Prediction markets are usually well calibrated in active markets: across many events priced around 70%, roughly 70% tend to happen. They are less reliable for cheap long shots, thin books, markets dominated by one large holder and ambiguous rules. On Nimb, judge one market by its spread, volume and holders, not by the platform's average record.
How accurate are prediction markets?
Accurate enough to take seriously, not accurate enough to take on trust. The research, mostly on older markets such as the Iowa Electronic Markets, Intrade and Betfair, says three things:
- Prices are informative. Market prices aggregate scattered information and often match or beat polls and expert forecasts (Wolfers and Zitzewitz, Journal of Economic Perspectives, 2004; Arrow et al., Science, 2008).
- Against polls, the record is close. Iowa market prices were nearer the final vote share than polls in most comparisons over 1988–2004 (Berg, Nelson and Rietz, International Journal of Forecasting, 2008); polls adjusted for their usual drift did as well or better in another study (Erikson and Wlezien, Public Opinion Quarterly, 2008).
- Calibration is good near the end and worse far out. Intrade prices were fairly well calibrated for events close to expiry, with a bias for events months away (Page and Clemen, Economic Journal, 2013).
For crypto-era venues the evidence is newer and thinner. A working paper on about $2.4B traded on the 2024 US presidential election across PredictIt, Kalshi and Polymarket found that accuracy varied by platform and type of market, and that the same outcome often moved differently on different venues day to day (Clinton and Huang, SocArXiv, 2025). Polymarket publishes its own accuracy figures on its site. Nimb does not quote a single accuracy percentage for Polymarket, because no published figure we have checked covers all markets, all horizons and all categories.
What does "calibrated" mean?
Calibration is how closely stated chances match outcomes: a well-calibrated market's 70% outcomes happen about 70% of the time. It is measured over many markets, never from one result.
It is a property of many forecasts, not one. A single market at 70% that resolves No is not evidence of anything; a thousand markets at 70% of which 50% resolve Yes is.
What does a calibration chart show?
A calibration chart groups resolved markets by their price at a fixed time before resolution, then plots the share that resolved Yes. The diagonal is perfect calibration. The numbers below are illustrative, not measured; they show the shapes to look for.
| Price group | Perfect | Well calibrated | Long-shot bias |
|---|---|---|---|
| 0–10% | 5% | 5% | 2% |
| 20–30% | 25% | 24% | 21% |
| 45–55% | 50% | 51% | 50% |
| 70–80% | 75% | 76% | 79% |
| 90–100% | 95% | 95% | 98% |
The usual single score is the Brier score: the average squared gap between the price and the result (1 for Yes, 0 for No). Lower is better; always saying 50% scores 0.25.
Why do cheap long shots tend to be overpriced?
The long-shot bias is the tendency for unlikely outcomes to be priced higher than how often they actually happen, so a 3% outcome may come true less than 3% of the time. Likely outcomes sit slightly below their real chance. It is one of the best-documented patterns in betting and prediction markets; in horse racing, it is better explained by people overweighting small chances than by a taste for risk (Snowberg and Wolfers, Journal of Political Economy, 2010).
On Polymarket, structure adds to it. A share at 3% is cheap to buy, but pushing it down means taking the other side: buying No at 97% ties up $0.97 per share until resolution to earn $0.03, before the spread and any fees. So prices near 0% and 100% move less than the news would justify, and a 3% price is a ceiling on the market's real view as often as an estimate.
Which biases distort prices?
| Bias | What you see | Why it happens |
|---|---|---|
| Long-shot bias | Outcomes at 1–10% happen less often than priced | Overweighting small chances; cost of selling cheap shares |
| Low liquidity | Wide spread, jumps on small orders | Few traders, little depth near the price |
| Single large holder | One wallet holds most of one side | One view or one deliberate push sets the price |
| Ambiguous rules | Price stuck between the two readings | Traders price the oracle's reading, not just the event |
| Fees and capital near 0% and 100% | Near-certain outcomes priced at 96–98%, not 99.9% | Money tied up until resolution, plus spread and fees |
| Who trades | Prices lean toward the views of crypto-native traders | The trader base is not a cross-section of the public |
When are markets wrong?
- When one trader can move them. In 2012, one large trader kept Intrade's presidential prices away from other venues for weeks (Rothschild and Sethi, Journal of Prediction Markets, 2016). Historical attempts to push betting prices usually faded quickly (Rhode and Strumpf, Journal of Economic Perspectives, 2004).
- When the information is not public. Markets aggregate what traders know. If nobody trading knows, the price is a guess with money behind it.
- When the event is rare or new. There is no base rate to anchor on, and calibration evidence from other events may not apply.
- When the question is far away. Prices months before resolution are noisier and more biased than in the last weeks.
- When the rules decide. A correct view of the world can lose to a reading of the rules (see how markets resolve).
How can I check a market's track record?
A single market has no track record until it resolves; what you can check is how much to trust its price today. Market reading is working out what a prediction market's price, flow and holders say about an event, and how much to trust it. On Nimb, for one market:
- Spread and depth. Is the spread under 2–4 points, with thousands of dollars near the price? See spread and order book depth.
- Volume. How much traded over the timeframe? A price set by $1M of trading means more than one set by $5k.
- Holders. Is one side held by many wallets or by one? Nimb nets out hedged positions so you see real exposure.
- Price level. Below 10% or above 90%, discount for the long-shot bias and the cost of capital.
- Rules. Is the resolution source clear and the deadline exact?
Nimb only shows markets that pass its gates: open for trading, with real volume on undecided outcomes and a spread of at most 4 points or $10k of depth. Passing the gates means the price is readable, not that it is right.
Nimb does not store resolved markets, so it publishes no calibration figures of its own, by category or overall.
What this cannot tell you
Calibration describes many markets, not the one in front of you. A well-calibrated platform still prices events at 70% that fail 3 times in 10, and you do not know in advance which ones. If you think a price is wrong, that may be an informational edge, which Nimb's docs call alpha; most apparent edge is noise or fees. Read Nimb's known limitations and how to read a move before you rely on any price.
Last reviewed 2026-11-07 · Data from Polymarket · Not financial advice · Nimb is not affiliated with Polymarket.