Define the first-half endpoint
The event is the score when the referee ends the first half, including first-half stoppage time under the accepted market. Confirm how abandonment, awarded results, neutral venues and incorrectly listed home teams are treated. Betfair's football rules and DraftKings' soccer rules are current operator examples.
IFAB's current Laws define match periods and the referee's allowance for time lost, but operator settlement still follows the accepted product contract.
Model a complete first-half vector
Estimate P(home leads), P(draw), and P(away leads) at half-time. The three adjusted probabilities must sum to one. A score-grid method should retain every first-half score and a tail for omitted counts.
| Input | First-half requirement |
|---|---|
| Goals | Event times and first-half score only |
| Lineups | Available before kickoff |
| Team strength | Estimated for the selected period |
| Tactical pace | Chronologically measured, not inferred after result |
| Competition | Stable period and data coverage |
The Dixon-Coles paper is primary football score-model research, but any adaptation to first-half scoring needs its own validation. Use time-ordered tests consistent with scikit-learn's TimeSeriesSplit documentation.
Worked three-leg check
Suppose illustrative first-half result probabilities for three matches are 0.48, 0.50 and 0.46. Under an explicit independence assumption:
Joint probability = 0.48 * 0.50 * 0.46 = 0.1104, or 11.04%
This calculation does not justify the estimates or a price. It shows why several individually plausible half-time outcomes form a much less frequent joint event.
Dependence and QA
OpenStax's independence condition must be checked across fixtures. Shared teams, competition incentives, weather and model parameters can connect first-half forecasts.
Test score mappers at 0-0, 1-0, 0-1, goals in first-half stoppage time, a goal after the half-time whistle, abandonment and a corrected goal time. Each fixture should map to one 1X2 state.
Next step
Use Half Time Full Time Betting for the next part of this topic.
Reconcile the first-half score grid
Row and column sums of a first-half home-goal by away-goal grid must reproduce the marginal goal distributions, and all cells plus the tail must sum to one. Then group cells into home lead, draw and away lead; those three result probabilities must also sum to one.
| Score | Half-time result |
|---|---|
| 0-0, 1-1, 2-2 | Draw |
| 1-0, 2-0, 2-1 | Home |
| 0-1, 0-2, 1-2 | Away |
Do not substitute a first-half clean-sheet rate for draw probability: 1-0 and 0-1 are clean-sheet states but not draws. Similarly, full-time comeback frequency is irrelevant to settlement after the half-time whistle.
For an accumulator, retain each match's complete vector even when one outcome is selected. This makes normalization and later reliability checks reproducible under scikit-learn's calibration guidance and prevents the chosen label from being modelled in isolation.
Continue learning
- Next guide: La Liga Accumulators
- Related guide: Ligue 1 Accumulators
Assumptions and limitations
The probabilities are illustrative. First-half goals are relatively sparse, tactical approaches change, and a full-time calibrated model is not automatically calibrated for half-time outcomes.

