How Football Predictions Work: Probabilities, Fair Odds and the Bookmaker Margin

A goal model starts from past results. Each team gets an attack rating and a defence rating, plus a home advantage for the league. Recent matches count more than old ones, and newly promoted teams start near the level of the league's weakest sides until they have played enough games.

For any fixture the ratings give expected goals for both teams. A scoreline grid (0-0, 1-0, 2-1 and so on) follows from those expected goals, and every market is read off that grid: home win, draw, away win, over/under goals, both teams to score, even the most likely exact score.

A 60% home win does not mean the home side 'should' win. It means that across many matches the model rates the same way, the home side wins about six times in ten. A model is good when its numbers are calibrated: its 60% calls really come true about 60% of the time.

Fair odds are 1 ÷ probability: 60% is 1.67, 25% is 4.00. Bookmaker odds include a margin, so their implied probabilities add up to more than 100%. For a home/draw/away market priced 2.00 / 3.50 / 4.00 the implied probabilities are 50% + 28.6% + 25% = 103.6%. Dividing each by 1.036 removes the margin: 48.3%, 27.6% and 24.1%.

On every PunditRival match page you see three things: the model's probability, the fair odds that follow from it, and the sharpest bookmaker's probability with its margin removed.