CrowdIntelDefinition
What is Market calibration?
How closely a market's prices match real-world frequencies — a market is well-calibrated if outcomes priced at 70% happen about 70% of the time.
In detail
Calibration measures whether prices mean what they say. A perfectly calibrated market resolves 70¢ contracts as winners 70% of the time, 20¢ contracts 20% of the time, and so on. Prediction markets are often better calibrated than polls or pundits because participants have money at stake, but they're not perfect — the tails show longshot bias, and thin markets drift. Calibration is measured by bucketing many resolved contracts by price and comparing implied probability to realized win frequency; the gap between the two is the miscalibration.
How CrowdIntel measures it
CrowdIntel builds a calibration curve across the entire resolved-market history: every contract is binned by entry price, and implied probability is plotted against realized frequency. Off-diagonal bins reveal systematic mispricing — overpriced longshots, cheap favorites. The same machinery underpins wallet scoring, since beating the calibrated baseline over a large sample is what separates edge from luck.
Frequently asked
What does it mean for a market to be well-calibrated?
Outcomes priced at X% happen about X% of the time. If 70¢ contracts win 70% of the time, the market is well-calibrated at that price.
Are prediction markets well-calibrated?
Generally better than polls, especially near resolution, but with tail biases — longshots overpriced, favorites slightly cheap. Calibration curves make the deviations visible.
How is calibration measured?
Bin many resolved contracts by price, then compare implied probability to the actual win rate in each bin. The distance from the diagonal is the miscalibration.
Related terms
- Longshot bias
The tendency for cheap, unlikely outcomes to be systematically overpriced and heavy favorites underpriced — a persistent inefficiency in betting and prediction markets.
- Prediction market
A market where participants trade contracts whose payout depends on the outcome of a future event — producing prices that aggregate crowd beliefs into probability estimates.
- P-value (trading)
The probability that an observed win rate could occur by chance alone if the trader had no edge — used to separate genuine skill from lucky streaks.
- Bayesian win rate
A shrinkage estimate of a wallet's true win rate that combines raw wins-over-bets with a prior — preventing a wallet with 9 wins out of 10 bets from being treated as a 90% sharpshooter.