This study uses a relative purchasing power parity (ppp) model—one that is an alternative to the ‘real exchange rate’—to construct a time-varying equilibrium exchange rate of six US Dollar exchange rates. A linear forecasting technique is then used to determine the horizon over which a relative ppp-based model outperforms a random walk in forecasting exchange rates out-of-sample. The ppp models presented are based on CPI, PPI, and a proxy for traded goods price indexes. For all the currencies examined, the relative ppp-based model displays some sizeable forecasting improvements over a random walk. The ability of the relative ppp model (based on the proxy for traded goods price indexes) to outperform the random walk is most consistent in predicting changes in the US Dollar–British Pound exchange rate. Moreover, the best forecasting results vis-a-vis the random walk are obtained over the post-Plaza Accord period. For example, the ppp model statistically significantly outperforms the random walk for the British Pound (based on the proxy for traded goods price indexes) and the German Mark (based on CPI) over a 1-month forecasting horizon. Copyright © 2010 John Wiley & Sons, Ltd.