Predictive regressions

When a rate of return is regressed on a lagged stochastic regressor, such a s a dividend yield, the regression disturbance is correlated with the regre ssor's innovation. The OLS estimator's finite-sample properties, derived he re, can depart substantially from the standard regression setting. Bayesian posterior distributions for the regression parameters are obtained under s pecifications that differ with respect to (i) prior beliefs about the autoc orrelation of the regressor and (ii) whether the initial observation of the regressor is specified as fixed or stochastic. The posteriors differ acros s such specifications, and asset allocations in the presence of estimation risk exhibit sensitivity to those differences. (C) 1999 Elsevier Science S. A. All rights reserved. JEL classification. C32; C11; G11.