Machine Learning for Equity Risk Premium Measurement

Summary

This paper compares machine learning methods for forecasting equity returns across the market time series and the cross section of stocks. It frames risk premium measurement as a prediction problem and describes how high-dimensional predictors, regularization, model selection, and efficient search can help address correlated signals and uncertain functional forms while limiting overfitting.

The excerpt reports that trees and neural networks performed best among the methods examined, with gains attributed to nonlinear relationships and predictor interactions. It gives out-of-sample Sharpe ratios for a neural network strategy on the S&P 500 and a value-weighted long-short stock strategy, comparing the market strategy with buy-and-hold and describing the stock strategy as surpassing leading regression-based approaches. The authors identify momentum, liquidity, and changes in volatility as shared signals. These results are specific to the study’s data and design; the excerpt does not provide the full methodology or establish that the findings generalize to other markets or periods.

Key ideas

• The paper treats equity risk premium estimation as a forecasting task in both the market time series and stock return cross section.
• Machine learning combines flexible high-dimensional models with regularization and efficient model search to control overfitting.
• Trees and neural networks show the strongest reported results, with gains linked to nonlinear effects and predictor interactions.
• Momentum, liquidity, and changes in volatility appear among the predictive signals recognized across methods.
• Reported strategy performance is out of sample but remains specific to the study’s setting and evidence.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.