Stambaugh Correlations and Redundant Predictors
48 Pages Posted: 17 Mar 2010
Date Written: March 15, 2010
We consider inference in a widely used predictive model in empirical finance. "Stambaugh Bias" arises when innovations to the predictor variable are correlated with those in the predictive regression. We show that high values of the "Stambaugh Correlation" will arise naturally if the predictor is actually redundant, but simply proxies univariate predictability. For such predictors even bias-corrected conventional tests will be severely distorted. We propose tests that distinguish well between redundant predictors and the true (or "perfect") predictor. An application of our tests does not reject the null that a range of predictors of stock returns are redundant.
Keywords: predictive return regressions, Stambaugh bias, ARMA models, predictive systems
JEL Classification: C22, G12
Suggested Citation: Suggested Citation