Forecasts: method and limits
Our forecasts are probabilities, not certainties. Here is how they are calculated and what they are actually worth.
Method
For each team we estimate attacking and defensive strength from goals scored and conceded this season, relative to the league average. Early in the season these strengths are pulled towards the average to avoid hasty conclusions.
Each team's expected goals combine its attack, the opponent's defence and the home advantage observed in the league. Win, draw and loss probabilities follow from a Poisson distribution.
Every forecast is frozen before the match and never changed afterwards. It is then compared with the actual result.
Limits
The model ignores injuries, suspensions, managerial changes, fixture congestion and line-ups. It assumes both teams' goals are independent, which slightly underestimates low-scoring draws. It relies only on the current season and on results the source has already published.
Statistical estimate, not betting advice. No gains are guaranteed.
Evaluation on this season's results
Each match is re-forecast using only results before its date, then compared with a naive baseline (frequencies of home wins, draws and away wins). A lower Brier score or log loss is better.
| Competition | Matches evaluated | Brier (model) | Brier (baseline) | Most likely outcome correct |
|---|---|---|---|---|
| Bundesliga (Germany) | 56 | 0.608 | 0.629 | 52% |
| Premier League (England) | 80 | 0.666 | 0.681 | 43% |
| LaLiga (Spain) | 99 | 0.588 | 0.658 | 48% |
| Ligue 1 (France) | 69 | 0.613 | 0.653 | 48% |
| Serie A (Italy) | 90 | 0.567 | 0.657 | 57% |
| Eredivisie (Netherlands) | 87 | 0.641 | 0.685 | 47% |
| Liga Portugal | 78 | 0.549 | 0.650 | 55% |
Samples are still small: these figures are indicative and may vary considerably.

