Forecasting GDP with many predictors in a small open economy: forecast or information pooling?

Abstract

This study compares two distinct approaches, pooling forecasts from single indicator MIDAS models versus pooling information from indicators into factor MIDAS models, for short-term Singapore GDP growth forecasting with a large ragged-edge mixed frequency dataset. We consider various popular weighting schemes in the literature when conducting forecast pooling. As for factor extraction, both the conventional dynamic factor model and the three-pass regression filter approach are considered. We investigate the relative predictive performance of all methods in a pseudo-out-of-sample forecasting exercise from 2007Q4 to 2020Q3. In the stable growth non-crisis period, no substantial difference in predictive performance is found across forecast models. In comparison, we find information pooling tends to dominate both the quarterly autoregressive benchmark model and the forecast pooling strategy particularly during the Global Financial Crisis. Supplementary Information The online version contains supplementary material available at 10.1007/s00181-022-02356-9.