Skip: A Small-sample Correction for Return Prediction with Valuation Ratios

Abstract

It has been known since Stambaugh (1986) that persistent valuation-ratio regressors lead to biased coefficients in small-sample return regressions. We propose a “Skip” estimator that corrects this bias by averaging estimates from sub-samples of non-consecutive observations. Each sub-sample simply skips adjacent observations and so weakens the bias-inducing link between the innovation in returns and the innovation in subsequent valuation ratios. The estimator does not require estimates of the unknown persistence and covariance parameters, and it facilitates inference since the t-statistic is now distributed approximately standard normal. Finally, it compares favorably to OLS in terms of mean square error and is easily implementable with standard software.

Keywords: equity premium, market timing, small-sample bias, predictability, valuation ratios

JEL Classification: G14