How accurate are prediction markets? On the best-studied questions, US elections, research going back decades finds market prices were usually closer to the result than polls taken at the same time. But a price is a probability, not a prediction of one outcome: a candidate at 70% should lose about three times in ten. This guide explains how to judge a market’s accuracy, what the research shows, and where markets go wrong.
What “accurate” means for a probability
A market that prices something at 70% isn’t wrong when it doesn’t happen. The real test is calibration: across many events priced at about 70%, roughly 70% should happen. Judging a market on one famous upset is like judging a weather forecaster on one rainy day.
The standard way to score probability forecasts is the Brier score, introduced by the meteorologist Glenn Brier in 1950. For yes-or-no events it’s the average of (forecast − outcome)², where the outcome is 1 if it happened and 0 if not. It runs from 0 (perfect) to 1, and lower is better:
| Market price | Outcome | Brier score |
|---|---|---|
| 70% | Happened | (0.70 − 1)² = 0.09 |
| 70% | Didn’t happen | (0.70 − 0)² = 0.49 |
| 50% | Either | 0.25, the score of a coin flip |
What the research shows
- Markets vs polls. Joyce Berg, Forrest Nelson and Thomas Rietz compared the Iowa Electronic Markets with 964 national polls across the five US presidential elections from 1988 to 2004. The market was closer to the eventual result 74% of the time, and it significantly beat the polls in every election when forecasting more than 100 days out (International Journal of Forecasting, 2008).
- A broad review. Justin Wolfers and Eric Zitzewitz’s 2004 survey found market forecasts were typically fairly accurate and beat most moderately sophisticated benchmarks. Examples included election markets that were more accurate than Gallup’s projections, and an internal market at Hewlett-Packard that forecast printer sales better than the company’s own methods.
Where markets go wrong
- Long shots. The same review found markets aren’t always well calibrated on small-probability events, so a 3% price isn’t necessarily a precise 3% chance.
- Thin markets. Markets on complex questions, or where insiders may know more, often fail to attract enough traders. A small market can be moved by one large order; check the volume shown on our tables.
- The rules, not the headline. A market settles on its written rules. If they define the question differently from the news, a “wrong” price may simply be pricing the rules. See how markets resolve disputes.
- Two venues can disagree. When Polymarket and Kalshi price the same question differently, they can’t both be exactly right. Our price-gap scanner shows where they differ now.
Checking the 2026 midterms
We recorded the House-control, Senate-control, Senate and governor markets on both venues at 4 p.m. ET on election day, November 3, 2026, plus the House districts both listed, then scored those prices against the results with the Brier score above. The race-by-race check is here: how accurate prediction markets were in the 2026 midterms. We update it as the last races are called.
FAQ
Are prediction markets more accurate than polls?
If the favourite loses, was the market wrong?
Is Polymarket or Kalshi more accurate?
What is a good Brier score?
Sources, checked November 19, 2026:
- Iowa Electronic Markets: Berg, Nelson and Rietz, “Prediction market accuracy in the long run” (International Journal of Forecasting, 2008)
- NBER Digest: Can markets predict the future? (on Wolfers and Zitzewitz) (Nov 1, 2004)
- Brier score, after G. W. Brier, “Verification of forecasts expressed in terms of probability”, Monthly Weather Review, 1950
Related: What are prediction markets? · Strategy basics · Polymarket vs Kalshi by category · Methodology