Third-Country Effects on the Exchange Rate

This paper attributes the exchange-rate disconnect puzzle, which we characterize as the low adjusted R-squared in short-horizon predictive regressions, to omitted `third-country' variables. Using a three-country DSGE exchange rate model, we identify channels through which shocks originating in Country 3 can generate substantial variability in the bilateral exchange rate between Countries 1 and 2. In Monte Carlo experiments, using the model as the data generating process, we find that including Country 3 variables in otherwise conventional two-country exchange-rate regressions result in sizable increases in explanatory power. Empirical analysis of data from the U.S. and eleven other countries support the predictions of the theory.