Summary

Arbitrage is the largest source of MEV on Ethereum, with CEX-DEX arbitrage (price synchronization from centralized to decentralized venues) dominating by value. Unlike sandwiching, arbitrage is often considered “useful MEV” — it improves price efficiency — but it imposes persistent loss-versus-rebalancing (LVR) costs on AMM liquidity providers. Sub-slot execution and Prop-AMMs are the leading approaches to reducing LVR without eliminating arbitrage.

CEX-DEX Arbitrage

Mechanism: Price updates on centralized exchanges (Binance, Coinbase, Bybit) happen before they propagate to on-chain AMMs. Arbitrageurs who observe the CEX price first can trade against the stale on-chain price before it updates.

CEX price: ETH = $2,000.50 (updated at T=0)
On-chain AMM: ETH = $2,000.00 (stale; updating at T = next block)

Arb at T+1ms: buy ETH from AMM at $2,000.00, sell on CEX at $2,000.50
Profit: $0.50 per ETH minus gas

Scale: CEX-DEX arb is the single largest MEV category by value. Top CEX-DEX arb bots are among the most profitable entities in crypto — and their backrun searches result in the highest-value blocks.

Who does it: Requires CEX API access, extremely fast execution (millisecond level), significant capital, and sophisticated infrastructure. This is concentrated among a few well-capitalized firms.

Timing Games and CEX-DEX Arb

CEX-DEX arbitrageurs are the primary victims and beneficiaries of timing games:

  • Victims: if a builder delays bid submission, the arb bot’s bundle (which was valid at second 0) may execute at a worse price at second 4
  • Beneficiaries: blocks built later contain more CEX-DEX arb value (more price updates to capture)

The research from the Cannes Blockspace Forum showed:

  • Shorter sub-slots (200ms–1s) reduce variance in CEX-DEX arb profit per trade
  • Reduced variance → more traders willing to participate → higher trading volume
  • Volume increase → more total fees, even if per-trade profit is flat
  • Simulation: multiple competing agents across different sub-slot configurations → aggregate block value increases with shorter sub-slots, driven by volume

LVR (Loss-Versus-Rebalancing)

AMM LPs continuously lose to CEX-DEX arbitrageurs:

  • AMM price = average of LP positions
  • Each time the “true” price moves away from the AMM price, arbitrageurs trade at the stale price
  • LP receives the stale price (a loss relative to the true price)
  • LVR measures the cost: total loss to informed trading as a % of LP capital

LVR formula: proportional to price volatility squared divided by the AMM’s fee tier. Higher volatility → more LVR; higher fees → more protection but less volume.

Impact: LVR makes AMM LP positions unprofitable in volatile markets. In practice, LPs either:

  1. Accept the loss as a cost of providing liquidity
  2. Exit when LVR > fee income (LP “death spiral” in high-volatility environments)
  3. Use Prop-AMMs (price-aware market makers) to avoid stale-price exposure

AMM Arbitrage

Mechanism: Price differences between on-chain DEXs (Uniswap, Curve, Balancer, etc.).

Uniswap ETH/USDC price: 2,000.00
Curve ETH/USDC price: 2,001.00 (slightly different due to different curve math)

Arb: buy ETH on Uniswap, sell on Curve
Profit: $1 per ETH minus gas

Scale: Smaller than CEX-DEX arb but still significant. Multi-hop arb (A→B on Uniswap, B→C on Balancer, C→A on Curve) can capture complex price differences.

On-chain visibility: AMM arb is more visible than CEX-DEX arb because all inputs are on-chain. Bots compete for the same opportunities; the one who pays the highest gas (or submits to the most favorable builder) wins.

Prop-AMMs and the “New Information First” Principle

Prop-AMMs (Proportional AMMs) address LVR by letting the AMM operator (a professional market maker) update on-chain prices before arbitrage can occur:

  1. Market maker observes CEX price update (off-chain)
  2. Market maker sends an oracle update transaction to the Prop-AMM
  3. The Prop-AMM updates its on-chain price to match CEX
  4. Subsequent trades execute at the correct price
  5. LVR is eliminated for this price movement

Prerequisite: the oracle update must land before any arbitrage trade exploiting the same price movement. This requires protocol-level ordering guarantees — the chain must treat oracle updates as higher priority.

Andrea Canidio (2026): “new information first” as a design principle for L1s aiming to support on-chain finance. Oracle updates should have protocol-level inclusion guarantees ahead of swaps.

Current problem (Prop-AMM Shenanigans): aggregators can display phantom Prop-AMM liquidity by querying at favorable moments, showing prices that have already been updated (better than available). Users see good prices that don’t materialize.

Liquidations

Mechanism: When a borrowing position’s collateral value falls below the required ratio, the position is eligible for liquidation — anyone can repay the debt and seize collateral at a discount (typically 5–15%).

Alice borrowed $1,000 USDC against 0.5 ETH ($2,500 collateral)
ETH price drops → collateral ratio falls below threshold
Bob liquidates: repays $1,000 USDC, seizes 0.5 ETH now worth $900 (after price drop)
Bob's profit: $900 - $1,000 = ... wait, Bob must find the 5-15% discount window

MEV angle: multiple bots compete to liquidate the same position. The one that gets their transaction in first wins. This creates gas wars and bundle competition in builders.

Scale: liquidations are the third largest MEV category. In high-volatility periods (large price moves), liquidation MEV spikes dramatically.

MEV Taxonomy by Value

TypeValue rankNatureUseful?
CEX-DEX arb1stPrice synchronizationYes (price efficiency)
AMM arb2ndPrice synchronizationYes (price efficiency)
Liquidations3rdDebt cleanupYes (system health)
Backrunning4thResidual value extractionNeutral
Sandwiching5thUser extractionNo (harmful)

Key Sources

  • What Emerged from the Blockspace Forum Workshop in Cannes (Apr 2026) — sub-slot execution; CEX-DEX arb volume research
  • What Network Effects Mean for L1s (Andrea Canidio, Mar 2026) — LVR; Prop-AMMs; “new information first” design principle
  • Sandwich Statistics (dataalways, Apr 2026) — MEV taxonomy; arb vs. sandwich comparison