Towards a taxonomy of learning dynamics in \(2\times 2\) games (Q2078051)

The authors study learning dynamics in $2\times 2$ games that are played repeatedly by boundedly rational players. Specifically, the examine outcomes for different parameterizations of experience-weighted attraction (EWA) [\textit{C. Camerer} and \textit{T.-H. Ho}, Econometrica 67, No. 4, 827--874 (1999; Zbl 1055.91504)] in different classes of $2\times 2$ stage games. EWA generalizes other widely used learning rules, and describes actual experimental play better than those other learning. Key parameters are the relative weights on experience (the history of previously experienced payoffs) and on attraction (the payoff from a particular action against the opponent's most recent action. The authors find that considering the full range of EWA parameters yields long-run behavior that in some cases is qualitatively different than under any of the learning rules that EWA generalizes.