The Value of Economic Constraints in Boosting Equity Premium Prediction

Abstract

This paper studies the role of three economically motivated constraints in estimating a boosted regression trees model: the sign restrictions on the relationship between equity premium and the predictors, the upper bound of equity premium predictability, and a modified loss function that shifts prediction toward positive equity premium. The empirical results show that simultaneously incorporating the three economic constraints helps the boosted trees model generate significant monthly out-of-sample R-squared above 1%, and the Sharpe ratios of market timing experiments based on the constrained predictions outperform the ones based on the historical mean forecasts.

Keywords: Economic constraints, equity premium prediction, boosted regression trees, weak signal-to-noise ratio.