Win rate is a lie: which Polymarket trader stats actually matter
A 95% win rate can lose money, and on Polymarket it often does. Win rate without entry price is meaningless — the win rate you need to break even is simply the price you paid — and in our scored data the stat barely predicts anything: across 1,052 actively-traded wallets, the correlation between win rate and realistic post-fee edge is just 0.17, and 36% of wallets with a win rate of 60% or better still have a negative realistic edge. Worse, the stat is for sale: farming-flagged wallets average a higher win rate than clean ones. Here's why the most-quoted number in copy trading tells you almost nothing, and what to read instead.
All wallet figures below are from CopyGrade's live scored set (active wallets with ≥20 trades in 90 days), pulled June 10, 2026 — methodology here, all published aggregates here.
Why win rate means nothing without the price
On a prediction market, the break-even win rate of any bet is its price. Buy YES at 20¢ and you only need to be right 20% of the time; buy at 95¢ and you must win 95% of the time just to tread water — before fees and slippage push the bar higher.
| You buy at | You need to win | A 60% win rate here is… |
|---|---|---|
| 20¢ | >20% of the time | spectacular |
| 50¢ | >50% of the time | good |
| 80¢ | >80% of the time | losing money |
| 95¢ | >95% of the time | losing money fast |
So the favourite-buyer posts a 90%+ win rate while bleeding — frequent small wins, rare wipeout losses — and the longshot specialist posts a 25% win rate while printing. Identical win rates, opposite outcomes; the stat doesn't even tell you the sign of the edge. This is the favourite–longshot structure of prediction markets, and it's why our data shows what it shows: win rate correlates with what a copier would actually keep at just ρ ≈ 0.17 — explaining roughly 3% of the variance in real edge.
Win rate is also the easiest stat to manufacture
The second problem is worse than the first: on a no-KYC platform, win rate is a purchasable statistic. A wallet trading against itself can paint nearly any record it wants — an independent Columbia study estimates about a quarter of Polymarket's all-time volume is consistent with wash trading — and a farmer curating a record to attract copiers optimises exactly the stats copiers screen on.
Our data shows the result directly: farming-flagged wallets average a 47% win rate; clean wallets average 30%. The dishonest cohort outscores the honest one on the number most copy-bot dashboards lead with. (Part of the clean cohort's lower figure is honest longshot-taking and positions that simply haven't resolved yet — which is precisely the point: an unmanaged win rate looks worse than a manufactured one.)
What actually predicts whether copying pays
The stats worth reading are the ones that price the bet, the risk, and the honesty — not the hit count:
| Stat | What it tells you |
|---|---|
| Realistic post-fee edge | What a copier keeps per unit staked after fees, slippage, latency — the only return number that's about you. The headline overstates it 20–40%. |
| Risk-adjusted return (Sharpe-style) | Whether the return came from skill or from surviving oversized bets. |
| Max drawdown + recovery | The worst stretch you'd have to sit through — the number to size from. |
| Sample size & independence | Hundreds of resolved, uncorrelated markets — or noise. |
| Category concentration | Where the edge lives, so you copy the edge and not the hobby. |
| Forensic cleanliness | Whether the record is real at all — the veto that overrides everything else. |
That table is, not coincidentally, the Copy Score's factor list. Win rate appears nowhere in it — we compute it, display it, and give it no weight a manufactured stat could exploit.
The one-line takeaway
Win rate answers "how often was this wallet right?" — a question whose answer is worthless without price, sample, and honesty attached. The question that pays is "what would I have kept, copying this wallet, through its worst stretch?" — and that's answered by post-fee edge, risk-adjusted return, and drawdown on a vetted, forensically clean record. The Copy Verdict computes all of it per wallet; the Copy Simulator replays it under your own conditions.
Figures above are a point-in-time pull from the live scored set (June 10, 2026; 1,052 active wallets), computed by the current model and recomputed on every sync; they'll move as coverage grows and the model is recalibrated in public. CopyGrade is analysis-only — it never executes trades or holds funds. Not financial advice.