LEARNING AND THE GREAT MODERATION*

We study a stylized theory of the volatility reduction in the U.S. after 1984—the Great Moderation—which attributes part of the stabilization to less volatile shocks and another part to more difficult inference on the part of Bayesian households attempting to learn the latent state of the economy. We use a standard equilibrium business cycle model with technology following an unobserved regime‐switching process. After 1984, according to Kim and Nelson (1999a), the variance of U.S. macroeconomic aggregates declined because boom and recession regimes moved closer together, keeping conditional variance unchanged. In our model this makes the signal extraction problem more difficult for Bayesian households, and in response they moderate their behavior, reinforcing the effect of the less volatile stochastic technology and contributing an extra measure of moderation to the economy. We construct example economies in which this learning effect accounts for about 30% of a volatility reduction of the magnitude observed in the postwar U.S. data.