      Computational Economics · Springer · Submitted 2026-03-06
    

Robust Covariance Estimation for Portfolio
Optimization under Systematic Market Disruptions

Under Review Computational Economics PFSE · SSRE 25% Breakdown Point

Stefano Blando · Alessio Farcomeni · 2026

Key Results

Abstract

Publication

Venue Computational Economics

Publisher Springer

IF 2.2 (2024)

Submitted 2026-03-06

Status Under Review

Downloads

📄 Publication Page

Data & Methods

📈 S&P 500 top-100 (2015–2025)

⚡ Row-wise systematic contamination

🔬 Monte Carlo · n=1000 reps

💥 Flash crash · COVID · Contagion

🏛️ Institutional daily rebalancing

The Problem: Row-wise Contamination

⚡ Flash Crash (May 2010)

Automated trading generated coordinated extreme movements across all assets in minutes. Entire observation vectors become contaminated simultaneously.

🏦 Monetary Policy Shock

Surprise announcements reprice all interest-rate-sensitive assets coordinately. Normal correlation structures break down during the event.

🦠 COVID-19 (March 2020)

VIX spiked to 82.7 (all-time high). Coordinated extreme returns across all asset classes — classic row-wise contamination, sustained for weeks.

💡 PFSE Solution

Contamination propagates through common factors. Apply MCD in k-dimensional factor space (k=5 vs p=100) for 32× speedup + 25% breakdown point.

The Parallel Factor Space Estimator

PFSE exploits a structural insight: systematic market disruptions propagate through common factors, not idiosyncratic components. Concentrating robust estimation in k-dimensional factor space (k≪p) achieves 25% breakdown point with 15–50× speedup.

"When a flash crash occurs, the coordinated extreme movements operate through systematic risk channels — the common factors — while idiosyncratic components remain relatively unaffected. This enables targeted robust estimation exactly where contamination concentrates."

Simulate Algorithm

Replay the 5-phase estimation on S&P 500 data
(n=2520 obs · p=100 assets · ε=10%)

Complexity

PFSE: O(T·n·p·k + n·k²)
k=5, T≈12 ≪ p=100–1000
MCD: O(h·n·p²)
infeasible p>200

Properties

📐 Breakdown point 25%

⚖️ Affine equivariant

📊 Consistent (factor + elliptical)

✅ Positive definite guaranteed

PFSE Estimation Log

Five-Phase Algorithm

Algorithm Parameters

Symbol	Parameter	Value	Notes

Robust Covariance Estimation for Portfolio
Optimization under Systematic Market Disruptions

The Parallel Factor Space Estimator

Performance Explorer

S&P 500 Out-of-Sample Backtest

Stress Testing & Economic Value

Robust Covariance Estimation for PortfolioOptimization under Systematic Market Disruptions

The Parallel Factor Space Estimator

Performance Explorer

S&P 500 Out-of-Sample Backtest

Stress Testing & Economic Value

Robust Covariance Estimation for Portfolio
Optimization under Systematic Market Disruptions