Shengwei You · 游盛巍

Abstract — Algorithmic stablecoin reserve controllers are vulnerable to regime-blind optimization when they calibrate on fair-weather data while ignoring tail events. The March 2020 BLACK THURSDAY collapse ($8.3 M in losses, 15 % peg deviation) exposed a key vulnerability: models such as the Stable Aggregate Stablecoin (SAS) systematically omit extreme volatility from covariance estimates, producing allocations that are optimal in expectation but catastrophic under stress. To address this gap, we present MVF-Composer, a trust-weighted Mean-Variance Frontier controller incorporating a Stress Harness for risk-state estimation. The Stress Harness drives multi-agent simulations that serve as adversarial stress-testers: heterogeneous agents execute protocol actions under crisis scenarios, exposing vulnerabilities before they manifest on-chain. A complementary trust-scoring mechanism down-weights manipulative agents, ensuring robustness against adversarial influence.

Across 1,200 scenarios with black-swan shocks, MVF-Composer reduces peak peg deviation by 57 % and recovery time by 3.1× versus SAS. The trust layer accounts for 23 % of stability gains. The system runs on commodity hardware and provides a reproducible framework for stress-testing DeFi reserves against tail risks.

Index Terms—Algorithmic Stablecoins, Decentralized Finance, Agent-Based Simulation, Stress Testing, Risk Management, Mean-Variance Optimization