Calibration

Calibration is the statistical property of a probability forecast being correct at the stated rate. A perfectly calibrated forecaster's 70% predictions come true exactly 70% of the time, their 90% predictions come true 90% of the time, and so on. Calibration is distinct from accuracy (getting individual predictions right) because it measures the reliability of your confidence levels across many predictions.

In prediction markets, a market is considered well-calibrated when the historical distribution of contract outcomes matches their trading prices. For example, if you looked at every contract that closed at 80¢ and found that 80% of them resolved YES, the market would be perfectly calibrated at that price level. Research on historical Polymarket data has generally found good calibration in liquid markets, particularly for near-term events with objective resolution criteria.

For individual forecasters, calibration is typically measured using a Brier score or log-score. These proper scoring rules reward accurate probability assignments rather than just binary right/wrong calls. Platforms like Metaculus display calibration charts for individual forecasters, showing how their stated probabilities have matched realized outcomes across hundreds of predictions.