Why Win Rate Is the Most Misleading Number on Polymarket
A 99.1% win rate made +$4.9k. An 18.5% win rate made +$39.6k. How Polymarket win rates get manufactured — and which metrics actually measure trader skill.
Spend ten minutes searching for the best Polymarket traders to follow and you will run into the same pitch everywhere: a wallet screenshot, a gaudy percentage, and a call to copy it. "This wallet has a 96% win rate." "He almost never loses." The number does all the selling. It feels like a skill stat — like a free-throw percentage for predicting the future.
Here is the problem: on launch-day data from our leaderboard, the #1 Polymarket trader ranked by win rate had won 99.1% of their resolved trades — and was up a grand total of +$4.9k. Meanwhile, a trader who won only 18.5% of the time was up +$39.6k. Eight times the result, at one-fifth the win rate.
Win rate is not a skill stat. On a prediction market, it is closer to a shopping receipt — a record of what kind of prices a wallet likes to buy, not how good its judgment is. This post walks through exactly how the number gets manufactured, what the data across ~2.8M wallets actually shows, and the metrics that hold up when you go looking for traders worth studying.
1. The seduction: why everyone advertises win rate
Win rate spreads because it is the easiest number to screenshot. It needs no explanation, it maps onto sports intuition ("they win nine times out of ten!"), and on Polymarket it can be pushed absurdly high — high enough that 90%+ wallets are genuinely common, which makes every copy-trading thread look like it has discovered a savant.
But ask the obvious follow-up — how much does a 96%-win-rate trader actually make per trade? — and the pitch falls apart. Because on a prediction market, your win rate is mostly a function of one decision: what price do you buy at?
2. The mechanism: manufacturing a 95% win rate for $0
Every Polymarket position is a share that pays $1.00 if you are right and $0.00 if you are wrong. The price you pay is, roughly, the market's probability. That single fact makes win rate trivially easy to manufacture.
Take a market where YES trades at 95¢ — a heavy favorite, priced at a 95% implied probability. Buy 100 shares for $95:
- If YES resolves: you receive $100. You won. You earned +$5.
- If NO resolves: your shares are worthless. You lost −$95.
Now repeat that trade 20 times, and suppose the market's prices were fair — the favorite really does come in 95% of the time:
| Count | P&L per trade | Total | |
|---|---|---|---|
| Wins | 19 | +$5 | +$95 |
| Losses | 1 | −$95 | −$95 |
| Net | 20 trades | 95% win rate | $0 |
A 95% win rate, and exactly zero dollars — before fees and slippage. One blown favorite erases nineteen wins. The famous "Polymarket 96% win rate" wallet is usually a variation of this: a high-volume favorite collector grinding nickels in front of a steamroller, whose headline stat says nothing about whether the nickels outrun the steamroller.
And the math gets worse with any mispricing. Suppose those 95¢ favorites were actually 93% likely — a 2-point overpay you would never notice from the trade log. Out of 100 trades, you now win 93 (+$465) and lose 7 (−$665), for −$200 across $9,500 staked — while your win rate still reads a dazzling 93%. The stat doesn't merely fail to detect the bleed; it actively disguises it as excellence. That is what makes win rate uniquely dangerous rather than just uninformative: it is the only popular metric that improves as a trader concentrates into exactly the trades where small mispricings hide.
This pattern even has a name from a century of betting-market research: the favorite–longshot bias. Favorites resolve in your favor most of the time by construction, so a wallet that exclusively buys them accumulates a spectacular win rate regardless of whether the trader has any edge at all. The win rate measures their price preference. It does not measure their judgment — judgment only shows up in whether the favorites they pick win more often than the price already implied.
3. The data: win rate vs. P&L across the leaderboard
If win rate measured skill, it should correlate with results. So plot one against the other:
The launch-day contrast we keep coming back to lives in that chart: the 99.1% win rate worth +$4.9k, and the 18.5% win rate worth +$39.6k. The high-win-rate cluster hugs the zero line — many small wins, occasional catastrophic losses, tight net results. Wallets with strong results show up across the entire win-rate spectrum, including far below 50%, because what drives their P&L is payoff asymmetry: paying 20¢ for things that happen more than 20% of the time.
The broader base rates make the point sharper. As reported by Yahoo Finance, roughly 70% of Polymarket traders lose money, and the top 0.04% of wallets captured most of the profits; a separate figure suggesting only about 12.7% of traders are profitable also circulates in press coverage. Treat the exact numbers with appropriate caution — methodologies differ — but the shape is consistent with what we see on-chain: the set of high-win-rate wallets and the set of wallets that actually perform overlap far less than the marketing threads imply.
To be fair to the stat: a high win rate is not proof a wallet is bad. Some favorite specialists really do have edge — they are just demonstrating it in the last few cents of price discovery, and you cannot tell them apart from the edge-less ones using win rate, because both groups post 95%+. The number simply doesn't discriminate where it matters. A metric that assigns nearly identical scores to a sharp closer and a coin-flipping favorite tourist is not a skill metric; it's a category label.
4. Case study: the twin wallets
The two wallets below are illustrative archetypes — composite profiles built from patterns we see repeatedly across the dataset, not specific real accounts. The numbers are illustrative, not measured.
Wallet A — "The Favorite Collector." 1,200+ resolved trades. Win rate: 96%. Average entry price: 97¢. Its trade log is a wall of green: +$3, +$6, +$4, +$5… punctuated every few weeks by a −$240. Net P&L drifts slightly above zero, with all of it at risk every time a "sure thing" isn't. This is the wallet that gets screenshotted.
Wallet B — "The Longshot Hunter." ~300 resolved trades. Win rate: 21%. Average entry price: 24¢. Its log is a wall of red: −$24, −$31, −$19… punctuated by +$310, +$280. It looks like a losing account at a glance. But it is buying 24¢ outcomes that resolve YES closer to a third of the time — and that gap between price paid and frequency delivered is the entire game.
Sort by win rate and Wallet A ranks near the top of the platform while Wallet B ranks near the bottom. Sort by calibration — how accurate their implied probabilities actually were — and the order flips. If you copy Wallet A expecting the win rate to protect you, you have signed up for its steamroller too.
5. What actually predicts persistence
So if win rate is noise, what holds up? Across our dataset, the metrics that behave like skill — stable for the same wallet over time, hard to fake with a price preference — are calibration metrics and execution metrics.
Calibration, measured by the Brier score. The question calibration asks is simple: when your trades say 70%, do those things happen 70% of the time? Every entry price is an implied probability — buy YES at 70¢ and you are asserting the event is more than 70% likely. The Brier score takes every resolved position, compares the implied probability to what actually happened (1 or 0), squares the error, and averages it. Lower is better. A coin-flip guesser who always says 50% scores 0.25; a perfect forecaster scores 0.
What makes Brier resistant to manufacturing is that it grades you relative to the difficulty of the claim. Buying a 95¢ favorite that wins barely moves your score — the market already said 95%, so you demonstrated almost nothing. The trader who repeatedly buys at 60¢ things that happen 75% of the time is demonstrating real, repeatable judgment, and Brier picks that up while win rate is busy crowning the favorite collector. On Polyrank we also benchmark each wallet's Brier against the market's own implied probabilities on the same markets — beating the market's calibration is the bar, not just beating a coin flip. (A full Brier-for-traders explainer is coming on this blog.)
And calibration persists. A win rate is a description of past shopping; a calibration score is a property of how someone processes information, and well-calibrated wallets in our data tend to stay well-calibrated across new markets and new quarters. That persistence is exactly what you want before you study anyone's positions — it is the difference between a track record and a hot streak.
Alpha-vs-mid, in one paragraph. Calibration measures judgment; alpha-vs-mid measures execution and timing. For every fill, we ask: did this trader get a better price than the market's midpoint at that moment, and did the price subsequently converge in their direction? One good fill is luck. Hundreds of fills that systematically beat the mid — measured trade-by-trade from raw Polygon transactions — is edge you can see forming in the data, long before resolution settles the score.
6. How to evaluate any wallet in 60 seconds
Next time a thread tells you to follow a wallet, run this checklist on its profile:
- Ignore the win rate. You now know what it measures.
- Check the average entry price. If almost every fill is above 90¢, the win rate is manufactured — judge the wallet as a favorite collector, not a forecaster.
- Check the Brier score against the market baseline. Is this trader more accurate than the prices they traded against?
- Check alpha-vs-mid. Do their fills systematically beat the midpoint, or are they paying up for entries?
- Put drawdown next to P&L. A result achieved while routinely risking ruin is a different result.
- Check the sample size — resolved markets, not trade count. Twenty resolved markets is an anecdote. A thousand fills in two markets is one opinion, repeated.
Every one of these is on a wallet's Polyrank profile — 57 skill metrics per wallet, and the leaderboard ranks by the ones that resist manufacturing.
7. Methodology & disclosures
The figures in this post come from Polyrank's pipeline: roughly 2.8 million wallets and ~35 million deduplicated trades spanning 3+ years of Polymarket history, reconstructed from raw Polygon blockchain data — fills, splits, merges, redemptions, and resolutions. Every number on a Polyrank profile traces back to a public Polygon transaction; nothing is self-reported. The 99.1%/18.5% comparison is a real pair of wallets from our launch-day leaderboard snapshot; the "twin wallets" in section 4 are labeled illustrative archetypes. Press statistics (the ~70% / 0.04% / 12.7% figures) are reported by third parties and cited as such.
Polyrank is an independent analytics product and is not affiliated with Polymarket. Everything here is read-only, non-custodial analysis of public blockchain data. None of it is investment advice, and no metric — including the good ones — promises anything about future results. We measure skill; we don't sell certainty.
Check any wallet before you trust it. Paste an address into the free wallet lookup — no login, no wallet connection — and see its calibration, alpha-vs-mid, and all 57 skill metrics in seconds.
Founding Trader: the first 250 seats get Pro for $19/mo (regular $39), locked for life.