Compensation Proposal: Distribute COMP to Affected Users in the DAI Liquidations

Proposals
89

I took a stab at modeling out this strategy.

The way I see it there are 4 questions we should reach agreement on in regards to compensation.

  1. What was the actual damages in raw numbers
  2. What percentage should be compensated (The proposal that @kybx86 suggested was intended to cover the liquidation incentive of 8%)
  3. What modifiers should be applied based on user behavior (@wario is suggesting modulating by how recursive the borrowing scheme is)
  4. How to distribute it to affected users. (@kybx86 proposed using COMP, other proposals involve DAI reserves. The most accurate to the true damages of the event would be to use the asset that was seized in false liquidations)

This is a strawman model that is intended to get the methodology for a direct computation of damages correct. It uses input values derived from @wario’s data, the minimum USD values for assets on the day of the exploit. A real computation would use on-chain data, but should be in the same ballpark. This was just to determine a good methodology and if we move forward we can work on using accurate data to determine the final raw damages result.

The model divides users into two classes. Those that would not have been liquidated using correct DAI prices and those that still would have. The damages for those that would have not been liquidated is the extra collateral that was taken due to the incorrect DAI price plus the liquidation penalty from the false liquidation. For those that would have been liquidated anyway, the damages are simply the delta between the liquidation penalty from the liquidation that would have occurred and the liquidation penalty that liquidation that did occur. Let me know if this sounds correct.

script: https://pastebin.com/DcsXXxcJ
output: https://pastebin.com/y9s6997E

The total raw damages from false liquidations comes to around $7.5M direct loss of user value. This model also takes into account DAI-DAI liquidations, which results in less direct damages for the “looper-whales”, which was an objection to the previous proposal, so that worked out nicely.

The counterfactual of who would have been liquidated is the trickiest part of this model. Using @wario 's data and coingecko prices, there are only 7 users that would have been liquidated if DAI prices were correct and only one had > 5 figures liquidated. Since this real computation of this counterfactual requires assuming no user transactions, then it may be fairer to ignore it as it will have relatively minimal impact. This is something to decide if we move forward with it.

Once we have a good method for computing raw damages, we can then decide, the other points (%, modifiers, distribution method).

Also per the discussion above, even if there was no manipulation (which seems extremely unlikely and can only really be determined through the legal system) then there was still an oracle failure leading to false liquidations (otherwise, why bother fixing it if its not broken?). Compensation is part of the fix.

6 Likes