The Brier score is a general measure of forecast error for binary outcomes. This page explains the concept with hypothetical examples; OUTCORA does not currently calculate this metric.
For a binary event, record a forecast probability p between 0 and 1 before the outcome is known. Set the outcome y to 1 if that event occurred and 0 otherwise. The error for one forecast is (p − y)². The Brier score is the average of those squared errors across the eligible, resolved observations. A lower score indicates smaller average squared error on the evaluated sample. The ECMWF verification material describes this probability-space mean squared error.
Consider an illustrative forecast of 60% for Up. If the official outcome is Up, the single-case squared error is (0.60 − 1)² = 0.16. If the official outcome is Down, it is (0.60 − 0)² = 0.36. A 50% forecast produces 0.25 in either of these single cases. One example cannot establish whether a forecasting process is well calibrated.
Compare the same event definition, outcome direction and settlement source. Fix each probability at a specified observation time before the outcome. State the full evaluation period, the number of eligible and resolved cases, the selection rule, any missing or unresolved data, and the model version. A market-probability baseline can be informative when it is recorded at comparable times for the same cases.
The Brier score is sensitive to both forecast error and the mix of events in the sample. It does not say whether a displayed difference was executable, whether fees could be covered or whether anyone earned a return. Cherry-picking only settled winners or changing the forecast after the fact invalidates the interpretation. This page contains no OUTCORA Brier score or system performance result.
