Building a prediction model for foreign exchange market prices has been remarkably challenging. The conventional model was not sufficiently flexible to address the time-varying nature of macrofinance risk exposures. Following Kelly et al. (J Financ Econ 134(3):501–524, 2019), we utilize Instrumented Principal Component Analysis (IPCA), a flexible factor model, to reduce the dimensionality of diverse information sets encompassing FX data and various financial risk factors, all while maintaining model tractability and accommodating time-varying betas. Our results demonstrate IPCA’s superior out-of-sample predictability compared to the random walk model, the PCA models, and conventional regression models, in terms of both statistical and economic significance. Furthermore, we find the unemployment gap and interest rate differential up to the medium term as crucial components for predicted foreign exchange returns.
