Exchange rate predictability: A variable selection perspective

Abstract

To enhance the exchange rate forecast ability, we adopt method for pooling forecasts from a large number of predictors, Bayesian Variable Selection. In pseudo out-of-sample forecasting, the Bayesian Variable Selection outperforms the random walk models and predicts the correct sign of exchange rate changes with higher than 60% accuracy at the short horizon. In sample analysis shows that critical predictors for exchange rates vary over time and differ across countries. It implies that not only the unstable relationship between the exchange rate and economic variables, but also the model uncertainty should be considered to the exchange rate forecasts. (JEL classi-fication: C11, C53, F31).