Cross-Sectional Expected Returns: New Fama-MacBeth Regressions in the Era of Machine Learning
57 Pages Posted: 13 Jun 2018 Last revised: 30 Apr 2023
Date Written: April 28, 2023
Abstract
We extend the Fama-MacBeth regression framework for cross-sectional return prediction to incorporate big data and machine learning. Extensions for improving cross-sectional return prediction include penalized regression, forecast ensembles, and random features to accommodate nonlinearities. We also develop tools for assessing cross-sectional return forecasts using the Fama-MacBeth approach, including a generalization of the popular out-of-sample R-squared statistic. Applying our new methods to predict cross-sectional stock returns using over 200 firm characteristics, we find that the Fama-MacBeth regression framework augmented by machine learning significantly improves cross-sectional return forecasts.
Keywords: Penalized regression, Forecast combination, Forecast Encompassing, Random features, Characteristic payoffs, Cross-sectional out-of-sample R-squared statistic
JEL Classification: C21, C45, C53, C55, C58, G12, G17
Suggested Citation: Suggested Citation