Forecasting Exchange Rates of Major Currencies with Long Maturity Forward Rates

This paper shows that error correction models assuming that long-maturity forward rates are stationary outperform the random walk in out of sample forecasting at forecasting horizons mostly above one year, for US dollar exchange rates against nine industrial countries’ currencies, using the 1990-2006 period for evaluating the out of sample forecasts. The improvement in forecast accuracy of our models is economically significant for most of the exchange rate series, and statistically significant according to a bootstrap test. Our results are robust to the specification of the error correction model and to the underlying data frequency.