Short-term exchange rate forecasting: A panel combination approach

Rossi (2013) finds that the lack of robustness in forecasting exchange rates is caused by the potential instability of model performance. We propose to combine individual exchange rate predictive models estimated via fixed effects estimation since fixed effects estimation generally eliminates the time-invariant latent factors and the combination of models can incorporate more information. We form our exchange rate forecasts by two combination methods, and the out-of-sample analysis shows that the forecasts made by our combination methods significantly outperform random walk models with or without drift for the majority of 11 currencies in recent decades. Additionally, we demonstrate that the superior performance of the combination methods is robust for various forecasting periods and areas.