The Performance of Forecast-Based Monetary Policy Rules under Model Uncertainty

Preprint
Publisher Website
Google Scholar
  • 1 January 2000
    • preprint
      • Published in RePEc
Abstract
In this paper, we consider whether monetary policymakers should adjust short-term nominal interest rates in response to inflation and output forecasts rather than to recent outcomes of these variables. The use of forecast-based rules has been advocated on the basis of transmission lags and other considerations, and such rules also provide a reasonably good description of the policy strategies of several inflation-targeting central banks. We address these issues using four different macro-econometric models of the U.S. economy (the Fuhrer-Moore model, the MSR model of Orphanides and Wieland, Taylor's Multi-Country Model, and the FRB/US staff model); all four models incorporate rational expectations and nominal inertia, but differ in many other respects. We begin by evaluating the performance of various forecast-based rules that have been proposed in the literature. We find that some of these rules yield relatively poor performance, and that a number of such rules fail to yield determinacy (that is, a unique rational expectations equilibrium) in at least one of the four models. Next, we determine the optimal set of forecast-based rules for each model (that is, the rules that trace out the inflation-output volatility frontier subject to an upper-bound on interest rate volatility). We find that even optimized forecast-based rules yield very small benefits compared with optimized outcome-based rules that respond to current inflation, the current output gap, and the lagged interest rate. In the case of rules that respond directly to inflation forecasts but not to the output gap, we find a substantial deterioration in performance, even as measured by a policymaker whose sole objective is to minimize inflation variability. Finally, rules that involve relatively short forecast horizons (less than one year ahead) are reasonably robust to model uncertainty; that is, when such a rule is optimized for one model, the rule also performs reasonably well in the other three models. H

This publication has 0 references indexed in Scilit: