Finite Sample Analysis of Predictive Regressions with Long-Horizon Returns

Abstract

In this paper, we provide an exact finite sample analysis of predictive regressions with overlapping long-horizon returns. This analysis allows us to evaluate the reliability of various asymptotic theories for predictive regressions in finite samples. In addition, our finite sample analysis sheds lights on the long outstanding question of whether a predictive regression with short or long-horizon returns is more powerful in detecting return predictability. Finally, we provide a simple bias-adjusted estimator of the slope coefficient as well as its estimated standard error for predictive regression with long-horizon returns. The resulting t-ratio of our bias-adjusted estimator has excellent size properties and dominates existing alternatives in the literature.