On the empirical relevance of correlated equilibrium

Absent coordinating signals from an exogenous benevolent agent, can an efficient correlated equilibrium emerge? Theoretical work in adaptive dynamics suggests a positive answer, which we test in a laboratory experiment. In the well-known Chicken game, we observe time average play that is close to the asymmetric pure Nash equilibrium in some treatments, and in other treatments we observe collusive play. In a game resembling rock-paper-scissors or matching pennies, we observe time average play close to a correlated equilibrium that is more efficient than the unique Nash equilibrium. Estimates and simulations of adaptive dynamics capture much of the observed heterogeneity across player pairs as well as dynamic regularities.