How Frequently Should We Re-Estimate DSGE Models?

Abstract

A common practice in policy making institutions using DSGE models for forecasting is to re-estimate them only occasionally rather than every forecasting round. In this paper we ask how such a practice affects the accuracy of DSGE model-based forecasts. To this end we use a canonical medium-sized New Keynesian model and compare how its quarterly real-time forecasts for the US economy vary with the interval between consecutive re-estimations. We find that updating the model parameters only once a year usually does not lead to any significant deterioration in the accuracy of point forecasts. On the other hand, there are some gains from increasing the frequency of re-estimation if one is interested in the quality of density forecasts.