This paper addresses the issue of using aggregate price indices for purchasing power parity (PPP) tests and the fitness of PPP as a model for exchange rate forecasting. Compared with consumer price index (CPI) and wholesale price index (WPI), the price index of traded-goods (TPI) appears to be a more appropriate price index for both PPP tests and exchange rate forecasting. It performs better in the tests and provides superior exchange rate forecasts, both within- and out-of-sample. While the half-life of the estimated deviations from PPP is about two years for the CPI and WPI-based real exchange rates, it is only one year for the TPI-based real exchange rates. (C) 2002 Elsevier Science Ltd. All rights reserved.