Skip to content
Crypto.world

Research

Open vs. Sealed: Auction Format Choice for Maximal Extractable Value

Block builders on Ethereum auction valuable transaction-ordering opportunities to competing "searchers," and these auctions move tens of millions of dollars per year.

Open vs. Sealed: Auction Format Choice for Maximal Extractable Value
arXiv
2603.16333
Published
2026-03-17
Authors
Aleksei Adadurov, Sergey Barseghyan, Anton Chtepine, Antero Eloranta, Andrei Sebyakin, Arsenii Valitov

AI summary

Overview

Research area: Market design and mechanism choice for Maximal Extractable Value (MEV) auctions on Ethereum — crossing auction theory (q-fin.TR) with empirical blockchain data analysis.

Technical level: Intermediate. The paper uses standard Bayesian Nash equilibrium derivations (first-order conditions, order statistics, the Milgrom–Weber linkage principle) and maximum-likelihood fitting of a log-normal distribution, but presents them as a structured comparison rather than a deep technical proof paper.

One-sentence scope: The paper asks which auction format a block builder should use to sell MEV opportunities, combining 2.2 million Ethereum MEV transactions with a Gaussian-copula model of affiliated searcher valuations to rank five auction formats by expected revenue.

What This Paper Is About

Block builders on Ethereum auction valuable transaction-ordering opportunities to competing "searchers," and these auctions move tens of millions of dollars per year. The authors ask a mechanism-design question that has received little attention: given what MEV values actually look like in practice, does the choice of auction format (English, Dutch, first-price sealed-bid, second-price sealed-bid, or all-pay) materially change how much revenue the auctioneer captures? The goal is to show that because MEV valuations are heavy-tailed, heterogeneous across MEV types, and affiliated (correlated across bidders), the classical Revenue Equivalence Theorem fails — and to quantify how much money the "wrong" format leaves on the table.

Key Contributions

  1. An empirical characterization of MEV value and competition. Using a dataset of 2.2 million transactions from three orderflow providers (Blink, Merkle, and MEV Blocker) covering September 2024 to August 2025, the authors show extracted values are log-normal (MLE: μ̂ = 1.102, σ̂ = 2.524) with extreme right-tail concentration, and that competition intensity — measured by a new "bribe percentage" feature — varies sharply by MEV type.

  2. An affiliation-aware equilibrium framework spanning five auction formats. Searchers' valuations are modeled as draws from a log-normal marginal linked by a Gaussian common factor, giving an affiliation parameter ρ, with equilibrium bidding strategies and expected revenues derived across a fine grid of bidder counts n and ρ.

  3. A quantitative ranking of formats and a dollar estimate of the stakes. English and second-price sealed-bid auctions strictly dominate Dutch and first-price sealed-bid formats for any ρ > 0, and the authors translate the resulting linkage gap into an estimated $10–18 million in foregone revenue over the sample period.

  4. A novel non-monotonicity in the affiliation parameter. At large n and high ρ, revenue peaks in the interior of the affiliation space and then declines, because near-perfect correlation collapses the order-statistic spread that generates competitive payments.

Main Findings

  • Log-normal values with a very heavy tail. Fitted parameters are μ̂ = 1.102 and σ̂ = 2.524; a one-standard-deviation upward shift in the latent signal multiplies extracted value by roughly e^2.524 ≈ 12.5. The empirical distribution spans from below 10⁻⁵ to above 10⁷ USDC.

  • Extreme concentration. The top 1% of transactions generate 68% of total revenue; the top 10% generate 90%. The Gini coefficient is 0.933. Mechanism choice is therefore decided by a thin set of high-value events.

  • Competition differs dramatically by MEV type. Sandwich attacks show an average bribe percentage of 95% (searchers bid away nearly all surplus), backruns 76%, naked arbitrage 67%, and liquidations 68%. The pooled average is 79%.

  • Scale and composition of the sample. Total extracted value is approximately $168.5M: sandwiches $66.5M across 890,967 transactions, naked arbitrage $52.6M across 915,194, backruns $28.2M across 405,701, and liquidations $21.2M across 4,759 transactions. Liquidations are rare but individually large (median $157.3 versus $3.01 pooled).

  • Revenue equivalence holds only without affiliation. Under independent private values (ρ = 0), all five formats satisfy revenue equivalence, confirmed by Monte Carlo verification.

  • A strict ranking emerges under affiliation. English and second-price sealed-bid dominate Dutch and first-price sealed-bid, which in turn dominate all-pay — consistent with the Milgrom–Weber linkage principle. At ρ = 0.5 with n = 5 bidders, English and SPSB yield revenues 17% higher than Dutch and FPSB, and all-pay falls 41% below the FPSB baseline.

  • The linkage gap is large and largest for few bidders. The gap is 14–28% at moderate affiliation (ρ = 0.5) and up to 30% for small bidder counts. Because small n corresponds to MEV categories like liquidations and naked arbitrage — precisely those where searchers retain the most surplus — format choice matters most where the money is.

  • Dollar impact on the observed data. Switching from a first-price to a second-price or English format would have raised auctioneer revenues by between 3.3% and 32.0%, or roughly $10–18 million over the sample period, with magnitude sensitive to the assumed ρ and n.

  • Non-monotonic revenue in ρ at high n. Revenue rises, peaks in the interior of the affiliation parameter space, and then falls as correlation approaches one, because the gap between the highest and second-highest valuations — the source of competitive payment — shrinks.

Methodology in Plain English

The authors proceed in three layers.

Empirical layer. They take a public dataset of MEV "bundles" (packaged transactions that searchers submit for inclusion), compute for each transaction the extracted value — the tip paid to the builder plus the searcher's own retained profit — and derive a bribe percentage (tip divided by extracted value) as a proxy for how fiercely that MEV category is contested. They then fit a log-normal distribution by maximum likelihood and describe concentration using a Pareto curve and a Gini coefficient.

Theoretical layer. They set up a standard single-object auction with n symmetric, risk-neutral searchers whose valuations follow that fitted log-normal distribution. For each of five formats they derive the symmetric equilibrium bidding function — for instance, the first-price bid equals the expected highest rival value conditional on winning, while the second-price format makes truthful bidding a weakly dominant strategy. They then derive expected revenue as a function of order statistics of the valuation distribution.

Affiliation layer and simulation. To break revenue equivalence, they let valuations share a common random factor (a Gaussian copula construction with correlation parameter ρ), so that when an opportunity is good for one searcher it tends to be good for others — matching the fact that all searchers observe the same on-chain state and price signals. Because closed forms are unavailable for log-normal order statistics, they compute expected revenues numerically over a fine grid of n and ρ, and then apply the resulting percentage gaps to the actual observed bribe totals to get dollar figures.

Why This Matters

For research: The paper provides an empirical case where a textbook result (Revenue Equivalence) fails for a measurable, economically important reason — affiliation — and it does so with real data rather than a stylized example. It also adds a new theoretical observation (the interior revenue peak in ρ) that qualifies the usual "more correlation is better for the seller" intuition behind the linkage principle.

Real-world applications:

  • Block builders and orderflow providers choosing a mechanism for selling bundle inclusion rights can use the format ranking and the 3.3%–32.0% revenue range as a prior for redesign.
  • MEV category segmentation: auction formats could be applied selectively by MEV type, since sandwich auctions (95% bribe rate, large n) and liquidation auctions (68%, small n) sit in very different regions of the (n, ρ) space.
  • Protocol and rollup designers picking auction rules — the paper discusses Arbitrum Timeboost (second-price sealed-bid) and Polygon Atlas (first-price sealed-bid) — can interpret their choices through this linkage-gap lens.
  • Market concentration analysis: the finding that the linkage gap is largest at small n implies that thin markets, where a handful of specialized searchers compete, are exactly where mechanism design has the most leverage.

Industry relevance: The estimated $10–18 million in foregone revenue over roughly one year of data is a concrete, order-of-magnitude figure for an industry that runs these auctions continuously. The bribe-percentage statistic also gives practitioners a simple competition-intensity diagnostic they can compute directly from bundle data.

Future Directions

  • Type-specific estimation. The authors explicitly flag that they pool all four MEV types into one valuation distribution for tractability, and propose estimating separate log-normal parameters and rerunning the revenue simulations per MEV category.
  • Better distributional fit. The pooled fit shows slight excess kurtosis relative to the fitted normal in the left tail (visible below log v ≈ −5), suggesting heavier-tailed or mixture specifications may be warranted.
  • Endogenizing n and ρ from data. The paper maps MEV types to qualitative ranges of n and ρ (large n for sandwiches, small n for liquidations) rather than estimating them; a natural next step is to identify these parameters empirically so the dollar estimates become tighter than the stated range.
  • Beyond static symmetric equilibrium. The related work cited in the paper finds that history-conditioned reinforcement-learning agents capture 49% of available profits alongside existing searchers and 81% when replacing the market leader, outperforming static game-theoretic bidding — raising the open question of how the format ranking changes under learning, latency constraints, and partial observability of the kind found in real bundle auctions.

Target Audience

Researchers and practitioners in market design, blockchain economics, and DeFi infrastructure: auction theorists interested in an applied affiliation setting with unusually clean data; block builders, relays, and orderflow providers deciding on mechanism design; and quantitatively inclined readers in the MEV/block-building ecosystem who want a rigorous but accessible treatment of why the auction format for extractable value is not a neutral choice.

Authors’ abstract

We study optimal auction design for Maximum Extractable Value (MEV) auction markets on Ethereum. Using a dataset of 2.2 million transactions across three major orderflow providers, we establish three empirical regularities: extracted values follow a log-normal distribution with extreme right-tail concentration, competition intensity varies substantially across MEV types, and the standard Revenue Equivalence Theorem breaks down due to affiliation among searchers' valuations.…

The full abstract on arXiv
Read the original paper