Paper: arXiv 2512.19986
Authors: Nikolaos Iliopoulos
Abstract
Metaheuristic algorithms for cardinality-constrained portfolio optimization require repair operators to map infeasible candidates onto the feasible region. Standard Euclidean projection treats assets as independent and can ignore the covariance structure that governs portfolio risk, potentially producing less diversified portfolios. This paper introduces Covariance-Aware Simplex Projection (CASP), a two-stage repair operator that (i) selects a target number of assets using volatility-normalized scores and (ii) projects the candidate weights using a covariance-aware geometry aligned with tracking-error risk. This provides a portfolio-theoretic foundation for using a covariance-induced distance in repair operators. On S&P 500 data (2020-2024), CASP-Basic delivers materially lower portfolio variance than standard Euclidean repair without relying on return estimates, with improvements that are robust across assets and statistically significant. Ablation results indicate that volatility-normalized selection drives most of the variance reduction, while the covariance-aware projection provides an additional, consistent improvement. We further show that optional return-aware extensions can improve Sharpe ratios, and out-of-sample tests confirm that gains transfer to realized performance. CASP integrates as a drop-in replacement for Euclidean projection in metaheuristic portfolio optimizers.
Complexity vs Empirical Score
- Math Complexity: 7.0/10
- Empirical Rigor: 8.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper introduces a novel, covariance-aware projection operator with explicit mathematical derivation and financial interpretation (tracking error), demonstrating high mathematical sophistication. It is backed by extensive empirical validation on S&P 500 data (2020-2024), including statistical significance tests, ablation studies, out-of-sample tests, and clear integration guidance for practical implementation.
Research Flowchart
flowchart TD
A["Research Goal"] --> B["Data & Inputs"]
B --> C["Standard Euclidean Projection<br>Baseline Repair Operator"]
B --> D["CASP Methodology<br>Two-Stage Repair Operator"]
D --> E["Covariance-Aware<br>Projection"]
D --> F["Volatility-Normalized<br>Selection"]
C & E --> G["Outcomes & Findings"]
F --> G
subgraph Key Findings
G --> H["Lower Portfolio Variance<br>Without Return Estimates"]
G --> I["Volatility Selection<br>Drives Main Reduction"]
G --> J["Robust & Significant<br>Improvements"]
end
style A fill:#e1f5fe
style D fill:#f3e5f5
style H fill:#e8f5e8
style I fill:#e8f5e8
style J fill:#e8f5e8
Evidence Card
What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card
Proposes
The paper proposes Covariance-Aware Simplex Projection (CASP), a two‑stage repair operator that selects assets using volatility‑normalized scores and projects weights with a covariance‑aware geometry aligned to tracking‑error risk.
Source sentence
This paper introduces Covariance-Aware Simplex Projection (CASP), a two-stage repair operator that (i) selects a target number of assets using volatility-normalized scores and (ii) projects the candidate weights using a covariance-aware geometry aligned with tracking-error risk.
Assumes 6
- Transaction costs are 10 basis points per unit of turnover. execution
Source sentence
Transaction Costs. We assume 10 basis points per unit of turnover, consistent with institutional trading costs for liquid large-cap equities.
- Covariance matrix is estimated from daily returns and shrunk 10% toward the identity matrix. model
Source sentence
The covariance matrix is estimated from daily returns and stabilized via a 10% shrinkage toward the identity matrix (in the spirit of Ledoit–Wolf [13]) to ensure numerical robustness.
- All parameter estimation uses only the training period data. data
Source sentence
All parameter estimation (expected returns, covariance matrix) uses training data only.
- The framework operates in a single‑period setting. model
Source sentence
The current framework is single-period; extension to multi-period optimization ...
Show 2 more
- Expected returns are annualized mean log returns. data
Source sentence
Expected returns are computed as annualized mean log returns.
- ESG quality scores are constructed from governance risk signals and sector‑level environmental/social proxies. data
Source sentence
... and (ii) sector-level environmental/social proxies reflecting typical sector ESG profiles.
Data
2020 – 2024 · S&P 500 stocks · equities
- S&P 500 daily adjusted closing prices market-prices public
Source sentence
Daily adjusted closing prices for 100 S&P 500 constituents are obtained via Yahoo Finance from January 2, 2020 to November 29, 2024 (1,237 trading days).
Yahoo Finance Refinitiv / LSEG (Datastream, Eikon, Tick History)
Costs
Modeled
10 basis points per unit of turnover, applied as a flat cost in performance calculations.
Source sentence
Transaction Costs. We assume 10 basis points per unit of turnover, consistent with institutional trading costs for liquid large-cap equities.
Finds 4
- CASP‑Basic reduces portfolio variance by 15.7% relative to Euclidean repair, with strong significance (p < 10⁻⁵⁴). performance
Source sentence
CASP‑Basic achieves 15.7% variance reduction without return information. Relative to VolNorm+Euc, the Ω‑metric projection geometry contributes an additional 1.4% incremental variance reduction
- Volatility‑normalized selection alone (VolNorm+Euc) yields a 14.3% variance reduction versus Euclidean baseline. performance
Source sentence
VolNorm+Euc achieves 14.3% variance reduction, showing that volatility-normalized selection alone is a strong baseline even with Euclidean projection.
- RA‑CASP attains a 32.0% higher best Sharpe ratio than Euclidean (1.137 vs 0.861), statistically significant (p = 0.0001). performance
Source sentence
RA‑CASP achieves 32.0% higher best Sharpe ratio than Euclidean (1.137 vs 0.861
- RA‑CASP shows slightly lower turnover (0.548 vs 0.565) and higher net Sharpe after transaction‑cost adjustment (0.552 vs 0.261). performance
Source sentence
RA‑CASP achieves slightly lower turnover (0.548 vs 0.565) and, due to higher gross Sharpe, remains favorable under the cost proxy (0.552 vs 0.261).
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 64 related papers). Coverage is partial, so this is neither replication nor refutation.
Code
No repository by the authors found in the paper.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.
Cited References (35)
- The Effects of ESG Scores and ESG Momentum on Stock Returns and Volatility: Evidence from U.S. Markets (2025) — link
- Intelligent Many-objective portfolio optimization using hybrid deep learning and evolutionary algorithm approach for advanced decision-making (2025) — link
- Deep reinforcement learning assisted surrogate model management for expensive constrained multi-objective optimization (2025) — link
- Non-linear shrinkage of the price return covariance matrix is far from optimal for portfolio optimisation (2021) — link
- Responsible investing: The ESG-efficient frontier (2020) — link
- Investment and Asset Pricing with ESG Disagreement (2020) — link
- Risk Return Trade-Off in Relaxed Risk Parity Portfolio Optimization (2020) — link
- Analytical nonlinear shrinkage of large-dimensional covariance matrices (2020) — link
and 27 more
- Bilevel Cutting-plane Algorithm for Solving Cardinality-constrained Mean-CVaR Portfolio Optimization Problems (2020)
- Why has the equal weight portfolio underperformed and what can we do about it? (2020) — link
- Aggregate Confusion: The Divergence of ESG Ratings (2020) — link
- Metaheuristics for rich portfolio optimisation and risk management: Current state and future trends (2019) — link
- A comprehensive review of deterministic models and applications for mean-variance portfolio optimization (2019) — link
- Reprint of: Mahalanobis, P.C. (1936) “On the Generalised Distance in Statistics.” (2018) — link
- Mean-variance efficiency of optimal power and logarithmic utility portfolios (2018) — link
- Portfolio selection problems with Markowitz’s mean–variance framework: a review of literature (2017) — link
- A new efficiently encoded multiobjective algorithm for the solution of the cardinality constrained portfolio optimization problem (2016) — link
- Optimally chosen small portfolios are better than large ones (2016)
- Multi-objective grey wolf optimizer: A novel algorithm for multi-criterion optimization (2016) — link
- Grey Wolf Optimizer (2014) — link
- A fast and elitist multiobjective genetic algorithm: NSGA-II (2002) — link
- Heuristics for cardinality constrained portfolio optimisation (2000) — link
- Portfolio Selection (1971) — link
- CAPITAL ASSET PRICES: A THEORY OF MARKET EQUILIBRIUM UNDER CONDITIONS OF RISK* (1964) — link
- Multi-period mean-variance fuzzy portfolio optimization model with transaction costs (2018) — link
- A Fast Projection onto the Simplex and the l 1 Ball (2015)
- Fundamental Portfolio Construction Based on Mahalanobis Distance (2013) — link
- Global Sustainable Investment Alliance (2024)
- Selection by individual Sharpe ratios, Euclidean projection
- CASP-Basic: Volatility-normalized selection, Ω -metric projection
- A.2 Ω -metric Projection Implementation For the Ω -metric projection, we solve the 𝐾 -dimensional
- Return-aware selection, Ω -metric projection (no return term)
- Convergence is guaranteed within 𝑂 ( log ( 1 / 𝜖 )) iterations for tolerance 𝜖
- Centre De Referència En Economia Analítica Barcelona Economics Working Paper Series Working Paper Nº 92 Honey, I Shrunk the Sample Covariance Matrix Honey, I Shrunk the Sample Covariance Matrix
- Selection by | 𝑧 𝑖 |/ Ω 𝑖𝑖 , Euclidean projection (selection-only baseline)
0 of 35 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.