A prediction market is accurate if its prices come true as often as they claim. When Polymarket prices an outcome at 70 cents, it is saying there is a 70% chance. So we checked: across every yes or no market that resolved recently, when the price said 70%, did the outcome actually happen about 70% of the time? It does, closely, almost all the way up and down the scale.
One number sums up how well prices matched outcomes: the Brier score. It's the average squared gap between the price and what happened, so smaller is better. A market that priced every outcome at exactly 0 or 1 and was always right would score zero. A market that priced everything at a coin-flip 50% would score 0.25. Polymarket came in at 0.064 across 15,243 markets, close to the perfect end of the scale.
That's what it means for a prediction market to be accurate. Its prices come true as often as they say they will.
Each dot on the chart is a group of markets that had a similar price a week before they closed. Left to right is what the price said. Bottom to top is how often those markets actually came true. A market that's perfectly accurate puts every dot on the dotted line, and Polymarket sits close to it the whole way up.
The same result as a table. Read the first two number columns together: what the market priced, and the share of those markets that actually happened. The bar fills to the share that came true, and the dark marker sits at the price, so a market on target has its marker at the end of its fill.
| Price band | Priced | Came true | Outcome against price |
|---|---|---|---|
| 0 to 10% | 2% | 1% | |
| 10 to 20% | 14% | 12% | |
| 20 to 30% | 24% | 21% | |
| 30 to 40% | 35% | 30% | |
| 40 to 50% | 44% | 31% | |
| 50 to 60% | 55% | 56% | |
| 60 to 70% | 65% | 66% | |
| 70 to 80% | 75% | 80% | |
| 80 to 90% | 85% | 89% | |
| 90 to 100% | 96% | 97% |
The misses aren't random. Outcomes priced under 50% came true 1.9 points less often than their price, on average. Outcomes priced over 50% came true 2.2 points more often. So longshots ran a little expensive and favorites a little cheap, a pattern known as the favorite-longshot bias. The widest miss is the 40 to 50% band, priced at 44% and true 31% of the time.
For a trader, that's the useful part. A price that's right on average can still be off in a direction you can plan for.
We took every yes or no market that resolved between August 2026 and November 2026, read its price one week before it closed, and sorted the markets into ten bands by that price. For each band we compared the average price to the share that resolved Yes. A week out is far enough that the answer usually isn't obvious yet, so it's a real test of the price.
Most markets are longshots priced near zero, which is why the first band holds the most markets, and they came true only 1% of the time. The favorite, the side priced above half, won 92% of the time. Across the whole sample the price sat within 1.9 pts of how often the outcome actually happened.
If the price is already a good forecast, you make money only by knowing something the price doesn't. Three things follow.
A share at 70¢ pays $1 if it happens, so you're paying for a 70% chance. You profit over time only if your own chance is higher than the price. The payout calculator shows what a trade returns.
A small edge can vanish once you pay to take an order. Check the fee schedule for the market before you count a gap as profit.
Because the market is accurate on average, a trader who beats it is worth finding. The leaderboard grades them with Skill Score, so you can tell a record from a lucky run.
Yes, by the measure that matters for a prediction market: its prices come true about as often as they claim. Across 15,243 resolved markets, an outcome priced at 70% happened about 73% of the time, and the price sat within 1.9 pts of the result on average. The favorite won 92% of the time.
It means the market is pricing a 70% chance the outcome happens. The 70 cents is both the price of a share that pays a dollar if it happens, and the probability the market is putting on it. Our check shows those prices hold up.
The side priced above 50% a week before the close won 92% of the time across 15,243 markets that closed between August 2026 and November 2026.
The misses aren't random. Outcomes priced under 50% came true 1.9 points less often than their price, on average. Outcomes priced over 50% came true 2.2 points more often. So longshots ran a little expensive and favorites a little cheap, a pattern known as the favorite-longshot bias. The widest miss is the 40 to 50% band, priced at 44% and true 31% of the time.
We took the price one week before each yes or no market closed, grouped markets into ten price bands, and compared each band's average price to the share that actually resolved Yes. The sample is every such market that closed between August 2026 and November 2026, refreshed daily.
The average squared gap between the price and what happened, so lower is better. Always being right scores zero and a coin flip scores 0.25. Polymarket scored 0.064.
A price carries money at risk and moves the moment anything changes, while a poll is a snapshot. What we can show here is that Polymarket's own prices are well calibrated: they come true about as often as they say.
More free tools in resources.
Book a demo and we will show you the full price history behind every resolved market.