Forecasting exchange rates using multivariate threshold models

This paper investigates the ability of a broad range of non-linear time series models to forecast the EUR/USD exchange rate. Using a variant of the well-known Dornbusch (Dornbusch, R. 1976. Expectations and Exchange Rate Dynamics. Journal of Political Economy 84: 11611176.) model to guide the specific choice of covariates, we find improvements over the random walk for all time horizons considered. While the improvement in forecasting accuracy is rather muted at the critical 1-month ahead horizon, accuracy increases seem to be more pronounced for longer-term forecasts. In addition, we account for model and specification uncertainty by applying several combination rules. Along this dimension our results suggest that we can still improve upon the single best performing model by a large extent.