New post on https://t.co/j5UaKGaEf3!
The Origins of MEV: Systematic Attribution of Arbitrage Opportunity Creation at Scale
By:
- Andrei Seoev
- Dmitry Belousov
- Anastasiia Smirnova
- Ksenia Kurinova
- Aleksei Smirnov
- Denis Fedyanin
- Yury Yanovich
🔗 https://t.co/Zee4wPFBQd
Highlights:
- The paper formalizes the “MEV opportunity attribution” problem: for an executed atomic arbitrage transaction with profit Π, compute how much of Π is causally attributable to each prior candidate transaction (plus any profit attributable to the pre-block state).
- Deterministic EVM execution makes exact counterfactuals feasible: you can replay the chain and re-run an arbitrage after removing (or truncating before) specific candidate transactions to measure their marginal impact, rather than relying on statistical inference.
- Empirically, the “single-source” hypothesis holds strongly: 96.7% of atomic arbitrage opportunities are dominated by one source transaction (one transaction accounts for >70% of total positive Shapley value), with only 3.3% showing genuinely tied multi-source creation.
- Four attribution methods present clear speed/accuracy trade-offs at scale: simulation is the recommended primary method (high coverage, ~12.3 ms/event); coefficient-based attribution is a very fast screening heuristic (~0.8 ms/event) but less accurate; Shapley (often via Monte Carlo) serves as ground truth but is far slower (~2.1 s/event); bot-data provides partial, real-time external validation but limited coverage.
- MEV creation is highly concentrated and protocol-dependent: a small fraction of transactions create most opportunities; concentrated-liquidity AMMs dominate opportunity creation (e.g., Uniswap V3 58.0%, Algebra 29.6%, Uniswap V4 28.9% participation) despite Uniswap V2 having higher overall volume, and a new measurable efficiency metric shows ~1.6 successfully executed arbitrages per opportunity-creating transaction on average.
ELI5:
MEV (Maximal Extractable Value) is the money bots can make by quickly trading when prices on different exchanges/pools on the blockchain get out of sync. This paper asks a simple question: when a bot makes profit from an arbitrage trade, which earlier transaction caused the price mismatch that made the profit possible? Because blockchains are deterministic, the authors can “replay” history and also replay it with certain transactions removed to see what would have happened. They test several ways to assign credit/blame for creating arbitrage opportunities (fast heuristics, full simulations, and a ‘fair split’ method called Shapley values). They find that almost always, one earlier transaction is mainly responsible for creating the opportunity, and that opportunity creation is concentrated in a small set of protocols—especially concentrated-liquidity AMMs like Uniswap V3.
1/4 Since deployment, Homelander has internalized over $15k in MEV on one of our integrated DEXs.
Value that would have otherwise been captured by external searchers, now back where it belongs.
This is what early looks like, here's how it works🧵
🧵Everyone talks about who extracts MEV, but we asked who creates it.
We analyzed 1M+ blocks on Polygon to trace arb txs to their source and find who's responsible for creating MEV.
Our paper "The Origins of MEV" got featured in Flashbots MEV Letter 🤖 https://t.co/ruMJCxPTRi
@zerodev_@Eli5defi This is the right question and almost nobody is building the answer yet. When agents are routing, the MEV surface explodes because you have predictable, high-frequency flow that's trivially observable.
Just posted about our latest paper on the Flashbots forum!🤖
We set out to answer a question MEV research has largely left open: which transactions actually create arbitrage opportunities, and by how much?
We submitted two grant proposals to Flashbots Research Program around this. Both were declined, reviewers concluded that combinatorial explosion in real blocks would make the approach infeasible (We built it anyway😂)
We formalized MEV opportunity attribution as a causal inference problem, built four complementary attribution methods, and evaluated them across 1M+ blocks on Polygon.
Key findings:
- 96.7% of opportunities trace to a single source transaction
- Simulation-based attribution: 91.7% accuracy at 12.3ms per event
- Top 5% of protocols generate 73% of MEV opportunities, Uniswap V3 participates in 58% of opportunity-creating transactions
Full discussion on the Flashbots forum: https://t.co/6SWujsqjle
Paper on arXiv: https://t.co/ZOXTj4eWGW
A huge integration is coming! @QuickswapDEX community voted 100% to integrate Homelander🔥
More details and what the vote actually reflects about where DeFi is heading in our latest Medium post👇🏼
https://t.co/DYNpkZ13zc
Jared from subway just sandwiched @VitalikButerin: https://t.co/H9mP1CTa8y
this is the most Ethereum thing that has ever happened, MEV doesn't care who you are🥪