Covariance-Aware Simplex Projection for Cardinality-Constrained Portfolio Optimization

Battle Count: 74

Highly relevant to quantitative trading and portfolio management. CASP provides a principled, drop-in replacement for Euclidean projection in metaheuristic portfolio optimizers, directly addressing the NP-hard cardinality-constrained portfolio optimization problem. The tracking-error interpretation gives clear financial meaning to the repair geometry. Results on S&P 500 data with walk-forward validation demonstrate practical applicability. The method is immediately deployable in institutional portfolio construction workflows, particularly for risk-averse mandates where return forecasts are unreliable. The ESG integration and transaction cost analysis further enhance practical relevance.

Moderate complexity. Stage 1 (volatility-normalized selection) is O(N) via partial sorting. Stage 2 (Ω-metric projection) requires solving a K-dimensional convex QP via SLSQP or active-set methods, O(K^3) worst case. The algorithm is parameter-free for CASP-Basic (only needs covariance matrix). RA-CASP adds two hyperparameters (λ, γ) requiring grid search. Integration as a drop-in replacement for Euclidean projection in existing metaheuristic frameworks is straightforward. Main implementation challenges: numerical stability of covariance matrix (regularization needed), efficient QP solving for large K, and proper handling of box constraints on the simplex.

The paper provides detailed algorithm pseudocode (Algorithm 1), parameter settings (K=15, bounds [2%,15%], risk-free rate 4.5%), data sources (Yahoo Finance, 100 S&P 500 constituents, 2020-2024), and statistical tests. However, no GitHub repository is mentioned, and the ESG score construction methodology (sector-level proxies) is somewhat ad hoc. The covariance matrix estimation uses 10% shrinkage toward identity. Full reproducibility would require access to the exact data processing pipeline and MOGWO implementation details.