"It is well known that augmenting a standard linear regression model with variables that are correlated with the error term but uncorrelated with the original regressors will increase asymptotic efficiency of the original coefficients. We argue that in the context of predicting excess returns,...