Learning with Expert Advice

Abstract

Surveys of inflation forecasts show that expectations combine forward-looking and backward-looking elements. This contradicts conventional wisdom: In the presence of rational agents, adaptive agents would be driven out of the market. In our paper, we rationalize this finding in an equilibrium framework. Our model has two types of agents, one having rational expectations and the other using adaptive learning. The proportion of these agents in the population evolves according to their past forecasting performance. We show that even an underparameterized learning algorithm survives competition with rational expectations.