Maximizing the Sharpe Ratio: A Genetic Programming Approach
55 Pages Posted: 13 Jan 2021
Date Written: November 7, 2020
While common machine learning algorithms focus on minimizing the mean-square errors of model fit, we show that genetic programming, GP, is well-suited to maximize an economic objective, the Sharpe ratio of the usual spread portfolio in the cross-section of expected stock returns. In contrast to popular regression-based learning tools and the neural network, GP can double their performance in the US, and outperform them internationally. We find that, while the economic objective plays a role, GP captures nonlinearity in comparison with methods like the Lasso, and it requires smaller sample size than the neural network.
Keywords: Machine Learning, Genetic Programming, Cross-sectional Returns, Predictability
JEL Classification: G12, G14, G15
Suggested Citation: Suggested Citation