Exchange rate forecasting with DSGE models

We run a real exchange rate forecasting "horse race", which highlights that two principles hold. First, forecasts should not replicate the high volatility of exchange rates observed in sample. Second, models should exploit the mean reversion of the real exchange rate over long horizons. Abiding by these principles, an open-economy DSGE model performs well in real exchange rate forecasting. However, it fails to forecast nominal exchange rates better than the random walk. We find that the root cause is its inability to predict domestic and foreign inflation. This shortcoming leads us toward simpler ways to outperform the random walk.