Estimation and Inference in Predictive Regressions
Abstract
In this paper, we analyze feasible bias-reduced versions of point estimates for predic- tive regressions: The plug-in estimates, which are based on the augmented regressions proposed by Amihud and Hurvich (2004) and Amihud, Hurvich and Wang (2010), and the grouped jackknife estimate by Quenouille (1949, 1956). These estimates are easily obtained by the least squares or the instrumental variable methods. We also derive the correct standard errors associated with these point estimates. The methods thus allow for a unified inferential framework, where point estimates and statistical inference are based on the same methods. Using the new estimates, we investigate U.S. stock returns and find that some variables, which have not been statistically detected as useful predic- tors in the literature, are able to predict stock returns. Because of their nice properties, our methods complement the existing statistical tests for predictability to investigate the relations between stock returns and economic variables.
