How Football Prediction Models Work: Poisson to Simulation

Most football prediction models follow the same outline: estimate how strong each team is, turn that into how many goals each side should score, convert the goals into probabilities for every result and, when needed, play the rest of a season out many times. This guide walks through the building blocks in that order, from the Poisson distribution to simulation, and shows where each one appears on TACTIX.

How Do Elo Ratings Rate Teams?

Elo is a rating system borrowed from chess. After every match each team’s rating moves according to the result and the strength of the opponent: beating a strong side earns more than beating a weak one, and an upset moves both ratings further. It gives a single, constantly updated measure of team strength that football prediction models can build on.

Why Do Models Use the Poisson Distribution?

The Poisson distribution describes how often independent events happen in a fixed period. Goals fit it reasonably well, so once a model has an expected number of goals for each team, it can work out the chance of that team scoring none, one, two or more. Combining the two teams’ chances gives the probability of every scoreline, and from those, of a home win, a draw or an away win.

What Does the Dixon-Coles Model Correct?

Dixon-Coles is a refinement of the Poisson approach. Plain Poisson treats the two teams’ goals as unrelated, which gets low-scoring results slightly wrong, so the Dixon-Coles adjustment corrects the chances of scores such as a goalless draw or a one-all, and it gives recent matches more weight than old ones. The TACTIX season projections price every remaining fixture with a Dixon-Coles engine.

What Is a Monte Carlo Simulation?

A Monte Carlo simulation runs a model many thousands of times with random outcomes drawn from its probabilities, then counts how often each result happens. For a league, that means playing out every remaining fixture again and again and recording where each club finishes. The supercomputer page and the predicted tables are built this way, and each shows how many seasons were simulated.

What Are Expected Points?

Expected points, or xPts, are the league points a team would have earned on average given the chances created and conceded in each match, rather than the actual results. A side well above its xPts has probably had luck at one end or both; one below it has probably been unlucky. That makes xPts a useful early guide to which tables will change.

What Is Pythagorean Expectation?

Pythagorean expectation estimates the points a team should have from its goals scored and conceded, using a formula borrowed from baseball and adjusted for football’s draws and low scores. Like expected points, it flags teams whose results have run ahead of or behind their underlying performance, using goals instead of chances.

How Does In-Play Win Probability Work?

In-play win probability is a live estimate of each result’s chance that updates with the score, the clock and events such as a red card. Early in a match the numbers stay close to the pre-match view; late on, a single goal can move them sharply. It is the same pre-match model, re-run with less time left.

Which Football Prediction Models Does TACTIX Use?

TACTIX uses several of these building blocks in one model. A team-strength score sets each side’s expected goals, a Poisson goal matrix with a Dixon–Coles correction prices every goal market, and Elo ratings feed the three-way model behind the headline match result. A stacking layer sets two goal markets, both teams to score and over/under 2.5 goals, and each of these learned layers is kept only where it beats the base model on matches it had not seen. The output is calibrated market by market. The methodology sets out the details, the guide to prediction model metrics explains how the results are scored, and the accuracy page shows the record.

18+. Information and analysis only, not betting advice. If gambling stops being fun, free and confidential help is at BeGambleAware.org.