MEV in DeFi: A Game-Theoretic View of Attacks and Mitigations
The 20-Second Summary
Maximal Extractable Value (MEV) turns transaction ordering into an adversarial market. This paper models the MEV supply chain (searchers → builders → validators) as a three-stage game of incomplete information and shows why competition pushes rational actors toward aggressive extraction that can reduce overall welfare. It then analyzes mitigation mechanisms (commit–reveal and threshold encryption) and validates the theory against Ethereum on-chain data.
The Problem
MEV exists because block producers (or their delegated builders) can reorder, insert, and censor transactions. In DeFi, that power translates into attack patterns like sandwiching (buy before you, sell after you), generalized frontrunning, and liquidation races. The downstream effects aren’t just “someone else profits”: MEV can increase congestion and gas costs, degrade user execution quality (slippage), and concentrate power in a small set of highly optimized actors.
A core challenge is that MEV is not one actor’s fault. It is a supply chain with specialized roles:
- Searchers detect opportunities and create bundles.
- Builders aggregate bundles and construct blocks.
- Validators/proposers select blocks to propose for consensus.
If you only model one layer in isolation, you can miss how incentives propagate through the entire pipeline.
Our Approach: A Three-Stage Game
We model the supply chain as a three-stage game with incomplete information, derive Perfect Bayesian Nash Equilibria (PBNE) for major MEV vectors (including sandwich attacks), and use the equilibrium to reason about market structure. A key takeaway is that the market is well-described by Bertrand-style competition: under intense competition, bids get pushed toward value, which creates a prisoner’s-dilemma-like outcome where rational actors extract aggressively even when it harms system welfare.
Mitigations We Analyze
Two families of mitigations are analyzed because they target the information advantage MEV relies on:
-
Commit–reveal ordering: users first commit to a hash of the transaction, then reveal later. In the paper’s model, a reveal delay of $k$ reduces latency-based MEV as \(\text{MEV}_{\text{new}} \le \text{MEV}_{\text{old}}\,e^{-\lambda k},\) where $\lambda$ captures the arrival rate of invalidating opportunities.
-
Threshold encryption / encrypted mempools: transaction contents remain hidden until inclusion. Under the paper’s proposition, frontrunning/sandwiching is eliminated if the decryption threshold exceeds the adversary’s control (typically $t \approx 2n/3$).
What We Measured (On-Chain)
To validate the model and ground parameter choices in observed behavior, we analyze Ethereum data from Jan 1, 2024 to Jun 1, 2024 using public MEV-Boost relay data (Flashbots, Ultra Sound, Agnostic, bloXroute) and DEX volume from DeFiLlama.
Key Results
The paper’s empirical validation connects theory to observed market behavior:
- MEV-Boost payment skew: the mean MEV payment is reported as 0.020 ETH, heavily skewed by outliers.
- Builder concentration: the top ten builders capture >63.5% of total MEV payments in the reported data.
- Competition intensity: an empirically observed bid-to-value ratio of 0.889 aligns with the theoretical curve for roughly $n \approx 9$ effective competitors, and remains stable (about 8.5–9.5) across months.
- Welfare loss estimate: using daily DEX volume of USD 1.5B and a baseline MEV-induced slippage loss rate of 0.2%, the cumulative welfare loss over the sample period is estimated at about USD 456M. (Sensitivity spans from USD 228M at 0.10% to USD 1.14B at 0.50%.)
Limitations and Next Steps
The model targets major MEV vectors and information-based mitigations; other MEV classes (cross-domain MEV, cross-rollup MEV) can require architectural changes (e.g., sequencing services or builder-network redesign). A practical next step is extending the analysis to those settings and testing how mitigation parameters trade UX latency (e.g., commit–reveal delay) for measurable reductions in extractable value and welfare loss.