How our football predictions work

1. The data

Results, half-time scores and bookmaker odds for 24 leagues: Premier League, Championship, League One, League Two, La Liga, Serie A, Bundesliga, Ligue 1, Eredivisie, Primeira Liga, Belgian Pro League, Süper Lig, Scottish Premiership, Greek Super League, MLS, Brazilian Série A, Argentine Liga Profesional, J1 League, Allsvenskan, Eliteserien, Danish Superliga, Austrian Bundesliga, Swiss Super League and Ekstraklasa. In the Champions League, Europa League and Conference League teams come from different leagues, so their probabilities come from the bookmakers' margin-free odds.

2. The goals model

For each league we fit a Dixon-Coles model, a refinement of the Poisson goals model that corrects for low-scoring results. Matches from the last two years are used, weighted so that a match from a year ago counts half as much as today's. Team ratings are pulled 30% towards the league mean, and promoted teams with few matches start at bottom-quartile strength. Separate models cover the first and second half.

3. From ratings to probabilities

The ratings give expected goals for both teams. From them the model builds a grid with the probability of every scoreline up to 10 goals, and adds up the grid for each market: 1X2, double chance, over/under 0.5 to 4.5 goals, both teams to score, half-time result and half-time/full-time.

4. Choosing the daily tips

Bookmaker odds contain a margin. We remove it to get the market's fair probability, which in our backtest was a better estimate than the model alone, so tips are ranked by it. Only selections with odds between 1.35 and 2.10 qualify, each match gets at most one tip, and the most likely picks fill VIP (up to 5), then Best (4), then Free (3).

5. How we test it

Before a rule goes live we replay past seasons match by match: the model only sees what was known before each kickoff, picks the tips exactly as it would today and is scored on the real results. Rules that do not hold up in this test are not used. A backtest is not a promise of future results.

6. Why you can trust the numbers

A prediction is only useful if its probabilities are honest. We check that outcomes rated at a given probability come in about that often, and that the model beats simple league averages on every market we measure. The selection rules are fixed in advance and applied the same way every day, so no tip depends on a gut feeling.

Frequently asked questions