Evaluating the predictability of exchange rates using long horizon regressions

Since the breakdown of the Bretton Woods agreement, researchers have used a wide variety of structural models to try to predict exchange rate movements. Unfortunately, finding consistent evidence that these models outperform a random walk has proven elusive. In this paper we investigate the impact different methods of inference may have had on these conclusions. Using p-values based on recently developed tests of forecast accuracy and encompassing, as well as q-values designed to mitigate multiple testing problems, we provide stronger evidence consistent with these models having superior predictive ability. Our results suggest that previous studies ’ inability to detect predictive ability may have been influenced by the statistics used and the manner in which they were employed