Cooperative teaching and learning of actions

This paper studies a novel game-theoretic setting: players may acquire new actions over time by observing the opponent’s play. We model this scenario as finitely repeated games where players’ action sets are private information and may endogenously expand over time. Three main implications emerge from this framework and its equilibria. First, players may target a payoff vector for the long run and voluntarily “teach” one another the actions needed in early periods. The action profile will be learned and sustained as long as each action is available to either player. Second, when no payoff target is prefixed, the players can always obtain or approximate strict ex-post efficiency via bilateral teaching and learning. Third, an alternative economic argument now exists for seemingly irrational cooperative behavior in games with finite horizon. For instance, fully rational players can play a cooperative equilibrium even if the stage game remains a Prisoner’s Dilemma for everyone.