Machine Learning in Portfolio Optimization: A Comparative Study of Hierarchical Risk Parity and Traditional Allocation Methods

Abstract

The Hierarchical Risk Parity (HRP) is an asset allocation technique, using clustering algorithms and machine learning to obtain and optimize the return and risk of a portfolio. This report evaluates the HRP approach in relation to more conventional allocation techniques used for decades, such as the minimum variance, equal weight and inverse variance. The context of the portfolio technique comparison is conducted in the Swedish stock market, more specifically the OMX Stockholm GI index. The time frame of the reports is taking account of the last 20 years, a time period that has experienced several financial crises, recessions and depressions, such as the financial crisis in 2008 and the covid crisis in 2022. All industries in the OMX index have been evaluated, in addition with sector specific and crisis specific analyses. The results, aligning with existing previous research in the field of HRP of Lopez de Prado (2016), indicate numerous advantages with the HRP approach. In several areas existing the HRP has outperformed conventional optimization methods in following areas, such as robustness, risk reduction, return increment. The HRP advocates and provides an optimal balance between risk mitigation and return, even in the circumstances of crises and sector specific environments.