Home / Wiki / F / Forecast Calibration

Forecast Calibration

Trading and SpeculationUpdated 16 Nov 2026

Forecast calibration is the degree to which stated probabilities match what happens. A forecaster is well calibrated if, across all the occasions on which a probability of 30 percent was given, the event occurred about 30 percent of the time, and likewise at every other level. It is checked by grouping forecasts by stated probability and comparing each group with its observed frequency. It is also one component of the Brier score.

Calibration is not the same as skill. Always quoting the long-run average frequency is perfectly calibrated but uninformative, so a good forecaster must also separate likely events from unlikely ones. In a prediction market, the price of an event contract is read as an implied probability, and studies test whether contracts priced at a given level pay out at that rate. Weather services apply the same test to each ensemble forecast.

Last updated 16 Nov 2026