As a Quantitative Analyst within the Miura Financial Research team, I contributed to a rigorous empirical study designed to test whether advanced portfolio construction methodologies could consistently outperform the S&P 500 benchmark. Our objective was to move beyond theoretical assumptions and evaluate how different allocation models perform when applied to the eleven GICS sectors of the S&P 500 (using SPDR ETFs) over a highly turbulent period from mid-2018 to May 2025.
To ensure a robust analysis, we implemented a diverse range of optimization frameworks. We tested traditional approaches like Mean-Variance, Minimum Variance, and Equally Weighted portfolios against more sophisticated risk-based models, including Risk Parity, Volatility Parity, and the machine-learning-based Hierarchical Risk Parity (HRP) introduced by De Prado. To simulate real-world investment conditions, we avoided static backtests in favor of a rolling-window approach, utilizing estimation horizons of 12 and 18 months with both monthly and quarterly rebalancing frequencies.
My analysis focused on evaluating these strategies using strict risk-adjusted metrics, including the Sharpe Ratio, Sortino Ratio, and the Probabilistic Sharpe Ratio (PSR), while employing block bootstrapping to account for the serial correlation in returns. We also tested for market timing skills using the Treynor-Mazuy and Henriksson-Merton models.
The results of our study were revealing and highlighted the limitations of classical financial theory during regime shifts. We found that widely accepted models like Mean-Variance often failed to deliver superior risk-adjusted returns compared to the naive Equally Weighted strategy, which frequently ranked as the top performer. Our data showed that during periods of market stress—specifically the COVID-19 crash of 2020 and the 2022 downturn—correlations between sectors spiked significantly, eroding the diversification benefits that sophisticated models rely upon. Ultimately, our regression analysis demonstrated that none of the strategies were able to generate statistically significant positive alpha relative to the benchmark. This project underscored the difficulty of generating excess returns purely through sector allocation and demonstrated that in highly correlated, volatile market regimes, complex optimization can sometimes underperform simple heuristic diversification.