How football prediction algorithms actually work

Published

Search for football predictions and you’ll find two kinds of sites: ones that shout “92% win rate!” without ever showing their working, and ones that quietly do statistics. This article explains what the statistical kind actually does — including ours, whose entire track record is public.

Step 1: turn results into ratings

Every serious prediction model starts the same way: compress a team’s history into a small number of ratings. The classic approach gives each team two numbers — an attack rating (how good they are at scoring) and a defence rating (how good they are at preventing goals) — estimated from a few seasons of results.

Two details matter enormously:

  • Home advantage is real and has to be modelled explicitly. Across the 14,000+ matches in our eight-league dataset (2021–2026), home teams won 43.6% of games and away teams 31.1%. Ignore that asymmetry and every prediction skews.
  • Recency weighting. A result from two seasons ago says much less about a squad than last month’s. Good models decay the weight of old matches exponentially, so ratings track what a team is now.

Step 2: turn ratings into goals

Ratings become predictions through a goal model. The workhorse of the field is the Poisson distribution: given how many goals a team is expected to score against this opponent at this venue, Poisson gives the probability of them scoring exactly 0, 1, 2, 3… (full explainer here).

Combine the home side’s goal distribution with the away side’s and you get a grid of every possible scoreline with a probability attached: 1-0 might be 9%, 1-1 might be 11%, and so on.

Step 3: turn goals into markets

Once you have that scoreline grid, every market is just arithmetic:

  • Match result (1X2) — sum the probabilities where home goals exceed away goals, where they’re level, and where away exceeds home.
  • Over/Under 2.5 — sum the scorelines with three or more total goals.
  • Both teams to score — sum the scorelines where neither team is on zero.

This is why a single well-built goal model can price many markets consistently, while sites that “predict” each market separately routinely contradict themselves.

What the good models add

The vanilla Poisson model has a known flaw: it underestimates low-scoring draws, because in reality the two teams’ goal counts aren’t fully independent (think of both sides settling for 0-0). The Dixon-Coles correction (1997) — which our model uses — adjusts exactly those cells of the grid. Others add Elo-style ratings as a second signal, or expected-goals data where available.

What no model can do

An honest list, because this niche rarely offers one:

  • No model beats the closing betting market consistently without private information. The market aggregates thousands of sharp opinions; in our own pre-launch backtest, the de-vigged closing odds scored better than our model (Brier 0.555 vs 0.578 over 2,071 matches). We publish that number rather than hide it.
  • Promoted teams are hard. With no top-flight history, any model starts them near league average and learns on the fly.
  • One-off events are invisible. A red card in the 3rd minute, a cup hangover, a manager sacked overnight — statistical models see none of it until it shows up in results.

How to judge any prediction site (including us)

Ask three questions: Do they show probabilities, or just picks? Do they publish every past prediction, graded, misses included? Do they explain their method? If the answer to any of these is no, you’re reading marketing, not modelling.

We built FootInsights to pass that test by construction: predictions are recorded before kick-off and graded automatically, and the track record only ever grows.