INTEGRAL ASSESSMENT OF BORROWER CREDITWORTHINESS IN DECENTRALIZED FINANCE BASED ON ON-CHAIN DATA: MODEL AND EMPIRICAL VALIDATION

Abstract

Introduction. Decentralized finance (DeFi) has created an on‑chain segment of collateralized lending where risk is commonly managed via collateral heuristics (LTV, health factor) and automated liquidations. Yet realized liquidation risk also depends on liquidation execution (market liquidity and transaction costs), observable borrower behavior, and network exposures arising from composability.

Purpose. To develop a formalized integral on‑chain creditworthiness model for DeFi borrowers and empirically test its ability to discriminate liquidation risk relative to the one‑factor LTV benchmark.

Methods. We construct a domain‑based indicator system (collateral quality, position buffer, behavioral discipline, network exposures, and liquidation conditions) and build an Integral Creditworthiness index (IC) using orientation and min–max normalization, entropy‑based information weights, and hybrid weighting. Discriminatory power is evaluated via logit models and ROC/AUC on a 30% test split within a controlled mock‑simulation design calibrated to Aave/Compound parameters (n=600).

Results. The IC index delivers superior out‑of‑sample liquidation discrimination compared with the LTV baseline (AUC=0.764 vs 0.683; Accuracy=0.778 vs 0.756) while preserving interpretability through a transparent domain structure. Scenario stress tests for collateral price shocks ΔP∈{−10%, −20%, −30%} show a monotonic increase in predicted liquidation probability from risk classes A to E, and sensitivity analysis confirms metric stability for alternative hybrid‑weighting parameters λ∈{0.5, 0.7, 0.9}.

Prospects. Future work should calibrate IC on real on‑chain panels with labeled liquidations across market regimes, incorporate oracle and MEV risk drivers, and extend the framework toward PD/LGD‑type parameterization for DeFi lending portfolios.