Self-tuning experience weighted attraction learning in games

Abstract

Self-tuning experience weighted attraction (EWA) is a one-parameter theory of learning in
\ngames. It addresses a criticism that an earlier model (EWA) has too many parameters, by
\nfixing some parameters at plausible values and replacing others with functions of experience
\nso that they no longer need to be estimated. Consequently, it is econometrically simpler
\nthan the popular weighted fictitious play and reinforcement learning models.
\nThe functions of experience which replace free parameters “self-tune” over time, adjusting
\nin a way that selects a sensible learning rule to capture subjects’ choice dynamics. For
\ninstance, the self-tuning EWA model can turn from a weighted fictitious play into an averaging
\nreinforcement learning as subjects equilibrate and learn to ignore inferior foregone
\npayoffs. The theory was tested on seven different games, and compared to the earlier parametric
\nEWA model and a one-parameter stochastic equilibrium theory (QRE). Self-tuning
\nEWA does as well as EWA in predicting behavior in new games, even though it has fewer
\nparameters, and fits reliably better than the QRE equilibrium benchmark.