A partial identification framework for dynamic games

Empirical work on strategic interactions is often subject to the critique that equilibrium selection assumptions drive the results. We develop a framework for partially identifying parameters of dynamic games without equilibrium selection assumptions. Our framework relies on incentive compatibility constraints that incorporate game theoretical results on equilibrium payoff sets to bound the unknown continuation payoffs. We apply this framework to identify cost parameters in three dynamic games where collusion is a potential outcome. The identified set demonstrates the ease of sustaining collusion with patient firms, in low demand and when monitoring is perfect, and can also be used to detect collusion.