Polymarket accuracy · updated every 6 hours
How accurate is Polymarket?
When Polymarket priced an outcome at about 45% a day before close, it happened about 45% of the time.
When Polymarket says X%, does it happen X% of the time?
Price 24 hours before close against how often that outcome happened. Dots on the dashed line are perfectly calibrated; dot size is the number of markets. Hover a dot for its figures.
Price band → how often it happened
| Band | Avg price | Happened | Markets |
|---|---|---|---|
| 0–10% | 1.5% | 0.5% | 1,732 |
| 10–20% | 15% | 12% | 237 |
| 20–30% | 25% | 22% | 424 |
| 30–40% | 35% | 37% | 430 |
| 40–50% | 45% | 45% | 836 |
| 50–60% | 55% | 60% | 664 |
| 60–70% | 64% | 69% | 258 |
| 70–80% | 74% | 76% | 182 |
| 80–90% | 85% | 90% | 137 |
| 90–100% | 97.5% | 99.2% | 662 |
A dot above the line means outcomes in that band happened more often than the price implied; below it, less often.
Accuracy by category
Brier score 24 hours before close, with its 95% range. Lower is better; 0.250 is a coin flip. A figure is shown once at least 100 markets have resolved.
| Category | All markets | All markets, compared with a coin flip | Contested (5–95%) | Contested, compared with a coin flip |
|---|---|---|---|---|
| Politics | 0.042 ± 0.006 n = 1,626 | 0.129 ± 0.015 n = 504 | ||
| Sports | 0.207 ± 0.005 n = 2,686 | 0.215 ± 0.005 n = 2,587 | ||
| Geopolitics | 0.045 ± 0.023 n = 142 | Not enough resolved markets yet (n = 37) | ||
| Crypto | 0.041 ± 0.009 n = 886 | 0.166 ± 0.030 n = 207 | ||
| Economy | Not enough resolved markets yet (n = 62) | Not enough resolved markets yet (n = 5) | ||
| Tech & Science | Not enough resolved markets yet (n = 62) | Not enough resolved markets yet (n = 1) | ||
| Culture | Not enough resolved markets yet (n = 15) | Not enough resolved markets yet (n = 0) | ||
| Esports | Not enough resolved markets yet (n = 83) | Not enough resolved markets yet (n = 64) | ||
| Other | Not enough resolved markets yet (n = 0) | Not enough resolved markets yet (n = 0) |
Sports matches are priced close to kick-off and many are genuinely close to coin flips, so a higher score there is expected. Markets without a category are counted as Other.
Does it get more accurate closer to the end?
Yes, on this sample. On the same 1,277 markets priced at all three points, the Brier score was 0.057 a week before close, 0.045 a day before and 0.030 an hour before.
Only markets with a price at all three points are compared, so the trend is not distorted by short markets that never had a 7-day price. This sample is mostly Politics markets (67%).
What is a Brier score?
The squared gap between a forecast and what happened. A market priced at 80% that resolves Yes scores (0.8 − 1)² = 0.04; if it resolves No, it scores (0.8 − 0)² = 0.64.
Averaged over many markets, 0 is perfect and lower is better. Always saying 50% scores exactly 0.25, whatever happens, which is why 0.25 is the coin-flip baseline.
Methodology
How this is measured
Each market is scored on Polymarket's midpoint price (halfway between the best bid and best ask) for its first listed outcome, 7 days, 24 hours and 1 hour before close. “Close” is the earlier of the actual close and the listed close date. Each price is the last hourly price Brierpoint had already captured at that moment, never a later one, so no hindsight enters the score.
How the outcome is decided
From the winner Polymarket recorded. If the scored outcome won, it counts as happened (1); if another outcome of the same market won, it did not (0). Brier = (price − outcome)², so scoring the other side gives the same number.
What's left out
- Markets with no recorded winner, or whose winner matches none of their outcomes.
- Markets that closed before 21 September 2026, when these closing prices start.
- Markets without a price at a given point are left out at that point only.
How categories are assigned
From Polymarket's own tags, mapped to Brierpoint's categories. Older closed markets that were missing tags were fetched again from Polymarket once, in November 2026. Anything still without a category is counted as Other.
Why some figures say “not enough”
A figure is published only once at least 100 resolved markets support it. The ± range is a 95% range: 1.96 × the standard deviation of the scores ÷ √n.
Limits of this analysis
- It covers only markets Brierpoint tracks, not every Polymarket market.
- Long-shot markets, priced near 0% and almost never happening, make the all-markets figure look better than it is. The contested column leaves them out.
- Sports matches are priced near kick-off and are genuinely close to coin flips.
Questions
Should I trade on these numbers?
No. This page measures how Polymarket's crowd price compared with outcomes after the fact. It is analytics, not financial advice, and says nothing about any market still open.
Does this measure Brierpoint's own forecasts?
No. Every figure here scores Polymarket's market price, not anything Brierpoint produces.
Why is there a separate contested column?
Markets priced near 0% or 100% a day out are easy to score well on. The contested column keeps only markets priced between 5% and 95%, where the crowd was genuinely uncertain, which is a harder test.
How often is this updated?
Every 6 hours, as more markets close and resolve.
Is Brierpoint part of Polymarket?
No. Brierpoint is an independent analytics site that reads Polymarket's public data.