Exchange Rates and Monetary Fundamentals: What Do We Learn From Long-Horizon Regressions?

Abstract

Long-horizon regression tests are widely used in empirical finance, despite evidence of severe size distortions. This paper introduces a new bootstrap method for small-sample inference in long-horizon regressions. A Monte Carlo study shows that this bootstrap test has much smaller size distortions than conventional long-horizon regression tests. It is also shown that long-horizon regression tests do not have power advantages against economically plausible alternatives. The apparent lack of higher power at long horizons suggests that previous findings of increasing long-horizon predictability are more likely due to size distortions than to power gains. The use of the bootstrap method is illustrated by analyzing whether deviations from monetary fundamentals help predict changes in four major exchange rates. In contrast to earlier studies, the test provides only weak evidence of exchange rate predictability and no evidence of increasing long-horizon predictability. Many of the differences in results can be traced to the implementation of the test.