Research Finds Aave’s Risk Models Underestimate Correlated Liquidations by 301 Times

MakeCryippreferred on

• A permutation-test analysis of 2,123 days of price data across nine Aave collateral assets found four-asset same-day liquidation triggers occur 301 times more often than independence assumptions predict
• Five-asset co-triggering events are underestimated by an even larger margin, roughly 6,889 times more often than expected under independent risk modeling
• The worst clustering days identified were May 11, 2022, during the Terra/Luna collapse, and November 8, 2022, during the FTX collapse

A detailed quantitative research post appeared on Aave’s governance forum on November 9, 2026, presenting evidence that the protocol’s current approach to setting liquidation thresholds systematically underestimates how often multiple collateral assets end up getting liquidated on the same day together. The analysis draws on 2,123 days of historical price data spanning nine different assets accepted as collateral on Aave, using a 2,000-shuffle permutation test to measure how often liquidation-triggering price moves across multiple assets coincide compared to what pure statistical independence between those assets would predict.

The headline finding is stark: four-asset same-day liquidation triggering events were observed 57 times in the historical data, compared to just 0.19 times that independence-based modeling would predict, a 301-fold underestimate. The gap widens further for five-asset co-triggering events, which the analysis finds occur roughly 6,889 times more often than an independence assumption would suggest, indicating that the correlation between different assets’ liquidation risk grows substantially more severe as more assets are considered simultaneously.

Aave’s current risk architecture, like that of most major DeFi lending protocols, sets liquidation thresholds on a per-asset basis, implicitly treating each collateral type’s liquidation risk as largely independent of the others. This research directly challenges that assumption, arguing that real-world market stress events tend to hit many assets at once rather than one at a time, meaning a protocol’s aggregate liquidation risk during a genuine crisis could be far higher than per-asset modeling alone would suggest.

The two days identified as the worst historical examples of this clustering effect are instructive: May 11, 2022, during the collapse of the Terra/Luna ecosystem, and November 8, 2022, during the unraveling of FTX, both widely recognized as among the most severe systemic shocks the broader crypto market has experienced. Both events triggered sharp, simultaneous price declines across a wide range of assets rather than an isolated move in just one.

The post’s author proposes replacing the current per-asset liquidation threshold approach with a joint, correlation-aware model that would more accurately price in the risk of simultaneous liquidations across multiple assets during genuine market-wide stress events, rather than treating each asset’s liquidation risk as if it existed in isolation from the others. The full research post is published on Aave’s own governance forum.