Fundamental Factors and Stock Returns - Based on Machine Learning Methods

Completed date: 2022-12-21

This paper utilized 207 fundamental and volume-price factors in the US stock market from January 1985 to October 2022, and employed 10 machine learning algorithms, including linear regression, penalized linear regression, tree methods, and neural networks, to synthesize factor signals and construct investment portfolios. Empirical findings demonstrated that machine learning algorithms effectively discerned the relationship between anomalies (factors) and returns. With a 1-year training window and monthly rebalancing, long-short portfolios yielded average annualized returns between 16.5% and 22.8%, with Sharpe ratios ranging from 0.69 to 1.43. Adjusting the training window to 3 months and 24 months resulted in annualized returns and Sharpe ratios spanning from 11.9% to 19.6%, 0.57 to 1.18; and from 4.96% to 21.2%, 0.59 to 1.54, respectively.