Evaluating DSGE Models for Monetary and Fiscal Policy Analysis

This paper evaluates the suitability of current DSGE models for monetary policy analysis from a Bayesian perspective, recognizing that these models may be useful even if they are misspecified along some dimensions, using the framework laid out in Faust and Gupta (2009). The paper argues that practical monetary policy analysis deals with determining how intended policy should be revised in light of the structural interpretation of incoming news. Standard DSGE models are linear and Gaussian, so this analysis reduces to i) the variance-covariance properties of one-step ahead reduced-form forecast errors and ii) Kalman gains relating those forecast errors to unobserved structural shocks. Standard model evaluation tools assess the degree to which the model matches all aspects of the reduced form of the data. The paper provides new tools for evaluating both the forecast error properties and the structural implications. Finally the paper illustrates this approach using the noted Smets-Wouters model (AER, 2007 ), finding strengths and some important shortcomings.