Soccer Over/Under Parlays: Lines and Goal Models

Inputs: define the total

DraftKings' soccer rules provide one current operator example of total-goals and period settlement. Record:

• match or team total;
• first half, regulation time or another period;
• exact line and over or under direction;
• whole, half or split-line structure;
• goals included or excluded under the market rule;
• accepted price and timestamp.

Settlement states

Quarter lines, where offered, split the stake between adjacent lines. Use Goal Line Betting Explained for complete half-win and half-loss arithmetic.

Transform score probabilities into a line probability

A goal expectation is not itself an over probability. A model must assign probability across score or total-goal states. OpenStax describes the Poisson distribution, while Dixon and Coles provide football-specific Poisson-regression research with sample and model limitations.

For an illustrative total-goal distribution:

Over 2.5 probability = 0.23 + 0.28 = 0.51, or 51.00%

Under 2.5 probability = 0.08 + 0.17 + 0.24 = 0.49, or 49.00%

The table is hypothetical and deliberately complete; a live model must preserve more score detail for handicaps, team totals and related legs.

Worked cross-match parlay

Suppose three totals are accepted at decimal 1.91, 1.85 and 2.00.

Combined price = 1.91 * 1.85 * 2.00 = 7.067

Gross return on $10 = 10 * 7.067 = $70.67

If independently validated probabilities are illustratively 0.54, 0.56 and 0.50:

Joint probability under independence = 0.54 * 0.56 * 0.50 = 0.1512, or 15.12%

Raw break-even probability = 1 / 7.067 = 0.1415, or 14.15%

OpenStax's independence condition must be justified. Shared leagues, weather, tactical priors and model errors can link totals from different fixtures.

Validation

Freeze features at the decision timestamp and test on later matches. scikit-learn's TimeSeriesSplit documentation explains chronological separation, and its calibration guidance explains probability reliability.

Validate the exact line probabilities, not only mean goals or whether selected overs won. Compare with a base-rate model and a stated market-implied benchmark; report coverage, uncertainty and excluded matches.

Dependence audit

Following OpenStax's distinction between independent and related events, treat result, BTTS, team totals and match totals from one match as shared score states. Across matches, common model and data errors remain possible. Use one score-state distribution for same-game legs instead of multiplying marginal total probabilities.

Continue learning

• Next guide: Soccer Same-Game Parlays
• Related guide: Soccer Parlay Apps in 2026

Assumptions and limitations

The distributions, prices and probabilities are illustrative. Poisson assumptions can fail through dependence, changing lineups, red cards, tactical states, competition differences and data drift. A calibrated historical model does not guarantee a favourable current price or a winning parlay.