Predictive Regressions with Time-Varying Coefficients

We evaluate predictive regressions that explicitly consider the time-variation of coefficients in a comprehensive Bayesian framework. This allows for fast and consistent adjustment of regression coefficients to changes in the underlying economic relationships (e.g., changes in the regulatory environment) as we document explicitly for the coefficient of the dividend yield. For monthly returns of the S&P 500 index, we demonstrate statistical and, especially, economic evidence of out-of-sample predictability. In both cases, the proposed framework outperforms regressions with constant coefficients. One explanation for this improvement is the proposed methodology’s ability to identify periods with high or low prediction uncertainty. JEL Classifications: G12, C11