PREDICTING HOW PEOPLE PLAY GAMES - REINFORCEMENT LEARNING IN EXPERIMENTAL GAMES WITH UNIQUE, MIXED STRATEGY EQUILIBRIA

We examine learning in all experiments we could locate involving 100 p eriods or more of games with a unique equilibrium in mixed strategies, and in a new experiment. We study bath the ex past (''best fit'') des criptive power of leaning models, and their ex ante predictive power, by simulating each experiment using parameters estimated from the othe r experiments. Even a one-parameter reinforcement learning model robus tly outperforms the equilibrium predictions. Predictive power is impro ved by adding ''forgetting'' and ''experimentation,'' or by allowing g reater rationality as in probabilistic fictitious play. Implications f or developing a low-rationality, cognitive game theory are discussed.