Modern Portfolio Theory (MPT), developed by Harry Markowitz, transformed investment practices by wisely balancing risk and return. Nonetheless, its efficacy wanes in fluctuating financial markets due to its dependence on historical data and fixed assumptions. This paper investigates incorporating Machine Learning (ML) techniques into the traditional Markowitz optimization framework to enhance portfolio construction and risk management processes. It highlights the use of supervised learning for forecasting asset returns, unsupervised learning for asset clustering, and reinforcement learning for adjusting portfolios dynamically. An empirical analysis utilizing recent U.S. market data reveals that ML models improve risk assessment, asset selection, and adaptive portfolio allocation. Techniques such as linear regression, clustering algorithms, and principal component analysis (PCA) facilitate superior forecasting and portfolio design in various market environments. The research also shows that ML can enhance Sharpe ratios in specific market conditions compared to conventional MPT. ML increases portfolio flexibility and robustness by aligning predictive modeling with optimization objectives. This evolving methodology lays the foundation for a more responsive, data-informed investment strategy in today’s finance, providing a viable alternative to the limitations of traditional models.
Junhao Xu (Thu,) studied this question.