Federal Reserve Bank

We provide a methodology for estimating time-varying weights in optimal prediction pools, building on the work by Geweke and Amisano (2011, 2012). We use this methodology to investigate the relative forecasting performance of DSGE models with and without financial frictions from 1992 to 2011 at various forecast horizons. Our results indicate that models with financial frictions produce superior forecasts in periods of financial distress, which may partly explain why macroeconomists may have neglected models with financial frictions prior to the Great Recession. At the same time we show that even in ‘tranquil ’ periods the weight on the financial friction model is far from zero, and that using real time estimates this weight increases before the Lehman crisis