From match history to useful probabilities.
Each eligible future fixture receives a pre-match forecast. The forecast tier tells you how much match-specific evidence supports it.
Use only pre-match information
Forecasts can use completed match history, team and competition context, corners and prices captured before the prediction cutoff. Nei future result can influence a published forecast.
- Completed results provide the scoring history for each team and competition.
- Market prices are used only when they were available before the prediction was made.
- Every displayed prediction was recorded before kickoff, so live and finished matches keep their original pre-match view.
Forecast every fixture without pretending every forecast is equal
MatchFormula uses three clearly separated forecast tiers. All matches remain available in the main prediction list, while Top recommendations include only qualifying full-model forecasts.
Enough reliable competition and team history is available for the complete score-distribution model. Only forecasts that pass the quality checks can enter Top recommendations.
When history is too thin or the full score model fails a safety check, a fresh pre-match 1X2 reference is converted into a coherent, conservative score forecast.
When neither sufficient history nor a safe recent 1X2 reference is available, the page shows a broad prior-based forecast. It is useful for coverage, not a high-confidence recommendation.
Estimate the full score distribution
The full model fits a time-decayed Dixon–Coles model on up to four years of completed matches. It estimates each team's scoring rates and produces one consistent score matrix rather than unrelated headline picks. When safe pre-match prices are available, the published 1X2 view combines 8% model probability with 92% margin-free market probability.
Derive each prediction view carefully
Hjemme, draw and away probabilities always form one normalized three-way distribution.
Over/under 2.5, expected goals and correct score come from the same score matrix.
The chance that both teams score is summed from score outcomes where each side has at least one goal.
1X, 12 and X2 are derived directly from the published 1X2 probabilities; they are overlapping views, not a separate three-way market.
A separate model uses team and competition corner history to estimate expected corners and an over/under line.
Record, review and evaluate
Predictions are recorded before kickoff and compared with the final result after the match. Performance is judged over chronological samples using log loss, ranked probability score and calibration—not a cherry-picked win rate or short-term ROI.
Open model performance →