Learning, regime switches, and equilibrium asset pricing dynamics

Abstract

We examine the autocorrelation properties of equilibrium returns when the growth rate of dividends, consumption, or GNP follows a Markov switching process. In contrast to earlier studies, we assume that the agent learns the current growth state using observed dividends and Bayes' rule. For the model with switching in the mean growth rate we find that learning significantly alters the pattern of returns, often generating positive rather than negative autocorrelation. For the model calibrated to GNP with switching in the variance of the growth rate, we show that learning induces negative autocorrelation.