Journal of Economic Perspectives—Volume 30, Number 4—Fall 2016—Pages 131–150

New Directions for Modelling Strategic

Behavior: Game-Theoretic Models of
Communication, Coordination, and
Cooperation in Economic Relationships
Vincent P. Crawford

H
alf a century ago, before the game-theoretic revolution that began in
the 1960s and 1970s, economics largely lacked the tools to analyze stra-
tegic interactions. There was clearly a perceived need for such tools,
and considerable excitement had greeted the publication of von Neumann and
Morgenstern’s Theory of Games and Economic Behavior (1944, 1947, 1953). But despite
the initial excitement, for several decades game theory remained mostly a branch of
mathematics, whose economic applications were the work of a few pioneers, such as
Nash (1950, 1953), Schelling (1960), Shapley and Shubik (1954, 1971), and Shubik
(1959). Some economists, making a virtue of presumed necessity, claimed that ques-
tions involving strategy or information were unimportant. A memorable example
is Rothschild’s (1973, p. 1283) quoting of a “prominent” colleague: “The friction
caused by disequilibrium and lack of information accounts for variations in the
numbers we observe at the fifth or sixth decimal place. Your stories are interesting
but have no conceivable bearing on any question of practical economic interest.”
Finally, in the 1960s, 1970s, and 1980s, game theory began to change the land-
scape of economics. If economists from that time could examine a modern graduate
microeconomics text (such as Mas-Colell, Whinston, and Green 1995—still thoroughly
“modern”), they would find their theories of market competition transformed
beyond recognition, with rich, explicit game-theoretic analyses of preemption and

■ Vincent P. Crawford is Drummond Professor of Political Economy, University of Oxford,

and Fellow of All Souls College, Oxford, England. He is also Distinguished Professor Emeritus
and Research Professor, University of California, San Diego, La Jolla, California. His email
address is [email protected].
†
For supplementary materials such as appendices, datasets, and author disclosure statements, see the
article page at
http://dx.doi.org/10.1257/jep.30.4.131 doi=10.1257/jep.30.4.131
132 Journal of Economic Perspectives

entry deterrence; signalling and screening with asymmetric information; competition

via explicit and/or implicit contracts; and platform and network competition. They
would also find unfamiliar but flourishing subdisciplines on game-theoretic topics
such as auctions; bargaining and coordination; agency and contract theory; strategic
communication; social choice; public goods; cooperation in long-term relationships;
and design of markets and other institutions. Such analyses, whose strategic aspects
had made them seem intractable, now make up most of the microeconomics core
in leading graduate programs. In the 21st century, game theory has fulfilled a large
part of its promise, giving systematic, illuminating analyses of many central ques-
tions. Indeed, game theory has also begun to unify the rest of the social sciences,
transforming parts of political science, computer science, and evolutionary biology—
though not yet having as much effect on anthropology, sociology, or psychology.
Although most of the research that revolutionized game theory was done by
economists, the revolution was not primarily a question of economics coming to
game theory. Rather, game theory and economics coevolved, with game theory
supplying a precise and detailed language for describing strategic interactions and
a set of assumptions for predicting strategic behavior, while economics contributed
questions and intuitions about behavior against which game theory’s predictions
could be tested and improved. In the process, the research frontier shifted from
the earlier stages of figuring out how to model economic interactions as games
and getting the logic of rational strategic behavior right to a later emphasis on
relaxing unnecessary restrictions and refining behavioral assumptions. As game
theory enriched economics, economics drove adaptations of game theory’s assump-
tions and methods, transforming it from a branch of mathematics with a primarily
normative focus into a powerful tool for positive economic analysis with a mainly
descriptive or predictive focus.
In this paper, I discuss the state of progress in applications of game theory in
economics and try to identify possible future developments that are likely to yield
further progress. To keep the topic manageable, I focus on a canonical economic
problem that is inherently game-theoretic, that of fostering efficient coordination
and cooperation in relationships, with particular attention to the role of communica-
tion. I thus favor microeconomics, omitting important macroeconomic applications
of game theory such as Summers (2000), Garcia-Schmidt and Woodford (2014),
and Evans and McGough (2015), whose discussions of financial crises and expecta-
tions formation nonetheless touch on some of the game-theoretic issues discussed
here. I also favor noncooperative game theory, omitting notable successes of coop-
erative game theory.1 I further narrow the focus to problems specific to game theory

1
The established terms “noncooperative” and “cooperative” game theory are misnomers, in that,
paradoxically, noncooperative game theory is better suited to explaining (as opposed to assuming)
cooperation than cooperative game theory. Noncooperative game theory starts with a detailed model
of the structure of a game and makes specific assumptions about how rational players will respond to
it. Cooperative game theory starts instead with a general description of the structure, sidestepping most
details, and makes general assumptions intended to characterize the possible outcomes of frictionless
bargaining among rational players. A notable economic application of cooperative game theory is the
 Vincent P. Crawford 133

by assuming that individuals are rational in the decision-theoretic sense of choosing

strategies that are best responses to consistent beliefs.
I begin with an overview of noncooperative game theory’s principal model of
behavior, Nash equilibrium, henceforth shortened to equilibrium. I next discuss the
alternative “thinking” and “learning” rationales for how real-world actors might
reach equilibrium decisions. I then review how equilibrium has been used to
model coordination, communication, and cooperation in relationships, and discuss
possible developments. Throughout the paper, I make no attempt at comprehen-
sive coverage or referencing, with apologies to those whose work is slighted.

The Notion of Equilibrium in Noncooperative Game Theory

Equilibrium is defined as a combination of decision rules or strategies, one for

each decision maker or player, in which each player’s strategy maximizes her/his
personal expected utility or payoff given the strategies of others who are deciding
in the same way. The generality, tractability, and precision of equilibrium analysis
have made it the method of choice in most economic applications of game theory
(Myerson 1999). However, equilibrium goes well beyond the notion of rationality
of individual decisions in that it requires a particular relationship among players’
strategies. How players’ strategies might come to be in equilibrium is a difficult
question, which is still on the research frontier and which is intimately related to the
question of how players can foster coordination and cooperation in relationships,
as explained below.
Consider the game in Figure 1, in which the players choose their moves simul-
taneously and it is assumed that the game’s structure is known to the players as
common knowledge, in the sense that each player knows the structure, including what
the other knows; knows that the other knows the structure; and so on.
This game has a unique equilibrium, in which the Row player (whose payoffs are
in the lower-left corners of the cells of the matrix) chooses the strategy Middle and
the Column player (whose payoffs are in the upper-right corners) chooses Center.
To see this, note that if Row chooses Middle, Column will look across the choices
of Left, Center, or Right, and see that Center then has the highest payoff (its two is
better than the zeros for Left or Right). Further, if Column chooses Center, Row will
look across the choices of Top, Middle, or Bottom, and see that Middle then has the
highest payoff (its two is better than the zeros for Top or Bottom). The outcome of
{Middle, Center} is therefore an equilibrium. The reader can confirm that, starting

theory of matching markets, which uses game-theoretic notions like the core to model the outcomes
of competition, with or without prices, among heterogeneous traders. For more on matching theory
and applications, in the context of the Nobel Memorial Prize in Economic Sciences awarded to Lloyd
Shapley and Alvin Roth in 2012, see Economic Sciences Prize Committee of the Royal Swedish Academy
of Sciences (2012).
134 Journal of Economic Perspectives

Figure 1
Equilibrium and Rationalizability: A Dominance-Solvable Game

Column player

Left Center Right

0 5 3
Top
Row player 7 0 0
0 2 0
Middle
5 2 5
7 5 3
Bottom
0 0 7

Note: This game has a unique equilibrium, in which the Row player (whose payoffs are in the lower-left
corners of the cells of the matrix) chooses the strategy Middle and the Column player (whose payoffs are
in the upper-right corners) chooses Center. A strategy choice is strictly dominated by another if it yields
a strictly lower payoff regardless of what choice another may make. The game in Figure 1 is dominance-
solvable: Row knows that Column is rational, and thus knows that Column will not play Right, which is
strictly dominated by Center. In turn, Column knows that of the remaining choices, Row will not play
Bottom, which is strictly dominated by Middle once Column’s strategy Right is eliminated. Next, Row
knows that of the remaining choices, Column will not play Left, which is strictly dominated by Center
once Row’s strategy Bottom is eliminated. The fourth step then leads precisely to the {Middle, Center}
equilibrium.

from any other cell, holding one player’s choice constant, the other would prefer
to switch to a different choice, so {Middle, Center} is the only equilibrium.2
However, just knowing that {Middle, Center} is the unique equilibrium is not
enough to ensure that rational players will make those choices. Suppose players
have possibly probabilistic beliefs about each other’s strategy choices. Then in
the game in the first panel, a rational Row will play Middle only if Row’s beliefs
assign high enough probability to Column playing Center. Conversely, if Row’s
beliefs assign high probability to Column’s choosing Left or Right, then Row
will be tempted to play Top or Bottom. By contrast, a rational Column will never
play Right, because for Column that choice is strictly dominated, meaning that for
Column, Right yields a strictly lower payoff than Center, without regard to Row’s
strategy choice. But a rational Column might play Left, if Column’s beliefs assign
high probability to Row’s choosing Bottom.
How can this ambiguity of rationality-based predictions be resolved?3 One
common approach is to strengthen the rationality assumption by making players’
rationality (in addition to the structure of the game) common knowledge, in the sense
that all players are rational, all know that all are rational, and so on ad infinitum.

2
I ignore randomized, or mixed, strategies throughout the paper, and they are irrelevant to the points I
make here.
3  
Manski (2003) has argued that economists should be tolerant of ambiguous predictions or as he calls
them, incomplete models. However, his main focus is on modelling individual decisions. In games, ambig-
uous predictions of individual decisions frequently “multiply up” to create severe ambiguity of predicted
game outcomes (Aradillas-Lopez and Tamer 2008).
 New Directions for Modelling Strategic Behavior 135

Figure 2
Equilibrium and Rationalizability: A Unique Equilibrium without Dominance
Column player

Left Center Right

0 5 7
Top
Row player 7 0 0
0 2 0
Middle
5 2 5
7 5 0
Bottom
0 0 7

Note: Like the game of Figure 1, this game also has the unique equilibrium of {Middle, Center} (Row
player choses Middle and Column player chooses Center). However, this problem cannot be solved by
iterated strict dominance.

Common knowledge of rationality does in fact yield a unique prediction in the game
in Figure 1, which is dominance-solvable—meaning if players eliminate their strictly
dominated strategies, and after that, their strategies that become strictly dominated
once others are eliminated, and so on, the game gradually reduces to one in which
only the unique equilibrium choices remain. The logic of the argument works like
this: Row knows that Column is rational, and thus knows that Column will not play
Right, which is strictly dominated by Center. In turn, Column knows that of the
remaining choices, Row will not play Bottom, which is strictly dominated by Middle
once Column’s strategy Right is eliminated. Next, Row knows that of the remaining
choices, Column will not play Left, which is strictly dominated by Center once Row’s
strategy Bottom is eliminated. The fourth step then leads precisely to the {Middle,
Center} equilibrium. In dominance-solvable games whose players have more strate-
gies, such epistemic reasoning may go on even longer before reaching equilibrium.
Now consider the game in Figure 2. It also has a unique equilibrium: If Row
plays Middle, then the best choice for Column is Center; and if Column plays
Center, the best choice for Row is Middle. But in that game, no choice is strictly
dominated for either player, and so, even with common knowledge of rationality,
epistemic logic alone does not narrow the possibilities down to a single outcome.
In fact, for any strategy combination in this game, one can construct a “tower”
of beliefs to show that it is consistent with common knowledge of rationality. A
rational Row, for instance, might play Top because of a belief that Column will play
Left (hoping for the high payoff of 7), while a rational Column might play Left
because of a belief that a rational Row will play Bottom (hoping for the high payoff
of 7). Some beliefs that are consistent with common knowledge of rationality lead
to the equilibrium, but most do not.
More generally, Bernheim (1984) and Pearce (1984) showed that common
knowledge of rationality, with no further restrictions on beliefs, implies only that
each player’s strategy is rationalizable, which can be iteratively defined as follows. A
1-rationalizable strategy is one for which there is some profile of others’ strategies
136 Journal of Economic Perspectives

that makes it a best response; a 2-rationalizable strategy is one for which there is
a profile of others’ 1-rationalizable strategies that makes it a best response; and
so on. A rationalizable strategy is then one that is k-rationalizable for all k. In the
game in Figure 1, the choices of Middle for Row and Center for Column are both
­4-rationalizable, referring to the four steps in which the players eliminate various
choices via iterated strict dominance; and four rounds of iterated strict dominance
identify the unique equilibrium. In the second game (Figure 2), all of each player’s
strategies are k-rationalizable for all k, and the equilibrium cannot be identified by
iterated strict dominance, even though it is also unique.
The dominance-solvability of Figure 1’s game is atypical in applications, as
indeed is the uniqueness of the equilibrium in the games of Figure 1 and 2. As the
examples suggest, equilibrium is a much stronger behavioral assumption than the
rationalizability that follows from common knowledge of players’ rationality. It also
requires that players’ beliefs be coordinated.

Equilibrium via Thinking or Learning

If epistemic arguments based on common knowledge of rationality do not

justify the coordination of players’ beliefs or strategies required for equilibrium,
how might it be justified? Assuming for simplicity that the structure of the game is
common knowledge and that the players know that each other is rational, it is useful
to sort applications into two groups according to the most plausible rationale for
equilibrium: “thinking” applications, in which players can plausibly reason their way
to an equilibrium; and “learning” applications, in which it is plausible that players
who adjust their strategies adaptively will converge over time to some equilibrium.
The rationale for assuming equilibrium affects the credibility of assuming equilib-
rium in applications, so I now discuss each approach.

Thinking Applications
In thinking (as opposed to learning) applications, players play a game with
no prior experience with analogous games. If assuming equilibrium is justified,
it must then be because players can reason their way to equilibrium beliefs and
strategy choices. In theory, this is possible if there is a commonly known principle
that focuses players’ beliefs on a unique prediction, because in the standard frame-
work such common knowledge implies that their beliefs must be the same, and
therefore, given rationality, in equilibrium. (For a good introduction to epistemic
game theory in this journal, see Brandenberger 1992.) In this view, equilibrium
becomes an equilibrium in beliefs, in which rational players’ beliefs are statistically
correct, given the best responses they imply.
Applications for which the thinking justification for equilibrium is behavior-
ally plausible are limited because in all but the simplest games the reasoning it
requires is dauntingly complex. In Figure 1’s dominance-solvable game, such
reasoning requires four iterative rounds; and in the second, Figure 2, game, finding
 Vincent P. Crawford 137

the equilibrium requires what is called “fixed-point reasoning,” whereby players’

strategy choices are justified as best responses to others’ choices in a two-way recur-
sion. (That is, one player’s choice is a best response to the other’s, and vice versa;
dominance reasoning is also recursive, but only one-way.) In experiments that elicit
subjects’ initial responses to games, and that separate fixed-point and other kinds
of strategic reasoning, subjects rarely follow fixed-point reasoning or indefinitely
iterated dominance (Crawford, Costa-Gomes, and Iriberri 2013, Section 3). It is some-
times suggested that experienced decision makers will nonetheless use fixed-point
reasoning when the stakes are high, but I have yet to find even anecdotal evidence
that quants or artificial intelligence analysts of poker use reasoning that subtle.
Equilibrium reasoning becomes still more complex when the game has multiple
equilibria. The logic of epistemic equilibrium-in-beliefs requires a selection among
equilibria because a player who is unsure which equilibrium others have in mind
will not generally find it rational to play her or his part of any particular equilib-
rium. Further, many important applications have multiple strict equilibria, in which
each player has a strict preference for a strategy given others’ strategies; in which
case, unique equilibrium selection requires common knowledge of a complex coor-
dination refinement, designed (unlike most equilibrium refinements) to discriminate
among such strict equilibria. The leading examples of such refinements are from
Harsanyi and Selten’s (1988) classic work A General Theory of Equilibrium Selection
in Games, which is part of the work for which they shared the 1994 Nobel prize in
economics with John Nash. Their notion of payoff-dominance favors equilibria whose
payoffs are not Pareto-inferior to those of other equilibria. Their alternative notion
of risk-dominance favors equilibria with (roughly) larger “basins of attraction,” that is,
larger sets of beliefs that make their strategies best responses. Harsanyi and Selten
showed that a logically consistent theory could be built on those foundations, with
added tie-breaking devices, to select a unique equilibrium in any (finite matrix)
game. It was a major achievement to show that such a theory could be constructed;
but it rests on some unavoidably arbitrary choices and is complex enough to render
it far from compelling, behaviorally (for discussion, see Aumann’s “Foreword” to
Harsanyi and Selten’s 1988 book). But compellingness is essential in applications
that involve thinking about multiple equilibria.4
How do people choose their strategies in thinking applications if fixed-point or
indefinitely iterated dominance reasoning or equilibrium selection principles are
too complex to focus their beliefs on an equilibrium? It may seem unlikely that any

4
Some theorists believe the problem of equilibrium selection via thinking in games with multiple strict
equilibria is settled by “global games” analyses (Carlsson and van Damme 1993). Such analyses add
privately observed payoff perturbations to the original game in a way that makes the game dominance-
solvable, and in simple coordination games makes the risk-dominant equilibrium in the unperturbed
game the unique equilibrium. Although such analyses provide a systematic way to analyze how the infor-
mation structure influences equilibrium selection, I believe they do not provide a conclusive argument
for selecting the risk-dominant equilibrium, because the payoff perturbations are artificially introduced,
and a behaviorally implausibly high number of rounds of iterated dominance are often needed to reach
equilibrium in the perturbed game.
138 Journal of Economic Perspectives

alternative model can predict observed behavior systematically better than a rational
expectations notion such as equilibrium, or that such a model could be identified
from among the enormous number of possible models. However, a growing body
of experimental work surveyed in Crawford, Costa-Gomes, and Iriberri (2013,
section 3) shows that subjects’ initial responses to games often follow simple level-k
(Costa-Gomes, Crawford, and Broseta 2001; Costa-Gomes and Crawford 2006) or
cognitive hierarchy (Camerer, Ho, and Chong 2004) rules, in which players anchor
their beliefs in a naive model of others’ responses to the game and then adjust their
beliefs by thinking through a small number of iterated best responses, a number
which varies across players but with a stable population distribution. Such rules are
decision-theoretically rational, and in sufficiently simple games they mimic equi-
librium strategies. In more complex games, such rules may lead to outcomes that
deviate systematically from equilibrium. Importantly, level-k or cognitive hierarchy
models predict not only that deviations from equilibrium will sometimes occur, but
also which settings are likely to evoke them, the forms they are likely to take, and
their relative frequencies. When applied to games with multiple equilibria, with
estimated population frequencies of rules, they predict selection among equilibria
(or not), while avoiding the complexity of coordination refinements.
The literature on strategic thinking in initial responses to games is evolving
rapidly, and level-k and cognitive hierarchy models are mentioned here not as
the last word, but to illustrate that structural nonequilibrium models of strategic
thinking are possible, and can be helpful.

Learning Applications
In learning applications, players have ample prior experience with closely anal-
ogous games. The learning process is modelled as repeated play of a given game,
with the game that is repeated called the stage game and each stage game normally
with a different partner.5 Players’ choices are modelled as adaptive learning, in which
they adjust their stage-game strategies over time in ways that increase their own
stage-game payoffs on the (usually false) assumption that others’ stage-game strat-
egies will continue as before. In adaptive learning models, players have a strong
tendency to converge to some equilibrium in the stage game. There are few general
theoretical results, but there is strong experimental support for such convergence.
Even if learning assures convergence to some equilibrium, nonequilibrium
strategic thinking often remains relevant. Suppose that only long-run outcomes
matter, but the stage game in the applications has multiple equilibria, as in many
important applications. Then all we need from game theory is a reliable prediction
of the prior probability distribution of the possible equilibrium outcomes. But with

5
Unless players’ partners vary across the stage games, repeated-game strategies are relevant, and it is
implausible that players focus on their choices of stage-game strategies, stage by stage, as opposed to
thinking about the effectiveness of their alternative repeated-game strategies. I ignore the literature on
rational learning models, in which players are assumed to play an equilibrium in the repeated game that
describes the entire learning process, because this approach seems less useful than adaptive learning
models in applications (for example, Crawford 2001, Section 6.4.4).
 New Directions for Modelling Strategic Behavior 139

multiple equilibria, learning dynamics are normally history-dependent, so people’s

initial responses influence limiting outcomes, as do the structures of their learning
rules (Van Huyck, Cook, and Battalio 1997; Crawford 1995, 2001; Camerer and Ho
1998, 1999).6 Moreover, even if the stage game has a unique equilibrium, analo-
gies across games are rarely as close in applications as adaptive learning models
assume, and analysis of the thinking needed to learn from imperfect analogies has
just begun (Rankin, Van Huyck, and Battalio 2000; Van Huyck and Battalio 2002;
Cooper and Kagel 2003; Samuelson 2001).

Communication, Coordination, and Cooperation in Economic

Relationships

To perform well, an economic relationship must solve one or more strategic

problems. For example, players may face incentive problems that encourage them not
to cooperate, even though cooperating would be in both of their interests, as in
the well-known Prisoner’s Dilemma game discussed further below. Players may face
assurance problems that make it seem too risky to play the strategies that would lead
to efficient equilibria, as in the Stag Hunt game discussed below. Players may also
face bargaining/coordination problems that make it difficult to coordinate on one of
multiple efficient equilibria.7 To be useful in applications, game theory must offer
theoretically coherent and behaviorally credible analyses of these kinds of problems.
In the standard approach to these problems, in one-shot interactions all three
problems are solved (or not) within an equilibrium, sometimes augmented by prin-
ciples that govern selection among multiple equilibria. In the standard approach to
repeated interactions, all three problems are again solved (or not) within an equi-
librium, now with refinements like subgame-perfect equilibrium applied to the game
that describes the entire relationship.8 Solving strategic problems may look quite
different in situations of static play or repeated play—although if a game is repeated
with fixed rather than varying partners, it becomes a game that is effectively static,

6
Some theorists consider the problem of equilibrium selection via learning to be settled by analyses of
“long-run equilibria” (Kandori, Mailath, and Rob 1993; Young 1993). But those analyses achieve equilib-
rium selection by modelling the dynamics of learning as ergodic and passing to the limit as randomness
in the dynamics becomes negligible. Neither feature seems realistic, nor do the results seem to corre-
spond closely to equilibrium selection in the lab or the field (Crawford 2001).
7
Here I follow Schelling (1960) and Roth (1987; see also Crawford 1997, Section 5.3) in suggesting that
most real bargaining is best modelled as unstructured and is then primarily a coordination problem, not
a problem that is resolved via delay costs in the subgame-perfect equilibrium of a game with artificially
imposed timing of offers and counteroffers (Rubinstein 1982).
8
A subgame is any part of a game that remains after part of it has been played. A subgame-perfect equilibrium
is an equilibrium strategy profile that induces an equilibrium in every subgame. In effect, subgame-
perfect equilibrium adds a time-consistency requirement to the notion of equilibrium. This notion can
be generalized to games with asymmetric information, via notions called “sequential equilibrium” or
“perfect Bayesian equilibrium” (Mas-Colell, Whinston, and Green 1995, chapters 8–9).
140 Journal of Economic Perspectives

Figure 3
Prisoner’s Dilemma

Cooperate Defect

Cooperate 3 5

3 0
0 1
Defect
5 1

with players making a one-time choice among strategies that describe how they will
act as the game unfolds.
If players can communicate during their interactions, that is usually modelled
via “cheap talk” messages, which involve no direct payoff consequences and have no
power to commit players to actions (Crawford and Sobel 1982; or in this journal,
Farrell and Rabin 1996).
In this section, I begin to explore new directions for modelling communica-
tion, coordination, and cooperation in relationships. I first explain the standard
approaches, and then suggest alternative directions that seem likely to be feasible
and potentially useful.

Coordination and Cooperation in Long-Term Relationships

Most existing work in repeated games seeks to identify ways to support coopera-
tion in a subgame-perfect equilibrium of the infinitely repeated Prisoner’s Dilemma
or some other well-behaved repeated game (Fudenberg and Maskin 1986). Consider,
for instance, the version of the Prisoner’s Dilemma in Figure 3. The best symmetric
outcome arises if both players choose Cooperate. However, Defect is a dominant
strategy for each player in that Defect yields each player a strictly higher payoff than
Cooperate whether or not the other player chooses Cooperate. Is there a way to
support cooperation in equilibrium in relationships based on the Prisoner’s Dilemma?
First suppose that the Prisoner’s Dilemma in Figure 3 is played repeatedly for
a potentially infinite number of times by the same two players; and after any given
number of plays the conditional probability of continuing remains bounded above
zero, so it never becomes common knowledge that any particular period will be the
last one in the relationship. Assume that players will choose a strategy to maximize
their payoffs added across plays of the game, but downweight future payoffs with a
discount factor (because otherwise undiscounted payoffs over an infinite horizon
are not well-defined). A relatively low discount factor means that the players do
not place much weight on future payoffs. In that case, only repeated choices of
{Defect, Defect} are consistent with equilibrium. By contrast, a high discount factor
means that the players place high weight on future payoffs; and that is enough
to make {Cooperate, Cooperate} consistent with subgame-perfect equilibrium. For
example, both players could follow the “grim trigger” strategy “Cooperate until the
 Vincent P. Crawford 141

other player Defects, then Defect forever,” which happens to be a subgame-perfect

­equilibrium and yields the outcome {Cooperate, Cooperate} in every period.
Why is it important that the game be played a potentially infinite number of
times? Suppose instead that the Prisoner’s Dilemma in Figure 3 is played repeatedly
for a commonly known, finite number of times by the same two players. Assume again
that players’ preferences are defined by the addition of players’ payoffs across plays
of the game, with or without discounting. The unique equilibrium then entails both
players choosing Defect in every period. During the last period of the game—which
is known in advance—Defect is a dominant strategy for both players. Knowing that,
in the second-to-last period, the players recognize that they cannot avoid Defect-
Defect in the last period, and Defect is therefore a conditionally dominant strategy
for both. Working backward, both players will see that Defect is a dominant strategy
in every period.
Most existing work in repeated games assumes that players’ beliefs will focus,
with certainty, on a particular subgame-perfect equilibrium as common knowledge,
and seeks to characterize the “Folk Theorem” set of outcomes, those which are
consistent with some such subgame-perfect equilibrium (for more on the Folk
Theorem, see the Wikipedia entry: https://en.wikipedia.org/wiki/Folk_theorem_
(game_theory)). However, with a potentially infinite horizon, the set of equilibria
is usually enormous. In the Prisoner’s Dilemma in Figure 3, consider the strategy
combination: “Row initially Cooperates and then alternates between Defect and
Cooperate, and Column always Cooperates—in each case until either player devi-
ates (that is, Row Defecting when not supposed to, or Column Defecting at all), in
which case both players Defect from now on.” In this asymmetric strategy combi-
nation, Row does better and Column worse than in the symmetric equilibrium
described above (where both use the grim-trigger strategy), but if the discount
factor is high enough, then the punishments for deviations are costly enough to
make it a subgame-perfect equilibrium as well. And there are many others.
This multiplicity of equilibria is an important difficulty because, in applications,
uncertainty about one’s partner’s strategic thinking is of the essence. The complexity
of repeated-game equilibria and the general difficulty of equilibrium selection make
the thinking justification especially implausible here. And in real long-term rela-
tionships, players’ opportunities for learning about the effectiveness of alternative
repeated-game strategies are limited (but see Dal Bó and Fréchette 2011 for some
intriguing experimental evidence on learning repeated-game strategies).
Applications of such repeated-games analyses must confront a number of issues,
of which I mention four. First, Folk Theorem equilibria are normally supported by
extreme punishments even for tiny deviations. (The punishments are often taken
to be more extreme than necessary to support cooperation because that gives a
cleaner characterization of the Folk Theorem set.) Imagine a relationship slightly
more complicated than a repeated Prisoners’ Dilemma, whose players start out
with different beliefs about their repeated-game strategies: For example, one player
might believe they are playing “Cooperate until a player defects, then defect forever,”
while the other believes they are playing the asymmetric strategy combination
142 Journal of Economic Perspectives

described in the previous paragraph. In this case, they will deviate while intending
to cooperate, and the trigger strategies meant to support their cooperation will end
cooperation. This brittleness suggests that in applications, people will favor strate-
gies that are more robust to deviations. There are few such analyses, but see Porter
(1983), van Damme (1989), and Friedman and Samuelson (1994).
A second issue involves the ambiguity of predictions associated with the extreme
multiplicity of equilibria in repeated games analyses. This ambiguity has been a serious
impediment to empirical applications, and I believe that it has slowed the co-evolution
of theory, experiment, and empirics that has been such a powerful engine of progress
in other parts of game theory. Perhaps surprisingly, there seems reason to hope that
closer attention in theoretical analyses to the need for strategies to be robust will,
as a side benefit, help reduce ambiguity of predictions of players’ behavior. Recent
experimental work by Blonski, Ockenfels, and Spagnolo (2011), Breitmoser (2015),
and others suggests the possibility of better and more precise theory.
A third issue is that long-term relationships enable strategic teaching, in which a
player whose future cooperation with current partners is worth preserving may try to
benefit by deviating from a short-run payoff-maximizing strategy in a way that could
influence others’ future beliefs and choices (Camerer, Ho, and Chong 2002). For
instance, in the repeated Prisoner’s Dilemma game of Figure 3, Row would benefit
if it were possible to teach Column to play the asymmetric equilibrium described
above (Row initially Cooperates and then alternates between Defect and Cooperate,
and Column always Cooperates—in each case until either player deviates) rather
than the symmetric equilibrium in which both players follow the “grim trigger”
strategy, which shares the surplus equally. Row could try to teach Column by devi-
ating from the latter equilibrium, risky as that is. Such considerations highlight the
importance of robustness, but are assumed away in a standard equilibrium analysis.
Van Huyck, Battalio, and Beil’s (1990) experiments with two-person minimum-
effort coordination games, like the Stag Hunt game presented below but with
seven symmetric Pareto-ranked equilibria, provide an intriguing example of stra-
tegic teaching. When their subjects played the games in fixed pairs, but with only
one repetition per play, many of them adjusted their current decisions to try to
teach their partners to coordinate more efficiently, and 12 of the 14 subject pairs
converged via various routes to the most efficient equilibrium. (By contrast, subjects
in the analogous treatment with random re-pairing of partners did not try to teach
their partners, and had significantly worse outcomes.) The puzzle is how did they
learn enough about the effectiveness of their alternative repeated game strategies to
play the efficient stage-game equilibrium? In Crawford (2002), I suggested that Van
Huyck et al.’s results might be explained by a strategic teaching model like that of
Camerer, Ho, and Chong (2002), in which some players are adaptive learners while
others are forward-looking and sophisticated in the sense of best responding to the
correct mixture of adaptive and sophisticated subjects.
The fourth and last issue I will mention here is that in most standard repeated-
game models, players who are well-informed about the structure of the game have
nothing to communicate in equilibrium, so such models imply no substantive role
 New Directions for Modelling Strategic Behavior 143

Figure 4
Stag Hunt

Stag Rabbit
9 8
Stag
9 1
1 7
Rabbit
8 7

for communication (for a recent exception, see Awaya and Krishna 2016). Yet
communication appears to interact in important ways with the phenomena just
discussed, and to play an essential role in real relationships.

Using Communication to Foster Coordination and Cooperation

Humans appear to be uniquely capable of using language to build, commu-
nicate, and counterfactually manipulate mental models of the world and of other
people. This capability has a powerful influence on how people structure and main-
tain their relationships, and on what they can accomplish in them. Yet existing
models of collusion and cooperation assign a limited role to communication. For
example, most repeated-games analyses imply that firms can accomplish as much
via tacit collusion as with communication. Why, then, does American antitrust law
bother to prohibit firms from communicating about pricing and output decisions
(Genesove and Mullin 2001; Andersson and Wengström 2007)? Presumably, when
an agreement has gone awry, communication is an important aid to understanding
what went wrong and restoring the relationship. Better models of how communica-
tion helps are needed.
To begin to explore these issues, consider the well-known Stag Hunt game,
which traces back to a scenario laid out by Rousseau (1754 [1973]), who wrote:

If a deer was to be taken, everyone saw that, in order to succeed, he must abide
faithfully by his post: but if a hare happened to come within the reach of any
one of them, it is not to be doubted that he pursued it without scruple, and,
having seized his prey, cared very little, if by so doing he caused his compan-
ions to miss theirs.

A two-player Stag Hunt game with a set of payoffs is shown in Figure 4. The
game has two pure-strategy equilibria, “all-Stag” and “all-Rabbit.” All-Stag is better
for both players than all-Rabbit, and is therefore “payoff-dominant” and a pref-
erable equilibrium using one of the Harsanyi and Selten (1987) criteria. But as
Rousseau’s scenario suggests, how can the two players build trust that they will stay
at their posts so that each can get a stag, rather having one of them deviate and try
to bag a Rabbit? Rabbit also has a fairly large payoff, and there are far larger sets of
144 Journal of Economic Perspectives

players’ beliefs that make Rabbit a best response. For the payoffs given in Figure 4, a
player finds it optimal to play Rabbit if the belief is that the player’s partner will play
Rabbit with probability at least 1/7, while it is optimal to play Stag only under the
belief that partner will play Stag with probability at least 6/7. Thus, using another
of the Harsanyi and Selton criteria for choosing between equilibria, the all-Rabbit
equibrium is “risk-dominant.”
Experiments suggest that if people play Stag Hunt with no opportunity to
communicate, a large majority of them will play Rabbit, as in other settings with
a strongly risk-dominant equilibrium (Straub 1995). But now imagine, following
Aumann (1990; see also Farrell 1988), that Stag Hunt is to be played only once,
but that before play, one player, the sender, must send a clear message about the
sender’s intended strategy, Stag or Rabbit. As already noted, in game theory such
communication is usually modelled via cheap talk messages, which are nonbinding
and have no direct payoff consequences. Even so, such a message might benefit
the sender by influencing the receiver’s choice (Crawford and Sobel 1982; Farrell
and Rabin 1996).
Aumann (1990) notes that whether or not the sender plans to play Stag, the
sender prefers that the receiver play Stag (9 > 1 and 8 > 7). He argues that for this
reason, the receiver will infer that the sender’s message is self-interested and the
message can convey no information to the receiver, so that the outcome will be
the same as without communication. Aumann’s argument is related to Farrell and
Rabin’s (1996) distinction between messages that are “self-committing” in that if
the message convinces the receiver, it’s a best response for the sender to do as he
said; and those that are “self-signaling” in that they are sent when and only when
the sender intends to do as he said. In this case, a message of intention to play Stag
is self-committing, but not self-signaling. Aumann’s argument is correct as a matter
of logic, yet many of us would expect most senders to send and play Stag, and most
receivers to play Stag as well. This conclusion is confirmed in most experiments
(Cooper, DeJong, Forsythe, and Ross 1992; Charness 2000; Ellingsen, Östling, and
Wengström 2013; but see Clark, Kay, and Sefton 2001).
One reason for the discrepancy has to do with Aumann’s (1990) exclusive
reliance on the logic of equilibrium, even though the multiplicity of equilibria,
with one payoff-dominant and another risk-dominant, seriously undermines the
thinking justification for equilibrium. When uncertainty about others’ thinking is
of the essence, it is unlikely that intelligent people will interpret a sender’s message
as if there were no chance whatsoever that it would influence equilibrium selection
or whether players’ choices are even in equilibrium. Rabin (1994; see also Farrell
1987, 1988) relaxes the assumption that players’ beliefs are perfectly coordinated
on some equilibrium, using a combination of rationalizability and behaviorally
plausible assumptions about how players use language to analyze the process of
negotiating how to play one of a class of finite matrix games. He shows that if players
can communicate as long as desired, they will use their messages to agree on an
equilibrium that is no worse for either player than the worst Pareto-efficient equilib-
rium for that player—thus, for example, yielding all-Stag in Stag Hunt.
 Vincent P. Crawford 145

Ellingsen and Östling (2010) take a different nonequilibrium approach,

adapting the level-k model of Crawford (2003) to resolve some puzzles regarding
the comparative effectiveness of one- or two-sided communication in coordination
and other games.9 They show, among other things, that even one-sided commu-
nication may allow players to coordinate on a Pareto-dominant equilibrium in a
wide class of games including Stag Hunt, again resolving the puzzle. Notable recent
experimental work includes Andersson and Wengström (2012) and Cooper and
Kühn (2014), who study communication and renegotiation in two-stage games.
A more subtle reason for the discrepancy between Aumann’s (1990) prediction
and prevailing intuitions and experimental evidence on the effectiveness of commu-
nication in Stag Hunt may be his assumption that players are limited to a fixed list
of messages, as in most theoretical work on communication: in his case, strategy
labels whose meanings are assumed to be understood. Yet Stag Hunt is one of many
situations in which people, even if well-informed about the structure, might benefit
from a discussion more nuanced than stating an intention before deciding how
to play. A sender who could send an unrestricted natural-language message would
probably try to convey not only an intention but also a broader understanding of
strategic issues. A fuller message might say, trying to give the assurance needed to
support All-Stag: “I can see, as I am sure you can, that the best outcome in this game
would be for both of us to play Stag. But I realize that Stag is risky for you, as it is for
me. Despite the risk, I have concluded that Stag is a better bet for us. I plan to play
Stag, and I hope you will too.” In experiments such natural-language messages can
be very effective (Charness and Dufwenberg 2006, 2010).
How could game theory incorporate such richer communication? Relaxing the
standard assumption that people are limited to a fixed list of messages about inten-
tions or private information to allow “metatheoretical” messages like my quotation
is a theoretical challenge, and it seems difficult even to formalize an epistemic
thinking justification for equilibrium when there is uncertainty about the principles
of equilibrium selection, or even whether such principles ensure that players play
some equilibrium. Even so, the gains from understanding natural-language commu-
nication, and how it interacts with people’s other decisions, seem likely to be very
large. McGinn, Thompson, and Bazerman (2003) report experimental evidence
on how subjects use natural-language messages, which may help in devising better
theories (on this point, see also Valley, Thompson, Gibbons, and Bazerman 2002;
Weber and Camerer 2003; Charness and Dufwenberg 2006; Houser and Xiao 2011;
Burchardi and Penczynski 2014).
Better theories of communication will certainly include some elements of
players’ rationality and their knowledge of the rationality of others, but they cannot
be entirely epistemic. Rather, such theories are likely to combine rationality-based

9
In Crawford (2003), I studied deceptive preplay communication of intentions before a zero-sum two-
person game, which can happen in a plausible level-k model, but not in equilibrium. This level-k model
has a more plausible thinking justification than equilibrium and also has some experimental support
(Wang, Spezio, and Camerer 2010).
146 Journal of Economic Perspectives

reasoning about the meaning of messages with empirically-based restrictions,

such as those used by Farrell (1987) and Rabin (1994) and, or perhaps like those
embodied in level-k rules like those studied by Ellingsen and Östling (2010).

Conclusion

Some economists seem less excited about game theory than during the period
in the 1960s, 1970s, and 1980s when the ability to analyze strategic interactions was
altering the landscape of many subfields of economics. But if the excitement over
game theory has in fact diminished, I do not believe it is because game theory has
ceased to be a major driving force in economics—quite the contrary!—it is mainly
because its centrality makes economists less aware of its presence. Modern econo-
mists’ relationship to game theory may resemble fully adapted aquatic creatures’
relationship to water: they are less aware of water than their amphibian ancestors,
for whom swimming was always a choice, but also more agile in their new medium.
That said, if game theory is to continue as a major force for progress in
economics, it must continue to co-evolve with economic applications and incorpo-
rate the empirical knowledge they provide, rather than pursuing an inwardly focused
agenda. In this paper, I have tried to give some concrete illustrations of what that
might mean, critiquing existing game-theoretic approaches to the canonical problem
of using communication to foster coordination and cooperation in relationships and
suggesting some directions in which further progress might be made.

■ I thank Colin Camerer, Gary Charness, Miguel Costa-Gomes, Martin Dufwenberg, Nagore
Iriberri, Robert Östling, H. Peyton Young, the editors, and especially Joel Sobel and Tore
Ellingsen for their helpful discussions and comments. My research received primary funding
from the European Research Council under the European Union’s Seventh Framework
Programme (FP7/2007-2013) / ERC grant agreement no. 339179. The contents reflect only
my views and not the views of the ERC or the European Commission, and the European Union
is not liable for any use that may be made of the information contained therein. I also thank
All Souls College, Oxford, and the University of California, San Diego, for research support.
 Vincent P. Crawford 147

References

Andersson, Ola, and Erik Wengström. 2007. “Do Repeated Games.” Journal of Economic Theory
Antitrust Laws Facilitate Collusion? Experimental 104(1): 137–88.
Evidence on Costly Communication in Duopolies.” Camerer, Colin F., Teck-Hua Ho, and Juin-Kuan
Scandinavian Journal of Economics 109(2): 321–39. Chong. 2004. “A Cognitive Hierarchy Model of
Andersson, Ola, and Erik Wengström. 2012. Games.” Quarterly Journal of Economics 119(3):
“Credible Communication and Cooperation: 861–98.
Experimental Evidence from Multi-Stage Games.” Carlsson, Hans, and Eric van Damme. 1993.
Journal of Economic Behavior and Organization 81(1): “Global Games and Equilibrium Selection.” Econo-
207– 219. metrica 61(5): 989–1018.
Aradillas-Lopez, Andres, and Elie Tamer. Charness, Gary. 2000. “Self-Serving Cheap
2008. “The Identification Power of Equilibrium Talk: A Test of Aumann’s Conjecture.” Games and
in Simple Games.” Journal of Business & Economic Economic Behavior 33(2): 177–94.
Statistics 26(3): 261–83. Charness, Gary, and Martin Dufwenberg. 2006.
Aumann, Robert. 1990. “Nash-Equilibria Are “Promises and Partnership.” Econometrica 74(6):
Not Self-Enforcing.” In Economic Decision-Making: 1579–1601.
Games, Econometrics and Optimisation, edited by Charness, Gary, and Martin Dufwenberg. 2010.
J. J. Gabszewicz, J.-F. Richard, and L. Wolsey, “Bare Promises: An Experiment.” Economics Letters
201–206. North Holland. 107(2): 281–83.
Awaya, Yu, and Vijay Krishna. 2016. “On Clark, Kenneth, Stephen Kay, and Martin Sefton.
Communication and Collusion.” American Economic 2001. “When Are Nash Equilibria Self-Enforcing?
Review 106(2): 285–315. An Experimental Analysis.” International Journal of
Bernheim, B. Douglas. 1984. “Rationalizable Game Theory 29(4): 495–515.
Strategic Behavior.” Econometrica 52(4): 1007–1028. Cooper, David J., and John H. Kagel. 2003.
Binmore, Ken, John McCarthy, Giovanni Ponti, “Lessons Learned: Generalizing Learning Across
Larry Samuelson, and Avner Shaked. “A Backward Games.” American Economic Review 93(2): 202–207.
Induction Experiment.” Journal of Economic Theory Cooper, David J., and Kai-Uwe Kühn. 2014.
104(1): 48–88. “Communication, Renegotiation, and the Scope
Blonski, Matthias, Peter Ockenfels, and for Collusion.” American Economic Journal: Microeco-
Giancarlo Spagnolo. 2011. “Equilibrium Selection nomics 6(2): 247–78.
in the Repeated Prisoner’s Dilemma: Axiomatic Cooper, Russell, Douglas V. DeJong, Robert
Approach and Experimental Evidence.” American Forsythe, and Thomas W. Ross. 1992. “Communi-
Economic Journal: Microeconomics 3(3): 164–92. cation in Coordination Games.” Quarterly Journal of
Brandenburger, Adam. 1992. “Knowledge and Economics 107(2): 739–771.
Equilibrium in Games.” Journal of Economic Perspec- Costa-Gomes, Miguel A., and Vincent P.
tives 6(4): 83–101. Crawford. 2006. “Cognition and Behavior in Two-
Breitmoser, Yves. 2015. “Cooperation, But No Person Guessing Games: An Experimental Study.”
Reciprocity: Individual Strategies in the Repeated American Economic Review 96(5): 1737–68.
Prisoner’s Dilemma.” American Economic Review Crawford, Vincent P. 1995. “Adaptive Dynamics
105(9): 2882–2910. in Coordination Games.” Econometrica 63(1):
Burchardi, Konrad B., and Stefan P. Penczynski. 103–143.
2014. “Out of Your Mind: Eliciting Individual Crawford, Vincent P. 1997. “Theory and Experi-
Reasoning in One Shot Games.” Games and ment in the Analysis of Strategic Interaction.”
Economic Behavior 84(1): 39–57. Chap. 7 in Advances in Economics and Econometrics:
Camerer, Colin, and Teck-Hua Ho. 1998. Theory and Applications, Seventh World Congress, vol.
“Experience-Weighted Attraction Learning in 1, edited by David Kreps and Kenneth F. Wallis.
Coordination Games: Probability Rules, Heteroge- Cambridge University Press (Reprinted in 2003 as
neity, and Time-Variation.” Journal of Mathematical chap. 12 in Advances in Behavioral Economics, edited
Psychology 42(2–3): 305–326. by Colin F. Camerer, George Loewenstein, and
Camerer, Colin, and Teck-Hua Ho. 1999. Matthew Rabin, Princeton University Press).
“Experience-Weighted Attraction Learning in Crawford, Vincent P. 2001. “Learning
Normal Form Games.” Econometrica 67(4): 837–74. Dynamics, Lock-in, and Equilibrium Selection in
Camerer, Colin F., Teck-Hua Ho, and Juin-Kuan Experimental Coordination Games.” Chap. 6 in
Chong. 2002. “Sophisticated Experience-Weighted The Evolution of Economic Diversity, edited by Ugo
Attraction Learning and Strategic Teaching in Pagano and Antonio Nicita. Routledge. (With
148 Journal of Economic Perspectives

correction of the published version of Figure 6.2(b) 1994. “The ‘Folk Theorem’ for Repeated Games
(2b in the working paper) at http://econweb.ucsd. and Continuous Decision Rules.” Chap. 6 in Prob-
edu/%7Evcrawfor/9719.pdf.) lems of Coordination in Economic Activity, edited by
Crawford, Vincent P. 2002. “Introduction to James W. Friedman, in Recent Economic Thought
Experimental Game Theory.” Journal of Economic Series, vol. 35. Springer Verlag.
Theory 104(1): 1–15. Fudenberg, Drew, and Eric Maskin. 1986.
Crawford, Vincent P. 2003. “Lying for Strategic “The Folk Theorem in Repeated Games with
Advantage: Rational and Boundedly Rational Discounting or with Incomplete Information.”
Misrepresentation of Intentions.” American Econometrica 54(3): 533–554.
Economic Review 93(1): 133–49. García-Schmidt, Mariana, and Michael Wood-
Miguel Costa-Gomes, Crawford, Vincent P., and ford. 2014. “Are Low Interest Rates Deflationary?
Bruno Broseta. 2001. “Cognition and Behavior in A Paradox of Perfect-Foresight Analysis.” http://
Normal-Form Games: An Experimental Study.” www.columbia.edu/~mw2230/GSW.pdf.
Econometrica 69(5): 1193–1235. Genesove, David, and Wallace P. Mullin. 2001.
Crawford, Vincent P., Miguel A. Costa-Gomes, “Rules, Communication, and Collusion: Narrative
and Nagore Iriberri. 2013. “Structural Models Evidence from the Sugar Institute Case.” American
of Nonequilibrium Strategic Thinking: Theory, Economic Review 91(3): 379–98.
Evidence, and Applications.” Journal of Economic Harsanyi, John C., and Reinhard Selten. 1988. A
Literature 51(1): 5–62. General Theory of Equilibrium Selection in Games. MIT
Crawford, Vincent P., and Joel Sobel. 1982. Press.
“Strategic Information Transmission.” Econometrica Houser, Daniel, and Erte Xiao. 2011. “Clas-
50(6): 1431–51. sification of Natural Language Messages Using a
Dal Bó, Pedro, and Guillaume R. Fréchette. Coordination Game.” Experimental Economics 14(1):
2011. “The Evolution of Cooperation in Infinitely 1–14.
Repeated Games: Experimental Evidence.” Kandori, Michihiro, George J. Mailath, and
American Economic Review 101(1): 411–429. Rafael Rob. 1993. “Learning, Mutation, and Long
Economic Sciences Prize Committee of the Run Equilibria in Games.” Econometrica 61(1):
Royal Swedish Academy of Sciences. 2012. 29–56.
“Stable Allocations and the Practice of Market Manski, Charles F. 2003. Partial Identification of
Design.” Scientific Background on the Sveriges Probability Distributions. Springer-Verlag.
Riksbank Prize in Economic Sciences in Memory Mas-Colell, Andreu, Michael D. Whinston, and
of Alfred Nobel 2012. http://www.nobelprize. Jerry R. Green. 1995. Microeconomic Theory. Oxford
org/nobel_prizes/economic-sciences/laure- University Press.
ates/2012/advanced-economicsciences2012. McGinn, Kathleen L., Leigh Thompson, and
pdf. Max H. Bazerman. 2003. “Dyadic Processes of
Ellingsen, Tore, and Robert Östling. 2010. Disclosure and Reciprocity in Bargaining with
“When Does Communication Improve Coordina- Communication.” Journal of Behavioral Decision
tion.” American Economic Review 100(4): 1695–1724. Making 16(1): 17–34.
Ellingsen, Tore, Robert Östling, and Erik Myerson, Roger B. 1999. “Nash Equilibrium
Wengström. 2013. “How Does Communication and the History of Economic Theory.” Journal of
Affect Beliefs?” June 13. http://perseus.iies. Economic Literature 37(3): 1067–82.
su.se/~rob/papers/Ellingsen_et_al_2013.pdf. Nash, John. 1950. “The Bargaining Problem.”
Evans, George W., and Bruce McGough. 2015. Econometrica 18(2): 155–162.
“The Neo-Fisherian View and the Macro Learning Nash, John. 1953. “Two-Person Cooperative
Approach.” http://economistsview.typepad.com/ Games.” Econometrica 21(1): 128–40.
economistsview/2015/12/the-neo-fisherian-view- Pearce, David G. 1984. “Rationalizable Strategic
and-the-macro-learning-approach.html. Behavior and the Problem of Perfection.” Economet-
Farrell, Joseph. 1987. “Cheap Talk, Coordina- rica 52(4): 1029–50.
tion, and Entry.” RAND Journal of Economics 18(1): Porter, Robert H. 1983. “Optimal Cartel Trigger
34–39. Price Strategies.” Journal of Economic Theory 29(2):
Farrell, Joseph. 1988. “Communication, Coor- 313–38.
dination, and Nash Equilibrium.” Economics Letters Rabin, Matthew. 1994. “A Model of Pre-Game
27(3): 209–214. Communication.” Journal of Economic Theory 63(2):
Farrell, Joseph, and Matthew Rabin. 1996. 370–91.
“Cheap Talk.” Journal of Economic Perspectives 10(3): Rankin, Frederick, John Van Huyck, and Ray
103–118. Battalio. 2000. “Strategic Similarity and Emergent
Friedman, James W., and Larry Samuelson. Conventions: Evidence from Similar Stag Hunt
 Vincent P. Crawford 149

Games.” Games and Economic Behavior 32(2): Financial Crises: Causes, Prevention, and Cures.”
315–37. American Economic Review 90(2): 1–16.
Roth, Alvin E. 1987. “Bargaining Phenomena Valley, Kathleen, Leigh Thompson, Robert
and Bargaining Theory.” Chap. 2 in Laboratory Gibbons, and Max H. Bazerman. 2002. “How
Experimentation in Economics: Six Points of View, Communication Improves Efficiency in Bargaining
edited by Alvin E. Roth. Cambridge University Games.” Games and Economic Behavior 38(1):
Press. 127–55.
Rothschild, Michael. 1973. “Models of Market van Damme, Eric. 1989. “Stable Equilibria and
Organization with Imperfect Information: Forward Induction.” Journal of Economic Theory
A Survey.” Journal of Political Economy 81(6): 48(2): 476–96.
1283–1308. Van Huyck, John, and Raymond Battalio.
Rousseau, Jean-Jacques. 1754 [1973]. “A 2002. “Prudence, Justice, Benevolence, and Sex:
Discourse on the Origin of Inequality.” In The Evidence from Similar Bargaining Games.” Journal
Social Contract and Discourses (translated by G. D. of Economic Theory 104(1): 227–46.
H. Cole), 27–113. London: J. M. Dent & Sons, Ltd. Van Huyck, John B., Raymond C. Battalio, and
Also at http://www.constitution.org/jjr/ineq.htm. Richard O. Beil. 1990. “Tacit Coordination Games,
Rubinstein, Ariel. 1982. “Perfect Equilibrium in Strategic Uncertainty, and Coordination Failure.”
a Bargaining Model.” Econometrica 50(1): 97–109. American Economic Review 80(1): 234–48.
Samuelson, Larry. 2001. “Analogies, Adapta- Van Huyck, John B., Joseph Cook P., and
tion, and Anomalies.” Journal of Economic Theory Raymond C. Battalio. 1997. “Adaptive Behavior
97(2): 320–66. and Coordination Failure.” Journal of Economic
Schelling, Thomas C. 1960. The Strategy of Behavior and Organization 32(4): 483–503.
Conflict. Harvard University Press. von Neumann, John, and Oskar Morgenstern.
Shapley, Lloyd S., and Martin Shubik. 1954. “A 1944, 1947, 1953. Theory of Games and Economic
Method for Evaluating the Distribution of Power Behavior. Princeton University Press.
in a Committee System.” American Political Science Wang, Joseph Tao-yi, Michael Spezio, and
Review 48(3): 787–92. Colin F. Camerer. 2010. “Pinocchio’s Pupil: Using
Shapley, Lloyd S., and Martin Shubik. 1971. Eyetracking and Pupil Dilation to Understand
“The Assignment Game I: The Core.” International Truth-telling and Deception in Sender-Receiver
Journal of Game Theory 1(1): 111–30. Games.” American Economic Review 100(3):
Shubik, Martin. 1959. Strategy and Market Struc- 984–1007.
ture: Competition, Oligopoly, and the Theory of Games. Weber, Roberto A., and Colin F. Camerer.
John Wiley and Sons, Inc. 2003. “Cultural Conflict and Merger Failure: An
Straub, Paul G. 1995. “Risk Dominance and Experimental Approach.” Management Science
Coordination Failures in Static Games.” Quarterly 49(4): 400–415.
Review of Economics and Finance 35(4): 339–63. Young, H. Peyton. 1993. “The Evolution of
Summers, Lawrence H. 2000. “International Conventions.” Econometrica 61(1): 57–84.
150 Journal of Economic Perspectives
This article has been cited by:

1. Yufei Wang, Mangirdas Morkūnas, Jinzhao Wei. 202⒋ Strategic Synergies: Unveiling the Interplay of
Game Theory and Cultural Dynamics in a Globalized World. Games 15:4, 2⒋ [Crossref]
2. Gürbüz AYDIN, Hakan KARABACAK. 202⒊ Oyun Teorisi Perspektifinden Çatışma Yönetim
Stratejilerinin Karşılıklı Etkileşimi ve Denetçilere Yönelik Bir Uygulama. Süleyman Demirel
Üniversitesi Vizyoner Dergisi 14:38, 607-62⒌ [Crossref]
⒊ Marina Agranov, Jeongbin Kim, Leeat Yariv. 202⒊ Coordination with Differential Time Preferences:
Experimental Evidence. SSRN Electronic Journal 71. . [Crossref]
⒋ Marina Agranov, Jeongbin Kim, Leeat Yariv. 202⒊ Coordination with Differential Time Preferences:
Experimental Evidence. SSRN Electronic Journal 71. . [Crossref]
⒌ Duk Gyoo Kim, Hee Chun Kim. 2022. Probability matching and strategic decision making. Journal
of Behavioral and Experimental Economics 98, 101850. [Crossref]
⒍ Scott Barrett, Astrid Dannenberg. 2022. The Decision to Link Trade Agreements to the Supply of
Global Public Goods. Journal of the Association of Environmental and Resource Economists 9:2, 273-30⒌
[Crossref]
⒎ Morgan Tait. 2022. Passive Revolution? Or Just Philosophical Confusion? Why ‘Mainstream
Economics’ Needs No Paradigm Shift. SSRN Electronic Journal 84. . [Crossref]
⒏ Chen Cao, Xueyun Chen. 2021. Can Industrial Integration Improve the Sustainability of Grain
Security?. Sustainability 13:24, 136⒙ [Crossref]
⒐ Christopher Roby. 2021. Can loss framing improve coordination in the minimum effort game?.
Journal of Economic Interaction and Coordination 16:3, 557-58⒏ [Crossref]
⒑ Hussein Waleed Hussein, Dr. Abdulnaser Alag Hafedh. 20⒛ REFLECTION OF STRATEGIC
BEHAVIORS IN SCENARIO PLANNING. International Journal of Research in Social Sciences and
Humanities 10:4, 141-14⒋ [Crossref]
⒒ Vincent P. Crawford. 20⒚ Experiments on Cognition, Communication, Coordination, and
Cooperation in Relationships. Annual Review of Economics 11:1, 167-191. [Crossref]
⒓ Tore Ellingsen, Robert Östling, Erik Wengström. 20⒙ How does communication affect beliefs in
one-shot games with complete information?. Games and Economic Behavior 107, 153-181. [Crossref]
⒔ Duk Gyoo Kim, Hee Chun Kim. 20⒙ Mixing Propensity and Strategic Decision Making. SSRN
Electronic Journal . [Crossref]
⒕ Atanu Saha. 20⒙ Strategic Deterrence: An Application to Biologic Pharmaceuticals. SSRN Electronic
Journal . [Crossref]