How much of Polymarket's volume is wash trading?
Update, July 2026: the third edition of our data report puts the active-wallet flag rate at 60% of 1,649 wallets (51% severe). The June figures below stand as the dated snapshot they were; the denominator lesson is unchanged.
About a quarter of it, by the best independent estimate. A Columbia University working paper (Sirolly, Ma, Kanoria & Sethi, November 2025) classifies roughly 25% of Polymarket's all-time trading volume as consistent with wash trading, and flags 14% of the 1.26 million wallets it examined. Our own forensic coverage of the actively-traded population reads worse: 53% of the 1,297 wallets under CopyGrade coverage carry a farming flag. The two numbers measure different things — and the gap between them is exactly why copiers should care.
All figures here are estimates from behavioural classification, not confessions; the Columbia paper is a working paper, not yet peer-reviewed, and our own numbers are model output, recalibrated in public. Precision varies; the direction doesn't.
What did the Columbia study actually find?
The study's network-based algorithm looks for trading patterns whose economics only make sense as self-dealing — volume manufactured between coordinated wallets rather than genuine price-taking. The headline findings:
- ~25% of all-time volume classified as consistent with wash trading.
- 14% of 1.26M wallets flagged as participants.
- Wildly uneven by category: Sports ~45% of volume, Elections ~17%, Politics ~12%, Crypto ~3% — with single-week peaks hitting ~95% (Elections, late March 2025) and ~90% (Sports, October 2024).
- Strongly time-varying: weekly wash share peaked near 60% in December 2024, collapsed below 5% by May 2025, then resurged to ~20% by October 2025.
| Category | Estimated wash share of volume |
|---|---|
| Sports | ~45% |
| Elections | ~17% |
| Politics | ~12% |
| Crypto | ~3% |
| Snapshot | Estimated weekly wash share |
|---|---|
| December 2024 (peak) | ~60% |
| May 2025 (trough) | below 5% |
| October 2025 (resurgence) | ~20% |
Why would anyone wash trade a prediction market?
Because volume is the currency of two separate games. The study points at the obvious incentives — no KYC and little fee friction make manufactured volume cheap — and at the big one: airdrop farming. Polymarket confirmed the POLY token in October 2025, and expected token allocations reward activity. Manufactured volume is a bid for free money; the October resurgence in the data is exactly what that incentive predicts.
The second game is the one we cover: manufactured credibility. Wash volume inflates the activity, win rates, and PnL that leaderboards rank by and copiers screen on. A wallet that trades against itself can paint nearly any record it wants — and a great-looking record is precisely the lure that turns copiers into exit liquidity.
Why is CopyGrade's 53% so much higher than the study's 14%?
Different denominators, different nets — and the difference is the lesson:
- Active wallets vs. all wallets ever. The study's 14% spans 1.26M wallets across Polymarket's history, most long dormant. Our 53% covered the 1,297 wallets actively trading at the June snapshot (60% of 1,649 by the July re-score) — and dishonest behaviour concentrates exactly where the activity (and the audience) is. On Polymarket's own leaderboard, our flag rate hits 72%.
- Wash trading vs. farming. Wash trading is one signature. Our farming detectors also catch iceberg accumulation, decoy clusters, stealth merges, and copy-bait patterns — schemes aimed at copiers specifically, which a volume-only lens undercounts.
Read together: across the whole history of the platform, manufactured volume is a quarter of the total — and among the wallets a copier would actually shortlist today, the flag rate is around half. The full breakdown of our number is in what percentage of Polymarket wallets are farming?
When the track record is literally staged
Manufactured volume corrupts a record from the inside. There's a cheaper attack that skips the trading entirely. In June 2026, the Wall Street Journal reported that Polymarket had paid social-media creators to post videos of enormous winning bets that, the Journal said, were never actually placed — many filmed on near-perfect clone sites built to mimic Polymarket's interface rather than on the live exchange. The Journal reported it reviewed 1,105 TikTok videos and found that roughly 70% (778) showed bets staged on these dummy pages, and that for more than half of the clips appearing to show a winning bet, the same wager would in reality have been a loss. Combined, the videos allegedly depicted about $1.9 million in winnings that never happened; in one, a creator "won" $100,000 on a $1,000 bet that, the Journal said, lost money for every one of the roughly 50 accounts that placed it on the real platform. Polymarket did not dispute the findings and said it had begun a "comprehensive audit" of its promotional content; the reported campaign also drew a letter from two U.S. senators asking the CFTC to investigate.
Set that beside the wash-trading numbers and it's the same lesson twice. Wash trading fakes a record by manufacturing volume; a staged-video campaign fakes one by manufacturing content — with no trades placed at all. Both target the same soft spot: the screenshots, win rates, and "look what I made" clips a prospective copier reads as proof. A win rate is already a lie when it's stitched from real-but-cherry-picked trades; it's pure fiction when the trades were never placed.
What neither attack can touch is the one record that isn't a marketing artifact: the on-chain tape. Every genuine Polymarket position settles publicly on-chain — which is precisely what a due-diligence tool reads and a promo video never has to. A clone site can display any figure it likes; it can't retroactively write that figure to the chain under a real wallet's address. That's why our scoring ignores the promotional surface entirely and works only from settled, public trades — the single version of a trader's history that nobody can stage.
What this means if you copy trade
Treat volume, win rate, and leaderboard rank as claims, not facts — on this platform, a quarter of the raw material behind those statistics is estimated to be self-dealing. The screens that survive manufactured records are behavioural: post-fee edge a copier would actually keep, risk-adjusted return, and a forensic farming check that vetoes the rest. That's the work the Copy Score automates per wallet, and the base rates it produces are sobering: about one active wallet in 75 is genuinely worth copying.
CopyGrade is independent and analysis-only — it never executes trades or holds funds. External figures above are the cited authors' estimates; our scores are our documented opinion from public data. Not financial advice.