Real-Time Out-of-Sample Exchange Rate Predictability

This paper revisits the long-standing Meese and Rogoff puzzle by examining the importance of real-time data for exchange rate forecasting. Most of the existing literature on exchange rate predictability uses recent historical data, which are not available to the public at the time the forecasts are made. This paper evaluates short- and long-horizon out-of-sample exchange rate predictability using Purchasing Power Parity (PPP) and Taylor rule fundamentals for 16 OECD currencies during the post-Bretton Woods era. Comparing the results with real-time and revised data, the evidence of short-run exchange rate predictability with Taylor rule models is stronger with real-time data. The models with Taylor rule fundamentals outperform the naïve no-change model at the 1-quarter horizon for 8 out of 16 currencies vis-à-vis the U.S. dollar with real-time data and for 6 out of 16 currencies with revised data, with the strongest evidence coming from specifications that incorporate heterogeneous coefficients. The evidence of short-run predictability is much stronger with Taylor rule models than with conventional purchasing power parity model regardless of which type of data is used. The out-of-sample performance of both PPP and Taylor rule fundamentals improves at longer horizons, with PPP model performing best in the long run. At the 16-quarter horizon, the models with Taylor rule fundamentals outperform the random walk for 10 out of 16 currencies vis-à-vis the U.S. dollar with either type of data, while the PPP model outperforms the naïve no-change model for 13 out of 16 currencies with real-time data and for 11 out of 16 currencies with revised data. Key Words: