Thread by @flashfish0x

Before my current life I was deep into MEV (back before it was called MEV). I feel it was long enough ago that the omertà has expired.

So here is the first of some threads on ancient alpha. MEV from > a year ago.

First up - Compound liquidations

On Compound if you take one step over the line you can be liquidated for 50% of your entire collateral at a 5% discount.

For example. If you have $200k collateral you can be liquidated for $100k at the price of $95k.  $5k profit minus costs to the liquidator

The transaction is fairly simple and can be done with only gas costs. Flashloan the debt token, liquidate the target and swap their collateral to pay off the flashloan.

Your account can be liquidated when the oracle says you have too much debt to be covered by your collateral.

Prior to August 2020, Compound's oracle was centralised and updated by a privileged tx. As a liquidator we would monitor the pending transaction mempool looking for the oracle update.

When we see it we would liquidate any account which was profitable.

Now onto the fun stuff. Only the first person to liquidate after the oracle update wins.

At the time miners would sort same price txs randomly. If you sent a liquidation at the same gas price as the oracle update there was a 50% chance of being placed after the update

Now we're into a maths problem with some game theory.

Each liquidator would spam txs at the same gas price as the oracle update. Hoping to win the liquidation without spending too much on failed txs.

Below is a screenshot showing some of this spam. Two liquidators are competing, spamming cheap-to-fail txs with green winning the prize.

So how much do you choose to spam?

Well we know the revenue from the liquidation, but not the odds of winning as that depends on others' spam

The block isn't finalised until it is mined. So every moment that the block is not mined, more spam enters the mempool. Changing the odds of the spam already there.

You cannot cancel. Once you commit your spam the cost is sunk.

It is in everyone's interest to collaborate - the less spam the less of the revenue goes to miners - but how?

It was a constant game. Our algo would monitor the mempool deciding what to do as spam came in.

Then after it was over we'd analyse and adjust our tactic for next time

On August 11th 2020 Geth released v1.9.19. Changing the game. Same priced txs would now be ordered by arrival times.

Meaning that if the block was mined by a 1.9.19 geth miner the first tx to arrive after the oracle update would win

This makes the game theory even more complicated. Suddenly geolocation and speed of getting txs out becomes much more important. While also playing the old game.

10 days later 80% of miners were running geth with 20% of them on 1.9.19.

On August 17th everything changed again. Compound moved to a more decentralised oracle system. Initially relying on an offchain Coinbase oracle.

This was the end of spam. Instead you monitor coinbase and combine the on-chain oracle price update with your liquidations transaction

The change meant that instead of sending lots of transactions. You would send one tx. Then monitor the mempool to see if anyone tried to liquidate with a higher gas price than you, and if they do you outbid them.

I call this gas racing and it formed the foundation of a lot of MEV competition before private mempools.

You each consistently outbid each other until someone bails (replaces their liquidation tx with a cheap transaction) or the block gets mined and a winner is declared.

To be competitive you need the simplest and cheapest liquidation transaction possible. No flash loans. No swaps.

The goal was be able to race to a higher price than others while squeezing out some profit.

It wasn't long after this that I stopped competing for liquidations on Compound. There were greener pastures elsewhere and the competition plus capital requirements were too high.