[Paper Review] Cross-Rollup MEV: Non-Atomic Arbitrage Across L2 Blockchains

Krzysztof Gogol, Johnnatan Messias+3 more|arXiv (Cornell University)|Jun 4, 2024

Transport and Economic PoliciesBusiness, Management and Accounting3 citations

TL;DR

This paper empirically analyzes cross-rollup MEV arbitrage on Ethereum Layer-2 blockchains, revealing over 500,000 unexploited arbitrage opportunities lasting 10–20 blocks on rollups like Arbitrum, Base, Optimism, and zkSync Era. By introducing a modified MAV (Maximum Arbitrage Value) metric to avoid double-counting, the study finds that LVR metrics overestimate arbitrage by a factor of five, with actual arbitrage profits ranging from 0.03% to 0.25% of trading volume.

ABSTRACT

This study quantifies the potential non-atomic MEV on Layer-2 (L2) blockchains by measuring the arbitrage opportunities between cross-rollup and DEX-CEX. Over recent years, we observe a shift in trading activities from Ethereum to rollups, with swaps on rollups occurring 2-3 times more frequently, albeit with lower trade volumes. By analyzing the costs of swap on L2s and price discrepancies cross-rollup and DEX-CEX, we identify more than 500 000 unexplored arbitrage opportunities. In particular, we find that these opportunities persist, on average, for 10 to 20 blocks, necessitating the modification of the Loss Versus Rebalancing (LVR) metric to prevent double-counting. Our findings indicate that the arbitrage opportunities in Arbitrum, Base, and Optimism range between 0.03% and 0.05% of the trading volume, while in the ZKsync it fluctuates around 0.25%.

Motivation & Objective

• To investigate the dynamics of arbitrage and swap behavior on Ethereum rollups compared to Ethereum mainnet.
• To quantify unexploited price discrepancies between DeFi AMMs on rollups and centralized exchanges (CEXs), particularly Binance.
• To assess the impact of faster block times and lower gas fees on arbitrage frequency and profitability in the L2 ecosystem.
• To propose and validate a revised metric (MAV) that avoids double-counting arbitrage during sustained price differences, improving accuracy over traditional LVR.
• To inform the design of future MEV auctions in decentralized rollup sequencers by analyzing cross-rollup arbitrage potential.

Proposed method

• The authors analyze WETH-USDC pool swaps on Uniswap v3 across Ethereum, Arbitrum, Base, Optimism, and zkSync Era.
• They identify price discrepancies between rollup AMMs and Binance CEX prices to detect unexploited arbitrage opportunities.
• A block-level analysis tracks price differences over time, measuring duration and magnitude of arbitrage windows.
• The study introduces a modified Maximum Arbitrage Value (MAV) metric that selects only the highest arbitrage value per block to prevent double-counting during sustained price gaps.
• The LVR (Loss vs. Rebalancing) metric is recalibrated using MAV to reflect actual arbitrage profits, correcting for overestimation.
• Empirical data is collected from on-chain transaction logs and price feeds, focusing on swap volumes, block times, and fee structures.

Experimental results

Research questions

• RQ1 How do swap frequency and volume on rollups compare to Ethereum mainnet, and what factors drive these differences?
• RQ2 What is the total number and duration of unexploited arbitrage opportunities between rollup AMMs and centralized exchanges?
• RQ3 How does the traditional LVR metric overestimate actual arbitrage profits, and what adjustments are needed for accurate measurement?
• RQ4 What role do faster block times and lower gas fees on rollups play in enabling more frequent but smaller arbitrage opportunities?
• RQ5 What are the implications of cross-rollup arbitrage for future MEV auction design in decentralized sequencers?

Key findings

• Over 500,000 unexploited arbitrage opportunities were identified across rollups, persisting on average for 10 to 20 blocks.
• Arbitrage on Arbitrum, Base, and Optimism generates 0.03% to 0.05% of daily trading volume in profits, while zkSync Era sees 0.25% due to persistent price discrepancies.
• Swaps occur 2–3 times more frequently on rollups than on Ethereum, but average trade volumes are about five times lower.
• Block times significantly affect swap frequency: Base sees a swap every third block, Optimism every fifth, and Arbitrum every tenth, despite higher total swap counts.
• The LVR metric overestimates actual arbitrage profits by a factor of five due to repeated counting of overlapping arbitrage events.
• The proposed MAV metric, which selects the highest arbitrage value per block, provides a more accurate and stable measure of true arbitrage activity.