Optimizing Portfolio Construction: Hierarchical Risk Parity (HRP) and Modified HRP Against Traditional Methods

Abstract

Inspired by the seminal work of Marcos López De Prado (2016) and Marat Molyboga (2020), this thesis critically assesses the efficacy of Hierarchical Risk Parity (HRP) and Modified Hierarchical Risk Parity (MHRP) in contrast to a wide range of conventional portfolio creation methods. HRP constructs diversified portfolios using an algorithm that combines graph theory and machine learning techniques. The performance of HRP and MHRP is assessed against equally weighted, minimum-variance, maximum Sharpe ratio, inverse-variance, inverse-volatility, and equal risk contribution portfolios. Utilizing a comprehensive dataset from the S&P 500 index ranging from January 1993 to December 2023, we employ a quantitative methodology with a rolling walk-forward analysis to assess their performance and robustness in quasi-real market conditions. The study reveals that while HRP and MHRP exhibit promising features for risk-adjusted performance, they do not consistently outperform all the traditional methods examined based on returns or (adjusted) Sharpe ratio. They nonetheless produce portfolios with low annualized volatility, which is only surpassed by the minimum variance portfolio. This outcome may be appealing to certain investors. The findings indicate that the potential of HRP and MHRP, while important, may not entirely line with the optimistic outcomes indicated by López De Prado (2016) and Molyboga (2020). This study extends the discussions about advanced portfolio strategies by highlighting the necessity for extensive empirical evaluations to determine their real-world feasibility and performance.