Machine Learning for Enhanced Portfolio Stability: Entropy and Clustering Insights

We introduce a novel machine learning approach for identifying equities suitable for constructing stable minimum variance portfolios, as compared to an index portfolio. We explore the effects of using correlation-based and entropy-based distance metrics on clustering and stability analysis. Notably, the interplay between entropy-based and correlation-based clustering outcomes is complex; we examine this intricacy, highlighting the practical advantages of the entropy-based metric. Our research demonstrates the ability of clustering to detect changes in the stationarity of portfolio-specific excess returns, which has implications for stability monitoring. We also examine the question of optimal cluster count, with consequences for improving the stability of minimum variance portfolios.