Firm Characteristics and Expected Stock Returns
62 Pages Posted: 13 Jun 2018 Last revised: 8 Aug 2019
Date Written: August 6, 2019
Complementing the widely used conventional multiple regression approach — which can suffer from overfitting with a large number of predictors — we propose a combination Lasso (C-Lasso) approach to improve out-of-sample forecasts of cross-sectional expected stock returns via shrinkage. Using 99 firm characteristics and an out-of-sample period spanning more than four decades, an approach that blends conventional and C-Lasso forecasts delivers unbiased estimates of the cross-sectional dispersion in expected returns. Similarly, combining spread portfolios formed from conventional and C-Lasso forecasts generates substantial performance gains. Our results indicate that more characteristics matter for cross-sectional expected returns than previously believed, due to time-varying characteristic premia.
Keywords: Cross-sectional expected stock returns, Characteristic premia, Forecast combination, Lasso, Forecast encompassing, Fama-MacBeth regression
JEL Classification: G11, G14
Suggested Citation: Suggested Citation