Model calibration · back-test
Premier League.
For each probability band the model produces, what fraction of outcomes actually occurred? A well-calibrated model has the bars hugging the diagonal — meaning when it says "65%", the event hits ~65% of the time.
10-20%
Δ -14.4
20-30%
Δ +1.9
30-40%
Δ -1.0
40-50%
Δ -5.9
50-60%
Δ +2.8
60-70%
Δ +37.1
Expected (model)Actual hit rate380 matches × 3 outcomes
| Probability bucket | N outcomes | Hits | Expected | Actual hit rate | Δ (actual − expected) |
|---|---|---|---|---|---|
| 10-20% | 31 | 1 | 17.6% | 3.2% | -14.4 pp |
| 20-30% | 572 | 151 | 24.5% | 26.4% | +1.9 pp |
| 30-40% | 194 | 65 | 34.5% | 33.5% | -1.0 pp |
| 40-50% | 203 | 79 | 44.8% | 38.9% | -5.9 pp |
| 50-60% | 130 | 74 | 54.1% | 56.9% | +2.8 pp |
| 60-70% | 10 | 10 | 62.9% | 100.0% | +37.1 pp |