has the move) would prefer the larger coali-
tion to form. (For details, see Appendix XII.9.) This selection device pro-
duces a unique equilibrium coalition structure which can be characterized
according to Bloch (1997, pp. 338ff.) by a Fibonnaci decomposition. A
Fibonnaci decomposition is derived from a sequence of Fibonnaci
numbers where f0 1, f1 2 and fn fn1 fn2. One starts by choosing the
largest Fibonnaci number equal to or smaller than N. Denote this
Fibonnaci number f nk. Then one looks for the largest Fibonnaci number
equal to or smaller than Nf nk. This process continues until f ikN.
For instance, consider the Fibonnaci numbers 1, 2, 3, 5, 8, 13, 21 and so
on. Thus if N4, the largest Fibonnaci number is 3 and N 31. Hence,
with some abuse of notation, the coalition structure is {3, 1} where the
numbers indicate the coalition size. If N20 the coalition structure would
be {13, 5, 3} and for N8 the grand coalition forms. Thus, whenever N is
not a Fibonnaci number rather asymmetric coalition structures form.
In contrast to Ray and Vohra, however, it may well be argued that a
player, though s/he may not bother about exhibiting a positive externality
on other players by choosing the larger coalition size if s/he is indifferent
between two strategies, s/he may be concerned about another issue.
Suppose there is a slight probability that a player, say k, who follows a
player say, i, in the sequence of moves, unintentionally makes a small error
by choosing a wrong strategy. In the terminology of Selten (1975) a player
306 Game theory and international environmental cooperation
may make a small tremble.29 Given this possibility, it might be in the inter-
est of player i to choose a strategy which is immune to such an error.30
For instance, suppose N3 and country 1 proposes the grand coalition.
Alternatively, it could propose a singleton coalition, knowing that it is in
the interest of country 2 to propose a coalition comprising countries 2 and
3 which country 3 will accept, that is, *2/3({{1}, {2, 3}}) *2/3({{1}, {2},
{3}}). According to Bloch, countries 2 and 3 will accept the proposal of a
grand coalition since *2({{1, 2, 3}})*2({{1,3}, {2}}) and *3({{1, 2,
3}}) *3({{1,2}, {3}}). However, suppose that, say, country 2 rejects the
proposal by mistake and proposes itself as a singleton coalition, after which
countries 1 and 3 will form a coalition. Since *1({{1, 2, 3}})*1({{1},
{2,3}}) *1({{1, 3}, {2}}) country 1 may be on the safe side by proposing
itself as a singleton coalition instead of the grand coalition, and the result-
ing equilibrium coalition structure is {{1}, {2, 3}}.
As is evident from Table 15.2, this procedure also delivers a unique equi-
librium in the example. In fact, denoting Bloch’s equilibrium set by CSE(B),
that of Ray and Vohra CSE(V/R) and the trembling hand equilibrium by
CSE(F), it is easy to see that CSE(B) CSE(V/R) CSE(F).
An obvious advantage of the SMUG is that the formation process is
explicitly modeled. In contrast to the simultaneous move games, there is an
explicit story of how a coalition builds and how players coordinate on an
equilibrium. It seems plausible for many games that there is an initiator at
the beginning of the game who proposes a coalition which is in his/her
interest. This is particularly true in asymmetric games. For instance, in an
asymmetric emission game it should be expected that governments with a
higher environmental preference will approach other governments with a
similar high preference. Governments with a lower preference may form
coalitions among themselves. Membership is only offered to outsiders if all
coalition members unanimously agree. Thus, Bloch’s framework is the
most sophisticated among the sequential move coalition formation models.
It seems promising for future research to derive the coalition structures in
an asymmetric emission game by applying his concept.31
A first step in this direction has been undertaken by Finus and
Rundshagen (1999). The authors investigate an issue linkage game which is
a mix of a positive and negative externality game. The positive externality
part of the game concerns the reduction of global emissions emanating
from the production of a homogeneous good which is sold in the domestic
market and exported abroad. The negative externality part of the game
concerns the formation of a customs union where the members abolish
tariffs within the union but protect their market against outsiders. Since the
authors assume that firms may migrate to other countries if the environ-
mental policy of governments is too strict, the endogenous plant location
Coalition models III 307
NOTES
1. Bloch (1997) and Yi (1997) provide an excellent survey of most concepts discussed in this
chapter. They have applied these concepts to the provision of a public good. Their
example could be interpreted as a global emission game defined in reduction space.
2. The equilibrium valuation is, apart from the normal and extensive form, a third possibil-
ity for representing a non-cooperative game. It may also be called the coalitional repre-
sentation of games. It corresponds to the characteristic function in cooperative games
(see Section 13.3).
3. So far not much progress has been made in endogenizing the sharing rule in the coali-
tion formation process. An exception is Ray and Vohra (1999). They provide some vin-
dication for the equal sharing rule if countries are symmetric.
4. This term refers to the fact that, although all players have the same strategies and payoff
308 Game theory and international environmental cooperation
spaces at the beginning of the play, asymmetric coalition structures may ex post emerge
during the formation process.
5. Obviously, for symmetric countries any kind of transfers are therefore not relevant to the
coalition formation process. This is at least true as long as one assumes that transfers can
only be paid among coalition members but not to outsiders. We shall make this assump-
tion in what follows.
6. It is interesting to note that in neither of the papers quoted below is this inconsistency
recognized.
7. Note that this definition does not imply that the coalition structure itself is an equilib-
rium. It only implies that a coalition maximizes the payoffs of its members for a given
coalition structure.
8. Coalition games without spillovers are discussed in Konishi et al. (1997).
9. To avoid confusion: so far we have called transboundary emissions a negative external-
ity. Now, in this chapter it is more convenient to keep with the tradition of Yi (1997).
Hence, we call an emission game a positive externality game since we think in terms of
emission reductions. A deviation from Nash equilibrium emissions by a coalition implies
an emission reduction.
10. Conditions C1–C3 apply in general to positive externality games. Condition C4 may not
hold in general but is true for payoff function (13.1). A general proof may be found in
Yi (1997).
11. We state the proposition with respect to payoff function (13.1) since the subsequent dis-
cussion is illustrated by this example. This procedure saves space, though a more general
proof for a larger class of payoff functions is certainly possible but more involved. See
Yi (1997) for a proof in a game on the provision of a public good.
12. Thus the CPNE is identical to a renegotiation-proof equilibrium in a two-player game.
See Chapter 6.
13. Unfortunately, there are no conditions which guarantee the existence of an SNE or a
CPNE.
14. For an extension to cover correlated strategies also, see Moreno and Wooders (1993).
15. One reason may be that some consumers prefer to buy an environmentally friendly car
and some consumers prefer a cheaper car without a cat. Thus the market is segmented
and makes heterogeneous production attractive.
16. Of course, country 3 would like to deviate to a3 to gain a payoff of 6 instead of 0.
However, since player 3 is not among the original deviators, this deviation is not consid-
ered by the CPNE concept. See the discussion on p. 290.
17. Of course, strictly speaking a dynamic definition would be required. However, for con-
venience we follow the somewhat sloppy notation and definition of Bloch (1997) and Yi
(1997). For an extended definition in a supergame framework, see Stähler (1996). He
applies his definition to a global emission game comprising three countries.
18. Hence, one can immediately conclude that for N 9 {{1, 2, 3, 4}, {5, 6, 7}, {8, 9}} cannot
be an NE coalition structure under the open-membership rule.
19. An application of the open-membership game to the formation of custom unions in
international trade (negative externality game) may be found in Yi (1996).
20. An application to the formation of custom unions in international trade (negative exter-
nality game) may be found in Burbidge et al. (1997).
21. For a more formal definition, see Chwe (1994, pp. 302ff.).
22. Though the FCS concept considers sequential deviations, the coordination of an equi-
librium coalition structure occurs by assumption simultaneously. This is why the concept
is classified as a simultaneous move game.
23. A concept similar to the FS concept has been applied by Ecchia and Mariotti (1997) in
the context of a global emission game.
24. For a full account of the concept the reader is referred to Ray and Vohra (1997).
25 . Alternatively, one could assume that the formation process starts from the singleton
coalition and that coalitions gradually become larger until no further enlargement is
stable.
26. See Chapter 4 for the definition of the extensive-form representation of a game.
Coalition models III 309
27. Note that, since payoffs are received at the end of the game, the game comprises (accord-
ing to the terminology of this book) only one period but of several stages V. See
Chapter 4.
28. Recall that a continuation strategy has been defined in Chapter 4 and applied in
Definition 4.9 to characterize a subgame-perfect equilibrium in a repeated game.
29. This does not imply that this player is irrational but only that s/he may choose a wrong
strategy with probability where lim 0. For details of the motivation of this concept,
see Selten (1975).
30. An issue for future research is to formalize the idea of the trembling hand sequential
equilibrium in the SMUG.
31. Bloch (1995) applied his concept to the coalition formation in Cournot and Bertrand oli-
gopoly where firms may form associations in order to reduce production costs or costs
of R&D (negative externality game). Yi (1996) applies the sequential formation game to
the formation of custom unions (negative externality game).
16. Summary and conclusions
After important terms, the notation of this book and the structure of game
theory had been laid out in Chapter 2, we started out in Chapter 3 by ana-
lyzing four simple matrix games: the prisoners’ dilemma, the chicken, the
assurance and the no-conflict games. The analysis first focused on two
countries only (Sections 3.2–3.5) and was then extended to N countries.
Depending on the cost–benefit structure of an abatement policy, either no
cooperation, cooperation among some countries or cooperation among all
countries could be explained. Though these matrix games are rather simple
by their nature, basic features of the coalition models in Chapters 13–15
could already be depicted.
In Chapter 3 it was also shown that though there is no external coordi-
nator endowed with the power to enforce an IEA, and hence countries must
play a correlated Nash equilibrium, coordination may improve upon the
non-cooperative outcome. This result was demonstrated with the help of
the chicken game. It was pointed out that future research should derive con-
ditions under which coordination is possible and that it should scrutinize
whether it is possible to transform a game such that coordination can be
applied more effectively. Moreover, retrospectively of the coalition forma-
tion models, it seems promising to look at the effect of coordination in an
N-country world. Due to the complexity of this issue, it seems promising
to look first at some of the reduced coalition formation games of
Chapter 15. Due to their simple structure, the simultaneous move games
seem particularly suited for such an undertaking. For instance, in the open-
membership game and the exclusive membership game, one could deter-
mine the set of correlated coalition-proof Nash equilibria and find out
whether large coalitions could be established as a self-enforcing outcome.
Of course, since we established that the farsighted coalitional stability of
Chwe is a particularly convincing concept, it would also be interesting to
investigate whether this concept could be extended to correlated strategies
too.
In Chapters 4–7 the framework was extended to cover dynamic games.
For finitely repeated games Chapters 4 and 6, and for infinitely repeated
games Chapters 5 and 7 traced the historical development of equilibrium
concept refinements. It became evident that the steps in this development
were closely related to more sophisticated definitions of credible threat
310
Summary and conclusions 311
NOTE
1. Bernheim et al. (1987) define their concept of a coalition-proof equilibrium only for static
and finite dynamic games. Farrell and Maskin’s (1989a) concept of a weakly and strongly
renegotiation-proof equilibrium is defined for two countries only or for N countries but
where only single deviations are considered. Stähler (1996) combines both concepts and
adapts them to a supergame framework. However, his analysis is restricted to three coun-
tries only.
Appendices
I CHAPTER 3: APPENDIX
Proof: For the proof of Proposition 3.1 it is helpful to note the following
relations. If two events, say A and B, occur independently with probability
p(A) and p(B), then the probability that they occur jointly is given by
p(A B)p(A)·p(B). If events A and B do not occur independently, then
the following relation applies:
p(AB)
p(A/B) (I.1)
p(B)
where p(A/B) reads as ‘probability that event A occurs given event B
occurred’ (see, for example, Rasmusen 1989, pp.54ff). For instance, in
Matrix 3.9 the probability that strategy a11 is played is z1 z2. Therefore, the
probability that strategy combination a21 is played, given player 1 plays a11,
is therefore p(a21/a11)z1/(z1 z2). Moreover, we find p(a22 /a11)z2 /(z1
z2) and so on.
Now, let payoffs be given by the General Payoff Matrix 3.2 and the cor-
related distribution z by Matrix 3.9. Assume for notational convenience a
symmetric game so that a1 a2 a and so on.1 Then, if the coordinator
recommends country 1 to play a1, its expected payoff is given by the two
LHS terms in inequality (I.2). Alternatively, if country 1 deviates, it receives
an expected payoff represented by the two terms on the RHS of this
inequality. For stability, it is thus required that the expected payoff when
complying is at least as high as when defecting:
z1 z z z ca
a· b· 2 c· 1 d· 2 ⇔z1· z . (I.2)
z1 z2 z1 z2 z1 z2 z1 z2 bd 2
The second possibility to be considered is that the coordinator recom-
mends playing the second strategy to country 1. For this to be an equilib-
rium recommendation the following inequality must hold:
317
318 Appendices
z3 z z z ca
c· d· 4 a· 3 c· 4 ⇔z4 z3· . (I.3)
z3 z4 z3 z4 z3 z4 z3 z4 bd
By the same token, it is straightforward to find the following equilibrium
conditions for country 2:
ca
z1· z (I.4)
bd 3
ca
z4 z2· (I.5)
bd
where we may recall that cabd and 2a bc2d holds in a chicken
game. Moreover:
must hold by definition. Since we look for an equilibrium with a high aggre-
gate payoff, z4 0 is assumed. Thus, equations (I.3) and (I.5) become non-
binding and we are left with two inequalities. Moreover, since strategy
combination (a1, a2) delivers the highest aggregate payoff in this game, z1
should be chosen as large as possible subject, however, to the two con-
straints (I.2) and (I.4). Hence z2 is chosen equal to the LHS term in (I.2)
and z3 equal to the LHS term in (I.4). Substituting this information into
(I.6) and solving for z1, z2 and z3 respectively, gives:
bd ca ca
z1 , z2 , z3 (I.7)
b d 2c 2a b d 2c 2a b d 2c 2a
from which the aggregate equilibrium payoff ( i (zN* )2a·z1 (bc)·z2
(bc)·z3 )
2(bc c2 ca da)
i (z*) (I.8)
b d 2c 2a
follows. In any of the two NE in pure strategies, aggregate payoffs are given
by i (s* )bc and in the mixed strategy equilibria aggregate payoffs
( i ( pN* )2·a·p2 2·(bc)(1p)·p2·d·(1p)2 ) are computed to be:
2(bc da)
i ( pN* ) (I.9)
bdca
which follows from p*1 p*2 p(bd)/(bcad). Then:
2(2a b c)(b d )
i (zN*) i (sN*) 0 (I.10)
2(c a) b d
Appendices 319
2(c a)2(c d )
i (zN*) i ( pN* ) 0 (I.11)
(2c 2a b d )(c a b d )
where we make use of the relations cab d and 2a bc 2d in a
chicken game.
For the example in Matrix 3.3 where a4.6, b2.2, c5.2, d2 we there-
fore have z*1 1/7 and z*2 z*3 3/7 and i(zN*)7.656.
II CHAPTER 4: APPENDIX
s ij if in any ht(0) … ht(t1)(sj,sj ), ht(0) … ht(t1)(sj,sj
k ),
j )
t0 … T, or
ht(0) … ht(t1)(sjj,sj
ht(t* 1) … ht(t1)(s
j , sj
N(1))
tt* 1 … T
i si in t0 (II.1)
si if in any ht(0) … ht(t1)(si,si )
t0 … t*
siN(1) if in any ht(0) … ht(t1)(si,si )
t0 … t*
∧ ht(t* 1) … ht(t1)(sN(1)
i
N(1))
tt* 1 … T
),si
player deviates at any time tt0 prior to t*, that is t0 t*, the threat involves
playing the bad NE strategy combination until the end of the game. Thus,
deviation triggers a punishment until the end of the game.
This punishment strategy is played against player j on four occasions:
1. All players play the agreed equilibrium strategy up to time tt* except
player j.
2. Player k deviates in a previous round and all players punish player k
except player j.
3. Punishment has started already in a previous round and therefore will
be continued.
4. All players play the good equilibrium strategy except player j.
t* T t 01 T
t0
ti *i
tt*1
i i
t N(1)
t0
ti *i ti0 D
i
tt 01
ti N(2)
i
. (II.2)
Rearranging gives:
t* t* T
ti0 *i
t 01
t *
i i
t0 D
i i
t 01
t N(2)
i i (
t*1
t
i
N(1) N(2))
i i (II.3)
t* T
t01
ti( *i N(1)
i N(1)
i N(2)
i ) (
t*1
t
i
N(1) N(2))t0( D *)
i i i i i (II.4)
Appendices 321
or t* T
t 01
ti( *i N(1)
i ) (
t 01
t
i
N(1) N(2))t0( D *)
i i i i i (II.5)
or making use of the formulae given in (4.2) and (4.3) in the text we get:
1 it*t * 1 Tt
0 0
(II.6)
Dividing through by it0 and rearranging leads to:
0 0
i[(1t*t
i )( *i N(1)
i )(1Tt
i )( N(1)
i N(2)
i i i ) (II.7)
)](1i )( D *
or
0 * 0 N(1)
i[ *i N(1)
i t*t
i ( i N(1)
i ) N(2)
i N(1)
i Tt
i ( i N(2)
i )
Di *] D *.
i i i (II.8)
D
i i
*
i if t0 t*. (II.10)
i i i i Tt N(2)
D * N(1) N(2) 0
i ( N(1)
i i )
For large T (II.9) and (II.10) become (see the Annex below):
Di i
*
i min
*i N(1) ∧
iI (II.11)
i i i N(2)
D * N(1)
i
i i
Di i
*
i min
N(1) *i N(2) ∧
i I. (II.12)
i i
D N(2) i i i
That is, the RHS terms in (II.11) and (II.12) constitute upper bounds for
the RHS expressions (II.9) and (II.10). Notice that (II.11) and (II.12) can
generally be satisfied for any *i (s) N(2)
i
i I and i close to 1 and there-
fore i can be a subgame-perfect strategy. It remains to be shown that *i (s)
is the average payoff of such a strategy provided T→ (and i →1). For this
suppose t0, t* and i are given. Further let tR i be the smallest integer value
of Tt* that satisfies (II.9) or (II.10), whichever is appropriate, and define
tR:max tR R
i . That is, t is the minimum time span required at the end of the
game, such that the trigger strategy i is an SPE, that is, TTmin t* tR.
Then, it is possible to choose t* TtR and the fraction of periods in
which the cooperative strategy is played in relation to T is given by:
322 Appendices
T tR tR
t*/T⇔ ⇔1 ⇒1 for T→. (II.13)
T T
Hence, for large T the fraction of periods in which the cooperative outcome
will be obtained approaches 1. Hence, the formula of the average payoff as
given in (4.5) in the text applies (assuming T→ and 1 … T) and
*i (s) is the average payoff to player i from such a strategy.
To adapt the proof for sequential move games the stage game strategies
sN(1)
i and sN(2)
i have simply to be replaced by subgame-perfect strategies
SPE(1)
si and sSPE(2) respectively.
i
Finally, we have to discuss games where N(1) i N(2)
i ( SPE(1)
i SPE(2)
i )
does not hold for all players but where one or some stage game equilibria
are good equilibria for some players but at the same time bad equilibria for
some other players. Then in the last periods Tt* each good equilibrium
must be played at least once, otherwise punishment is not a deterrent. A
typical sequence at the end of play would be (B) it*1, it*2, …, it*m,
(B) (G)
where the superscript B stands for bad equilibrium, G for good equilibrium
and m for the length of the sequence. This implies that if a player deviates,
say, at time t0 t*, punishment might not hurt this player immediately but
m rounds later, because at times t* 1 to t* (m1) a bad equilibrium from
his/her point of view would have been played anyway. Nevertheless,
Theorem 4.4 holds as long as T is large enough.
To see this, note first that for large discount factors a player deviates, if
at all, at time t0 t*. Then, s/he nets a gain of Di i . Punishment involves
*
(II.14)
Annex
To show that for large T (T →) the RHS terms in (II.9) and (II.10) are
bounded from above by the RHS expression in (II.11) if *i N(1) i and by
(II.12) if N(2)
i * N(1). Suppose first that N(2) * N(1) and t0 t* is
i i i i i
true. Then, we have to show that:
Di i
* D
i i
*
(II.16]
D
i i
N(2) D N(2) t*t ( * N(1)) Tt 0( N(1) N(2))
0
i i i i i i i i
for T → or:
0 * 0 N(1)
t*t
i ( i N(1)
i )Tt
i ( i N(2)
i ) 0 (II.18)
0 *
for T →. Since t*ti ( i N(1)
i )0 due to *i N(1)
i by assumption and
0 N(1) 0
i ( i i )→0 for large T (Tt
Tt N(2)
i →0 for T→ because i 1),
0 *
(II.18) is satisfied. If tt0, t*ti ( i N(1)) becomes( * N(1))0,
i i i
and, again (II.18) holds.
Next consider *i N(1)i and t0 t*. Then, we have to show that:
D
i i
*
D
i i
N(2) * N(1)
i i
D
i i
*
(II.19)
D
i i
N(2) t*t 0( *i N(1)) Tt N(2)
0
i i i ( N(1)
i i )
if T → or:
0 * 0 N(1)
*i N(1)
i t*t
i ( i N(1)
i )Tt
i ( i N(2)
i ) 0 (II.21)
0 *
if T→. Note that *i N(1) i 0 by assumption and *i N(1)i t*t
i ( i
0
N(1)
i )0 because i 1. Again, Tt ( N(1) N(2)) approaches zero
i i i
from below. Taken together, this shows that (II.21) holds if t0 t*.
Moreover, if we let t0 →t* in (II.21) we can always find a T large enough so
324 Appendices
0 * 0 N(1)
that *i N(1)
i t*t
i ( i N(1)
i )→0 Tt
i ( i N(2)
i )→0. This com-
pletes the proof.
Phase 3 If there was no deviation in phase 2 for tPi (if player i was pun-
ished) or tPj (if any player j among the punishers was punished) periods,
then play strategy si for the rest of the game which gives a payoff of *i
to each player i, except to the last deviator j in phase 2 who receives only
*j , *j *j . Moreover, * ( *1, …, N*)IR and therefore ( *1 , *2
, …, *j , N* )IR, too. If any player deviates in phase 3, restart
phase 2.
Phase 1
In this phase each player receives a payoff of *i in each stage. If s/he devi-
ates s/he can net a payoff of at most U
i in the first period. Then, s/he is mini-
maxed for tPi periods and receives a payoff of zero. From period tPi 1 until
perpetuity s/he receives a payoff of *i if s/he complies in phase 2. Since
*i M
i 0 by assumption, compliance is more attractive than deviation
in phase 2. Thus, the maximal gain from a deviation in phase 1 is given by:
tPi it i 1 * *
P
U
i 0·
t1
it *i ·
ttPi1
it *i ·
t0
it ⇔ U
i
1 i
· i i . (III.1)
1 i
Appendices 325
U
i · i i ⇔ U · i . (III.2)
1 i 1 i i
1 i
Equation (III.2) can generally be satisfied for sufficiently big discount
factors and sufficiently long punishment durations. To see this, note that
1 it i 1 P
P
Phase 2
of the game if s/he complies. Alternatively, if s/he does not fulfill her pun-
ishment obligations in phase 2, s/he receives at most a payoff of U j in the
first period and then s/he is minimaxed him- or herself for tjP periods (phase
2 starts anew) which, as argued above, s/he will accept in order to reach
phase 3 finally. Therefore, the discounted payoff stream of a deviation in
phase 2 is given by:
tjP jt j 1
P
U
j 0·
t1
tj j*·
tt jP 1
j ⇔ j
t U
1 j j
· *. (III.4)
P
Since j 1 and hence jti 1 1 holds, an upper bound for this deviation
payoff is given by:
*j
U
j . (III.5)
1 j
Thus, the maximum gain from a deviation in phase 2 does not exceed:
326 Appendices
t i 1
P
*
jt ( *j )
j
U
j M(i)· jt ⇔
1 j j
t0 P
tt i
1 jt i
P
t i
P
U * j .
j ( j j
M(i)) (III.6)
1 j 1 j
P
Now let j →1, then (1jti )/(1j ) approaches tPi from below. That is, the
first two terms are finite. However, the third term goes to minus infinity for
j →1, so that (III.6) becomes negative. Thus if the discount factor is suffi-
ciently close to 1, deviation in phase 2 does not pay a punisher.
Phase 3
Last but not least, we have to check for the incentive to deviate in phase 3.
In this phase a player who was punished previously in phase 2 will not
deviate in phase 3 because otherwise phase 2 is started again. Whereas com-
pliance in phase 3 gives him/her a payoff of *i , deviation gives him/her at
most a payoff of U i in the first period and subsequently the payoffs of
phases 2 and 3. Thus, the gain from a deviation is given by:
it i 1 * * 1 it i 1 *
P P
U
i · i i ⇔ Ui · i (III.7)
1 i 1 i 1 i
which is equivalent to (III.2) and negative by (III.3). For a punisher j who
complied with his/her punishment obligations in phase 2, the incentive to
deviate in phase 3 is given by:
jt j 1
P
*
j
U · j* j (III.8)
1 j 1 j
which is obviously smaller than (III.7) (of course, i has to be replaced by j
in (III.7)) and hence deviation does not pay either.
Taken together, deviation does not pay in phases 1, 2 or 3, provided dis-
count factors are close to 1. Choosing *i sufficiently close to M i and
letting →0, any payoff vector * IR can be obtained with *i → M i at
the limit of i →1.
IV CHAPTER 7: APPENDICES
IV.1 Appendix 1
P
In Section 7.1.1 we claim that for i close to 1 (implying ti i →1) (7.3) and
(7.4) in the text reduce to (7.7).
Appendices 327
Proof:
(a) Assume si si (sj ). Then D i i and from (7.4) i i follows.
* P *
IV.2 Appendix 2
1* C1 ( p1,q11) (IV.1)
π2
C2
c2 A
a2
C1
ΠIR
B
d2
d1 a1 c1 π1
IV.3 Appendix 3
In the particular example * (3.2, 3.2) and in the first phase R(1) ( p11
p1, q11 0)(1.4, 4.4). Moreover, let I( p11 p,q11 q) denote the ‘inter-
mediate payoff tuple’ of the second phase. Then, the average continuation
punishment payoffs to players 1 and 2 are given by:
P P P
1P (11t1 ) R(1)
1 1t1 (11) 1I1t1 1 *1 .
P P P
2P (12t1 ) R(1)
2 2t1 (12) 2I2t1 1 *2 . (IV.9)
Now there are two possibilities for a deviation during the punishment,
either in phase 1 or phase 2. Hence, compliance with the punishment
(repentance phase) must be ensured in both phases:
must hold. (IV.10) is the familiar condition (7.3) in the text which ensures
that deviation in phase 1 does not occur. (IV.11) is a similar condition for
phase 2. Condition C2 states that the average payoff to player 1 in phase 2
and subsequently resuming cooperation must be greater than deviating in
phase 2 (recall p1 1 is a dominant defection strategy in this game) and then
starting punishment anew.
Of course, deviation during the cooperative phase should not pay either
(see inequality (7.4) in the text):
Finally, we have to take care of condition (7.5) in the text. Using (IV.9) we
have:
P P P
C4: *2 (12t1 ) 2R(1) 2t1 (12) 2I 2t1 1 *2 . (IV.13)
We start solving for imin by assuming t1P 1, p11 p1 and q11 q1/3 in
phase 2. Then 1I 2, 2I 4, C1 ( p1, q11 0)2 and C1 ( p1, q11 1/3)
2.8. Consequently, condition C1 implies 1 0.5, C2 1 0.359 and C3
1 0.5. C4 is always satisfied for any 0 2 1. Hence, min
1 (WRPE)0.5
and by symmetry min2 (WRPE)0.5. It is easily checked that if we substi-
tute 1 0.5 into (IV.9) 1P 1M 2 from which min i (WRPE)min i (SPE)
follows.
330 Appendices
V CHAPTER 8: APPENDIX
In Proposition 8.3 we claim that issue linkage improves upon the chances
for cooperation if issues are substitutes in governments’ objective function
and that cooperation becomes more difficult if issues are complements.
Proof: Since Proposition 8.3 focuses on the Pareto-efficient cooperative
stage game strategy tuple (a1, a2), min
i (SPE)min
i (SRPE) by Theorem 7.4.
Sub-case 1
Governments are cooperating on issue 1 if:
i
ui(ai1)(1i )ui(ci1)⇔ u (a )ui(ci1)ui(ai1)
i I (V.1)
(1 i ) i i1
holds. If the new issue 2 is viewed as a separate game, independent of game
I, cooperation can be sustained in game II provided:
Collecting terms show that (V.7) is equivalent to (V.5) which we have signed
already. Hence, again, if issues are substitutes (complements) linking
makes cooperation easier (more difficult) on issue 1.
Sub-case 2
Next we have to clarify how issue linkage affects issues 1 and 2 if on issue
1 cooperation cannot be sustained on its own but on issue 2 it can. In this
case we may write condition (V.1) as:
ui (ci1) ui (ai1)
i i (V.8)
ui (ci1)
where i is some positive number which might be interpreted as a slack var-
iable. Accordingly, the issue linkage condition (V.3) becomes a strict
inequality:
332 Appendices
ui (ai )
A: . (V.12)
ui (i ai )
Since ci1 ai1 and ci2 ai2 we have to show that A is decreasing in the payoff
ai . For this we look at some typical concave utility functions for which
2ui /xik2 0 and 2ui /xi1xi2 0 is true. For instance, consider the utility
function ui ai 12 a2i . Then we find for (V.12):
1
ai ai2
2 A 2(1 )
A: ; 0. (V.13)
1 a (2 ai)2
ai (ai)2 i
2
For other typically frequently used utility functions, such as the exponential
function ui (1/)(1/)e(ai ) or the logarithmic function ui ln(1ai ),
A/ai 0 holds as well and therefore (V.11) holds for a great class of utility
functions with a substitutional relation between issues and homogeneous
payoffs in the two games. This implies that linking issues also improves
upon the condition for the existing issue 1 even though cooperation was not
possible if this issue was viewed as a single game. A similar procedure
would establish that the opposite holds if issues are complements.
Appendices 333
Sub-case 3
This sub-case is symmetric to sub-case 2 and therefore a proof is omitted.
Sub-case 4
Here we assume that cooperation cannot be sustained on both issues if they
are viewed as single issues. Thus:
are true, which of course also applies to the sum of (V.17) and (V.18). Thus,
introducing a slack variable i, we have alternatively:
This condition has to be compared with the issue linkage condition in the
form (V.3). Upon substitution we derive:
334 Appendices
Sub-case 1
Cooperation on both single issues can be sustained. That is, if both games
are not linked, condition (V.2) holds in game II and by analogy in game I
we require:
ui (ai1,ai2 )(1i )ui(ci1,ai2)iui(ai2 ) 0. (V.23)
If (V.2) and (V.23) hold, this implies that (V.2)(V.23) must hold too. Thus,
to analyze the effect issue linkage has on the stability requirements we have
to find out whether (V.2)(V.23) or (V.3) is more stringent. For this we
compute (V.3)((V.2)(V.23)):5
i
[u (a )ui (ai2 )ui (ai1,ai2 )][ui (ci1,ai2 )ui (ai1,ai2 )]
(1 i ) i i1
[ui (ci1,ci2 )ui (ai1,ci2 )] (V.24)
which is easily signed. If both issues are substitutes, then the first term is
positive and the same applies to the difference between the second and the
third terms. By the same token, the opposite holds if both issues are com-
plements, and hence issue linkage has a positive (negative) effect on the
stability of contracts if issues are viewed as substitutes (complements) in
the objective function of governments.
Sub-cases 2, 3 and 4
Next consider the sub-case where cooperation on one single issue, say 2, can
be sustained but not on issue 1. This is identical with sub-case 2 above, for
Appendices 335
which the impact of issue linkage has been shown. Of course, by symmetry
if issue 1 can be sustained as a single issue but not issue 2, sub-case 3 above
applies. Finally, the sub-case where cooperation on single issues cannot be
sustained is equivalent to sub-case 4 above, and we are done.
VI CHAPTER 9: APPENDICES
VI.1 Appendix 1
Proof: The inverse of the slope of country 1’s reaction function may be
written as:
1 e2 1 1 1e e
1 1 (VI.2)
r1 e1 1 1e e
1 2
where the last term on the RHS may be replaced by (VI.2). Then after some
basic manipulations we derive:
2e2 1e e e · 1e e 2 2 1e e e · 1e e · 1e e 1e e e · 1e e 2
1 1 1 1 2 1 1 2 1 1 1 2 1 2 2 1 1
(VI.4)
e21 ( 1e e )3
1 2
where:
1e1e1 1,
1 1e1e1e1 1
1
, 1e1e2 ,
1 1e1e1e2 1
,
VI.2 Appendix 2
We claim in Section 9.7 that country 1’s (country 2’s) indifference curve (iso-
welfare curve) is concave (convex) with respect to the origin. We prove this
exemplarily for country 1’s indifference curve. A proof for country 2 would
follow exactly along the same lines.
where the same notation as in Appendix VI.1 has been used. In order to
show concavity of this function we compute the second-order derivative:
1e 1e
I1
2e2
1
1e 2
1
1e 2
· e
2
(VI.7)
e21 e1 e2 e1
where the last term on the RHS may be replaced by (VI.6). Proceeding in a
similar way as in Appendix VI.1 we arrive at:
2e2 1e e · 21e 2 1e e · 1e · 1e 1e e · 21e
1 1 2 1 2 1 2 2 2 1
(VI.8)
e21 ( 1e )3
2
(VI.9)
2e2 2 112
1 1 0
e 1
2 ( 1 )3
VII.1 Appendix 1
In Section 10.2, note 3, we indicate that the SOC may not hold in general
and therefore they must be assumed to hold. Moreover, it is claimed that if
the SOC hold, then a unique equilibrium exists.
Proof: Taking the second derivative of the objective function of the leader
as stated in (10.1) leads to:
from which it appears that all terms are negative except the last term which
might be positive if 2ej /ei2 0. That is, if the second-order derivative of
the reaction function is negative the whole term in (VII.1) could be posi-
tive. From Section 9.3 it is known that strictly concave reaction functions
cannot be ruled out on theoretical grounds. Hence, to guarantee a unique
maximum, either convex reaction functions have to be required or the effect
of the last term in (VII.1) has to be of minor importance. (For quadratic
benefit and damage cost functions the last term vanishes since 2ej /ei2 0.)
If 2ej /ei2 0 the damage cost function becomes concave instead of
convex.6 Then if ii, where the last three terms in (VII.1) have been
summarized to i, the payoff function will still be concave and we get an
interior solution. Only if ii would the SOC be violated (that is,
i0) which therefore has to be ruled out by assumption.
If the SOC hold, the leader’s objective function is strictly concave and
his/her equilibrium strategy is unique. Since the follower’s profit function is
also strictly concave as established in note 3 of Chapter 9, the uniqueness
of the Stackelberg equilibrium follows.
VII.2 Appendix 2
j j where country i is the leader and j the follower. The strict inequal-
ST N
ity sign cannot be used since a boundary Nash equilibrium has not been
j 0, then, obviously, there is no incentive for the Stackelberg
ruled out. If eN
leader i to expand emission beyond eiN and eiNeiST, ejN eSTj and ek ek
ST N
follows. For all other cases, however, the strict inequality sign holds.
To prove the above-stated relation, assume eSTi ei (which requires ei 0)
N N
338 Appendices
i ei would be true. Then from 1ej /ei 0, ek ek
instead, eST N ST N
tion functions. Since 1ej /ei 0 country j reduces its emissions less
than country i expands its emissions and hence ekST ekN follows.
It should be obvious that the leader must gain (provided eN j 0) other-
wise he/she could choose eSTi e N. The follower’s loss is obvious since due
i
to ekST ekN and eST
j ej damage will be higher and benefits lower than
N
VII.3 Appendix 3
Nash eq.:
i
ei
e Ni
i
ei
e N
k
0; Social opt.
i
ei
e Ni
i
ei
e N
k
i
ei
e N
k
0.
(VII.3)
Appendices 339
From (VII.3), together with the assumptions with respect to the second-
order derivatives, it is evident that emissions in the Nash equilibrium are
too low from a global point of view.8,9 Hence, the relation derived in the pre-
vious sections for the case of filterable externalities is reversed and
ekN ekS is proved.
In the Stackelberg equilibrium the leader’s FOC are given by:10
i i i ej
0. (VII.4)
ei ei ej ei
() () ()·()
Since the last term in (VII.4) is negative, marginal damages are less valued
by the Stackelberg leader and s/he will choose higher emissions than in the
Nash equilibrium. The follower, due to his/her positively sloped reaction
function, will respond by increasing emissions. Hence, aggregate emissions
in the Stackelberg equilibrium exceed those in the Nash equilibrium, that
is, ekN ekST.
VII.4 Appendix 4
At the end of Section 10.4, in note 20 we mention that the FOC and the
SOC may not be satisfied in a negative conjectural variation equilibrium.
Proof: The FOC for an interior conjectured equilibrium are:
i (1k
i i )0. (VII.5)
(a) β (b) β
i i
φi φi
φi
φi βi βi
VII.5 Appendix 5
eA
i e I and (eA) (eI) hold provided eI eN
i I.
i i i i i
1. The FOC in the initial situation (ri 0) for any eiI eN
i
i I are
(recalling ei (1ri )·eiI):
i ei i ei
0⇔ i ·eiI i ·eiI 0 (VII.6)
ei r ei r
i i ei ei
where the equality sign holds in the (interior) Nash equilibrium and
for any greater emission levels. The FOC of an offer (assuming an
interior solution) may be written as (see also (10.10) in the text):11
i ei i ek
0⇔ i ·eiI i ·(eiI ·ejI )0. (VII.7)
ei r ek r
i i ei ei
Since i /ei (eiI)i (eiI ejI )/ei
ri and 0 and recalling i0,
the offer will always be positive, ri 0. Clearly, if initial emissions are
below those in the Nash equilibrium the sign in (VII.6) becomes
and ri 0 cannot be deduced any more by comparing (VII.6) with
(VII.7). The optimal offer may involve ri 0.
Appendices 341
(VII.8)
(VII.9)
The last two equalities follow from ei (1ri )·eIi and ek (1ri )·eIi
(1·ri )·eIi .
3. Suppose that eIi emax i and eIj emax
j . If * 0, implying that country j
would not contribute to abatement, country i would propose ri such
i ei (ej
that eA max). For any * 0, country j contributes something to
VIII.1 Appendix 1
In Section 11.4 we claim that to ensure strict concavity of the payoff func-
tions with respect to a uniform tax ti, third-order effects must be sufficiently
small (see assumption A2 in (11.7)).
Proof: The first derivative of payoffs with respect to ti is given by:
i i ei i ek
(VIII.1)
ti ei ti ek ti
where i /ti 0 in the optimum. The sufficient condition for strict concav-
ity is 2 i /t 2i 0
ti [0, t max
i ]. Therefore, we require:
(VIII.2)
342 Appendices
1 2 1
(ii) () 3 ·i
0. (VIII.3)
i i i
i
The first term is positive because i0 and i0 by assumption A1. Since
i 0
i is possible if (0)tt
i i, we have to require i
to be sufficiently
small in this case to ensure that the whole term is positive. (However, for
t (0),
i we set ei 0 and then ei /ti 0 and 2ei /t2i 0 in (VIII.3) and
because i 0 and i0 hold, concavity follows immediately.)
VIII.2 Appendix 2
In this appendix we provide equilibrium emissions under the quota and tax
regimes, the Nash equilibrium, the social optimum and the minimax emis-
sion tuples for the payoff functions in (11.8). All relevant information is
provided in Table VIII.1.
Equilibrium emissions in the Nash equilibrium, social optimum and
when a country is minimaxed are derived as described in Chapter 11;
however, they are now based on payoff functions (11.8).
From the table it is evident that the following non-negativity conditions
(abbreviated NNCi ) have to be imposed:
That is, NNC1 follows from requiring eSi 0, NNC2 from eN i 0 and NNC3
from eM(i)
i 0
i I where the upper bound of the strategy space is, again,
assumed to be e0i which follows from (e0)0. Using NNC and NNC we
i i 1 2
can show that emissions under the tax and quota regimes are positive.
Moreover, note that the proposals under the quota regime are given by:
( )( )
r1 ,
22 2 22 2 2 2
( )( )
r2 (VIII.5)
22 2 22 2 2 2
Appendices 343
from which it follows that r1 r2 if . That is, r1 rA if ,
r2 rA if and r1 r2 rA if . The equilibrium emissions follow
i (1r )·ei .
from eQ A N
VIII.3 Appendix 3
Proposition 11.3 claims ekQ ekN, ekT ekN, ekQ ekS and ekT ekS.
Proof: To prove Proposition 11.3 we need the following lemma as prelimi-
nary information:
Lemma VIII.1
Let k* {r*, t*} denote a global welfare maximizing uniform policy level
which solves max (k)[ k(ei (k), ej (k))], i (ei (k), ej (k)) a strictly concave
function in k, and let ki and kj be the bargaining proposals for countries
i and j. Then either k* ]ki, kj [ or k* ki kj and therefore kA k*.
Proof: The idea of the proof is to show that the maximum of the sum of
two strictly concave functions is located in the interval of the maxima of
the two single functions. Concavity of the payoff functions under both
regimes has been established in Section 11.4. Hence, i /k0
k ki
i /k0
kki . We show ki k* kj or ki kj k* by contra-
diction, assuming, first, ki, kj k* which implies:
k
k k*
i
k k*
j
k k*
0. (VIII.6)
0 0
Second, we assume ki , kj k* which implies:
k
k k*
i
k k*
j
k k*
0. (VIII.7)
0 0
However, (VIII.6) and (VIII.7) should both be zero evaluated at k* due
to the FOC. Of course, kA k* is then an immediate implication of the
LCD decision rule.
e1 Tax regime e2
344
e1 Social optimum e2
e1 Nash equilibrium e2
d( ) d( )
345
e1 Minimax e2
d( 1)
1 is minimaxed d
1
d( )
2 is minimaxed d
Note: The symbols imply: REj, j {1, 2}, defines parameter regions ( and ). The subscript a stands for adjustment. e1 and e2 are emissions of
country 1 and country 2.
346 Appendices
country i and hence ei ei . This is evident from comparing the FOC in the
T N
(interior) Nash equilibrium (NE) with those when deriving a tax proposal
ti:
i i ei i ei i i ei i ek
0; 0 (VIII.8)
ti
N ei ti
N ei ti
N ti ei ti ei ti
and noting
i ei
ei ti
N
i ek
ei ti
ti and i0.
1. ti tN
j ⇒ej ej . Since ti ti implies ei ei , ej ej follows.
T N N T N T N
2. ti tN
j
(a) j(ej )j(ei , ej )⇒country j does not adjust.
T T T
2(a) and 2(b) imply eTj eN j and hence j(ej )j(ej ). From the assump-
T N
tion j(eTj )j(ekT ) in 2(a), j(eT )j(ekN ) follows, which is only true
if eTj eN j . In 2(b) country j adjusts according to its reaction function.
Since eiT eN i and the slopes of the reaction functions are less than 1 in abso-
lute terms, we conclude eN i ei ej ej (ei ) and therefore ei
T N T T T
ekQ ekS can be shown for the functions in (11.8), using the information
of Table VIII.1:
: ekQ ekS
where (VIII.9) and (VIII.10) are positive for 1 and zero for
1.
ekQ ekS can be shown by assuming slightly different net benefit func-
tions than in (11.8). For instance, consider:
1 c 1 c
1 b1e1 e21 (e1 e2 )2, 2 b2e2 e22 (e1 e2 )2
2 2 2 2
(VIII.11)
and assume b1 5, b2 10, c1, 1 and 11/5. Then eN 1 10/7, e2
N
15/7, eS1 0 and eS2 50/21. The proposals under the quota regime are given
by r1 15/2911/32r2 and hence r2 rA, which implies eQ i 75/32
50/21eSi .
VIII.4 Appendix 4
Since r 0, Q
A
i i
i follows by strictly concave payoff functions, which
N
tA t*. Only case 2(b) is less straightforward because (11.3) in the text does
not hold after adjustment has been conducted. If eTi eN i were to hold
before adjustment takes place, then Ti N i would already be true
without adjustment (see the arguments above). Additionally, since the non-
bottleneck country j conducts the adjustment voluntarily, this country
must gain from such an action and, by i /ej 0, country i as well.
348 Appendices
βi',φi' βj',φj'
φi'(ei,ejT ) φj'(eiT,ej)
βi' (ei) βj' (ej)
B E
C D
ei ei holds after adjustment has taken place, which is known from
T N
i(ei ) is true (see Figure VIII.1). The superscript eq stands for ‘equivalent’
N
with eN i ei ej and a stands for adjustment. Since ei ei , ej gives emis-
T eq T N eq
ej ej follows from the fact that adjustment takes place and that the reac-
N T,a
tion functions are downward-sloping in emission space with slope less than
j ej must hold, otherwise ei ei is not
1 in absolute terms. Finally, eeq T T N
possible and the result for ei ei derived above would apply.
T N
eTi e eq eT,j a eT,j a
A BCD D DE
(VIII.12)
is true, where the second relation simply follows from the fact that adjust-
ment is conducted voluntarily and the first is a consequence of the fact that
the marginal opportunity costs of abatement ‘become more equal’ so that
cost efficiency potentials are realized.
Appendices 349
socially optimal if and only if rS r* rA. That ri rj does not imply
Qk kS is simply demonstrated by example. For instance, choose for
the payoff function (11.8), 1, then r1 r2 r*. (VIII.9) and
(VIII.10) above reveal that eQk ekS and therefore Qk kS follows.
VIII.5 Appendix 5
Q
i >i
i P I
N
Tj >Nj
Consider the cases 1, 2(a) and 2(b) in Appendix VIII.3. Case 2(b) implies
adjustment by country j. Since we show below that Ij(tA) 0 implies
Ti N i and by Proposition 11.4. k k holds, j j must be true.
T N T N
holds due to j /ei 0, j(ei ,ej ) j(ei ,ej ) must be true. Case 1 implies
T T N N
an even higher tax ti tN j and since tj ti, tj is true by assumption that
N
payoff functions with respect to the tax rate and Tj N j for ti tj as dem-
N
onstrated above.
Ti >≤ N
i
This statement is simply proven by example. For instance, choose
1/2, 1/4, d100, b10, c1 for the functions in (11.8), then T1 N
1
and T2 N2 . For 1, 1 and d, b, c as assumed above, i i
i
T N
I.
350 Appendices
Ti ≥M i
Note that M i is derived from max i(ei,ej
max) and that t0 implies eT emax.
j j
Any proposal ti by country i, however, is derived with the information
eTj emax
j for any ti 0 and hence Ti M i must hold by i /ej 0.
Ti <N A
i if Ij(t )<0
i ei and ej :ej(ei )ei despite adjustment by coun-
First, note that eN T T,a T N
j(ei , ej ), j(ej )j(ej ), which cannot be true by Ii (ei ,ej )0 and
N N T,a N T T,a
i ,ej )0. Next, we show that any emission tuple e for which ej ei
Ii (eN N T T N
(ej ei ) and ei ei is true implies i (e ) i (e ). For this, note the fol-
T,a N T N T N
lowing relations:
ej
ri
eT
eN rj
IiN
ei
VIII.6 Appendix 6
Write down the FOC from which a biased proposal is derived to note that
i ti and that the implied emissions are exactly those in a Stackelberg
tstr
equilibrium. Then most claims of Proposition 11.7 follow immediately
from Proposition 10.2. Only (1) kT(tstr
i ) k (ti ); (2) j (t ) j and (3)
T T str M
(1): First, note that ekT(tstr i )ek ek (ti ) from Proposition 10.2 and
N T
i(ei )j(ej ) and a biased proposal will lead to i(eTi (tstr
T T,a
i ))
j(eTj (tstr
i )). That is, the biased proposal creates (increases) a (the) gap
between marginal opportunity costs of abatement in the two countries and
therefore leads to a (more) inefficient allocation of emission reduction
burdens (see CEA condition (11.2)). Together with higher aggregate emis-
sions resulting from a biased proposal, implying higher damages, global
welfare must have decreased through the strategic proposal. (The case of ti
tA implying eTi 0 and eTj 0 or eTi 0 and eTj 0 (corner solutions) can
be discarded because they obviously provide no incentive for the strategic
mover i to increase emissions since eTj (t str
i )0 is not possible.)
(2): M
j follows from max i(ei
max,e ) whereas T(e(tstr)) ST follows from
j j j
max i(ei ,ej ) where ei ei by the definition of the strategy space and
ST ST max
hence ST
j j by j /ei 0.
M
1 c
*2 b de12 e122 (e11 e12 )2. (IX.1)
2 2
Solving (IX.1), using country 1’s best reply, e11 (bdce12)/(bc), gives
because of (12.27) in the text:
where:
d(b22 2bc c22 bc2) (b c) ·A
e12,min
b22 2bc c22 bc2
A: b(b3d 22 2b2 *2 2cb2d 22 3cb2d 2 2cb *2 4cb *2 c2bd 22 2c2 *2).
In order to obtain a solution for e12 A must be positive, which is true if:
bd 2(b2 2cb 3cb c2)
*2 . (IX.3)
2(b2 cb 2cb c2)
X.1 Appendix 1
1 1
Ji () NJ
j ()0 if 0 and Ji () NJ
j () 0 if 1
N 1 N 1
(X.3)
d( 1 N )
i ()ej
ẽNJ ()0 if ()*
N
(2 2 2 1 2 2 2N)2
from which it is evident that NJ j j for (or nn ) must hold.
N * *
Together with i i
(or nn ) as shown below in (3), it follows
J N * *
2 Ji (n)/n2 0
n [0, N]; see Figure 13.1(a), again, n* N* and Ji N i
at n*. (Hence the claims in (2) above that Ji (n*U) Ji (n*L) and therefore N*
354 Appendices
Since ei ej ej and signatories and non-signatories suffer from the
J* N NJ*
implies kN k* and k* kS follows from the fact that only for 1
ẽiJ()eSi and ẽ jN J()eSj (and that the socially optimal emission allocation
is unique for payoff functions of type (3)).
* I I
(4) <0, 3 <0, 4 <0 for N→
Compute the difference functions C2(n) and C3(n) and let N →. Then it is
found that:
2
2(2 1) *U 22
2(2 1) *L
N→
n*L , n N→ nN→ 1 (X.6)
2 2
where C2 () 0 if n() n*U N→ C3 () 0 if n() nN→ and where
*L
n*L N* n*U must hold for the stability of a coalition. It is easily checked
that n*L
N→ /0 and n N→ /0 hold and therefore N /0 is true.
*U *
X.2 Appendix 2
In this appendix we prove that the social optimum associated with the
transfer scheme described in (13.27) lies in the core provided one of the
three properties mentioned in Proposition 13.5 holds (see Chander and
Tulkens 1995, 1997).13
Proof: As preliminary information we need the following lemma (Chander
and Tulkens 1997, Proposition 4):
Lemma X.1
The partial-agreement equilibrium (PANE) as described by Definition
14.6 has the following properties: (1) there exists a PANE; (2) the equi-
librium emission vector is unique; (3) ẽk ẽiJ ẽNJ
j ek ; and (4)
N
eNj ẽ NJ
j IJ in the PANE.
j
Proof: Properties (1) and (2) are proved by applying the theorems as laid
out in Section 9.4. Property (3) is proved by contradiction. Suppose
ẽk ekN were true. This implies (a) i(ẽ)i(ẽiJ)i(eN
i )i(e )
N
Appendices 355
Note first that for linear damage cost functions non-signatories have a
dominant strategy and hence eJN ẽNJ
j j I . Moreover, ei ẽi and ei ẽi
J S J N J
i I hold.
J 14
and
follows (since max kI k kI k* w(I) by definition) where the equal-
ity sign in (X.12) is implied by (X.8).
(X.10) is shown by noting that:
356 Appendices
implies:
iI Ji (e S)
iIJ ˆ i iIJi (ẽiJ)iIJi(eS) [ (eS)iIi (ẽiJ)]
iIi (e S) iI i i
(X.14)
or
iI Ji (e S)
iIJ ˆ i iIJi(ẽiJ )iIJi(ẽ) ·A;
iIi (e S)
ˆ j j(eSj )j(eS)t̂j
j (e S)
j(ẽjNJ)j(eS) [jIj(eSj )jIj(ẽjNJ)] (X.16)
jIj (e S)
and:
*j j(eSj )j(eS)t*j
j (e S)
j(eN S
j )j(e ) [jIj(eSj )jIj(eN
j )] (X.17)
jIj (e S)
which implies:
j (e S)
j (ẽjNJ) [jIj (eSj )jIj (ẽjNJ)]
jIj (e S)
j (e S)
j (eN
j ) [jIj (eSj )jIj (eN
j )] (X.18)
jIj (e S)
according to (X.11). It easily checked that (X.18) in fact holds by noting that
j ẽj
j I due to linear damage cost functions and ei ẽi i I as
eN NJ J N J J
established above, and hence j (ẽj )j (ej ) and jIj (ẽj )jIj (ej ) is
NJ N NJ N
true.
Appendices 357
Case 2:
IJ
I, | IJ | ≥2, kPIJk(eS)≥i(eN )
i P IJ
For the subsequent proof we need the following lemma (see Tulkens and
Chander 1997, Section 6, Proposition 5):
Lemma X.2
Under the assumption of case 2, ẽi eSi
i I and e N
i ẽi
i I .
J J
Proof: First, we show ẽi eSi . Suppose instead ẽi eSi were true.
Then:
i IJ: i(ẽiJ)kIJk(ẽ)kIJk(eS)kIk(eS)i(eSi )
j IJ: j(ẽ N
j )j(ẽ)j(e )j(ej )
J S S
to i0.
Next we have to show that * lies in the core. As in case 1, this is shown by
contradication. Thus, exactly the same arguments which have been pre-
sented when going from (X.8) to (X.12) apply. Consequently, it has to be
shown that (X.10) and (X.11) hold in the present case 2.
(X.10) implies:
iIJi(ẽiJ)iIJi(ẽ )
iIJi (ẽiJ )iIJi (ẽ)iIJi (ẽ)iIJi (eS)iIJi(eS)(ẽeS)
iIJi (ẽiJ )iIJi (eS)iIJi(eS)(ẽeS) (X.21)
or:
iIJi (e S)
iIJt̂i iIJi(eSi )iIJi (ẽiJ) (eS)(ẽeS) (X.23)
iIi (e S) iI i
or, using t̂i as defined in (X.9):
i (e S)
iIJi(eSi )iIJi(ẽiJ ) J ( (eS)iIi(ẽi ))
iI iIi (e S) iI i i
iI Ji (e S)
iIJi (ẽiJ) (eS)(ẽeS) (X.24)
iIi (e S) iI i
after further manipulation one derives:
Using the FOC iIi(eS)i(eSi ) (and the original notation) we have:
must be true (iIJi(eSj ) on both sides has been dropped). Using (X.9), the
LHS term may be written as:
jIJj(eSj )jIJt̂j
j (e S)
jIJj (ẽjNJ) (jIj (eSj )jIj (ẽj ))
jIJ jIj (e
S)
Appendices 359
j (e S)
⇔ jIJj (eN
j ) (jIj (eN
j )jIj (ej ))
S
jIJ jIj (e )
S
j (e S)
jIJj (ẽjNJ)jIJj (eN
j ) (jIj (eN
j )jIj (ẽj ))
jIJ jIj (e
S)
j (e S)
(jIj (eN
j )jIj (ẽj ))
jIJ jIj (e
S)
j (e S)
(jIJj (ejN)jIJj(ẽjJ))
jIJ jIj (e
S)
j (e S)
⇔ jIJj (eSj )jIJtj* 1 jIJ jIj (e )
S ( jIJj (ẽj )jIJj (ej ))
NJ N
j (e S)
(jIJj (ẽjJ )jIJj (ejN)). (X.29)
jIJ jIj (e
S)
Note that the first term in brackets is obviously positive; the second term
in brackets as well, since ẽjNJ ejN
j IJ by Lemma X.1, and the third term
in brackets is negative since ẽjJ ejN
j IJ, as has been established in
Lemma X.2 above. Taken together, (X.10) ((X.20)) and (X.11) ((X.28)) have
been shown to be true in case 2, which completes the proof.
X.3 Appendix 3
In this appendix we prove that for the transfer scheme (13.30) in the text the
imputation as defined in Proposition 13.6 lies in the core at each stage t.
Proof: We first have to establish that the iterative process is individually
rational; that is, each country gains along this process: ! i,t 0
t and i
I (Germain et al. 1995, Theorem 2).
360 Appendices
Note that:
which, assuming a linear damage cost function, that is, i diek (implying
!i,t di!ek,t), implies:
where g stands for gains or surplus from cooperation. From (X.8) it follows
that:
where:
which, given:
sion vector et* and the PANE with respect to the coalition IJ has to be
indexed by time t. Both PANE and the local global optimum are defined
with respect to constraint (13.29) in the text.
XI.1 Appendix 1
2i(2N 2 N 1 2i )
ri
42 4N 4 8i 4iN N 2 2N 4Ni 1 4i 4i2
iNdbc
ti (XI.6)
b icN
for which ri /i0 and ti /i0 is true. Hence country 1 is the bottleneck
under the LCD decision rule and by substituting i1 in (XI.6), the agree-
ment according to the LCD decision rule, that is, rA r1 and tA t1 is deter-
mined. From this global emission reduction (based on initial emissions eiI
eN i as referred to in Table 14.1), r
Q1 and rT1, can be computed to be (rQ1
2(2N 2 N 1) N1
rQ1 , rT1 . (XI.7)
42 8N 4 N 2 2N 1 2( N)
A globally optimal solution, given the constraint of a uniform application
of the instrument, leads to:
(N 2 1)(6 1) N2 1
rQ* , rT* (XI.8)
122 6N N 2 1 6N 2 2 N 2 N
where rT* rS*. Substitution of i(N1)/2 into (XI.6) leads to the median
country proposal for which it is easily checked that this implies global emis-
sion reductions as given in (XI.8).
For a sub-coalition the proposal of a potential signatory i is derived as
follows. In the quota regime, two pieces of information are used. First,
country i knows that countries 1 to N0 choose their initial emission level eNi .
Therefore, emissions of all non-signatories are given by eNJ "0 eN
j j1 j
where j INJ{1, 2, ..., N0}. Second, if countries N0 1 to N comply with
country i’s proposal, ri, then signatories’ emissions will be ekJ kN
N eN·
01 k
(1ri ) where i, k I {N0 1, . . . , N}. Consequently, aggregate emis-
J
sions are given by ekQ ekNJ ekJ. Using payoff function (14.1) in the
text, it is straightforward to compute a signatory i’s proposal to be:
2Ni(2N 2N0 N 20 NN0 N 2i 2)
ri (XI.9)
A
where ri /i0 holds. That is, the signatory with the lowest index number
is the bottleneck country according to the LCD decision rule, which is
country N0 1.
Appendices 363
XI.2 Appendix 2
To test whether an emission tuple e* (e*1,e*2, ..., eN*) of the grand coalition
is a WRPE if →1 the following algorithm is used:
A. Set i1.
B. Determine the amount !ei by which all countries i have to increase
their emissions during the punishment of country i in order to fulfill
(12.1) in the text.
B.I Determine the reaction function of country i, ei (ei ).
i e* !e
B.II Insert ei i i in (12.1) and solve for !ei !emin.
Define !ei:!emin where is a sufficiently small positive
number.
C. Check whether (12.2) in the text can be satisfied for all countries i if
they jointly increase emissions by the amount !ei .
C.I Set eii 0 in (12.2) to determine the outer boundaries of the
WRPE emission space.16 Hence, ei i e* !e where !e has
i i i
been determined in step B.II.
C.II For all ji let eji:ej* !ej with !ej the amount by which
country j increases its emission to punish i. Further, define
#min:0 and #max:0.
C.III Repeat for all ji:
i in (12.2). Solve (12.2) for !e to have !e [!e
Insert ei j j j,min,
!ej, max] or !ej $. In the first case define #min:#min !ej,min
and #max:#max !ej,max. In the second case stop C.III and
define #min:0 and #max:0.
C.IV If, and only if !ei [#min, #max], then country i can be pun-
ished by WRPE strategies. If this is the case continue with C.V;
otherwise stop.
364 Appendices
C.V Define i:i1. Repeat all steps starting from B. If the algo-
rithm runs for all i, i {1, ... , N}, then e* is a WRPE.
XI.3 Appendix 3
A. Set i 1.
B. Set i:i% where % is a suitable step size (for example, %0.01 or %
0.001).
C. Determine the amount !eai,min by which all countries i have to
increase their emissions during the punishment of i in order to satisfy
(12.24) in the text.
C.I Determine the best deviation in the normal phase of country i,
ei (e*i ), to have D
i .
C.II Insert ei i e* !e in (12.24) and solve for !e !ea
i i i i, min.
D. Check whether there exist eii and !ei !eai,min, so that (12.23) and
(12.2) in the text can be satisfied jointly.
D.I Define eii:0 in (12.23) and (12.2) in the text.
D.II Solve (12.23) for !ei !ebi, min. Define !ei :max {!eai, min,
!ebi, min}.
D.III Let eji:ej* !ej where !ej is the amount by which country j
increases its emission to punish i. Moreover, let ekieii ei i
XI.4 Appendix 4
The test of stability of a sub-coalition if →1 proceeds along the same lines
as laid out in Appendix XI.2. First, the amount !ei by which the signato-
ries have to increase their emissions to punish a non-signatory to satisfy
(12.1) is determined. Then, it is checked whether the signatories can
provide such a !ei under the restriction that (12.2) is met for each signa-
tory. Accordingly, the same procedure is applied in the case where a signa-
tory breaches the contract.
from which it is evident that ei /ci 0 (since 0). That is, members of
larger coalitions emit less than members of smaller coalitions. Since all
countries suffer equally from damages, members of smaller coalitions
receive a higher payoff than those of larger coalitions (C2). Next consider
that two coalitions i and j merge. Then for payoff funtion (13.1) and sym-
metric countries – with slight abuse of notation – equilibrium emissions are
given in the initial situation by:
ci ·ei (ci )cj ·ej (cj )(ci 1)·ei (ci 1)(cj 1)·ej (cj 1)
cj ci 1
2c 0, ci cj . (XII.6)
Nb
This has two implications. First, members of the ‘old’ coalition i are better
off if a member leaves to join coalition j. On the one hand, global emissions
decrease through such a move; on the other hand, members of coalition i
increase emissions after the move. Thus, the remaining members in coali-
tion i must be better off (C3). For the deviator two effects must be consid-
ered. On the one hand, s/he has to carry a higher abatement burden in the
new coalition which decreases his/her benefits. On the other hand, global
emissions decrease, implying lower damage. To sign the overall effect, we
compute payoffs of country k which left coalition i before and after the
accession to coalition j. We find:
Remark: Note in addition the following relations which are useful for estab-
lishing the incentive profile in Appendix XII.2:
c2(4·ci 2· cj 3)
j (cj ,c) j (cj 1,c) 0. (XII.8)
2N 2b
The members of the coalition j may gain or lose if a member of coalition i
joins. The larger coalition j is compared to coalition i, the more likely will
a member of coalition j gain from the accession. If the country which joins
Appendices 367
XII.2 Appendix 2
In this appendix we derive the incentive profile of (ex ante symmetric) coun-
tries with payoff function (13.1). The profile is derived for N [3, 4, 5].
Note that for symmetric countries only the coalition structure as such and
not the membership matters. (That is, for instance, {{1, 2},{3}} is identical
to {{1}, {2, 3}} and{{1, 3},{2}}.)
Three countries
There are three permutations to be considered: {{1, 2, 3}}, {{1, 2},{3}} and
{{1},{2},{3}} (Table XII.1).
From the profile it is evident that all three permutations are stand-alone
stable.
Four countries
There are five permutations to be considered: {{1, 2, 3, 4}}, {{1, 2, 3},{4}},
{{1, 2},{3, 4}}, {{1, 2},{3},{4}} and {{1},{2},{3},{4}}. Except for the first
permutation, all permutations are stand-alone stable (Table XII.2).
Table XII.2 Incentive profile of countries (N4)
Change of permutation 1 2 3 4 Reason
{{1, 2, 3, 4}} → {{1, 2, 3},{4}} 1, 2, 3: (XII.10); 4: (XII.9)
{{1, 2, 3, 4}} → {{1, 2}, {3, 4}} 1–4: (XII.10)
{{1, 2, 3, 4}} → {{1, 2}, {3},{4}} 1, 2: C1; 3, 4: (XII.10)
→
368
{{1, 2, 3, 4}} {{1}, {2}, {3},{4}} 1–4: repeated appl. of (XII.9) and (XII.10)
{{1, 2, 3}, {4}} → {{1, 2}, {3, 4}} 1, 2: (XII.8); 3: C4; 4: C3
{{1, 2, 3}, {4}} → {{1, 2}, {3},{4}} 0 1, 2: (XII.10); 3: (XII.9), 4: C1
{{1, 2, 3}, {4}} → {{1}, {2}, {3},{4}} 1, 2, 3: (XII.9), (XII.10), C1; 4: C1
{{1, 2}, {3, 4}} → {{1, 2}, {3},{4}} 1, 2: C1; 3, 4: (XII.10)
{{1, 2}, {3, 4}} → {{1}, {2}, {3},{4}} 1–4: (XII.10), C1
{{1, 2}, {3}, {4}} → {{1}, {2}, {3},{4}} 1, 2: (XII.10); 3, 4: C1
Appendices 369
Five countries
There are seven permutations to be considered: {{1, 2, 3, 4, 5}}, {{1, 2, 3,
4},{5}}, {{1, 2, 3}, {4, 5}}, {{1, 2, 3},{4},{5}}, {{1, 2},{3, 4},{5}}, {{1,
2},{3},{4},{5}} and {{1},{2},{3},{4},{5}} (Table XII.3).
R1: Payoff to 1 and 2 are the same as to 3 and 4 one row above. Payoff
to 3 and 4 are the same as to 1, 2 and 3 one row above. The implication for
5 follows from: *5({{1, 2, 3, 4},{5}}) *5({{1, 2},{3, 4},{5}}) *5({{1,
2},{3, 4, 5}}) where the inequality sign follows from C1 and the equality
sign from (XII.9).
R2: Payoffs to 1, 2, 4 are the same as one row above. Payoff to 5 is worse
than one row above by C2. The implication for 3 follows from *3({{1,
2},{3},{4, 5}}) *3({{1, 2, 3},{4, 5}}) *3({{1, 2, 3, 4},{5}}) where the
equality sign follows from (XII.9) and the inequality sign from (XII.8).
R3: The coalition structure is the same as one row above. Payoffs to 1 and
2 are the same as to 3 one row above. Payoff to 3 is the same as one row
above. Countries 4 and 5 are worse off than one row above by C2.
XII.3 Appendix 3
In this appendix we derive the NE, CPNE and SNE in the open-member-
ship game. Each country makes an announcement (number) mi (m is the
collection of announcements) and countries with the same announcement
form a coalition.
N3
m(1, 1, 1) ⇒{{1, 2, 3}}: NE. m(1, 1, 2)⇒{{1, 2},{3}}: NE since 3 is
indifferent to announcing m1. m{1, 2, 3}: no NE since by announcing
m1 2, 1 can improve upon its payoff.
Both NE – {{1, 2, 3}} and {{1, 2},{3}} – are Pareto-efficient and are
therefore an SNE. From CSNE CCPNE CNE it follows that these two
SNE coalition structures are also a CPNE.
N4
m(1, 1, 1, 1)⇒{{1, 2, 3, 4}}: no NE since it is not stand-alone stable.
m(1, 1, 1, 2)⇒{{1, 2, 3},{4}}: no NE since a country i, i {1, 2, 3},
Table XII.3 Incentive profile of countries (N5)
Change of permutation 1 2 3 4 5 Reason
{{1, 2, 3, 4, 5}} → {{1, 2, 3, 4},{5}} 1–4: (XII.10); 5: (XII.9)
{{1, 2, 3, 4, 5}} → {{1, 2, 3},{4, 5}} 1, 2, 3: (XII.10); 4, 5: (XII.9)
{{1, 2, 3, 4, 5}} → {{1, 2, 3},{4},{5}} 1, 2, 3: C1 ({4}{5}) and
(XII.10) ({1, 2, 3}{4, 5});
4, 5: twice appl. of (XII.9)
{{1, 2, 3, 4, 5}} → {{1, 2},{3, 4},{5}} 1–4: (XII.10), ({1, 2}{3, 4}
and ({1, 2, 3, 4}{5});
5: {3, 4}{5} ((XII.9) and
{1, 2}{3, 4, 5} ((XII.10))
{{1, 2, 3, 4, 5}} → {{1, 2},{3},{4},{5}} 1, 2: C1 ({3}{4}{5}) and
370
N5
m(1, 1, 1, 1, 1) and m(1, 1, 1, 1, 2) are not stand-alone stable and there-
fore no NE. m(1, 1, 1, 2, 2)⇒{{1, 2, 3},{4, 5}}: NE since for 1, 2 and 3
announcing mi 2 or mi 3, i {1, 2, 3}, implies a payoff loss or leaves
these countries indifferent; for 4 and 5 announcing mj 1 or mj3 implies
a payoff loss. m(1, 1, 1, 2, 3): no NE since m4 3 improves upon 4’s payoff.
m(1, 1, 2, 2, 3)⇒{{1, 2},{3, 4},{5}}: NE since mi 2 or mi 4, i {1, 2}
implies a payoff loss to i, mi 3 leaves i indifferent; by symmetry the same
holds for j {3, 4} and m5 1 leaves 5 indifferent. m(1, 1, 2, 3, 4): no NE
since m4 2 improves upon 4’s payoff. m(1, 2, 3, 4, 5): no NE since m2 1
improves upon 2’s payoff.
There is no SNE since both NE are Pareto-inefficient.
{{1, 2, 3},{4, 5}} is a CPNE since forming smaller coalitions is not prof-
itable. 1, 2 and 3 are indifferent to joining {4, 5}. A member of {4, 5} has
no incentive to join {1, 2, 3}. If both coalitions merge to form the coalition,
further deviations will occur. {{1, 2},{3, 4},{5}} is no CPNE since {3, 4}
would like to merge with {5}, which is a CPNE as laid out above.
XII.4 Appendix 4
In this appendix we derive the NE, CPNE and SNE in the exclusive mem-
bership & game. Each country makes an announcement regarding a coali-
tion. A coalition only forms by unanimous agreement. If one country
deviates, the coalition breaks apart.
Since a coalition structure can emerge from different proposals, we shall
only investigate below whether an equilibrium combination of announce-
ments exists leading to a particular coalition structure.
N3
{{1, 2, 3}}: NE since a deviation leads to {{1},{2},{3}} and leaves all coun-
tries worse off. {{1, 2},{3}}: NE since if 1 and 2 make different proposals
Appendices 373
this leads to {{1},{2},{3}} and leaves both countries worse off; 3 has no
incentive to make a different proposal. {1, 2, 3}: NE by definition: if each
country proposes a coalition by itself, no country can form another coali-
tion by the unanimity rule.
The NE coalition structures {{1, 2, 3}} and {{1, 2},{3}} are SNE since
they are Pareto-efficient.
From CSNE CCPNE CNE it follows that these two SNE coalition struc-
tures are also a CPNE. However, {1, 2, 3} is not a CPNE since it is Pareto-
dominated by the other two CPNE.
N4
It is easily checked that all permutations constitute an NE (see the discus-
sion of N3). That is, {{1, 2, 3, 4}}, {{1, 2, 3},{4}}, {{1, 2},{3, 4}}, {{1,
2},{3},{4}}, {{1},{2},{3},{4}} are all NE. Only {{1, 2, 3, 4}}, {{1, 2,
3},{4}} are SNE since they are efficient NE.
These coalitions structures are therefore also a CPNE. Since all other
permutations are Pareto-dominated by these coalition structures no other
CPNE exists.
N5
It is easily checked that all permutations constitute an NE. That is, {{1, 2,
3, 4, 5}}, {{1, 2, 3, 4},{5}}, {{1, 2, 3},{4, 5}}, {{1, 2, 3},{4},{5}}, {{1,
2},{3, 4},{5}}, {{1, 2},{3},{4},{5}}, {{1},{2},{3},{4},{5}} are an NE.
Only {{1, 2, 3, 4, 5}}, {{1, 2, 3, 4},{5}} are SNE. All other NE are
Pareto-dominated by these two SNE.
Accordingly (see the discussion of N4 above), these are also the only
CPNE.
XII.5 Appendix 5
In this appendix we derive the NE, CPNE and SNE in the exclusive mem-
bership ! game. Each country makes an announcement regarding a coali-
tion in which it is a member. A coalition forms of those countries which
have made the same proposal even though not all countries which have pro-
posed it may eventually join the coalition. If one country deviates, the
remaining coalition members remain in the coalition.
Since a coalition structure can emerge from different proposals, we shall
only investigate below whether an equilibrium combination of proposals
exists leading to a particular coalition structure.
N3
{{1, 2, 3}}: NE since a deviation by 3 leads to {{1, 2},{3}} and leaves
3 indifferent. {{1, 2},{3}}: NE since 1 and 2 have no incentive to be in
374 Appendices
singleton coalitions and 3 has no incentive to merge with {1, 2}. {1, 2, 3}:
NE by definition: if each country proposes a coalition by itself, then any
other proposal leaves the coalition structure unaffected.
The NE coalition structures {{1, 2, 3}} and {{1, 2},{3}} are an SNE
since they are Pareto-efficient.
From CSNE CCPNE CNE it follows that these two SNE coalition struc-
tures are also a CPNE. {1, 2, 3} is not a CPNE since it is Pareto-dominated
by the other two CPNE.
N4
It is easily checked that all permutations ({{1, 2, 3},{4}}, {{1, 2},{3, 4}},
{{1, 2},{3},{4}}, {{1},{2},{3},{4}}, are an NE except {{1, 2, 3, 4}} which
is not stand-alone stable.
Only {{1, 2, 3},{4}} is an SNE since all other coalition structures are
Pareto-dominated by the grand coalition.
Thus {{1, 2, 3},{4}} is a CPNE too. Moreover, {{1, 2},{3, 4}} is a CPNE.
No country has an incentive to propose a coalition of three countries since
the proposed joining country and the coalition accepting the new member
would be worse off. The grand coalition, though it would raise all countries’
payoff, is subject to a further deviation. Any smaller coalitions would imply
a payoff loss. {{1, 2},{3},{4}} and {{1},{2},{3},{4}} are no CPNE since
these coalition structures are Pareto-dominated by the CPNE {{1, 2},{3,
4}}.
N5
It is easily checked that all permutations ({{1, 2, 3},{4, 5}}, {{1, 2,
3},{4},{5}}, {{1, 2},{3, 4},{5}}, {{1, 2},{3},{4},{5}}, {{1},{2},{3},{4},
{5}}) constitute an NE except {{1, 2, 3, 4, 5}}, {{1, 2, 3, 4},{5}} which are
not stand-alone stable. For instance, {{1, 2, 3},{4},{5}} is an NE since 1, 2
and 3 are indifferent to leaving the coalition and to operating as singletons.
A merger of 1, 2 or 3 with countries 4 or 5, as well as a merger of 4 and 5
requires a simultaneous change in at least two proposals which is not con-
sidered by the NE concept.
There is no SNE since all NE are Pareto-dominated by the grand coali-
tion.
{{1, 2, 3},{4, 5}} is CPNE since larger coalitions are not stable and there
is no incentive to form smaller coalitions. {{1, 2, 3},{4},{5}} is no CPNE
since it is Pareto-dominated by the CPNE {{1, 2, 3},{4, 5}}. {{1, 2},{3,
4},{5}} is CPNE since country 5 is indifferent to accession of a coalition of
two countries. Any coalition of four or five countries is not stable. {{1,
2},{3},{4},{5}} and {{1},{2},{3},{4},{5}} are not a CPNE since they are
Pareto-dominated by the CPNE {{1, 2},{3, 4},{5}}.
Appendices 375
XII.6 Appendix 6
N3
{{1, 2, 3}}: CS since no player is better off in another coalition. {{1,
2},{3}}: CS since 1 and 2 can only jointly deviate with 3. However, 3 is
indifferent to such an offer. {{1},{2},{3}}: no CS since all players prefer a
grand coalition.
{{1, 2, 3}} and {{1, 2},{3}} are AS (BS) by Cc C C. {{1},{2},{3}}:
no AS (BS) since all players prefer a grand coalition.
N4
{{1, 2, 3, 4}}: no CS since 4 prefers {{1, 2, 3},{4}}. {{1, 2, 3},{4}}: no CS
since 1 and 2 prefer {{1, 2},{3, 4}}. {{1, 2},{3, 4}}: no CS since all prefer
{{1, 2, 3, 4}}. {{1, 2},{3},{4}}, {{1},{2},{3},{4}}: no CS since all players
prefer {{1, 2, 3, 4}}.
{{1, 2, 3, 4}}: AS (BS) since any deviation may lead eventually to {{1},
{2}, {3}, {4}} which is Pareto-dominated. {{1, 2, 3}, {4}}: AS (BS).
Though 1 and 2 prefer {{1, 2},{3, 4}}, they have to reckon with {{1,
2},{3},{4}} which is a Pareto-inferior coalition structure for them. {{1,
2},{3, 4}}, {{1, 2},{3},{4}} and {{1},{2},{3},{4}}: no AS (BS) since the
grand coalition is a Pareto-superior coalition structure.
N5
{{1, 2, 3, 4, 5}} and {{1, 2, 3, 4},{5}} are not stand-alone stable and hence
no CS. All other permutations ({{1, 2, 3}, {4, 5}}, {{1, 2, 3}, {4}, {5}}, {{1,
2}, {3, 4}, {5}}, {{1, 2}, {3}, {4}, {5}} and {{1},{2},{3},{4},{5}}) are
Pareto-dominated by the grand coalition and therefore no CS. Similar
arguments as laid out for N4 establish that {{1, 2, 3, 4, 5}} and {{1, 2, 3,
4},{5}} are AS (BS).
XII.7 Appendix 7
In this appendix the coalition structures are derived by applying Chwe’s far-
sighted coalitional stability concept. For the definition of the largest con-
sistent set, CLCS, see Sub-section 15.4.4. An inequality sign below implies
strict Pareto-dominance.
N3
{{1},{2},{3}}{{1, 2},{3}} ⇔ {{1, 2, 3}}. Hence, CLCS {{1, 2, 3}},{{1,
2},{3}}.
376 Appendices
N4
{{1, 2, 3, 4}}{{1, 2, 3},{4}}. {{1, 2, 3},{4}} is not Pareto-dominated by
any other coalition structure and is therefore stable. {{1, 2},{3, 4}} is
directly dominated by the grand coalition and hence indirectly by {{1, 2,
3},{4}}. Since, however, {{1, 2, 3},{4}} is not preferable by all members of
the original coalition, a deviation is not deemed feasible and {{1, 2},{3, 4}}
is stable. {{1, 2},{3},{4}} and {{1},{2},{3},{4}} are not stable, since they
are Pareto-dominated by {{1, 2},{3, 4}}. Hence, CLCS {{{1, 2, 3},{4}},
{{1, 2},{3, 4}}}.
N5
A deviation from {{1, 2, 3, 4, 5}} to {{1, 2, 3, 4},{5}} is subject to a further
deviation to {{1, 2, 3},{4},{5}}, which is not in the interest of all original
coalition members. A deviation from {{1, 2, 3, 4, 5}} to {{1, 2, 3}, {4, 5}}
is also not in the interest of any country. Smaller coalitions are Pareto-dom-
inated by {{1, 2, 3, 4, 5}}.
{{1, 2, 3, 4},{5}} CLCS since *4({{1, 2, 3, 4},{5}}) *4({{1, 2,
3},{4},{5}}) *4({{1, 2, 3},{4, 5}}) *4({{1, 2, 3, 4, 5}}). {{1, 2, 3},{4, 5}}
and {{1, 2, 3, 4, 5}} deliver a higher payoff to country 4 than {{1, 2, 3,
4},{5}}. That is, country 4 leaves coalition {{1, 2, 3, 4},{5}} since it expects
that the grand coalition will eventually form. All other coalition structures
are Pareto-dominated by the grand coalition and therefore do not belong
to the largest consistent set. Hence, the CLCS {{1, 2, 3, 4, 5}}.
XII.8 Appendix 8
In this appendix the coalition structure is derived by applying the concept
of equilibrium binding agreements. For the definition see Sub-section
15.5.1.
N3
{{1},{2},{3}} is EBA by definition. {{1, 2},{3}} is EBA since it is not
blocked by {{1},{2},{3}}. {{1, 2, 3}} is also EBA since it is not blocked by
{{1, 2},{3}} and {{1},{2},{3}}. By applying Eff(CEBA) (extension to the
original concept), Eff(CEBA){{{1, 2},{3}}, {{1,2,3}}}.
N4
{{1},{2},{3},{4}} is stable by definition. {{1, 2},{3},{4}} and {{1, 2},{3,
4}} are stable since a further partition implies a payoff loss to all countries.
{{1, 2, 3},{4}} is stable since country 3 is indifferent to forming a singleton
coalition. {{1, 2, 3, 4}} is not stable since it is blocked by {{1, 2, 3},{4}}.
Hence, Eff(CEBA){{{1, 2, 3},{4}}, {{1, 2},{3, 4}}}.
Appendices 377
N5
{{1},{2},{3},{4},{5}} is stable by definition. {{1, 2},{3},{4},{5}}, {{1,
2},{3, 4},{5}}, {{1, 2, 3},{4},{5}}, {{1, 2, 3},{4, 5}} are stable since a par-
tition does not pay a country. {{1, 2, 3, 4},{5}} is not stable since country
4 has an incentive to form a singleton. {{1, 2, 3, 4, 5}} is stable since a devi-
ation from {{1, 2, 3, 4, 5}} to {{1, 2, 3, 4},{5}} leads eventually to {{1, 2,
3},{4},{5}} which is Pareto-dominated by the grand coalition. A deviation
by two players to {{1, 2, 3},{4, 5}} is also a Pareto-dominated partition.
Any smaller partitions are also Pareto-dominated and hence the grand
coalition is an efficient equilibrium binding agreement. Eff(CEBA){{1, 2,
3, 4, 5}}.
XII.9 Appendix 9
In this appendix the coalition structure under the sequential coalition for-
mation process is derived. We distinguish between Bloch’s original concept,
Ray and Vohra’s extension which assumes that if a country is indifferent
between two coalitions it accepts the larger coalition and Finus’s extension
of trembling-hand sequential equilibrium (see Sub-section 15.5.2 for
details).
N3
{{1},{2},{3}} is no sequential equilibria (SE) since if country 1 proposes
itself, country 2 will propose {2, 3}. {{1},{2, 3}} is an SE by the previous
argument. {{1, 2, 3}} is also SE since no country has an incentive to reject
the proposal.
Since country 1 is indifferent between {{1, 2, 3}} and {{1},{2, 3}} it pro-
poses {{1, 2, 3}} according to Ray and Vohra’s extension.
Since country 1 reckons that its proposal may be rejected by coincidence
and it may end up in {{1, 3},{2}} (since 2 proposes {2}), it proposes itself
and {{1},{2, 3}} is the equilibrium according to Finus’s extension.
N4
From the incentive profile it is known that if country 3 is in the position to
make a proposal (that is, {1, 2} has formed), it will suggest {3, 4}. If it is
country 2’s turn to make a proposal (that is, {1} has formed), it can choose
between {{1},{2, 3, 4}}, {{1},{2},{3, 4}} and {{1},{2, 3},{4}}. Since
*2({{1},{2, 3},{4}}) *2({{1},{2},{3, 4}}) *2({{1},{2, 3, 4}}), country 2
will propose either {2} or {2, 3, 4}. Thus, country 1 has the option between
(1) {1, 2, 3, 4}⇒{1, 2, 3, 4}; (2) {1, 2, 3}⇒{{1, 2, 3},{4}}; (3) {1, 2} ⇒{{1,
2},{3, 4}}; (4) {1}⇒(a) {{1},{2, 3, 4}} or (b) {{1},{2},{3, 4}}. From
country 1’s perspective option 1 dominates option 2. Option 1 dominates
378 Appendices
NOTES
11. A proof for asymmetric payoffs proceeds along exactly the same lines. Since the nota-
tion is rather messy and confirms Proposition 3.1, the proof is not reproduced here.
12. Multiple deviations by more than one player are ignored and not punished.
13. We assume that players also receive payoffs at time T.
14. Of course, for the entire game 2ui(ai1,ai2)(1i )2ui(ci1,ci2) 0
i I must hold, but
upon dividing through by 2 this is equivalent to (V.3).
15. More precisely, we subtract the LHS term of (V.2)(V.23) from (V.3).
16. i will be an increasing function in any case since 1 ki 0 implies that if the
Stackelberg leader increases emissions, ei, global emissions, ek, will increase, though ej
diminishes.
This last relation can also be used to prove that eST i ei is irrational since this would
17. N
imply ekST ekN, eST j ej and hence j i . By choosing ei ei the leader could at
N ST N ST N
then marginal damage in the NE would be higher than in the social optimum since
i(eNi )i(ei )j(ei )
i I due to j0 (jj /ei ). Then the FOC in the
S S
i ei
i I must
Nash equilibrium would require higher marginal benefits and hence eN S
be true, which contradicts the initial assumption eN i e S. The extension to the cases
i
of corner equilibria would confirm this result.
19. It is easy to see that SOC in the NE are satisfied. Since j /ei2 0 by assumption, this
2
12. It is only sensible to consider the possibility that the bottleneck country i (assuming a
non-biased proposal) makes a strategic proposal. If country j made a strategic proposal
this would imply two Stackelberg followers for which no equilibrium exists. To see this,
note that for a proposal tistr to be effective j(eTj (tistr)) j(ekT(tistr)) and i(eTi
(tistr))i(ekT (tistr)) must hold before adjustment is conducted so that the neighboring
country j adjusts and country i is in the lead. If country i is already the bottleneck
without strategic considerations j(eTj (ti )) j(ekT(ti )) holds for country j before it con-
ducts adjustment. Hence, any tjstr ti would lead to j(eTj (tjstr))j(ekT(tjstr)) and if the
proposal were to be effective i(eTi (tjstr)) i(ekT(tjstr)) would hold. This would basically
imply two Stackelberg followers, which is not possible.
13. All proofs related to Section 13.3 assume an interior Nash equilibrium and social
optimum.
14. Comparing the FOC in the PANE with that in the social optimum, assuming marginal
damages in country i to be given by di and the number of countries in IJ to be given by
N*, we find: i(ẽ iJ )kIJk(ẽ)N i 1di i 1di kIk(e )i(ei ) and hence
* N S S
ei ẽ i i I follows from i 0. Comparing the FOC in the PANE with that in the Nash
S J J
equilibrium, we find: i(ẽ iJ )kIJk(ẽ)N i 1di dik(e )i(ei ) and hence
* N N
eN
i ẽ J
i IJ by 0.
i i
15. To see this, note that from the FOC in the social optimum we have iIi(eS)i(eSi ).
Rearranging terms gives i(eSi )(iIẽ i iIeSi )iI(i(eSi )i(ẽ i )) which is satisfied
since i(eSi ) [i(ẽ iJ )i(eSi )]/[ẽ i eSi ] by the concavity of the benefit function (and ẽ i eSi
i I).
16. For the appropriateness of this assumption in the case of → 1, see Chapter 12.
References
Abreu, D. (1986), ‘Extremal equilibrium of oligopolistic supergames’,
Journal of Economic Theory, 39, 191–225.
Abreu, D. (1988), ‘On the theory of infinitely repeated games with dis-
counting’, Econometrica, 56, 383–96.
Abreu, D., D. Pearce and E. Stacchetti (1986), ‘Optimal cartel equilibrium
with imperfect monitoring’, Journal of Economic Theory, 39, 251–69.
Abreu, D., D. Pearce and E. Stacchetti (1993), ‘Renegotiation and symme-
try in repeated games’, Journal of Economic Theory, 60, 217–40.
Alho, K. (1992), ‘Bilateral transfers and lending in international environ-
mental cooperation’, Environmental and Resource Economics, 2, 201–20.
Andersson, T. (1991), ‘Government failure: the cause of global environ-
mental mismanagement’, Ecological Economics, 4, 215–36.
Arnold, V. (1984), ‘Umweltschutz als internationales öffentliches Gut:
Komparative Kostenvorteile und Verhandlungsgewinne’, Zeitschrift für
Wirtschafts- und Sozialwissenschaften, 104, 111–29.
Arnold, V. (1992), Theorie der Kollektivgüter, Munich: Franz Vahlen.
Aronson, A.L. (1993), ‘From “cooperator’s loss” to cooperative gain: nego-
tiating greenhouse gas abatement’, Yale Law Journal, 102, 2143–74.
Asheim, G.B. (1991), ‘Extending renegotiation-proofness to infinite
games’, Games and Economic Behavior, 3, 278–94.
Aumann, R.J. (1959), ‘Acceptable points in general cooperative N-person
games’, in A.N. Tucker and R.D. Luce (eds), Contributions to the Theory
of Games, Volume IV (Annals of Mathematics Studies 40), Princeton,
NJ: Princeton University Press, pp. 287–324.
Aumann, R. and L. Shapley (1976), ‘Long-term competition: a game-
theoretic analysis’, mimeo.
Ausubel, J.H. and D.G. Victor (1992), ‘Verification of international envi-
ronmental agreements’, Annual Review of Energy and Environment, 17,
1–43.
Avenhaus, R. (1992), ‘Monitoring the emissions of pollutants by means of
the inspector leadership method’, in R. Pethig (ed.), Conflicts and
Cooperation in Managing Environmental Resources: Microeconomic
Studies, Berlin: Springer, pp. 241–69.
Axelrod, R. (1984), The Evolution of Cooperation, New York: Basic Books.
Bac, M. (1996), ‘Incomplete information and incentives to free-ride on
380
References 381
Foley, D. (1970), ‘Lindahl solution and the core of an economy with public
goods’, Econometrica, 38, 66–72.
Folmer, H., P. von Mouche and S. Ragland (1993), ‘Interconnected games
and international environmental problems’, Environmental and Resource
Economics, 3, 313–35.
Førsund, F.R. and E. Naevdal (1994), ‘Trading sulphur emissions in
Europe’, in G. Klaasen and F.R. Førsund (eds), Economic Instruments
for Air Pollution Control, Economy and Environment, vol. 9, pp.
231–48.
Foster, B.A. (1993), The Acid Rain Debate: Science and Special Interests in
Policy Formation, Ames, IA: State University Press.
Friedman, A. (1994), ‘Differential games’, in R. Aumann and S. Hart (eds),
Handbook of Game Theory with Economic Applications, Amsterdam:
Elsevier, pp. 781–99.
Friedman, J.W. (1971), ‘A non-cooperative equilibrium for supergames’,
Review of Economic Studies, 38, 1–12.
Friedman, J.W. (1985), ‘Cooperative equilibria in finite horizon nonco-
operative supergames’, Journal of Economic Theory, 35, 390–98.
Friedman, J.W. (1986), Game Theory with Applications to Economics,
London and New York: Oxford University Press.
Fudenberg, D. and D. Levine (1992), ‘Maintaining a reputation when strat-
egies are imperfectly observable’, Review of Economic Studies, 59,
561–79.
Fudenberg, D. and E. Maskin (1986), ‘The folk theorem in repeated games
with discounting or with incomplete information’, Econometrica, 54,
533–54.
Fudenberg, D. and J. Tirole (1996), Game Theory, 5th edn, Cambridge, MA
and London: MIT Press.
Fudenberg, D., D. Levine and E. Maskin (1994), ‘The folk theorem with
imperfect public information’, Econometrica, 62, 997–1039.
Gardner, R. and E. Ostrom (1991), ‘Rules and games’, Public Choice, 70,
121–49.
Garrod, G. and K.G. Willis (1999), Economic Valuation of the Environment,
Cheltenham, UK: Edward Elgar.
General Accounting Office (GAO) (1992), International Environmental
Agreements Are Not Well Monitored, Washington, DC: United States
General Accounting Office, RCED-92-43.
Germain, M., P.L. Toint and H. Tulkens (1995), International Negotiations
on Acid Rains in Northern Europe: A Discrete Time Iterative Process,
CORE Discussion Paper no. 9556, Louvain: Center for Operations
Research and Econometrics, Université Catholique de Louvain.
Germain, M., P.L. Toint and H. Tulkens (1996), ‘Calcul économique
References 391
Owen, G. (1982), Game Theory, 2nd edn, New York: Academic Press.
Oye, K.A. (1986), ‘Explaining cooperation under anarchy: hypotheses and
strategies’, in K.A. Oye (ed.), Cooperation Under Anarchy, Princeton, NJ:
Princeton University Press, pp. 1–24.
Pearce, D., A. Markandya and E.B. Barbier (1989), Blueprint for a Green
Economy, London: Earthscan.
Pearce, D.G. (1992), ‘Repeated games: cooperation and rationality’, in J.-J.
Laffont (ed.), Advances in Economic Theory: Sixth World Congress of the
Econometric Society, Cambridge: Cambridge University Press, pp.
132–74.
Pethig, R. (1982), ‘Reciprocal transfrontier pollution’, in H. Siebert (ed.),
Global Environmental Resources, Frankfurt: Verlag Peter Lang, pp.
57–93.
Pethig, R. (ed.) (1992), Conflicts and Cooperation in Managing
Environmental Resources: Microeconomic Studies, Berlin: Springer.
Petrakis, E. and A. Xepapadeas (1996), ‘Environmental consciousness and
moral hazard in international agreements to protect the environment’,
Journal of Public Economics, 60, 95–110.
Ragland, S.E., L.L. Bennett and P. Yolles (1996), ‘Unidirectional problems
with international river basins: a game theoretic approach’, paper pre-
sented at the European Meeting of Resource and Environmental
Economists, Lisbon, 26–29 June.
Raiffa, H. (1982), The Art and Science of Negotiation, Cambridge, MA:
Harvard University Press.
Rasmusen, E. (1989), Games and Information: An Introduction, Oxford and
New York: Basil Blackwell.
Rasmusen, E. (1995), Games and Information. An Introduction to Game
Theory, 2nd edn, Oxford and Cambridge, MA: Basil Blackwell.
Ray, D. (1994), ‘Internally renegotiation-proof equilibrium sets: limit
behavior with low discounting’, Games and Economic Behavior, 6,
162–77.
Ray, D. and R. Vohra (1997), ‘Equilibrium binding agreements’, Journal of
Economic Theory, 73, 30–78.
Ray, D. and R. Vohra (1999), ‘A theory of endogenous coalition structure’,
Games and Economic Behavior, 26, 286–336.
Richer, J. and J.K. Stranlund (1997), ‘Threat positions and the resolution
of environmental conflicts’, Land Economics, 73, 58–71.
Rosen, J.B. (1965), ‘Existence and uniqueness of equilibrium points for
concave n-person games’, Econometrica, 33, 520–34.
Roth, A. (1979), Axiomatic Models of Bargaining, Berlin: Springer.
Roth, A. (ed.) (1985), Game-theoretic Models of Bargaining, Cambridge:
Cambridge University Press.
References 399
405
406 Index
finite dynamic games with discrete Stackelberg 152–4, 156, 162, 190
strategy space 42–61, 75–88 strategic matching and 168–71
conceptual framework 50–57 subgame-perfect (SPE) 45, 53–4,
examples and first results 43–50 149, 250
extensions 75–9 infinite dynamic games with
general results 57–61, 79–85 continuous strategy space
strongly perfect equilibrium 85–7 194–5, 202–4
infinite dynamic games with Nash 46–7, 57–8, 59–60
continuous strategy space prisoners’ dilemma 76–8
194–217 sequential move unanimity game
discount factors close to 1 194–209 304–5
discount factors smaller than 1 strongly perfect (SSPE) 85–7,
210–17 101–2, 202–4
infinite dynamic games with discrete exclusive membership games 295–7
strategy space 63–73, 89–102 expectations
Folk Theorems 66–72 convergent 125
strongly perfect equilibrium 101–2 rational 26
strongly renegotiation-proof extensive form games 44
equilibrium 99–101 subgames in 46
weakly renegotiation-proof externalities
equilibrium 89–99 filterable 152–5
static representations of 150–52, negative externality games 63, 285,
250 286
positive externality games 10, 285–6
Ecchia, G. 308 transferable 155–7
Eichberger, J. 5, 9, 26, 27, 44, 52, 53, Eyckmans, J. 192
58, 61
emissions see global emissions game fairness, coalitions 238–9
Endres, A. 1, 4, 20, 102, 147, 173, 177, Fankhauser, S. 181
178, 193, 217, 254 Farrell, J. 89, 90, 91, 100, 101, 102, 316
equilibrium of the game 7–8 farsighted coalitions 243, 298–300
auction 164–6, 201 feedback strategies 13
bargaining 180, 183–90 Feeny, D. 40
coalition-proof equilibrium 75 Fees, E. 5
conjectural variation coalition 222 Felder, S. 255
in dominant strategies 23, 31 Feldman, A.M. 11
dynamic adjustment to socially Fibonacci decomposition 305
optimal steady-state filterable externalities 152–5
equilibrium 252–4 finite games 13
equilibrium binding agreement finite dynamic games with
(EBA) 300–302 continuous strategy space
Nash see Nash equilibrium 149–72
non-Nash/hybrid behavior 159–62 auctioning emission reductions
ratio 251 150, 162–6
renegotiation-proof equilibrium filterable externalities 152–5
(RPE) 75, 76, 79–85, 87–8 non-Nash/hybrid behavior 150,
strong 99–101, 108, 195–202, 157–62
210–16 strategic matching 167–71
weak 89–99, 195–202, 210–16, theory of reciprocity 172–3
258, 261–2, 268–9, 279 transferable externalities 155–7
Index 409