Learning, hypothesis testing, and rational-expectations equilibrium

Foster and Young (2003) provide a model of learning by hypothesis testing that spends almost all of the time approximating Nash equilibria of a repeated game. Here I extend this learning model to a macroeconomic setting, where agents' decisions are informed by hypotheses they hold regarding the economy. They periodically test these hypotheses against observed data, and replace them if they fail. Under certain conditions, agents who learn in this way spend a large fraction of the time approximating rational-expectations equilibria.