Estimating Time-Varying DSGE Models Using Minimum Distance Methods

Abstract

This paper uses kernel methods to estimate a seven variable time-varying (TV) vector autoregressive (VAR) model on the US data set constructed by Smets and Wouters. We use an indirect inference method to map from this TV VAR to time variation in implied Dynamic Stochastic General Equilibrium (DSGE) parameters. We find that many parameters change substantially, particularly those defining nominal rigidities, habits and investment adjustment costs. In contrast to the ‘Great Moderation’ literature our monetary policy parameter estimates suggest that authorities tried to deliver a low and stable inflation from 1975 onwards. However, the severe adverse supply shocks in the 70s could have caused these policies to fail.