A numerical analysis of the evolutionary stability of learning rules

In this paper, we define an evolutionary stability criterion for learning rules. Using simulations, we then apply this criterion to three types of symmetric 2×2 games for a class of learning rules that can be represented by the parametric model of Camerer and Ho [1999. Experience-weighted attraction learning in normal form games. Econometrica 67, 827-874]. This class contains stochastic versions of reinforcement and fictitious play as extreme cases. We find that only learning rules with high or intermediate levels of hypothetical reinforcement are evolutionarily stable, but that the stable parameters depend on the game.