Averaged predictions and the learning of equilibrium play

Abstract

Abstract The main objects here are noncooperative games in which all externalities occur via a one-dimensional variable. So-called mean-value iterates are used to approach Nash equilibrium. The proposed schemes generalize many received methods, and can be interpreted as learning taking place during repeated play. An important feature is that no player need be fully informed about the game structure. Particular examples include Cournot oligopolies and some nonatomic market games.