Learning, parameter variability, and swings in US macroeconomic dynamics

Recent studies show that the estimated parameters of rational expectations dynamic stochastic general equilibrium models of the business cycle are largely time-varying. This paper shows that assuming adaptive learning (rather than rational expectations) strongly reduces the estimated parameter variability of standard models (by around 75%). Moreover, the reduction in parameter variability induced by adaptive learning is much stronger for the subsets of parameters that control nominal price and wage rigidity and the subset of policy rule parameters (at 98% and 83%, respectively). Furthermore, our estimation results suggest that adaptive learning helps to explain the recent swings in the comovements between real and nominal US macroeconomic variables, but the swing in the relative weight of supply and demand shocks seems to be the most important driving force.