Exploration of Hierarchical Clustering in Long-only Risk-based Portfolio Optimization

Daniel Sjöstrand & Nima Behnejad

Student thesis: Master thesis

Abstract

Modern portfolio optimization methods have introduced new ways of allocating capital and have drawn the attention of scholars, practitioners, and the general public alike. The thesis aims to add to the empirical evidence on the impact and risk-based performance of hierarchical clustering portfolios in long-only risk-based portfolio optimization. This is achieved by analyzing and investigating the Hierarchical Risk Parity, Hierarchical Equal Risk Contribution, and Nested Clustered Optimization methods, and compare these from a risk-based perspective to several traditional optimization methods. The relative risk-based performance is assessed through Monte Carlo simulations using synthetic data as well as through a walk-forward backtest applied on historical S&P 500 data. Together, the methodology provides a broad view of the general performance, but also more focused insights into potential estimation error reduction and the impact of different clustering parameters. The combined empirical results do not provide conclusive support for any general performance gains from hierarchical clustering in portfolio optimization. The initial positive effects found in earlier studies for Nested Clustered Optimization are hypothesized to stem from the highly stylized and simplified assumptions applied. The results given in this thesis suggest that these initial positive effects diminish when applied to more realistic data. Furthermore, the results for Hierarchical Risk Parity and Hierarchical Equal Risk Contribution show results in line with previous studies by Raffinot (2018). It is concluded that they are performing reasonably well but underperform in comparison to several of the traditional portfolios on most risk-based performance dimensions included. The findings do not indicate any general increase in risk-based performance, but do, however, show promise in providing more control over the weight concentration. In conclusion, the authors find that clustering indicates some promising aspects, but that these are limited given the applied hierarchical methodology, and further research is warranted to reach more conclusive answers.

EducationsMSc in Finance and Investments, (Graduate Programme) Final Thesis
LanguageEnglish
Publication date2020
Number of pages140