Asset Heterogeneity in DeFi Oracle Vulnerabilities

I test whether information latency in decentralized lending protocols generates regime-dependent liquidation outcomes, using transaction-level data on 16,639 Aave V3 Ethereum mainnet liquidations of WETH, WBTC, LINK, and AAVE collateral between January 2023 and April 2026. The pre-registered binary specification produces a null result; the continuous-volatility refinement yields a small positive interaction opposite to the cascade-burnout prediction. Asset-by-asset estimates reveal cross-asset heterogeneity that the pooled regressions obscure: a joint Wald test rejects cross-asset homogeneity at the 5 percent level in the full sample (chi-squared with 3 df = 8.05, p = 0.045), with WBTC and LINK contributing the largest negative coefficients in the direction predicted by attenuation.

Regime-conditional refinements that would identify the underlying mechanism are statistically fragile: the heterogeneity does not survive a no-stress restriction, and the apparent strengthening in stress-only sub-samples is mechanically driven by within-block simultaneity in a single cascade window. The paper's two contributions are methodological: documenting that pooled cross-asset DeFi tests can systematically mask economically meaningful idiosyncratic risk (the masking effect), and documenting that regime-conditional inference on liquidation cascades is fragile to plausible variation in stress-window definitions and to within-block clustering in the cascade observations themselves. Both observations have direct implications for how empirical claims about DeFi cascade dynamics should be evaluated.