Learning and Equilibrium in Games

In the last ten years theory (e.g., Fudenberg and Levine, 1998) and empirical data fitting have provided many ideas about how equilibria arise in games or markets. This short chapter describes a very general approach to learning in games: "experience-weighted attraction" (EWA) learning. This approach strives to explain, for every choice in an experiment, how that choice arose from players' previous behavior and experience, using a general model which can be applied to most games with minimal customization and which predicts well out of sample. Sophisticated EWA includes important equilibrium concepts and many other learning models (simple reinforcement, Cournot, fictitious play, weighted fictitious play) as special cases (see Camerer and Ho, 1999; Ho, Camerer and Chong, 200 I, and cited references for details). The model therefore allows "one-stop shopping" for learning about the latest statistical comparisons of many different learning and equilibrium models (see Camerer, 2002, Chapter 6 for more details). The model can also be adapted to field applications in which strategies and payoffs are often poorly-specified (e.g., it has been used successfully to predict actual consumer choices of products like ice cream, see Ho and Chong, 1999).