How to Manage Portfolios in Different Clusters? An Unsupervised Machine Learning Approach

Rahman, Mohammad Mahbubur; Syed, Zeeshan; Akiotu, Cynthia Abiemwense; Eskandari, Rasol

Abstract

Using the daily data from 2010 to 2023 on global indices for 15 heterogeneous stocks’ prices, this study examines traditional portfolio theory by applying unsupervised machine learning, particularly the K-means clustering algorithm. This algorithm identifies several clusters, representing a specialised portfolio strategy in each cluster. Each cluster has demonstrated distinct risk-return characteristics and asset correlations. Crucially, this study has unveiled the potential of cluster-based portfolio management. Clusters are tailored to different risk profiles, exhibiting the potential for adaptive investment strategies especially pivotal in fluctuating markets.

• Working Paper Type: Preprint
• Deposit Date: Jun 14, 2025
• DOI: https://doi.org/10.2139/ssrn.5251712