Hierarchical portfolio stress-testing with flexible probabilities

Matthews, Nina

Hierarchical portfolio stress-testing with flexible probabilities

Thesis / Dissertation

2026

Publisher

University of Cape Town

Department

Department of Statistical Sciences

Faculty

Faculty of Science

Abstract

Traditionally, portfolio optimisation has heavily relied on historical data to predict future outcomes, with the embedded assumption that historical distributional properties persist over time. However, observations have shown that many of these properties are not necessarily preserved within financial time series. This research challenges these assumptions by adopting a non-parametric, scenario-based approach using the Historical Simulation with Flexible Probabilities (HS-FP) framework (Meucci [89]). The study extends this framework by integrating the Hierarchical Risk Parity (HRP) strategy (De Prado [29]) and the Mean-Variance (MV) approach (Markowitz [69]) to enhance portfolio resilience within a South African financial market context. The HS-FP methodology quantifies future market risk by mapping drivers into state- and time-conditioned probabilities, generating forward-looking estimates of returns and volatilities. This study reveals that while the Sharpe-Ratio maximising MV portfolio demonstrated stronger cumulative performance, HRP achieved superior risk-adjusted returns, highlighting its robust diversification and resilience to extreme market volatility. The integration of the HS-FP framework, while not consistently improving overall performance, illuminated the sensitivity of traditional methods to underlying assumptions and underscored the value of incorporating market regime considerations. The findings advocate for adaptive approaches that move away from rigid assumptions, proposing a systematic, scenario-based approach to regime capturing in order to enhance the real-world applicability of stress tests.

Keywords

portfolio optimisation

Hierarchical Risk Parity

Reference:

Collections

Masters

Full item page