Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Stablecoin Design with Adversarial-Robust Multi-Agent Systems via Trust-Weighted Signal Aggregation

Shengwei You, Aditya Joshi, Andrey Kuehlkamp, Jarek Nabrzyski

TL;DR

The paper targets tail-risk vulnerabilities in algorithmic stablecoins by addressing the 2020 depeg and Black Thursday failures. It introduces MVF-Composer, a trust-weighted Mean-Variance Frontier reserve controller augmented with a Stress Harness that uses adversarial multi-agent simulations to generate crisis scenarios, and a trust layer that down-weights signals from potentially malicious agents. Through 1,200 randomized stress scenarios, it demonstrates 57% reductions in peak peg deviation and 3.1× faster recovery under crises, with the trust component contributing about 23–72% robustness improvements depending on metrics, and a 60–80% reduction in adversarial influence. The approach includes stress-augmented covariance estimation, a modular and auditable trust mechanism, and an end-to-end pipeline executable on commodity hardware without on-chain oracles beyond standard price feeds, offering a reproducible framework for stress-testing DeFi reserve policies. The work generalizes beyond a single stablecoin and provides practical guidelines for deployment, ethics, and governance in tail-risk aware reserve design.

Abstract

Algorithmic stablecoins promise decentralized monetary stability by maintaining a target peg through programmatic reserve management. Yet, their reserve controllers remain vulnerable to regime-blind optimization, calibrating risk parameters on fair-weather data while ignoring tail events that precipitate cascading failures. The March 2020 Black Thursday collapse, wherein MakerDAO's collateral auctions yielded $8.3M in losses and a 15% peg deviation, exposed a critical gap: existing models like SAS systematically omit extreme volatility regimes from covariance estimates, producing allocations optimal in expectation but catastrophic under adversarial stress. We present MVF-Composer, a trust-weighted Mean-Variance Frontier reserve controller incorporating a novel Stress Harness for risk-state estimation. Our key insight is deploying multi-agent simulations as adversarial stress-testers: heterogeneous agents (traders, liquidity providers, attackers) execute protocol actions under crisis scenarios, exposing reserve vulnerabilities before they manifest on-chain. We formalize a trust-scoring mechanism T: A -> [0,1] that down-weights signals from agents exhibiting manipulative behavior, ensuring the risk-state estimator remains robust to signal injection and Sybil attacks. Across 1,200 randomized scenarios with injected Black-Swan shocks (10% collateral drawdown, 50% sentiment collapse, coordinated redemption attacks), MVF-Composer reduces peak peg deviation by 57% and mean recovery time by 3.1x relative to SAS baselines. Ablation studies confirm the trust layer accounts for 23% of stability gains under adversarial conditions, achieving 72% adversarial agent detection. Our system runs on commodity hardware, requires no on-chain oracles beyond standard price feeds, and provides a reproducible framework for stress-testing DeFi reserve policies.

Stablecoin Design with Adversarial-Robust Multi-Agent Systems via Trust-Weighted Signal Aggregation

TL;DR

The paper targets tail-risk vulnerabilities in algorithmic stablecoins by addressing the 2020 depeg and Black Thursday failures. It introduces MVF-Composer, a trust-weighted Mean-Variance Frontier reserve controller augmented with a Stress Harness that uses adversarial multi-agent simulations to generate crisis scenarios, and a trust layer that down-weights signals from potentially malicious agents. Through 1,200 randomized stress scenarios, it demonstrates 57% reductions in peak peg deviation and 3.1× faster recovery under crises, with the trust component contributing about 23–72% robustness improvements depending on metrics, and a 60–80% reduction in adversarial influence. The approach includes stress-augmented covariance estimation, a modular and auditable trust mechanism, and an end-to-end pipeline executable on commodity hardware without on-chain oracles beyond standard price feeds, offering a reproducible framework for stress-testing DeFi reserve policies. The work generalizes beyond a single stablecoin and provides practical guidelines for deployment, ethics, and governance in tail-risk aware reserve design.

Abstract

Algorithmic stablecoins promise decentralized monetary stability by maintaining a target peg through programmatic reserve management. Yet, their reserve controllers remain vulnerable to regime-blind optimization, calibrating risk parameters on fair-weather data while ignoring tail events that precipitate cascading failures. The March 2020 Black Thursday collapse, wherein MakerDAO's collateral auctions yielded $8.3M in losses and a 15% peg deviation, exposed a critical gap: existing models like SAS systematically omit extreme volatility regimes from covariance estimates, producing allocations optimal in expectation but catastrophic under adversarial stress. We present MVF-Composer, a trust-weighted Mean-Variance Frontier reserve controller incorporating a novel Stress Harness for risk-state estimation. Our key insight is deploying multi-agent simulations as adversarial stress-testers: heterogeneous agents (traders, liquidity providers, attackers) execute protocol actions under crisis scenarios, exposing reserve vulnerabilities before they manifest on-chain. We formalize a trust-scoring mechanism T: A -> [0,1] that down-weights signals from agents exhibiting manipulative behavior, ensuring the risk-state estimator remains robust to signal injection and Sybil attacks. Across 1,200 randomized scenarios with injected Black-Swan shocks (10% collateral drawdown, 50% sentiment collapse, coordinated redemption attacks), MVF-Composer reduces peak peg deviation by 57% and mean recovery time by 3.1x relative to SAS baselines. Ablation studies confirm the trust layer accounts for 23% of stability gains under adversarial conditions, achieving 72% adversarial agent detection. Our system runs on commodity hardware, requires no on-chain oracles beyond standard price feeds, and provides a reproducible framework for stress-testing DeFi reserve policies.
Paper Structure (98 sections, 18 theorems, 44 equations, 11 figures, 10 tables, 1 algorithm)

This paper contains 98 sections, 18 theorems, 44 equations, 11 figures, 10 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Trust-Weighted Signal Aggregation
  3. Adversarial Multi-Agent Simulation
  4. Application: Stablecoin Reserve Control
  5. Contributions
  6. Threat Model and Security Objectives
  7. System Model
  8. Adversarial Threat Model
  9. Adversary Capabilities.
  10. Adversary Limitations.
  11. Attack Taxonomy
  12. A1: Sybil Attacks
  13. A2: Signal Injection Attacks
  14. A3: Coordination Attacks
  15. A4: Timing Exploitation
  16. ...and 83 more sections

Key Result

proposition 1

Let $\mathbf{w}^{*}_{\textsc{SAS}{}}$ be the optimal weights under SAS with covariance $\boldsymbol{\Sigma}^{\mathrm{calm}}$. Define the fragility ratio $\kappa(\boldsymbol{\Sigma}^{\mathrm{calm}}, \boldsymbol{\Sigma}^{\mathrm{stress}}) \triangleq \lambda_{\max}(\boldsymbol{\Sigma}^{\mathrm{stress}} Intuitively, $\kappa$ bounds how severely realized portfolio risk can exceed model expectations whe

Figures (11)

  • Figure 1: Stress Harness Workflow and Trust Risk Vectors. Agents process market state and news, generating actions and psychology logs. The Trust Module aggregates these into a risk-weighted state, defending against adversarial signal injection and coordination attacks.
  • Figure 2: Agent Behavior and Risk Mapping. Specific behaviors for each archetype are mapped to potential system risks. The Trust Defense layer identifies anomalies such as coordination ($f_3$) and destabilization timing ($f_4$).
  • Figure 3: Rolling 20-day annualized volatility for stablecoin returns. The shaded region (2020) is excluded from SAS's covariance estimation. Peak volatility during the COVID-19 crash (March 2020) exceeded 30%, an order of magnitude higher than post-2021 levels. By omitting this regime, SAS calibrates to an unrealistically calm market.
  • Figure 4: Regime mismatch between calm-period and stress-period covariance matrices. The trace ratio $\mathrm{tr}(\boldsymbol{\Sigma}^{\mathrm{stress}})/\mathrm{tr}(\boldsymbol{\Sigma}^{\mathrm{calm}}) \approx 7.17\times$ implies that portfolios optimized for calm periods experience realized variance far exceeding model predictions. Left: SAS paper period (trace $= 0.0021$). Middle: Extended period including 2020 (trace $= 0.0151$). Right: Difference matrix shows "what SAS misses" -- the "fragility zone" where calm-period optimization underestimates total risk by over 600%. The 7.17 factor underestimation of total variance directly validates the theoretical fragility result (\ref{['prop:sas_fragility']}).
  • Figure 5: Optimal portfolio weights under each covariance regime. The SAS period (blue) assigns 60% to USDT based on its calm-period variance. When 2020 data is included (red), the optimizer diversifies away from assets that exhibited high crisis-period volatility.
  • ...and 6 more figures

Theorems & Definitions (26)

  • definition 1: Adversarial Influence Reduction
  • definition 2: Adversarial Detection Rate
  • definition 3: Robust Multi-Agent Aggregation
  • proposition 1: SAS Fragility
  • proposition 2: Bounded Adversarial Influence
  • proposition 3: Robustness Under Coordination
  • definition 4: Sigmoid Function
  • definition 5: Positive Semi-Definite Matrix
  • definition 6: Probability Simplex
  • theorem 1: Trust Score Boundedness and Monotonicity
  • ...and 16 more