Is Polymarket Accurate? What 401 Million Resolved Positions Say

"Is Polymarket accurate?" usually means one thing: when the market says 70%, does the thing happen about 70% of the time? That is a question about calibration, and it can be answered directly, because on Polymarket every price paid is an implied probability and every market eventually resolves to 0 or 1.

We index every Polymarket trade from public Polygon transactions. For each resolved position (one wallet, one market, one side) we know the average price the wallet paid and whether that side paid out. This post puts all of them in one table: 401,672,723 positions on 3,786,243 markets, entered by 3,223,940 wallets, resolved between December 2022 and 7 November 2026.

The calibration table

Each row groups the positions bought in a 10¢ price range. "Price" is the average price paid in that range, "happened" is the share of those positions whose side won, and the gap is the difference in percentage points. A perfectly calibrated market has a gap of zero in every row.

Two things stand out. First, the averages match: the side bought cost 55.96¢ on average and won 56.02% of the time. Second, the misses are not random. Below 50¢ every range happened less often than its price, and above 50¢ every range happened more often. That is the favourite-longshot bias, the oldest regularity in betting markets: long shots are a little overpriced and favourites a little underpriced. On Polymarket the effect is largest in the middle of each side, around 25¢ and 75¢, and small at the extremes, where the 0–10¢ and 90–100¢ ranges miss by less than half a point.

The same table, by the dollar

Counting positions treats a $5 position like a $500,000 one. Weighting each position by what was paid for it gives the calibration of the money rather than of the wallets, and the tilt roughly doubles. Positions bought between 20¢ and 30¢ cost an average of 25.5¢ per dollar of payout and paid out on 18.4% of the money, a gap of 7.1 points; positions bought between 70¢ and 80¢ cost 74.8¢ and paid out on 82.2% of the money, a gap of 7.4 points. The dollar-weighted Brier score is 0.129, lower than the position-weighted 0.143, because large positions sit more often at extreme prices: $26.2 billion of the $73.8 billion paid was paid at 90¢ or more.

Where the prices are most and least accurate

Polymarket's own market categories split the 401.7 million positions very unevenly. Crypto markets hold 65% of them and sports 23%.

35,622 positions on markets with no category are left out of this table.

The two numbers answer different questions. The Brier score mixes calibration with how lopsided the questions are: a market that trades at 95¢ scores well even if it is only roughly right, so categories full of near-certain outcomes look better on it. The calibration gap is the average distance between price and outcome rate across the ten price ranges, which is the direct reading of "when it says 70%, does it happen 70% of the time". By that reading, weather and sports markets were the best calibrated, crypto and politics the least.

The time between entry and resolution tells a similar story. 60% of all positions (241.4 million) were entered less than an hour before their market resolved.

The Brier score falls steadily with the horizon, but read it with care: positions entered weeks ahead were bought at more extreme prices on average, so part of that fall is the questions, not the prices. The calibration gap does not fall steadily. It is largest for the positions entered in the last hour and smallest for those entered one to seven days out.

What a Brier score of 0.143 means

A Brier score is the average squared distance between a probability and what happened: 0 is perfect, 0.25 is saying 50% about everything, and lower is better. Polymarket's 0.143 sits well below 0.25. A tougher baseline is a forecaster who knows only how often bought sides win overall (56%) and says 56% about everything; that scores 0.246, so the prices remove about 42% of that baseline's error. The Brier score guide explains the measure in full, including why a trader's own Brier score only means something next to the prices that trader paid.

Method

The data is every resolved position on Polymarket that Polyrank has indexed from public Polygon transactions: a position is one wallet's holding of one side of one market. For each one we take the wallet's average entry price for that side as the probability it paid for, and whether that side paid out at resolution as the outcome. Positions are grouped into ten 10¢ ranges by entry price; for each range we compare the average price with the share that paid out. The Brier score is the mean of (price − outcome)², by position and, in the dollar section, weighted by the amount paid. The mean calibration gap is the position-weighted average of the absolute gap across the ten ranges. Categories are Polymarket's own market categories (tennis and esports grouped with sports, stocks with finance); the horizon is the time from the wallet's first entry to the market's resolution.

Three groups are left out, all in the same direction of caution. Positions in voided markets have no winning side (615,955 position rows). Positions with no recorded entry (34.3 million) are left out because they were not bought at a market price: they were created by splitting collateral into both sides or received from another wallet, and carry a nominal 50¢. Including them would add a block of exactly 50¢ "forecasts" that say nothing about prices. And because each wallet's position counts once, a market traded by many wallets weighs more than a quiet one; this is the accuracy of the prices people actually paid, not of one closing price per market. Figures were read from Polyrank's production database on 7 November 2026, covering resolutions through 02:31 UTC that day.

FAQ

Is Polymarket accurate?+

On average, yes. Across 401.7 million resolved positions measured on 7 November 2026, outcomes happened within 1.7 percentage points of the price paid on average, and the prices score a Brier score of 0.143 against 0.25 for a coin flip. The misses lean one way: outcomes priced below 50 cents happened slightly less often than their price, and outcomes priced above 50 cents slightly more often.

When Polymarket says 70%, how often does it happen?+

Positions bought between 70 and 80 cents, at an average of 74.7 cents, won 78.3% of the time. Positions bought between 60 and 70 cents, at 64.7 cents on average, won 68.0% of the time. Favourites on Polymarket have happened a little more often than their prices said.

What is Polymarket's Brier score?+

0.143 across every resolved position with a recorded entry, using the price paid as the probability, and 0.129 when each position is weighted by the dollars paid for it. 0 is perfect and 0.25 is a coin flip. Measured on 7 November 2026 on 401.7 million positions.

Are Polymarket long shots overpriced?+

Slightly, in the 10 to 50 cent range. Positions bought between 20 and 30 cents, at 25.0 cents on average, won 20.7% of the time, and between 30 and 40 cents, at 35.0 cents, 31.4%. Below 10 cents the gap is under half a point. This is the favourite-longshot bias known from other betting markets.

Which Polymarket categories are the most accurate?+

By the calibration gap, the average distance between price and outcome rate across ten price ranges, weather (0.7 points) and sports (1.0) were the best calibrated, and crypto (2.1) and politics (1.8) the least, measured on 7 November 2026. Brier scores rank them differently because they also reward lopsided questions.

---

Market-wide calibration says the prices are close. A single wallet's calibration says whether its owner paid good prices or lucky ones: every wallet lookup shows that wallet's own version of the chart above, free, and the leaderboard ranks wallets on it.

Polyrank is an independent analytics product and is not affiliated with Polymarket. Everything here is read-only analysis of public blockchain data and is not investment advice.