We use factor-augmented predictive regressions to analyze the relation between nominal exchange rates and macroeconomic variables. Using a panel of 121 US macroeconomic time series, we estimate eight factors through principal component analysis. Those estimated factors have significant predictive power and can substantially improve the predictive power of purchasing power parity through both in-sample and out-of-sample analyses. The estimated macroeconomic factor, which co-moves with US stock market variables, has strong predictive power for nominal exchange rate fluctuations in the short run, while estimated factors, co-moving with interest rate spreads, government-issued bond yields and employment variables, have strong predictive power in the long run. Moreover, optimal factors selected by the BIC in the out-of-sample analysis differ greatly depending on the time points when forecasts are made. Finally, we show that factors extracted from a panel of 121 US time series data and those extracted from a panel of 215 Korean macroeconomic series together can predict a substantial portion of movements in the Korea–US bilateral exchange rate.
Read full abstract