Abstract

Self-tuning experience weighted attraction (EWA) is a one-parameter theory of learning in games. It replaces the key parameters in an earlier model (EWA) with functions of experience that “self-tune ” over time. The theory was tested on seven different games, and compared to the earlier model and a one-parameter stochastic equilibrium theory. The more parsimonious self-tuning EWA does as well as EWA in predicting behavior in new games, and reliably better than an equilibrium benchmark. The economic value of a learning theory is measured by how much more subjects would have earned in an experimental session if they followed the theory’s recommendations. Economic values for several learning and equilibrium theories were estimated (controlled for boomerang effects of following a model’s advice in one period, on future earnings). Most models have economic value. Self-tuning The power of equilibrium models of games comes from their ability to produce precise predictions using only the structure of a game and assumptions about players’ rationality. Statistical models of learning, on the other hand, often need data to calibrate free parameters, and use

Citations