Blockchain data from three NFT marketplaces suggest widespread fictitious trades
If you’re an investor active in or eyeing the nonfungible token or crypto world, you might want to familiarize yourself with the career of one Michael J. Meehan. A one-time ticket agent for Broadway shows, Meehan was the first person the Securities and Exchange Commission ever brought an enforcement action against under the Securities Exchange Act of 1934.
Meehan engaged in what were then known as matched orders — buying and selling the shares of Bellanca Aircraft to create the impression of active investor interest — helping push its stock up from $1.75 to $5.50 in a few months time in 1935. Then he profitably sold before the shares plunged.
After the 1929 stock market crash, and the creation of the SEC, American markets remained treacherous for the unsophisticated investor. Wash trading (effectively trading with oneself) and matched orders (coordinated trades among parties), each intended to create a false appearance of trading interest and market activity, have been illegal in mainstream U.S. financial markets since 1934. That is in part thanks to public outrage at tactics like Meehan’s.
In recent years, though, the practice has found new life in the loosely policed markets for NFTs and crypto, where the illusion of large trading volume and liquidity attracts, just as it did in Meehan’s days.
How to Detect Wash Trading?
In research published in Research Policy, University of Pennsylvania’s Brett Hemenway Falk, Oxford’s Gerry Tsoukalas and UCLA Anderson’s Niuniu Zhang, a Ph.D. student, develop and apply to modern markets a detection mechanism capable of estimating suspicious trading activity across these markets and also of pointing out particular trades for scrutiny.
In the paper, the authors examine sales recorded on three large nonfungible token, or NFT, marketplaces. The researchers developed a machine learning framework that uses on-chain data to flag individual transactions suspected of being wash trades.
To create the training data for the machine learning program, the researchers uniformly applied a same set of four on-chain filters across the three NFT marketplaces and treated the flagged transactions as proxy labels for wash trades. The models then learned to identify transactions resembling those labeled trades.
The researchers later adapted and applied their framework to historical Bitcoin trades from leaked records of Mt. Gox, a now defunct centralized exchange. The framework performed well in independently identifying what have been called out as fictitious trades and extended the potential utility of the authors’ model.
What makes the researchers’ framework useful is that it can move from a broad estimate of suspicious activity to a list of particular trades worth examining. Because the NFT version works from public blockchain records, regulators and other market watchers like journalists and investors could potentially use it to monitor on-chain trading continuously without waiting for a platform to provide private records. (In this study, the researchers tested the framework only on historical data, not a live monitoring system.)
The Liquidity Mirage
Averaged across the three NFT platforms, the researchers’ filters flagged about 38% of trades, and an astounding 60% of traded value, as carrying the fingerprints of wash trading. Deep volume on a market suggests both broad investor interest and that, when it’s time to sell, you’ll find buyers. If much of the volume is fictitious, however, selling can be difficult.
According to CoinGecko, as of August 2026, the market for cryptocurrencies and other fungible tokens, which are digital assets like Bitcoin that can be exchanged for one another, had a market capitalization of about $2.79 trillion. (NFTs, by contrast, are each distinct and not interchangeable even if from the same collection.) The entire global crypto market, all 19,000-plus tokens, trading around the clock with no closing bell, reports roughly $100 billion in volume a day, meaning there is a 3.6% turnover each day.
That compares to just 1.5% turnover for U.S. equities ($68.9 trillion in market cap and $1.03 trillion in daily traded volume).
That anomaly — the greater turnover among these new instruments — disappears, however, if one applies the 60%-of-value-appears-to-be-wash-trades broadly across these markets, an admittedly rough calculation.
Proving wash trading has always been tricky. The authors cite a November 2022 estimate from Forbes that roughly 83% of crypto trading took place on centralized exchanges, where trades are typically recorded on private internal systems rather than a public blockchain. These exchanges generally disclose little beyond the assets being traded and the size of each transaction.
NFTs, however, offer a rare corner of crypto where the curtain is already open on individual trade data. On the marketplaces studied, every sale posts publicly with buyer and seller addresses, price, token and time stamp attached.
Falk, Tsoukalas and Zhang collected transaction records from the following three NFT marketplaces:
- Their LooksRare dataset covers December 2021 through April 2023 and included 401,636 sales. LooksRare is a platform that processes relatively higher value NFTs and has a token reward system that has been criticized for incentivizing wash trades. A 2023 study, says the rewards, which are tied to trading activity, gave users a reason to manufacture volume.
- From the Blur marketplace, the researchers gathered data from October 2022 through March 2024, containing roughly 3.5 million sales.
- The bulk of the researchers’ data, though, came from OpenSea. It included 21.5 million sales from February 2022 through March 2024.
Friends in Ether
Additionally, the researchers collected tens of millions of incoming Ether transfers, a payment of Ether, the Ethereum blockchain’s native cryptocurrency, from one wallet to another. This allowed them to test whether a buyer and seller had both received funds from the same upstream address.
Falk, Tsoukalas and Zhang evaluated methods for detecting aggregate signs of trade manipulation and found that the existing statistical warning systems had limited usefulness in NFT markets. Clustering around round numbers showed only subtle differences in distinguishing between legitimate trades and wash-labeled trades. Benford’s Law, which notes the tendency for smaller leading digits like 1 and 2 to appear more often in naturally occurring datasets, proved ineffective as a diagnostic.
The researchers’ approach to labeling potential wash trades relied on four on-chain filters. One flagged trades in which the buyer and seller used the same wallet. Another identified cases in which two wallets passed NFTs from the same collection back and forth. A third filter flagged cases in which the same buyer acquired the exact same NFT three or more times. The fourth checked whether the buyer and seller had both received Ether from the same upstream account.
The researchers labeled trades flagged by any of the four filters as proxy wash trades. Each filter indicates possible manipulation rather than constituting a legal determination that wash trading occurred, a distinction that the authors are careful to make.
The proportion of suspicious trades looked remarkably steady across platforms, between roughly 31% and 41%, but the economic weight of flagged trades by marketplace was another story. On LooksRare, flagged trades account for nearly 96% of value traded, according to the estimates. On Blur, it was about 58%. On OpenSea it was about 26%.
Direct self-dealing was rare, while the common-funder filter flagged by far the largest number of transactions on each marketplace. This pattern suggests that much of suspected wash trading may involve choreography across multiple wallets.
Need to Manipulate Markets? Call Us
To see in more detail how wash trading is coordinated, consider a 2024 FBI sting. Undercover investigators launched a token called “NexFundAI” and approached specialized trading firms that were offering market manipulation as a service, using automated bots to fake bustling buyer demand.
The Justice Department charged 18 individuals and entities. The charging documents include what was allegedly said by those individuals when pitching their services: The employee of one firm offering trade manipulation explained, “The idea of volume generation is… so the token looks organic and looks live, and people get interested in trading it.” Another stated they would be using multiple wallets so the trading wouldn’t look “fake.” It’s to “create an illusion of massive buys and new holders,” according to the internal communications of a firm. And, finally, the founder of one of the firms said their objective was to find “other buyers from the community, people you don’t know about or don’t care about” because “we have to make [the other buyers] lose money in order to make profit.”
Today, we worry about human traders consciously planning wash trades. But with the growth of AI agents being used for trading, it’s not unthinkable that these agents might on their own organize and perform wash trading together in the near future. This topic was not discussed in the paper, but it’s one that could open a can of worms about who to actually prosecute given that the trading agents were never directly instructed to perform wash trading.
Exchanges Have Their Own Tells
After the data analysis, the researchers built a model for identifying wash trades, using the wash trade labels identified by their filters in lieu of proven wash trades.
The authors tested their methods against the work of others that, after much fine tuning, estimated the overall share of traded value labeled as wash trading to within a fraction of a percentage point of what was calculated by the researchers’ filter-based totals. Each marketplace appeared to have its own tell. Heavy buyer activity was an important input signal on Blur. Price and buyer activity stood out on LooksRare. Quick resales and sharp price changes were important clues on OpenSea.
With their framework set, the researchers turned away from NFTs to determine how their models would perform on fungible crypto assets such as Bitcoin. The retrained models were pointed at the leaked Mt. Gox Bitcoin logs from 2011 to 2013 and saw even better performance than with NFTs. The authors suggest that modeling a single fungible asset may be easier than modeling many varied NFT collections.
While blockchains were sold as a cure for opacity, this research suggests something subtler. They don’t prevent manipulation, but blockchains do preserve durable records of on-chain transactions that investigators can analyze. This can make suspicious on-chain behavior easier to detect at scale. However, the harder problem is separating manipulation from behavior that merely resembles it. For this reason, the researchers recommend their tool be used alongside legal judgment.
How Wash Trading Lures
Many active traders have alerts on their monitor for unusual moves in price or volume. Wash trading drives up the volume for an asset — let’s call it XYZ — and an alert goes off on the trader’s screen.
The trader goes to investigate XYZ, starting by pulling up the trading in the crypto asset and looking at the bid/ask screen. He sees many trades scrolling past on the trade feed. That suggests plenty of strangers did their homework and must know something about XYZ that he doesn’t. He checks the news and forums. XYZ is being discussed with lots of mentions of the increasing volume but only speculative ideas about what might be driving it. The trader is tempted to get in on the action.
What our trader can’t see is that one person with 10 wallets or a rented bot keeps selling to himself a few hundred times an hour. Every fake trade recorded does real things to the minds of traders watching. It may lift the last traded price, which is what traders’ screen shows them and causes FOMO. It increases the volume, which gets the asset ranked, listed, recommended and written about, giving it more legitimacy.
And it gives our trader an illusion about what, should he buy, he can then sell into. The “depth” he is relying on can disappear the moment the wash trader or his bot decides to stop trading. And as soon as other traders show up to be his exit liquidity, the wash trader may sell into that volume. The new holders of XYZ will suddenly discover bids for the asset disappear, meaning trying to sell it would mean a substantial loss.
Featured Faculty
About the Research
Falk, B. H., Tsoukalas, G., & Zhang, N. (2026). Detecting crypto wash trades via machine learning. Research Policy, 55, 105508.
