The Value of Economic Regularization for Stock Return Predictability

We propose a new approach for using stock market return predictors to maximize investor's utility gains by adding an economically motivated penalization term to the conventional OLS loss function. Our approach is computationally easy to implement and delivers superior out-of-sample (OOS) utility gains measured by the certainty equivalent (CE). Moreover, the new methodology demonstrates that the advantage relative to the OLS approach becomes larger when an investor considers higher moments of portfolio returns and has a larger degree of risk aversion coe cient. Our results point toward the importance of aligning the loss function with the OOS evaluation metric when using return predictor variables.