Fictitious play and ``gradient'' learning are examined in the context of a large population where agents are repeatedly randomly matched. We show that the aggregation of this learning behaviour can be qualitatively different from learning at the level of the individual. This aggregate dynamic...

We develop a general model of best response adaptation in large populations for symmetric and asymmetric conflicts with role-switching. For special cases including the classical best response dynamics and the symmetrized best response dynamics we show that the set of Nash equilibria is...

We analyze a repeated first-price auction in which the types of the players are determined before the first round. It is proved that if every player is using either a belief-based learning scheme with bounded recall or a generalized fictitious play learning scheme, then for sufficiently large...

Fictitious play is the classical myopic learning process, and games with strategic complementarities are an important class of games including many economic applications. Knowledge about convergence properties of fictitious play in this class of games is scarce, however. Beyond dominance...

Exploiting small uncertainties on the part of opponents, players in long, finitely repeated games can maintain false reputations that lead to a large variety of equilibrium outcomes. Even cooperation in a finitely repeated prisoners' dilemma is obtainable. Can such false reputations be...

Earlier experiments have shown that under little information subjects are hardly able to coordinate even though there are no conflicting interests and subjects are organised in fixed pairs. This is so, even though a simple adjustment process would lead the subjects into the efficient, fair and...

According to the standard definition, a Bayesian agent is one who forms his posterior belief by conditioning his prior belief on what he has learned, that is, on facts of which he has become certain. Here it is shown that Bayesianism can be described without assuming that the agent acquires any...

Fictitious play is the oldest and most studied learning process for games. Since the already classical result for zero-sum games, convergence of beliefs to the set of Nash equilibria has been established for some important classes of games, including weighted potential games, supermodular games...

We propose a new and simple adaptive procedure for playing a game: "regret-matching." In this procedure, players depart from their current play with probabilities that are proportional to measures of regret for not having used other strategies in the past. It is shown that our adaptive procedure...

This paper presents an alternative or enhanced approach to information acquisition in Cournot markets with stochastic demand in which the cost of information acquisition is endogenously determined by firms\222 information purchasing strategy. I propose a two-stage model in which in the first...