(Disclosure, I am the founder of B.Protocol)
Recently our community introduced a novel automated market maker algorithm that can handle liquidation of big amounts, and has automatic rebalance procedure.
We are proposing that compound protocol would deposit funds (potentially from the 2.8% seize fee) into a deployed contract of the system, which will offer all supported compound debts with 4% discount (over compound’s oracle).
Further, the contract will only support liquidations over compound.
After each liquidation, an automatic rebalance process will start.
Such setup will guarantee that if X amount of token C is deposited, then liquidations of up to X amount of C token are guaranteed to succeed, even if all of defi liquidity is temporarily drained.
Offering liquidity with 4% discount give some room to incentivise callers to the liquidation function. And fee that is being charged during the rebalance process.
The backstop automatic market maker (B.AMM) serves as a buffer of liquidity, and offer liquidity with premium (e.g., 4%) over market price (where the market price is determined according to chainlink price feed).
Once a liquidity is taken, the buffer position becomes imbalanced, and it offer the imbalanced inventory for sale, with discount over market price.
Where the market price is again determined by chainlink oracle, and the the actual discount is determined by a smart contract formula, which takes into account the inventory imbalance.
The authors of the whitepaper performed a simulation on how it would perform. The benchmark was set to be Binance futures USD-ETH market. Which had $1B of liquidations that month.
The figure below shows liquidations where the y axis units are 100 million USD, and the x axis is time in 30 minutes resolution
The whitepaper assumes certain delay in the DeFi liquidity recovery time, and shows how much will be needed to cover such volume of liquidations
Further, it assumes 5% premium, and shows that overall, in the month of April 2021, this strategy could have been profitable
A preliminary implementation (for a different platform) is available here.