Crypto Wash Trading: What the Research Actually Shows

There is no agreed figure for how much reported cryptocurrency volume is real. There are several published estimates, they were produced by credible parties, and they disagree by a wide margin. Any article that quotes one of them as settled fact is quoting a methodology it has not described.

The three most cited results are worth stating together, because seeing them side by side is more informative than any one of them alone. In 2019 Bitwise Asset Management filed research with the U.S. Securities and Exchange Commission concluding that roughly 95% of CoinMarketCap-reported bitcoin spot volume, across an 83-exchange sample, was fake or non-economic, and that only ten venues showed volume it considered entirely real. In 2022 Forbes ran its own assessment across 157 exchanges and put the figure at about 51%. In the academic literature, Cong, Li, Tang and Yang examined 29 centralised exchanges and found that wash trading averaged over 70% of reported volume on the unregulated ones, while the regulated venues in their sample behaved like conventional financial markets.

A 44-percentage-point spread between published estimates is not a rounding difference. It is the signature of three teams measuring different exchange samples, over different periods, with different definitions of what counts as fake — and because the samples overlap only partly, the headline figures are not directly comparable.

The disagreement is mostly definitional. “Wash trading” in the strict sense means trades where beneficial ownership does not change — an account trading against itself, or two accounts under common control trading with each other. “Non-economic volume” is a broader category that also captures incentivised trading, volume generated by exchange-run market-making programmes, and fee rebates structured so that trading at a loss is still profitable.

The narrower the definition, the smaller the headline number. The Forbes figure and the Bitwise figure are not two measurements of the same quantity that happen to differ; they are partly measurements of different quantities.

Sample construction matters just as much. An estimate built from the long tail of small venues will find more fabricated volume than one restricted to the largest exchanges, because the incentive to fabricate is strongest where listing rank and aggregator visibility are the primary source of new users.

None of these studies had access to exchange account records, so none of them could observe beneficial ownership directly. What they could do is test whether the statistical shape of reported trades matches the shape that genuine trading produces.

Real order flow has well-documented regularities. Trade sizes cluster at round numbers but not uniformly so; the distribution of first significant digits in trade sizes follows a predictable pattern; and the tail of the size distribution follows a recognisable law. Fabricated volume, especially when generated programmatically, tends to violate at least one of these regularities — sizes are too uniform, too round, or too evenly distributed across the digit range.

This is a powerful method and an inherently probabilistic one. It identifies volume whose statistical fingerprint is inconsistent with organic trading. It does not identify who generated that volume, and it does not establish intent. Every serious paper in this literature says so explicitly, and the honest way to report their results is to preserve that caveat rather than drop it.

• ▶First-significant-digit tests — genuine trade sizes follow a known digit distribution; generated sizes often do not
• ▶Trade-size roundness — an unusually high share of perfectly round sizes is a common artefact of automated wash trading
• ▶Tail-distribution tests — the largest trades in a real market follow a recognisable power law

Three things. The regulatory perimeter widened, enforcement moved from statistics to stings, and detection methods moved from distributional tests to machine learning trained on labelled trade-level features.

The regulatory shift matters because the strongest finding in the academic work is not the 70% figure itself — it is the gap between regulated and unregulated venues in the same sample and the same period. Where a venue is subject to market-surveillance obligations, the statistical anomalies largely disappear. That result is more durable than any single percentage, and it is the one that should inform where a trader chooses to execute. The perimeter has since widened: in the European Union, the Markets in Crypto-Assets Regulation prohibits market manipulation in crypto-assets and requires licensed venues to run surveillance for it, with those provisions applying since December 2024.

Enforcement has also become direct rather than statistical. In October 2024 the U.S. Department of Justice charged eighteen individuals and entities after the FBI created a token of its own and market-making firms agreed to generate volume for it. In March 2026 it charged ten foreign nationals tied to four market-making firms — Gotbit, Vortex, Antier and Contrarian — after a second undercover operation by the FBI and IRS Criminal Investigation, alleging that the defendants acted as illicit market makers, wash trading bureau-created tokens to inflate their volume and price. These are allegations until proven. What they establish independently of any verdict is that volume-for-hire is a service that exists, is sold openly enough for an undercover buyer to purchase it, and targets exactly the long-tail tokens and lightly supervised venues where the statistical studies find the most anomalous volume.

On the detection side, recent work applies tree-based and deep-learning models to trade-level features rather than aggregate distributional tests. Applied to NFT marketplaces, these methods have flagged around 38% of trades and 60% of traded value as wash trades, with wide variation between platforms. The technique generalises; the specific percentages do not transfer to spot exchange markets and should not be quoted as though they do.

Most of the useful signal is available from public exchange data if you know what to compare. The single most informative check is the relationship between reported volume and observable order-book depth.

Genuine volume is expensive to produce, because it consumes real liquidity. Fabricated volume is cheap, because it does not. A venue reporting volume far out of proportion to the depth resting in its book is describing a market that would not absorb the trades it claims to be processing. That ratio is checkable by anyone with access to both numbers, and how to measure depth properly is a short article of its own.

The second check is behavioural rather than statistical: does the reported volume respond to market events the way a real market does? Genuine volume spikes on catalysts and thins out overnight. Volume that is flat across sessions, or that fails to react to a large price move, is describing a market with no participants in it.

• ▶Compare volume to resting depth — a large ratio means volume that the visible book could not have supported
• ▶Compare across venues on the same pair — one venue diverging from the rest is more informative than any absolute number
• ▶Check the intraday shape — genuine volume has a session profile; generated volume often does not
• ▶Check whether spreads widen when volume spikes — in a real market under stress, they usually do

Vultax runs market-quality screens across the venues it connects to and surfaces them inside the market-quality domain of Vi IQ, its 0-100 composite. How the screens are constructed is documented in the methodology.

What that produces is a probabilistic risk reading on a market, derived from order-book and trade behaviour. It is not an identification of a participant, because Vultax has no access to exchange account records and no screen built on public market data can establish beneficial ownership.

The practical use is comparative and defensive. If a pair's market-quality reading is degraded at the moment you are about to size a position, that is a reason to check depth and spread before you commit, or to route the order somewhere else. It is not a verdict on the venue.

Every published estimate of crypto wash trading is a modelled inference from market data, not a count. The honest summary: a material share of reported volume on unregulated venues does not behave like organic trading, credible estimates of that share run from roughly half to nearly all depending on definition and sample, and the regulated-versus-unregulated gap is the most consistent finding across all of them. Anyone quoting a single precise figure without naming the study, the sample and the definition behind it is overstating what the research supports.