Title:Learning of Actions in Finitely Repeated Games
Presenter:Prof. Yangbo Song
Time:1.12 10：00—11：30
Location:A1038
Invited by: Prof.Zhiwei Cui
Abstract:This paper studies a novel setting in game theory: a player may learn new actions over time by observing the opponent's play. We investigate the impact of such learning behavior in the context of finitely repeated games. In contrast to related literature such as Kreps et al. (1982), we provide a framework with full rationality and consistent stage-game payoffs for sustaining cooperation, which bridges the gap between theories of finitely and infinitely repeated games. Even if rational cooperation is impossible without learning, for instance in a Prisoner's Dilemma, it can be sustained with approximate efficiency when players can learn from one another. Cooperation does not have to be endowed in each player's initial action set, but can be ``taught'' and enforced. When learning is imperfect, the set of sustainable payoffs is not continuous, in the sense that no equilibrium exists when learning is nearly perfect and the repeated games last for sufficiently many periods.
