Forecasting the nominal exchange rate movements in a changing world. The case of the U.S. and the U.K

Exchange rate forecasting has become an arena for many researchers the last decades while predictability depends heavily on several factors such as the choice of the fundamentals, the econometric model and the data form. The aim of this paper is to assess whether modelling time-variation and other forms of instabilities may improve the forecasting performance of the models. Paper begins with a brief critical review of the recently developed exchange rate forecasting models and continues with a real-time forecasting race between our fundamentals-based models, a DSGE model, estimated with Bayesian techniques and the benchmark random walk model without drift. Results suggest that models accounting for non-linearities may generate poor forecasts relative to more parsimonious and linear models.