On rational cooperation in finitely repeated games
Sigrid Suetens (Tilburg University)
November 17, 2017 | 12:00 - 13:00 | BLU-003
Two patterns are common across finitely repeated prisoner’s dilemma experiments. First, players use conditionally cooperative strategies. Second, cooperation unravels over time with the first defection period occurring later in the repeated game and cooperation starting off at a higher rate if the expected payoff of a cooperative strategy is relatively high compared to its risk. Broadly speaking, these patterns are consistent with error models, models of social preferences, and reputational models. The key difference between reputational models on the one hand and error and social preferences models on the other hand is that players update beliefs about their partner's type rationally in the course of a repeated game. This feature introduces a correlation between past actions of others and expectations about future actions, and may lead rational money maximizers to uphold a reputation for being a cooperative type for some time. It is an open question – there is no consensus in the literature – how important reputation building is behaviorally. We run an experiment using simple finitely repeated dilemma games designed to answer this question. We find that reputation building is crucial to explain the between-treatment differences in the behavior of experienced players in our experiment.