Estimating the Experience-Weighted Attractions for the Migration-Emission Game

Abstract

Players are unlikely to immediately play equilibrium strategies in complicated games or in games in which they do not have much experience playing. In these cases, players will need to learn to play equilibrium strategies. In laboratory experiments, subjects show systematic patterns of learning during a game. In psychological and economic models of learning, players tend to play a strategy more if it has been successful in the past (reinforcement learning) or that would have given higher payoffs given the strategies of other players (belief learning). This paper uses experimental data from the four sessions of a pilot experiment of a three-stage emission game to estimate parameters of experience-weighted-attraction (EWA) learning, which is a hybrid of reinforcement and belief learning models. In this estimation, we transform the strategy space for the three-stage game extensive form game to a normal form game. This paper also considers asymmetric information across players in estimating EWA parameters. In three of the four sessions, estimated parameters are consistent with reinforcement learning, which means that players tend to choose to strategies looking at past strategies that are more successful than the others. In the other session, estimated parameters are consistent with belief learning, which means that players consider forgone payoffs to update their beliefs that determine the probability of strategy choice.