Merging Structural and Reduced-Form Models for Forecasting

Abstract

Recent economic crises have posed important challenges for forecasting. Models estimated pre-crisis may perform badly when normal economic relationships have been disrupted. Meanwhile, forecasting, especially in central banks, is increasingly based on a suite of models, following two main approaches: structural (DSGE) and reduced form. The challenge remains to identify which model – or combination of models – is likely to make better forecasts in a changing environment. We explore this issue by assessing the forecasting performance of combinations of a medium-scale DSGE model with standard reduced-form methods applied to the Spanish economy and a reference period that includes both the great recession and the sovereign debt crisis. Our findings suggest that: (i) the mean reverting properties of the DSGE model cause it to underestimate the growth of real variables following the inclusion of crisis episodes in the estimation period; (ii) despite this, reduced-form VARs benefit from the imposition of an economic prior from the structural model; but (iii) pooling information in the form of variables extracted from the structural model with (B)VAR methods does not improve forecast accuracy. By analysing the quantiles of the predictive distributions, we also provide evidence that merging models can help improve the forecast in a context including crisis episodes.