Home › Polymarket odds › Accuracy study
Sample collected 2026-10-01 · re-collected weekly · data and method below
Across 1,912 resolved Polymarket markets from 225 of the highest-volume events that closed between 2025-10-02 and 2026-09-21, prices a week before close lined up with what happened: in 9 of 10 price bands the gap between the average price and the share of outcomes that happened is within normal sampling noise (the exception: the 0–10% band, where outcomes happened more often than priced, 4.0% against 1.7%). Outcomes priced at 50% or more a week out happened 75% of the time (246 favourites). The overall Brier score is 0.090, but that is flattered by long shots that were clearly going to lose; on the 614 genuinely uncertain markets (priced 10–90%) it is 0.202, against 0.250 for always saying 50%.
Each dot is a price band; dots on the dashed diagonal mean perfect calibration. Dot size grows with the number of markets.
| Price band | Markets | Average price | Resolved Yes | Difference |
|---|---|---|---|---|
| 0–10% | 1,236 | 1.7% | 4.0% | +2.3 pts |
| 10–20% | 167 | 14.2% | 19.2% | +5.0 pts |
| 20–30% | 115 | 24.5% | 26.1% | +1.6 pts |
| 30–40% | 78 | 34.8% | 42.3% | +7.5 pts |
| 40–50% | 70 | 45.1% | 45.7% | +0.6 pts |
| 50–60% | 71 | 54.3% | 60.6% | +6.2 pts |
| 60–70% | 49 | 65.0% | 73.5% | +8.5 pts |
| 70–80% | 41 | 74.7% | 63.4% | -11.3 pts |
| 80–90% | 23 | 85.1% | 87.0% | +1.8 pts |
| 90–100% | 62 | 96.1% | 95.2% | -0.9 pts |
On the 1,350 markets with a price at all three points, accuracy improves steadily as the close approaches: the Brier score is 0.087 a month before close, 0.064 a week before and 0.047 a day before (lower is better), and favorites won 65%, 82% and 83% of the time.
| Price taken | Markets | Brier score | Favorites won |
|---|---|---|---|
| 30 days before close | 1,350 | 0.087 | 65% of 162 |
| 7 days before close | 1,350 | 0.064 | 82% of 158 |
| 1 day before close | 1,350 | 0.047 | 83% of 185 |
Among categories with at least 20 genuinely uncertain markets, Politics & Elections markets were the most accurate a week before close (Brier 0.183 on 165 markets priced 10–90%) and Crypto markets the least (0.248 on 58). Some categories have only a few dozen uncertain markets, so read differences of a few hundredths as noise.
| Category | Markets | Brier score | Brier, uncertain markets (count) | Favorites won |
|---|---|---|---|---|
| Politics & Elections | 710 | 0.078 | 0.183 (165) | 81% |
| Sports | 573 | 0.134 | 0.207 (327) | 64% |
| Crypto | 305 | 0.074 | 0.248 (58) | 85% |
| Economy & Fed | 97 | 0.059 | 0.246 (23) | 86% |
| Tech & AI | 97 | 0.016 | 0.069 (9) | 100% |
| Culture & Entertainment | 83 | 0.023 | 0.058 (16) | 100% |
| Geopolitics & World | 47 | 0.171 | 0.282 (16) | 100% |
Brier scores use the price a week before close. Categories with more long shots score better overall, so the uncertain-market column (prices 10–90%) is the fairer comparison.
Within this sample, more trading made no clear difference: on genuinely uncertain markets (priced 10–90% a week out) the quarter with the least volume ($28.1K to $1.04M) scored 0.197 and the quarter with the most ($6.86M and up) scored 0.208. Every market here already traded at least $10,000, so this says nothing about thinly traded markets.
| Lifetime volume | Markets | Brier score | Brier, uncertain markets (count) | Favorites won |
|---|---|---|---|---|
| $28.1K – $1.04M | 478 | 0.094 | 0.197 (176) | 75% |
| $1.04M – $2.40M | 478 | 0.088 | 0.192 (118) | 74% |
| $2.41M – $6.85M | 478 | 0.087 | 0.209 (165) | 71% |
| $6.86M – $375.81M | 478 | 0.090 | 0.208 (155) | 78% |
Every market in the sample is in the free data: JSON · CSV (CC BY 4.0, credit “PolymarketTrader” with a link). The long shots that came in are on the biggest upsets page; how long markets take to settle is in the resolution-time study; how long shots are priced is in the longshot-bias study. Score your own forecasts with the Brier score calculator.
Explained in plain words: can a 90% favorite lose?
Quoting this study in an article, a paper or a newsletter? Copy this citation and keep the date, because prices move. Tips: how to cite Polymarket odds.
Polymarket View, “How accurate is Polymarket?”, October 6, 2026, https://polymarkettrader.com/odds/accuracy/ (data: Polymarket, CC BY 4.0).
Across 1,912 resolved Polymarket markets from 225 of the highest-volume events that closed between 2025-10-02 and 2026-09-21, prices a week before close lined up with what happened: in 9 of 10 price bands the gap between the average price and the share of outcomes that happened is within normal sampling noise (the exception: the 0–10% band, where outcomes happened more often than priced, 4.0% against 1.7%). Outcomes priced at 50% or more a week out happened 75% of the time (246 favourites). The overall Brier score is 0.090, but that is flattered by long shots that were clearly going to lose; on the 614 genuinely uncertain markets (priced 10–90%) it is 0.202, against 0.250 for always saying 50%.
On the 1,350 markets with a price at all three points, accuracy improves steadily as the close approaches: the Brier score is 0.087 a month before close, 0.064 a week before and 0.047 a day before (lower is better), and favorites won 65%, 82% and 83% of the time.
Among categories with at least 20 genuinely uncertain markets, Politics & Elections markets were the most accurate a week before close (Brier 0.183 on 165 markets priced 10–90%) and Crypto markets the least (0.248 on 58). Some categories have only a few dozen uncertain markets, so read differences of a few hundredths as noise.
Within this sample, more trading made no clear difference: on genuinely uncertain markets (priced 10–90% a week out) the quarter with the least volume ($28.1K to $1.04M) scored 0.197 and the quarter with the most ($6.86M and up) scored 0.208. Every market here already traded at least $10,000, so this says nothing about thinly traded markets.
A calibrated market's 70% outcomes happen about 70% of the time. The table groups outcomes by their price a week before close and shows how often each group actually resolved Yes.
We took the highest-volume Polymarket events that closed in the last 12 months, up to 15 of the most-traded outcome markets per event (at least $10,000 traded), read each market's price 7 days before it closed from Polymarket's public price history, and compared it with the final resolution. Markets that resolved 50-50 or had no price history at that time are excluded.
Polymarket View is independent and not affiliated with Polymarket. Educational analysis, not financial advice. All Polymarket odds · Statistics