Machine Learning and Factor-Based Portfolio Optimization

TL;DR

This paper explores how machine learning, especially autoencoder neural networks, can improve portfolio optimization by identifying latent factors that outperform traditional methods in risk minimization, particularly during volatile periods.

Contribution

It demonstrates that autoencoder-based latent factors and sparse methods enhance portfolio risk reduction compared to traditional estimators, especially in high volatility environments.

Findings

Autoencoder-derived factors show weaker links with characteristic-sorted portfolios.

Machine learning methods produce covariance and weight structures diverging from simpler estimators.

Autoencoder-based portfolios outperform benchmarks in risk minimization during volatile periods.

Abstract

We examine machine learning and factor-based portfolio optimization. We find that factors based on autoencoder neural networks exhibit a weaker relationship with commonly used characteristic-sorted portfolios than popular dimensionality reduction techniques. Machine learning methods also lead to covariance and portfolio weight structures that diverge from simpler estimators. Minimum-variance portfolios using latent factors derived from autoencoders and sparse methods outperform simpler benchmarks in terms of risk minimization. These effects are amplified for investors with an increased sensitivity to risk-adjusted returns, during high volatility periods or when accounting for tail risk.