Robustness in Foreign Exchange Rate Forecasting Models: Economics-based Modelling After the Financial Crisis

The aim of this article is to analyse the out-of-sample behaviour of a bunch of statistical and economics-based models when forecasting exchange rates (FX) for the UK, Japan, and the Euro Zone in relation to the US. A special focus is given to the commodity prices boom of 2007-8 and the financial crisis of 2008-9. We analyse the forecasting behaviour of six economic plus three statistical models when forecasting from one up to 60-steps-ahead, using a monthly dataset comprising from 1981.1 to 2014.6. We first analyse forecasting errors until mid-2006 to then compare to those obtained until mid-2014. Our six economics-based models can be classified in three groups: interest rate spreads, monetary fundamentals, and PPP with global measures. Our results indicate that there are indeed changes of the first best models when considering the different spans. Interest rate models tend to be better predicting using the short sample; also showing a better tracking when crisis hit. With the longer sample the models based on price differentials are more promising; however, with heterogeneous results across countries. These results are important since shed some light on what model specification use when facing different FX volatility.