# Per-Asset Liquidation Thresholds Underpredict Co-Liquidation by 301×

**URL:** <https://governance.aave.com/t/per-asset-liquidation-thresholds-underpredict-co-liquidation-by-301x/26131>\
**Category:** Risk\
**Created:** [November 9, 2026, 12:43am UTC](https://governance.aave.com/t/per-asset-liquidation-thresholds-underpredict-co-liquidation-by-301x/26131 "2026-11-09T00:43:56Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![oraclethaacat](https://avatars.discourse-cdn.com/v4/letter/o/5f8ce5/32.png) [@oraclethaacat](https://governance.aave.com/u/oraclethaacat)\
**Post date:** [November 9, 2026, 12:43am UTC](https://governance.aave.com/t/per-asset-liquidation-thresholds-underpredict-co-liquidation-by-301x/26131/1 "2026-11-09T00:43:57Z")

</div>

Aave V3 computes liquidation risk per asset. Each collateral asset has its own maximum LTV and liquidation threshold, set independently. The health factor is a weighted average. There is no term for correlation between assets.

I tested whether this independence assumption holds, using Aave’s own published parameters and public daily price data from Yahoo Finance. Nine major collateral assets, 2,123 trading days (September 2020 to October 2026).

Method: For each asset, mark the days where a position at maximum LTV would cross the liquidation threshold. Then apply a column-shuffle permutation test (B = 2,000): shuffle each asset’s column independently to destroy cross-asset dependence while preserving each asset’s marginal rate. Compare the real count of simultaneous liquidation triggers to the shuffled distribution.

Results:

- 4 assets hit same day: 57 real days vs 0.19 expected. 301× underpriced.
- 5 assets: 31 real days vs 0.004 expected. 6,889× underpriced.
- 6+ assets: 21 real days vs 0 in 2,000 shuffles. The independent model never produces this.

All 36 pairwise asset combinations are significant at p \< 0.001. Not one pair behaves independently under Aave’s own thresholds.

The worst days are the ones you’d expect: 2021-05-19 (all 9 assets), 2022-05-11 (Terra/Luna), 2022-11-08 (FTX).

The test takes 30 seconds and uses no API keys. The full methodology, results, and test script are at [The Short Version — The Rogue Wave](https://roguewavebook.com/short) — sections 17 and 18.

What needs to change: The per-asset threshold model must be replaced with a joint probability model that accounts for tail correlation across collateral assets. The mathematics exists — it was developed after 2008 specifically because the same independence assumption failed in CDO pricing. Portfolio-level liquidation thresholds must tighten as correlated assets are added, not average.

This is not a code bug. The code correctly implements the specified thresholds. The thresholds themselves are structurally inadequate for the correlation that exists in the data.