Skip to content
Crypto.world

Research

Deviations from Tradition: Stylized Facts in the Era of DeFi

Traditional exchanges run on limit order books where orders execute in microseconds, while decentralized exchanges run on automated market maker algorithms whose transactions are batched into blocks…

Deviations from Tradition: Stylized Facts in the Era of DeFi
arXiv
2510.22834
Published
2025-10-26
Authors
Daniele Maria Di Nosse, Federico Gatta, Fabrizio Lillo, Sebastian Jaimungal

AI summary

Overview

Research area: Market microstructure of decentralized finance (DeFi), specifically Automated Market Makers (AMMs) — empirical statistical analysis of high-frequency on-chain trading data.

Technical level: Intermediate. The paper assumes familiarity with basic market microstructure concepts (returns, autocorrelation, order flow, tick size) and introduces the Uniswap v3 mechanics needed to follow the argument; a mathematical appendix is provided for readers new to DEXs.

Scope (one sentence): An event-level empirical study of the 24 most active Uniswap v3 pools on Ethereum over 2023–2024, testing whether the classic stylized facts of prices, liquidity and order flow from traditional finance survive the move to concentrated-liquidity AMMs.

What This Paper Is About

Traditional exchanges run on limit order books where orders execute in microseconds, while decentralized exchanges run on automated market maker algorithms whose transactions are batched into blocks validated roughly every 12 seconds on Ethereum. The authors ask how this structural difference — plus the presence of Maximal Extractable Value (MEV) searchers who reorder, insert and remove transactions inside blocks — changes the well-documented statistical regularities of returns, liquidity and order flow. Their goal is to document which traditional stylized facts still hold, which break down, and to attribute each deviation either to the block-based market structure or to specific MEV strategies such as Just-in-Time liquidity provision, sandwich attacks and echo swaps.

Key Contributions

  1. First event-level stylized-facts analysis of Uniswap v3. The authors study the 24 most active pools (15 token pairs, pools with at least 40,000 events over 2023–2024) at event time, swap-time, tick-time and clock-time, rather than at the daily or hourly granularity used in prior DEX studies. They explicitly identify a gap: earlier work focuses mainly on price dynamics at coarse granularity and misses high-frequency microstructure effects.

  2. A taxonomy linking each statistical anomaly to a market-structure cause. Deviations from traditional finance are traced either to the block/mempool organization of order flow or to the activity of MEV searchers, rather than being left as unexplained empirical curiosities.

  3. Detection algorithms and prevalence estimates for MEV strategies. The paper provides explicit pseudo-code for detecting Just-in-Time (JIT) liquidity attacks and sandwich attacks, and measures how much of Uniswap v3 mint activity, minted liquidity and traded volume these strategies account for. It also documents and quantifies a previously unnamed pattern, "echo swaps".

  4. Concentration analysis of MEV searchers. The authors measure how few wallets dominate each MEV strategy, showing that although the searcher role is formally open to anyone, technical barriers concentrate activity in a small set of agents — with implications for who can realistically participate.

A secondary stated aim is to support the construction of realistic order-level simulators of DEX dynamics, analogous to what has been built over the last twenty years for centralized exchanges.

Main Findings

  • Fat tails at the single-swap level, unlike large-tick traditional stocks. Individual swap returns are leptokurtic, whereas for large-tick stocks in traditional finance they are not. D'Agostino-Pearson tests reject Gaussianity both in swap-time and tick-time. In swap-time, kurtosis decays linearly or super-linearly over the first few aggregated swaps, then follows a sub-linear power-law as the aggregation scale grows.
  • Echo swaps exist and are absent from traditional markets. These are two opposite-sign swaps by the same agent in the same block with no swap in between — effectively a sandwich attack with no victim. Their volume impact is usually below 1% of total swapped volume (17 of 24 pools), exceeding 5% only in the USDC-USDT 0.01% and DAI-USDC 0.01% pools. About 65% of them involve more than one pool, which the authors conjecture reflects sourcing tokens needed for MEV positions in other pools.
  • Multiple peaks in the return autocorrelation function. Traditional markets show a single peak at lag one (the bid-ask spread). Uniswap v3 shows extra peaks attributable to echo swaps and reverse arbitrageurs at lag 1, and to sandwich attacks at lags 2 and 3. The arbitrageur effect persists when measured in clock-time, and a similar pattern appears in the signed-volume ACF.
  • Herding dominates splitting in order flow. Order flow is correlated in both swap-time and tick-time, but the herding component is more important than order splitting — the reverse of what is observed in traditional markets — and herding remains dominant even when transactions from different pools are aggregated.
  • Concentrated liquidity creates discrete liquidity drops. Liquidity falls sharply because of the concentrated-liquidity mechanism, producing jumps in the marginal price and deviations from the centralized-exchange benchmark. These divergences are not profitable and can persist for minutes, yet they generate artificial volatility inside the pools.
  • Volatility and liquidity range are only weakly correlated. In traditional markets volatility and spread are strongly correlated; in these pools the link between volatility and the range over which liquidity is provided is very weak.
  • Transition probabilities between event types show MEV-related peaks. The probabilities of moving from one event type to another display several peaks that the authors associate with MEV strategies.
  • JIT liquidity is large but not rare. More than 20% of mint operations and more than 50% of minted liquidity in the data come from JIT activity; the liquidity share rises to 90% when restricting to the 0.05% and 0.3% fee tiers, though the authors note this is slightly biased because JIT ranges are usually very narrow. The median share of the swap fee captured by a JIT attacker exceeds 90% for most pools. JIT minted liquidity is far larger than passive liquidity, reflecting the narrowest possible tick range.
  • Fee tier acts as the price of protection. In low-fee pools, about 40% of total traded volume is tied to sandwich attacks, because sandwiching is cheaper there and JIT is less rewarding; in high-fee pools pure sandwich attacks become less frequent and JIT strategies more rewarding.
  • Sandwich structure and concentration. About 95% of detected sandwich attacks target a single intermediate swap, about 4.5% target two swaps, and the remainder total less than 1%; the largest number of swaps affected by a single attack is 11. In pure sandwiches the front-run and back-run amounts are almost equal, while in mixed sandwich+JIT attacks the back-run distribution shifts left because part of the exposure is unwound against the victim in the JIT leg.
  • MEV activity is highly concentrated in a few wallets. 26% of all JITs were carried out by an MEV bot operated by the user publicly known as jaredfromsubway (wallet 0xae2Fc483527B8EF99EB5D9B44875F005ba1FaE13); 13 accounts account for at least 2% of transactions, rising to 25 and 42 accounts at 1% and 0.5% thresholds. Mixed attacks are more concentrated still: jaredfromsubway is responsible for about 88%, only 3 attackers contribute more than 1% and 2%, and 5 contribute more than 0.5%. For pure sandwiches, about 64% were executed by jaredfromsubway, with 5, 8 and 24 agents contributing at least 2%, 1% and 0.5% respectively. For echo swaps, jaredfromsubway is the most active at 8%, with 8, 15 and 34 agents contributing at least 2%, 1% and 0.5%.
  • JIT providers are a distinct population. The authors find that JIT attackers appear separated from attackers using other strategies, while sandwich and mixed-attack wallets also perform echo patterns and vice versa — evidence that echo swaps are MEV-related. They also report that the JIT role has become less concentrated over time compared with the earlier dataset of June 2021 to January 2023, in which more than 70% of all JITs were made by a single user.

Methodology in Plain English

The authors download the complete on-chain event history of the 24 busiest Uniswap v3 pools on Ethereum for 2023 and 2024, using the eth_defi Python package and Uniswap v3's official Ethereum subgraph hosted by TheGraph. They group the pools into four families based on what is being traded: "Normal" pairs (a volatile asset against a stablecoin numeraire, e.g. USDC-WETH), "Volatile" pairs (two volatile assets), "Stable" pairs (stablecoin against stablecoin, price oscillating around 1) and "Synthetic" pairs (WETH or WBTC against a wrapped liquid staking token). For each pair a numeraire is fixed so that prices and returns are well defined.

Rather than sampling prices at fixed clock intervals, they build several clocks. A counter can tick once per event of any kind (event-time), once per swap only (swap-time), once per mint or burn, once per change in the discrete price tick (tick-time), or in ordinary physical time at scales from 30 seconds to 15 minutes. This matters because Uniswap v3 prices move in discrete ticks of size defined by ⌊log₁.₀₀₀₁ S⌋, and a tick change alters the virtual reserves and therefore the liquidity available to traders.

To attribute patterns to MEV, they write rule-based detectors that scan each block. A JIT attack is a mint and a later burn in the same block, from the same wallet, with identical tick bounds, amount and wallet, separated by one or more swaps. A sandwich is a swap followed by an opposite-sign swap from the same wallet, with at least one victim swap of the front-run's sign in between. Echo swaps are opposite-sign swaps by the same wallet in one block with nothing in between. They then measure how much of the volume, mint count and minted liquidity each strategy accounts for, and how concentrated each strategy is across wallets.

Statistical work includes D'Agostino-Pearson normality tests, autocorrelation functions of returns and signed volume across lags, transition probabilities between event types, and Q-Q plots of returns aggregated over n swaps, where n is set to the average number of swaps occurring in 15 minutes, 1 hour, 6 hours and 1 day. (The provided text is truncated while describing these Q-Q results, so the precise reported conclusion at the largest aggregation scales is not available in the excerpt.)

Why This Matters

Impact on research. The paper argues that the standard stylized-facts toolkit built for limit order books does not transfer cleanly to AMMs, and that MEV is a first-order driver of microstructure statistics rather than a side curiosity. It offers a replicable event-level methodology (open on-chain data, explicit detection pseudo-code) that other researchers can apply to other DEXs, chains and later Uniswap versions, and it supplies the empirical targets against which order-level DEX simulators can be validated.

Real-world applications:

  • Liquidity provision strategy: passive liquidity providers can see how much fee income JIT searchers capture (median above 90% per swap in most pools), and traders can read fee tier as an implicit price of protection against sandwiching.
  • Protocol and parameter design: evidence that low-fee pools concentrate roughly 40% of volume in sandwich-related swaps is directly relevant to choosing fee tiers and thinking about MEV mitigation.
  • Market surveillance and regulation: the concentration figures — one operator behind about 88% of mixed attacks — give regulators and exchanges a quantitative picture of who actually controls order flow on a nominally permissionless venue.
  • Execution and risk management for traders: knowing that echo swaps and sandwich activity create autocorrelation peaks at specific lags, and that concentrated-liquidity drops generate non-profitable price deviations lasting minutes, informs routing and slippage assumptions.

Industry relevance. Billions of dollars trade daily on DEXs, and Uniswap v3 is the most traded Uniswap version; findings about fee-tier-dependent attack profitability and about who dominates MEV map directly onto decisions made by market makers, aggregators, block builders and searchers.

Future Directions

  • Explain echo swaps. The authors state that the economic motivation behind echo swaps remains uncertain and explicitly leave a detailed exploration to future research; their conjecture is that echo swaps supply liquidity takers with tokens needed for MEV positions in other pools.
  • Extend to Uniswap v4. The paper notes that v4's "hooks" allow greater refinement of pool dynamics but that v4 was too recently released to provide sufficient data; since the skeleton of v3 and v4 is roughly the same, the authors expect their results to hold across both, a claim that needs testing.
  • Two-wallet attack detection. The detection algorithms assume the front-run and back-run come from the same wallet; the authors flag that a searcher could move proceeds to a second wallet, which may explain inconsistencies between sandwich counts reported by different data sources, and leave this to further work.
  • Cross-venue and cross-chain comparison. The study is restricted to Ethereum, chosen over Binance Smart Chain for its decentralization and censorship resistance, and to Uniswap v3; whether the documented deviations from traditional finance generalize to other blockchains and DEX designs is untested here.

Target Audience

Quantitative researchers and market-microstructure specialists moving into crypto; DeFi protocol designers and liquidity providers who need empirical evidence on MEV and fee-tier economics; MEV searchers, market makers and execution engineers; and regulators or surveillance analysts seeking a quantitative picture of concentration and order-flow manipulation on decentralized venues. Readers unfamiliar with AMM mechanics can rely on Appendix A, which sets out the Uniswap v3 mathematical formalism for newcomers.

Authors’ abstract

Decentralized Exchanges (DEXs) are now a significant component of the financial world where billions of dollars are traded daily. Differently from traditional markets, which are typically based on Limit Order Books, DEXs typically work as Automated Market Makers, and, since the implementation of Uniswap v3, feature concentrated liquidity. By investigating the twenty-four most active pools in Uniswap v3 during 2023 and 2024, we empirically study how this structural change in…

The full abstract on arXiv
Read the original paper