Convergence Rates in Resource Allocation Games

If Nash equilibrium corresponds to the long run outcome of a dy- namic process, its usefulness as a predictive tool may depend on the rate of convergence to equilibrium. This paper experimentally tests theoretical predictions about the rate of convergence to equilibrium in settings where agents simultaneously allocate resources between contests with complementary prizes. More responsive contest suc- cess functions give agents a stronger incentive to best respond, but learning models predict slower convergence to equilibrium under more responsive success functions because of the incentives agents face out of equilibrium. Consistent with learning model predictions, we observe slower convergence under more responsive success functions, suggest- ing that disequilibrium incentives contain useful information about the rate of convergence to equilibrium in empirical settings.