This research examines equilibrium selection in infinitely repeated games played by boundedly rational agents. The decision rules used by these players will be determined by an evolutionary process in which successful rules proliferate while unsuccessful rules die. The project first develops a model of boundedly rational players, building on recent work using finite automata. A dynamic evolutionary process will then be constructed, with particular attention devoted to the fact that with conventional definitions, evolutionarily stable strategies do not exist in repeated games. The research will then examine the limiting outcomes of this evolutionary process, which will embody both rules for action and rules for learning from experience during the game. Finally, the robustness of the results, especially to different specifications of environmental "trembles," will be explored. The effectiveness of game theory is limited by the "folk theorem," which established that in repeated games, virtually anything can be an equilibrium. The significance of the proposed research lies in its potential to yield game-theoretic predictions for repeated games, making game theory a more powerful analytical tool for a wide class of applications. In particular, the research should shed light on the evolution of learning rules among boundedly rational agents.
