Regressions of multiple-period changes in the log exchange rate on the deviation of the log exchange rate from its 'fundamental value' display evidence that long-horizon changes in log nominal exchange rates contain an economically significant predictable component. To account for small-sample bias and size distortion in asymptotic tests, inference is drawn from bootstrap distributions generated under the null hypothesis that the log exchange rate is unpredictable. The bias-adjusted slope coefficients and R[superscript]2's increase with the forecast horizon, and the out-of-sample point predictions generally outperform the driftless random walk at the longer horizons.

We study a Lucas asset-pricing model that is standard in all respects, except that the representative agent's subjective beliefs about endowment growth are distorted. Using constant relative risk-aversion (CRRA) utility, with a CRRA coefficient below 10; fluctuating beliefs that exhibit, on average, excessive pessimism over expansions; and excessive optimism over contractions (both ending more quickly than the data suggest), our model is able to match the first and second moments of the equity premium and risk-free rate, as well as the persistence and predictability of excess returns found in the data.