A model of learning and emulation with artificial adaptive agents

Abstract

We study adaptive learning in a sequence of overlapping generations economies in which agents live for n periods. Agents initially have heterogeneous beliefs, and form multi-step-ahead forecasts using a forecast rule chosen from a vast set of candidate rules. Agents learn in every period by creating new forecast rules and by emulating the forecast rules of other agents. Computational experiments show that systems so defined can yield three qualitatively different types of long-run outcomes: (1) coordination on a low inflation, stationary perfect foresight equilibrium; (2) persistent currency collapse; and (3) coordination failure within the allotted time frame.