Long-Run Expected Stock Returns

Abstract

We predict individual stock returns over horizons from 1 month to 10 years using machine learning. Cumulative stock returns are significantly predictable in the cross-section over all horizons. One-month ahead predicted returns explain only a quarter of the variation in 10-year predicted returns, suggesting that predicted returns at different horizons follow distinct dynamics. Predictors related to turnover and volatility are influential at all horizons. Momentum, cash flow and size related predictors are mostly important at shorter horizons, while dividend yield, value and long-term reversal related predictors are more important at longer horizons.

Keywords: Long-run discount rates, return predictability, cross-section of returns, machine learning, gradient boosting; variable importance, SHAP values

JEL Classification: G12, G17, C53