How accurate are prediction markets?
Every figure below is computed from markets that have already resolved, scored against what actually happened.
Every figure below is computed from markets that have already resolved, scored against what actually happened.
94.3%
4 hour accuracy
89.3%
1 month accuracy
0.0366
Brier score (4 hours)
How well market prices matched the eventual outcome at different points before resolution.
| Before resolution | Markets | Accuracy | Brier score |
|---|---|---|---|
| 1 hour | 140 | 95.7% | 0.0270 |
| 4 hours | 140 | 94.3% | 0.0366 |
| 12 hours | 140 | 92.9% | 0.0423 |
| 1 day | 140 | 92.1% | 0.0482 |
| 3 days | 140 | 89.3% | 0.0761 |
| 1 month | 112 | 89.3% | 0.0797 |
Grouping markets by the probability they quoted, then checking how often those markets actually resolved Yes.
Each band groups markets by the price they quoted. Bars of equal height mean the market was well calibrated — a market priced at 70% should resolve Yes about 70% of the time. Empty bands had no markets in the sample.
| Band | Markets | Expected | Resolved |
|---|---|---|---|
| 0–10% | 88 | 0.3% | 0.0% |
| 20–30% | 3 | 26.2% | 0.0% |
| 30–40% | 2 | 35.8% | 0.0% |
| 40–50% | 6 | 45.6% | 16.7% |
| 50–60% | 9 | 54.7% | 33.3% |
| 60–70% | 2 | 69.5% | 50.0% |
| 70–80% | 3 | 74.7% | 100.0% |
| 80–90% | 1 | 85.5% | 100.0% |
| 90–100% | 26 | 99.0% | 100.0% |
Lower is better. Brier scores show how close the price was to the truth, not just whether it picked the right side.
Bars are scaled against the worst quartile in this sample, so relative length shows which markets predicted best — not an absolute scale.
How the sampled markets actually settled.
25.0%
resolved Yes
35 resolved Yes · 105 resolved No · 140 markets
What these numbers are, and what they are not.
The sample.Resolved binary markets are pulled from Polymarket's API ordered by traded volume, filtered to those that took at least $10,000 in volume and stayed open at least three days, then sampled at an even stride across the volume range so the result is not only the largest markets. That left 140 markets scored, drawn from a pool of 592.
The measurement. For each market we read the price of the first outcome at a set of points before it closed, then compare that price with what actually happened. A market is accurate at a given point if the side priced above 50% is the side that won.
The Brier score is the mean squared error between the quoted probability and the outcome (1 or 0). Zero is perfect; 0.25 is what you would get by always saying 50/50. It rewards confidence only when that confidence is justified.
Known limits. This is a sample, not the whole history of Polymarket, and it skews toward liquid markets that ran for a while — thin or very short-lived markets are excluded because their prices are mostly noise. Markets that never traded near a horizon are dropped from that horizon only, so the sample size differs per row. Results are recomputed every few hours and will move as new markets resolve.
Figures describe Polymarket's historical record. They are not a forecast of future accuracy and are not financial advice.