Enhancing portfolio optimization with machine learning methods: A comparative study using commodity markets data

Portfolio optimization is the process of selecting an optimal portfolio of assets according to some objective, investor preferences, and constraints, typically optimizing the trade-off between risk and return. This study evaluates the effectiveness of traditional and novel machine learning portfolio optimization techniques by incorporating short selling, a design feature often overlooked in previous research. We employ historical commodity market data from seven commodity groups. The strategies investigated include Mean-Variance Optimization, Global Minimum Variance, Equal Weights, Maximum Diversification, Risk Parity, and Hierarchical Risk Parity. The findings suggest that allowing for short selling impacts the performance portfolio optimization strategies. Mean-Variance Optimization potentially increases returns but at the cost of greater volatility. Global Minimum Variance consistently exhibits stability and minimal risk, ideal for portfolio managers who adopt conservative investment strategies. Maximum Diversifying Portfolio and Risk Parity show moderate but resilient performance, and Hierarchical Risk Parity, despite its innovation, tends to be more volatile. Surprisingly, the Equal Weighted strategy holds its ground against more complex approaches, providing a viable option for those who value simplicity. This analysis highlights the importance of matching portfolio strategies with investor risk preferences, especially when integrating techniques like short selling.