Learning to Coordinate by Forward Looking Players

Experiments show that players with identical information and objectives have heterogeneous behavior rules. We formulate how different beliefs arise from the same information and analyze how heterogeneous belief formation affects learning outcomes of two-person two-action coordination games. With heterogeneous belief formation, it is possible to generate enough diversity in actions for eventual coordination, which is impossible under the ordinary adaptive learning model. When each player uses the same function to compute a belief over time, diversity of beliefformation functions within or across populations is sufficient for convergence to coordination. When players choose belief-formation functions as well as actions over time, diverse adjustment of belief-formation functions is sufficient. A risk-dominant equilibrium is more likely to emerge than a payoff-dominant equilibrium since the former is more likely to be predicted by the players. JEL classification number : C 73