Sparse Predictive Regressions: Statistical Performance and Economic Significance

Abstract

We propose and evaluate a variety of penalized regression methods for forecasting and economic decision making in a data-rich environment under parameter uncertainty. Empirically, we explore the statistical and economic performance across different asset classes such as stocks, bonds, and currencies, and alternative strategies within an asset class (e.g., momentum and value in the space of equity). The main result shows that penalties that both shrinkage the model space and regularize the remaining regression parameters, e.g. elastic net penalty, tend to outperform competing sparse and dense methodologies, both statistically and economically.

Keywords: Return Predictability, Empirical Asset Pricing, Machine Learning, Bayesian Methods, Penalized Regressions.

JEL Classification: C38, C45, C53, E43, G12, G17.