Learning aspiration in repeated games

We study infinitely repeated symmetric 2×2 games played by bounded rational agents who follow a simple rule of thumb: each agent continues to play the same action if the current payoff exceeds the average of the past payoffs, and switches to the other action with a positive probability otherwise. By applying the stochastic approximation technique, we characterize the asymptotic outcomes for all 2×2 games. In the prisoners’ dilemma game, for example, the players cooperate in the limit if and only if the gain from defecting against cooperation is “modest.”     

Finitely repeated games with social preferences

A well-known result from the theory of finitely repeated games states that if the stage game has a unique equilibrium, then there is a unique subgame perfect equilibrium in the finitely repeated game in which the equilibrium of the stage game is being played in every period. Here I show that this result does in general not hold anymore if players have social preferences of the form frequently assumed in the recent literature, for example in the inequity aversion models of Fehr and Schmidt (Quartely Journal of Economics 114:817–868, 1999) or Bolton and Ockenfels (American Economic Review 100:166–193, 2000). In fact, repeating the unique stage game equilibrium may not be a subgame perfect equilibrium at all. This finding should have relevance for all experiments with repeated interaction, whether with fixed, random or perfect stranger matching.     

Interbranch cooperation and the shadow of the future

Prisoners’ Dilemma (PD) experiments confirm and extend Axelrod’s (The evolution of cooperation. Basic Books, New York, 1984) Shadow of the Future hypothesis: subjects cooperate in infinitely repeated PD, but they also cooperate until near the end in finitely repeated PD. So the extended hypothesis is that cooperation depends on the probability of continued play. Observational tests of this hypothesis, or even applications, have been rare at best. Here we not only apply but test it for interbranch cooperation under separated-powers constitutions, specifically those of the American states, using the end of governors’ final terms as end points and the rate of overridden vetoes as the extreme case of a breakdown in interbranch cooperation. Controlling for a variety of factors, including divided government, we find support for the hypothesis, whose explanation of interbranch interaction fills a gap left open by Madison’s Federalist 51: how republican government can control itself when what is needed is “energy” more than safeguards.     

Rational cooperation in the finitely repeated prisoners' dilemma

A common observation in experiments involving finite repetition of the prisoners' dilemma is that players do not always play the single-period dominant strategies (“finking”), but instead achieve some measure of cooperation. Yet finking at each stage is the only Nash equilibrium in the finitely repeated game. We show here how incomplete information about one or both players' options, motivation or behavior can explain the observed cooperation. Specifically, we provide a bound on the number of rounds at which Fink may be played, when one player may possibly be committed to a “Tit-for-Tat” strategy.     

Moral Hazard and Private Monitoring

We clarify the role of mixed strategies and public randomization (sunspots) in sustaining near-efficient outcomes in repeated games with private monitoring. We study a finitely repeated game, where the stage game has multiple equilibria and show that mixed strategies can support partial cooperation, but cannot approximate full cooperation even if monitoring is “almost perfect.” Efficiency requires extensive form correlation, where strategies can condition upon a sunspot at the end of each period. For any finite number of repetitions, we approximate the best equilibrium payoff under perfect monitoring, assuming that monitoring is sufficiently accurate and sunspots are available. Journal of Economic Literature Classification Numbers: C73, D82.     

On Learning to Cooperate

This paper shows that experimentation can have the effect of inducing outcomes that are “far” from Nash in a learning model similar to that studied by D. Fudenberg and D. K. Levine (1993, Econometrica61, 547-573). The game under consideration is a version of a finitely repeated prisoner's dilemma, and the average length of players' lifetimes is arbitrarily large, but fixed. If the number of stages in the prisoner's dilemma is sufficiently large, then experimentation induces cooperation within the society on an arbitrarily large scale, even though the beliefs of almost all players are eventually within an arbitrary η>0 of being correct at all of the information sets in the game. Journal of Economic Literature Classification Numbers: C72.     

Collective choice under dichotomous preferences

Agents partition deterministic outcomes into good or bad. A mechanism selects a lottery over outcomes (time-shares). The probability of a good outcome is the canonical utility. The utilitarian mechanism averages over outcomes with largest “approval”. It is efficient, strategyproof, anonymous and neutral.We reach an impossibility if, in addition, each agent's utility is at least , where n is the number of agents; or is at least the fraction of good to feasible outcomes. We conjecture that no ex ante efficient and strategyproof mechanism guarantees a strictly positive utility to all agents, and prove a weaker statement.     

Reciprocity Game

This paper shows that reciprocity comes from the desire to cooperate in finitely repeated prisoner's dilemma game. Before playing the finitely repeated prisoner's dilemma game, players choose the reciprocity level and commit to it, and the reciprocity level is public information. There are T equilibria if the prisoner's dilemma game is repeated for T periods, and each equilibrium is associated with different levels of cooperation. Further, if players choose their reciprocity levels sequentially, then the most cooperative equilibrium will be the unique equilibrium. However, reciprocity does not matter for the one‐period game and the infinitely repeated game.     

The efficiency of competitive mechanisms under private information

We consider the efficiency properties of exchange economies where privately informed traders behave strategically. Specifically, a competitive mechanism is any mapping of traders’ reports about their types to an equilibrium price vector and allocation of the reported economy. In our model, some traders may have non-vanishing impact on prices and allocations regardless of the size of the economy. Although truthful reporting by all traders cannot be achieved, we show that, given any desired level of approximation, there is such that any Bayesian-Nash equilibrium of any competitive mechanism of any private information economy with or more traders leads, with high probability, to prices and allocations that are close to a competitive equilibrium of the true economy. In particular, allocations are approximately efficient. A key assumption is that there is small probability that traders behave non-strategically.     

Almost budget-balanced VCG mechanisms to assign multiple objects

Subsidy-free VCG mechanisms assign p identical objects to n agents. The efficiency loss is the largest ratio of budget surplus to efficient surplus, over all profiles of non-negative valuations. The smallest efficiency loss satisfies . If is bounded away from , converges to zero exponentially in n.Participation is voluntary in the optimal mechanism achieving if p=1, but not if p?2. Among voluntary mechanisms, the optimal efficiency loss is not significantly larger than if . But it does not converge to zero in n if .     

Building rational cooperation

Experiments have shown that people have a natural taste for cooperation. This paper takes a first step in understanding how formal and informal institutions might be designed to utilize these private tastes to facilitate more efficient economic interactions. We examine a twice-played prisoners’ dilemma in which the total of the stakes in the two periods is fixed, but the distribution of these stakes can be varied across periods. We verify experimentally that it is best to “start small,” reserving most of the stakes for the second period.     

Reputation and imperfect information

A common observation in the informal literature of economics (and elsewhere) is that is multistage “games,” players may seek early in the game to acquire a reputation for being “tough” or “benevolent” or something else. But this phenomenon is not observed in some formal game-theoretic analyses of finite games, such as Selten's finitely repeated chain-store game or in the finitely repeated prisoners' dilemma. We reexamine Selten's model, adding to it a “small” amount of imperfect (or incomplete) information about players' payoffs, and we find that this addition is sufficient to give rise to the “reputation effect” that one intuitively expects.     

Set-rationalizable choice and self-stability

Rationalizability and similar notions of consistency have proved to be highly problematic in the context of social choice, as witnessed by a range of impossibility results, among which Arrow?s is the most prominent. We propose to rationalize choice functions by preference relations over sets of alternatives (set-rationalizability) and introduce two consistency conditions, and , which are defined in analogy to Sen?s α and γ. We find that a choice function satisfies if and only if it is set-rationalizable and that it satisfies and if and only if it is self-stable, a new concept based on earlier work by Dutta. The class of self-stable social choice functions contains a number of appealing Condorcet extensions.     

Reciprocity in evolving social networks

We study the viability of conditional cooperation in a dynamically evolving social network. The network possesses the small world property, with high clustering coefficient but low characteristic path length. The interaction among linked individuals takes the form of a multiperson prisoners’ dilemma, and actions can be conditioned on the past behavior of one’s neighbors. Individuals adjust their strategies based on performance within their neighborhood, and both strategies and the network itself are subject to random perturbation. We find that the long-run frequency of cooperation is higher under the following conditions: (i) the interaction radius is neither too small nor too large, (ii) clustering is high and characteristic path length low, (iii) the mutation rate of strategies is small, and (iv) the rate of adjustment in strategies is neither too fast nor too slow.

Decentralized learning from failure

We study decentralized learning in organizations. Decentralization is captured through Crawford and Haller's [Learning how to cooperate: optimal play in repeated coordination games, Econometrica 58 (1990) 571-595] attainability constraints on strategies. We analyze a repeated game with imperfectly observable actions. A fixed subset of action profiles are successes and all others are failures. The location of successes is unknown. The game is played until either there is a success or the time horizon is reached. We partially characterize optimal attainable strategies in the infinite horizon game by showing that after any fixed time, agents will occasionally randomize while at the same time mixing probabilities cannot be uniformly bounded away from zero.     

Unique solutions for stochastic recursive utilities

We study unique and globally attracting solutions of a general nonlinear stochastic equation, widely used in Finance and Macroeconomics and closely related to stochastic Koopmans equations. The equation is specified by a temporal aggregator W and a certainty equivalent operator M. The main contribution of the paper is the introduction of the new class of Thompson aggregators. Other contributions of the paper are: (i) a detailed analysis of quasi-arithmetic operators M that generalize those of Kreps and Porteus (1978) [18]; (ii) a clarification of the nature and properties of the stochastic recursive preferences that underlie Koopmans equations.     

Monotonic and implementable solutions in generalized matching problems

We consider one-to-one matching problems when preferences are weak orders and search for (Maskin-)monotonic and implementable solutions. We show that for any coalition structure , the strong -core is monotonic if and only if it is the individually rational solution. We introduce a solution, which we call the “union strong -core”, and show that this solution is the minimal way to expand the strong -core recovering monotonicity. We also show that the union strong -core is implementable.     

Cooperation without Reputation: Experimental Evidence from Prisoner's Dilemma Games

This paper investigates cooperative play in prisoner's dilemma games by designing an experiment to evaluate the ability of two leading theories of observed cooperation: reputation building and altruism. We analyze both one-shot and finitely repeated games to gauge the importance of these theories. We conclude that neither altruism nor reputation building alone can explain our observations. The reputation model is inconsistent with play in both one-shots and finitely repeated games while the model with altruism is unable to explain observed play in the finitely repeated games.Journal of Economic LiteratureClassification Numbers: C72, C92.