Systemic Risk in DeFi: A Network-Based Fragility Analysis of TVL Dynamics

Battle Count: 173

The paper is primarily focused on systemic risk monitoring and macroprudential oversight rather than direct trading signal generation. However, the CFI as a slow-moving structural state variable could inform regime detection, portfolio rebalancing during high-fragility periods, and risk budgeting in DeFi-focused strategies. The RCS identifies structurally important protocol categories that may warrant position limits or hedging. The predictive relationship between CFI and future TVL volatility (significant at 7-30 day horizons) has some trading relevance for liquidity risk management.

Core pipeline involves: (1) TVL data collection and cleaning from DeFiLlama API, (2) log return construction with winsorization, (3) Ledoit-Wolf shrinkage correlation estimation over rolling windows, (4) computation of four network metrics, (5) PCA aggregation into CFI, (6) counterfactual node removal for RCS, (7) attack tests. Each step uses well-established statistical methods. Main complexity lies in data preprocessing (anomaly detection, category mapping) and ensuring numerical stability. No deep learning or complex optimization required.

Data sourced from public DeFiLlama API. Code and data available in an anonymous GitHub repository. Parameters (window size, thresholds, shrinkage settings) are clearly specified. Rolling window methodology is standard. However, the anonymous repository may limit long-term reproducibility.