Source-linked AI summary
Detecting and Quantifying Wash Trading on Decentralized Cryptocurrency Exchanges
Friedhelm Victor, Andrea Marie Weintraud
TL;DR
Wash trading on decentralized exchanges can inflate apparent volume and affect trader sentiment, but systematic DEX analysis was lacking. The paper models IDEX and EtherDelta trades as token-flow graphs to identify wash-trading structures. It finds $159 million in wash-traded volume and activity affecting more than 30% of tokens on both exchanges.
Problem
No systematic analysis of wash trading on decentralized exchanges existed, despite fake volume potentially influencing traders’ sentiment about virtual assets.
Source: §1 · pp. 1–2Method
The paper models account trades on IDEX and EtherDelta as weighted directed token-flow graphs and identifies matching-volume cycles and related structures.
Source: §1 · p. 1, §4.2 · p. 4Results
$159 million in wash trading was identified across IDEX and EtherDelta, and more than 30% of tradable tokens on each exchange were affected; 10% of EtherDelta tokens were almost exclusively wash traded.
Source: §6.5 · pp. 8–9, §8 · p. 10Takeaways & Limitations
The findings support the need for countermeasures applicable to decentralized systems, with frequent self-trades identified as potentially preventable.
Source: §1 · p. 2, §8 · p. 10The conservative parameters make the reported extent a lower bound, while relaxing them could also cause short-term speculative buybacks to be falsely identified as wash trades.
Source: §7 · p. 9
Abstract
from arXiv · showhide
Cryptoassets such as cryptocurrencies and tokens are increasingly traded on decentralized exchanges. The advantage for users is that the funds are not in custody of a centralized external entity. However, these exchanges are prone to manipulative behavior. In this paper, we illustrate how wash trading activity can be identified on two of the first popular limit order book-based decentralized exchanges on the Ethereum blockchain, IDEX and EtherDelta. We identify a lower bound of accounts and trading structures that meet the legal definitions of wash trading, discovering that they are responsible for a wash trading volume in equivalent of 159 million U.S. Dollars. While self-trades and two-account structures are predominant, complex forms also occur. We quantify these activities, finding that on both exchanges, more than 30\% of all traded tokens have been subject to wash trading activity. On EtherDelta, 10% of the tokens have almost exclusively been wash traded. All data is made available for future research. Our findings underpin the need for countermeasures that are applicable in decentralized systems.
1 INTRODUCTION
Decentralized exchanges avoid externally controlled custody but remain vulnerable to manipulative behavior, including wash trading that can inflate apparent volume and influence trader sentiment. The paper studies IDEX and EtherDelta, identifying account-level wash trading structures and quantifying a lower bound of the activity.
1 INTRODUCTION
- Decentralized exchanges offer token trading without relying on externally controlled services, but their growing complexity and vulnerability make it important to understand how they are used.Such understanding can help uncover manipulative behavior and inform future improvements.Source: §1 · p. 1
- Wash trading can involve colluding users or one user operating multiple accounts, creating apparent trading without changing positions or taking real market risk.Because Ethereum account creation is virtually cost-free and requires no identity information, multi-account wash trading is especially plausible.Source: §1 · p. 1
- Prior cryptocurrency wash-trading analysis focused on centralized exchanges and exchange-reported trades, where account-level information was unavailable.Decentralized-exchange trades are recorded on the ledger with involved accounts and carry a cost.Source: §1 · p. 1
- The study estimates a lower bound of $159 million in wash trading across IDEX and EtherDelta, with at least 30% of all tokens on both exchanges experiencing wash-trading activity.On EtherDelta, 10% of traded tokens were almost exclusively wash traded.Source: §1 · p. 2
- The paper presents a systematic analysis of wash trading on IDEX and EtherDelta, identifying both predominant one- or two-account structures and more complex patterns.The authors describe their findings as a lower bound of the activity.Source: §1 · p. 2
2 BACKGROUND
The background introduces blockchain accounts and token trading, contrasts centralized and decentralized exchanges, and distinguishes order-book and automated-market-maker DEX designs. It also gives the legal criteria used to characterize wash trades.
2 BACKGROUND
- Ethereum supports tokens traded against Ether and other tokens, while blockchain accounts use public and private keys to send transactions.A wallet is an account consisting of a public and private key pair.Source: §2 · p. 2
2.1 Centralized and Decentralized Exchanges
- Centralized exchanges hold users’ assets and conduct trading off-chain, whereas decentralized exchanges can enable non-custodial trading through smart contracts.On a centralized exchange, assets are returned to a user wallet when they withdraw.Source: §2.1 · p. 2
2.2 Types of Decentralized Exchanges
- The two main DEX designs are limit order books (LOB) and automated market makers (AMM); IDEX and EtherDelta are LOB-based Ethereum exchanges with off-chain order books and on-chain settlement.Both charge trading fees of 0.3% of traded amounts, with different fee allocation between maker and taker.Source: §2.2 · p. 2, §2.2.1 · p. 2
2.2.2 Automated Market Maker DEX.
- The paper focuses on LOB-based DEX because they have a history of over three years, while AMM DEX have recently gained popularity.The paper identifies Uniswap as an AMM example.Source: §2.2.2 · p. 2
2.3 Defining Wash Trades
- The paper’s wash-trade framing includes transactions that create the appearance of purchases and sales without market risk or a change in the traders’ positions.The cited CME guidance also emphasizes intent to execute fictitious trades and a wash result involving the same or similar price and common beneficial ownership.Source: §2.3 · p. 2
3 RELATED WORK
Related work covers market manipulation in cryptoassets, graph-based approaches to wash-trade detection, and cryptocurrency wash trading. The paper situates its account-level DEX analysis against prior studies that lacked such account information.
3.1 Cryptoasset Price Manipulation
- Prior cryptoasset-manipulation studies examined pump-and-dump schemes, Bitcoin price manipulation, suspicious Mt. Gox networks, and front-running attacks on DEX.The passage characterizes the evidence on DEX manipulation as more limited than work on centralized exchanges.Source: §3.1 · p. 3
3.2 Wash Trade Detection
- Wash-trade detection work includes graph clustering and nearest-neighbor methods for collusive cliques, as well as spectral clustering, Hidden Markov Models, and order-volume correlation statistics.Cao et al. identify trade subsets that leave traders’ positions unchanged by checking trade-set power sets, with O(2^n) runtime; later work considered order-book data.Source: §3.2 · p. 3
3.3 Cryptocurrency Wash Trading
- Prior cryptocurrency wash-trading studies focused on centralized exchanges and statistical indicators, rather than account-level analysis of individual traders.Cong et al. found that 26 unregulated exchanges fake up to 70% of their total trading volume; the paper analyzes DEX trades at the account level following legal definitions.Source: §3.3 · p. 3
4 DATASETS AND PREPARATION
The authors collect Ethereum trade data for IDEX and EtherDelta, preprocess it for token-level analysis, and model trades as weighted directed multigraphs. Their exploratory results show both accounts with many trades but few counterparties and trade-size distributions that appear natural.
4 DATASETS AND PREPARATION
- The dataset covers IDEX and EtherDelta Ethereum blocks, transactions, and events from 02/09/2017 to 05/04/2020, using contract calls or trade events to extract executed trades.The extracted information includes participating accounts, tokens, and trade amounts.Source: §4 · p. 3
4.1 Data Preprocessing
- Preprocessing converts large integer amounts to floating-point values, computes prices and joins U.S. Dollar values, removes missing or potentially unsuccessful trades, and retains Ether-token trades.These steps removed 3,612 IDEX trades and 3,095 EtherDelta trades.Source: §4.1 · p. 3
4.2 Modeling Token Trade Graphs
- A token trade graph is a directed multigraph whose vertices are trading accounts and whose edges represent trades directed from token seller to buyer and weighted by token volume.Matching-volume cycles can represent wash trading with no change in participants’ market positions, including self-trades and overlapping sub-cycles.Source: §4.2 · p. 4, Figure 1 · p. 4
4.3 High Level Insights
- The median ratio of trades per trade partner is 1, while the 99th percentile is 3.25; accounts with many trades but few partners may be suspicious.The DEX interfaces listing orders do not show account information, so traders usually do not know their counterparties when placing orders.Source: §4.3 · p. 4, Figure 2 · p. 4
- Trade-size distributions show higher Ether amounts traded less frequently and local peaks at round values, appearing consistent with natural trading behavior and likely unsuitable for detecting wash trading on DEX.The passage notes round-number peaks such as 5 Ether.Source: §4.3 · p. 4, Figure 3 · p. 4
5 WASH TRADE DETECTION
The detection method first narrows candidate traders using repeatedly observed strongly connected components, then searches their token-specific trades for subsets that leave each trader with approximately unchanged positions. It uses iterative volume matching across multiple time windows to keep the search tractable.
5 WASH TRADE DETECTION
- The method follows legal wash-trading criteria by seeking collusive trade sets that leave each participating trader’s position unchanged, using candidate generation followed by volume matching.Checking every possible trade subset has exponential complexity, so the method first narrows the trade sets to analyze.Source: §5 · p. 5
5.1 Candidate Set Generation with SCCs
- Repeatedly observed strongly connected components (SCCs) capture maximal groups of accounts linked by cycles, including cycles with sub-cycles, and serve as candidate sets.SCCs appearing at least 100 times are selected as candidates, indicating at least 100 round-trip trades.Source: §5.1 · p. 5, Figure 5 · p. 5
5.2 Volume Matching
- For each candidate SCC and token, the method seeks trade subsets whose matching volumes produce no overall position change for the participating accounts.The method formalizes account-level positions and describes the volume-matching algorithm as linear time.Source: §5.2 · p. 6, §5.2.1 · p. 6
5.2.1 Formalizing trader position changes for wash trade detection.
- A trade is represented as a transfer of volume from seller to buyer, and summing trades yields signed account positions; token identity is implicit because matching runs within one token.A wash-trade subset qualifies when each account’s resulting position is zero within a margin set to 1%, which accommodates imperfectly matched volumes and trade costs.Source: §5.2.1 · p. 6
5.2.2 Wash Trade Detection Algorithm.
- Rather than test all 2^n trade subsets, the algorithm examines trade sets iteratively, removing the last trade until the account-position condition is met or fewer than two trades remain.The algorithm checks whether each trader’s position is within the margin of the mean trade size.Source: §5.2.2 · pp. 6–7
- Identified structures differ across exchanges: EtherDelta is dominated by single-account self-trades, while IDEX also commonly has two-account structures.The figure reports 5,501 EtherDelta self-trade instances; complex structures are rare and almost absent on EtherDelta.Source: Figure 6 · p. 7
- The algorithm makes three passes using one-hour, one-day, and one-week windows, starting with the shortest; trades already labeled as wash trades are not checked again.The windows accommodate actors who may trade over different time periods, while short intervals are assumed more suitable for repeatedly inflating apparent market activity.Source: §5.2.2 · p. 7
6 QUANTIFICATION
The section quantifies detected wash trading on IDEX and EtherDelta across trading structures, affected tokens, token lifespans, and time. It reports $159 million in total wash-trading volume and substantial exposure among tradable tokens.
6 QUANTIFICATION
- The quantification uses candidate strongly connected components appearing at least 100 times, a 1% position-change margin, and 1-hour, 1-day, and 1-week matching windows.These parameters form the basis for the remaining quantification of wash-trading activity.Source: §6 · p. 7
6.1 Wash Trading Structures
- EtherDelta wash trades were predominantly self-trades, while IDEX had more two-account and three-or-more-account structures, often containing short sub-cycles.The authors suggest these complex structures may reflect simple low-level cycles distributed across multiple accounts; branched structures with sub-cycles longer than three did not occur.Source: §6.1 · p. 7
6.2 Affected Tokens
- 31.54% of IDEX tokens and 42.12% of EtherDelta tokens had at least one wash-trade set; Figure 7 shows at least 10% on each exchange had wash shares of 20% or more.Almost 10% of EtherDelta tokens were entirely wash traded, compared with less than 1% on IDEX.Source: §6.2 · p. 7, Figure 7 · p. 8
6.3 Wash Trade Activity in a Token’s Lifespan
- Wash trading was more frequent near the beginning and end of token lifespans, while a middle peak may reflect tokens that were entirely wash traded.The authors suggest early activity may generate awareness of new tokens and late activity may boost trading when activity is low.Source: §6.3 · p. 8, Figure 8 · p. 8
6.4 Wash Trade Activity Over Time
- Over $15 million in monthly wash volume peaked on IDEX in June 2018, and more than $12 million peaked on EtherDelta in January 2018; IDEX weekly wash share neared 100% in late 2017.Activity flattened sharply from mid-2018 on EtherDelta and mid-2019 on IDEX; both exchanges also showed weekly-share peaks in late 2017 and early 2018.Source: §6.4 · p. 8, Figure 9 · p. 8, Figure 10 · p. 9
6.5 Summary
- $159 million in wash-trading volume was identified across IDEX and EtherDelta.Wash trades were about 4% of IDEX trades and 1.3% of EtherDelta trades; over 30% of tradable tokens on each exchange were affected. EtherDelta involved over 5,000 accounts, while 659 accounts were responsible for over 200,000 IDEX wash trades; estimated fees totaled $478,134.Source: §6.5 · p. 9, Table 2 · p. 9
7 DISCUSSION
The authors treat $159 million as a conservative lower bound: relaxing thresholds could detect more wash trades but risks false positives. Proposed defenses face trade-offs: self-trade prevention misses other topologies, while partner limits may disrupt legitimate trades in small markets.
7 DISCUSSION
- Conservative detection parameters make the $159 million estimate a lower bound; relaxing thresholds could increase detections but risks false positives among short-term buybacks.The authors note that self-trades have no such ambiguity.Source: §7 · p. 9
- Wash trading may help token founders attract attention, while activity on IDEX and EtherDelta declined significantly after mid-2018 as AMM-based exchanges emerged.The authors suggest wash traders may also target AMM DEX, where users trade against liquidity pools.Source: §7 · p. 9
7.1 Potential Countermeasures
- Self-trade prevention does not address other trading topologies, and IDEX self-trades persisted through May 2020 despite its matching engine's claimed functionality.EtherDelta does not seem to implement self-trade prevention.Source: §7.1 · p. 9
- Limiting repeated trades with the same partners may disrupt legitimate trading in markets with few traders.IDEX introduced KYC in July 2019, coinciding with lower wash-trading volumes.Source: §7.1 · p. 9
- KYC procedures requiring official identity information are another way to make multi-account wash trading harder.The paper notes KYC was not widely adopted in decentralized finance.Source: §7.1 · p. 9
8 CONCLUSION AND FUTURE WORK
The study identifies substantial wash trading on IDEX and EtherDelta, with the reported figures representing a lower bound. Future work includes extending detection, examining other DEX models, and studying temporal and market effects.
8 CONCLUSION AND FUTURE WORK
- $159 million in wash trading was identified; more than 30% of tokens on both exchanges were affected, and 10% on EtherDelta were almost exclusively wash traded.Structures were mainly one- or two-account patterns, though more complex patterns also occurred.Source: §8 · p. 10
- Checking all contiguous subsets could extend volume matching, but would bring approximately quadratic complexity and require an efficient implementation or distributed computing.The authors also identify new detection algorithms as a future direction.Source: §8 · p. 10
- Future work includes adding temporality, examining order-book alternatives such as Uniswap, and studying effects on price development and token communities.The paper also flags loans, derivatives, and insurance as DeFi areas with potential for manipulation that remains poorly understood.Source: §8 · p. 10