Informational path dependency of learning in games: Collusion and competition in Cournot experiments
47 Pages Posted: 17 Oct 2017 Last revised: 19 Oct 2023
Date Written: October 19, 2023
People do not only face fixed informational environments in real-world interactions, but they may learn about the nature of the game, about themselves and others while playing repetitions of a game. In order to investigate how human learning behavior adapts in such situations, we conducted lab experiments on Cournot games where new, richer information concerning the underlying repeated interaction became available over time. To do so, we provided --along different informational paths-- more information as the game was repeated. We consider three main learning rules relevant in this setting, reinforcement learning, best-response dynamics and imitation. Our evidence confirms a natural mapping of information contexts onto predominant learning rules. Action and payoff feedback concerning oneself triggers reinforcement learning, action and payoff feedback concerning others triggers imitation, and information about the payoff structure of the game together with action feedback triggers best-response dynamics. Moreover, we identify important path dependencies in learning that have substantial effects in terms of aggregate convergence. In particular, feedback about others leads to imitative dynamics and overshooting of Nash equilibrium that do not revert when structural information is supplied subsequently.
Keywords: Learning, Imitation, Best-Response, Information, Cournot Competition, Heuristics, Cooperation
JEL Classification: C72, C92, D83
Suggested Citation: Suggested Citation