Arbitrage & Liquidation MEV

The earlier lessons painted MEV mostly as a mugging: sandwiches, front-running, the searcher who profits by hurting you. Fair — but it’s only half the story, and the smaller half. Most extracted value on Ethereum doesn’t come from preying on users at all. It comes from two activities the system needs: keeping prices consistent across venues, and keeping lending protocols from going insolvent. This lesson is about that load-bearing MEV — why it’s a feature, how the arithmetic of the profit works, and the uncomfortable bit where “benign” extraction starts behaving badly anyway.

You already know AMMs and DeFi lending — health factors, liquidations, the lot. So we go straight to the mechanics and the money.

DEX arbitrage: the price-equalizing machine

Analogy. Connect two water tanks at different levels with a pipe and water rushes from high to low until the levels match. DEX arbitrage is that pipe for prices. When the same asset sits cheaper on one pool than another, capital flows through trades from the cheap venue to the dear one until the prices level off. The searcher is just the pipe — and they get to keep a cup of the water for laying it.

Definition. DEX arbitrage is the act of buying an asset where it’s underpriced and simultaneously selling it where it’s overpriced, capturing the price difference. On a blockchain this happens in a single atomic transaction, so the buy and the sell either both land or both revert.

Worked example. Pool A prices ETH at $2000; pool B prices it at $2040. Ignore fees and slippage for the clean version: a searcher buys 1 ETH on A for $2000 and sells it on B for $2040.

$\text{gross profit} = \$2040 - \$2000 = \$40 \text{ per ETH}$

But the trades themselves move the prices. Buying ETH on A removes ETH from A’s reserves and adds USDC, walking A’s spot price up from $2000. Selling ETH on B adds ETH to B’s reserves, walking B’s price down from $2040. The searcher keeps buying-and-selling until the two prices meet somewhere in the middle — say $2020. At that point the spread is gone and there’s no profit left to take, which is exactly why the searcher stops. The leftover, after the prices have converged, is their fee for doing the system’s housekeeping.

The deep point: the profit and the public good are the same action. You cannot capture the $40 without simultaneously dragging the prices together. That’s why this is the canonical “benign” MEV — the extraction is the price correction. A world with no arbitrageurs is a world where pool prices drift apart and stay wrong.

When it matters

DEX arbitrage matters most when prices fragment — many pools for the same pair, lots of independent trading, volatile markets. After every large swap, after every new block, small misalignments open up and arbitrageurs immediately close them. Without this constant correction, the price you see on any single pool would be unreliable, and every protocol that reads an on-chain price (lending markets, derivatives, oracles) would inherit the error.

Atomic arbitrage and why blockchains make it special

Analogy. In tradfi, arbitrage means holding something for a moment — buy here, sprint to the other exchange, hope the price hasn’t moved by the time you sell. That gap is inventory risk. Atomic arbitrage is teleportation: the buy and sell happen in the same indivisible instant, so there’s no in-between to be exposed in.

Definition. Atomic arbitrage bundles every leg of the trade into one transaction that the EVM executes as an all-or-nothing unit. If the sell leg would not clear at a profit, the whole transaction reverts — the searcher pays gas for the failed attempt but never ends up holding the asset at a loss. No price-movement risk, no inventory risk, no overnight exposure. This is structurally impossible across two separate centralized exchanges, and it’s the single biggest reason on-chain arbitrage is so ruthlessly efficient.

It gets sharper with flash loans. A flash loan lets a searcher borrow a large sum within one transaction with no collateral, on the sole condition that it’s repaid before the transaction ends — and if it isn’t, the whole thing reverts as though it never happened. So the arbitrageur needs essentially no capital of their own: borrow millions, buy on A, sell on B, repay the loan, keep the spread, all atomically. Capital stops being the barrier to entry; speed and search quality become everything.

This is also where back-running ties back to Lesson 3. A large swap creates an arbitrage opportunity by knocking one pool’s price out of line. A searcher who lands a transaction immediately after that swap — back-running it — captures the resulting spread. Back-running a big trade is, mechanically, just atomic arbitrage triggered by someone else’s price impact. It’s why pending large swaps are so contested: the realignment behind them is free money for whoever gets there first.

CEX-DEX arbitrage: the non-atomic cousin

Analogy. Atomic arbitrage is teleportation; CEX-DEX arbitrage is a courier run. You buy on one side, then physically carry the goods to the other venue to sell — and during the carry, the market can move against you. The profit is real but so is the risk.

Definition. CEX-DEX arbitrage exploits price differences between a centralized exchange (Binance, Coinbase) and an on-chain pool. Crucially, it is not atomic: the on-chain leg and the off-chain leg live in different systems that cannot settle together in one indivisible step. The searcher executes one leg, then the other, and in between they are holding inventory and bearing real price risk on the leg that hasn’t closed yet.

Despite the risk, CEX-DEX arbitrage is one of the largest categories of MEV, because CEX prices update continuously while on-chain prices only update block-by-block — so the on-chain price is always a little stale relative to the CEX, and that staleness is a near-permanent edge for whoever can act on it fastest. It’s also a major source of toxic order flow to builders: a CEX-DEX arbitrageur who knows the “true” CEX price will pick off on-chain liquidity providers, and that flow is precisely what builders fight to capture (Lesson 6’s territory). The presence of inventory risk also means CEX-DEX arbitrage doesn’t fully self-extinguish the way atomic arbitrage does — it’s a feature and a steady tax on passive on-chain liquidity.

Liquidation MEV: the solvency machine

Analogy. A lending protocol is a pawnshop that lends against collateral. As long as your collateral comfortably covers your loan, all is well. But if the collateral’s value falls too far, the shop needs someone to step in fast, sell the collateral, and clear the bad loan before it goes underwater and the shop eats the loss. Liquidators are that someone — and the shop pays them a finder’s fee for showing up first.

Definition. When a borrower’s health factor drops below 1 (their collateral no longer adequately covers their debt), the position becomes liquidatable: anyone may repay part of the borrower’s debt and, in return, seize a chunk of the collateral at a discount. That discount is the liquidation bonus — the liquidator’s incentive. Searchers race to be the one who lands the liquidation, because the bonus is pure profit.

Worked example. A borrower has $10,000 of ETH collateral against a loan, and their health factor has just slipped below 1. The protocol offers a 8% liquidation bonus. A liquidator repays $5,000 of the borrower’s debt. In exchange, they’re entitled to seize collateral worth the repayment plus the bonus:

$\text{collateral seized} = \$5{,}000 \times (1 + 0.08) = \$5{,}400$

$\text{liquidator profit} = \$5{,}400 - \$5{,}000 = \$400$

The liquidator put up $5,000 to clear debt and walked away with $5,400 of ETH — a $400 gross profit (before gas and before selling the seized ETH, which itself carries price risk unless they hedge or do it atomically). The borrower eats the bonus as a penalty for letting their position go underwater. The protocol gets exactly what it wanted: a risky loan cleared before it became a bad debt the protocol itself would have to absorb.

This is necessary, benign MEV. If nobody liquidated underwater positions, lending protocols would accumulate bad debt and eventually become insolvent — depositors couldn’t be paid back. The liquidation bonus exists precisely to summon searchers to do this job promptly.

But the competition for that $400 is pure MEV. The moment a position crosses health factor < 1, a swarm of bots fights to be the liquidator. Often the trigger is an oracle price update — a transaction that posts a new, lower collateral price and pushes the borrower underwater. Searchers back-run that oracle update: the instant the new price lands, they fire the liquidation in the very next slot. The race is frequently a priority gas auction (bidding up fees to win ordering) or a sealed bid to a builder. The solvency service is a public good; the scramble to provide it is a textbook ordering game.

The blurry line: necessary, but still a scramble

Here’s the nuance, and it deserves to be stated without moralizing. Arbitrage and liquidations are good: they make prices accurate and keep protocols solvent. The system would be worse without them. And yet:

• The competition for these opportunities congests the chain. Dozens of bots submitting and reverting transactions to win one liquidation is real waste — failed transactions still consume blockspace and drive up fees for everyone.
• The profit mostly leaks upward. Through priority fees and direct bribes, the bulk of arbitrage and liquidation revenue ends up with builders and validators, not with users or even, in the long run, with the searchers (whose margins get competed to the bone).
• “Benign” extraction can still misbehave in practice. The same infrastructure that does honest back-running arbitrage can front-run and sandwich users. A liquidator racing for a bonus can grief rivals. The category label “benign” describes the function, not a guarantee about every actor’s behavior.

So the right mental model isn’t “good MEV vs. bad MEV” as two clean bins. It’s a spectrum: the function of arbitrage and liquidation is load-bearing and healthy, while the competition to perform it is a costly ordering game whose spoils flow to whoever controls block construction. That tension — useful work, extractive race, value captured upstream — is exactly the problem the next lesson’s machinery tries to manage.

We’ve now seen MEV in full: the predatory side (sandwiches, front-running) and the load-bearing side (arbitrage, liquidations), and the spectrum that connects them. We’ve also seen a recurring theme — the profit of all this activity keeps flowing upward, to whoever controls how blocks are built and ordered. That concentration is dangerous: if one party both decides ordering and captures the spoils, the chain’s neutrality is at stake. Lesson 6 — Proposer-Builder Separation, MEV-Boost, and mitigations — is about the machinery built to defuse exactly that: splitting block building from block proposing, auctioning blockspace transparently, and the design ideas (encrypted mempools, fair ordering, MEV redistribution) aimed at taming the race we just mapped.