Model Selection for Monetary Policy Analysis – Importance of Empirical Validity

Abstract

We investigate the importance of employing a valid model for monetary policy analysis. Specifically, we investigate the economic significance of differences in specification and empirical validity of models. We consider three alternative econometric models of wage and price inflation in Norway. We find that differences in model specification as well as in parameter estimates across models can lead to widely different policy recommendations. We also find that the potential loss from basing monetary policy on a model that may be invalid, or on a suite of models, even when it contains the valid model, can be substantial, also when gradualism is exercised as a concession to model uncertainty. Furthermore, possible losses from such a practice appear to be greater than possible losses from failing to choose the optimal policy horizon to a shock within the framework of a valid model. Our results substantiate the view that a model for policy analysis should necessarily be empirically valid and caution against compromising this property for other desirable model properties, including robustness.