Strategic Learning With Finite Automata Via The EWA-Lite Model

We modify the self-tuning Experience Weighted Attraction (EWA-lite) model of Camerer, Ho, and Chong (2007) and use it as a computer testbed to study the likely performance of a set of twostate automata in four symmetric 2 x 2 games. The model suggested allows for a richer specification of strategies and solves the inference problem of going from histories to beliefs about opponents' strategies, in a manner consistent with \belief-learning". The predictions are then validated with data from experiments with human subjects. Relative to the action reinforcement benchmark model, our modified EWA-lite model can better account for subject-behavior.