Mathematical prediction: what a statistical football model actually computes
TipsQuant editorial team · Updated
Many sites call their tips mathematical. This page explains what the word means when we use it, so that you can judge whether it means the same elsewhere. A mathematical prediction here is a probability computed by a model from past results, with every step fixed in advance and every tip settled afterwards without a hand on the scales. Below the explanation are today's summaries for the match winner, goals and both teams to score markets, and the scorelines the model rates most likely.
No matches meet our criteria for this day yet. The list refreshes every morning and again before kick-off.
The model in four steps
First, every team gets two numbers: an attack rating and a defence rating. They are fitted on roughly three seasons of finished matches, and recent results count for more, with a match losing half its weight after about eight months. European and African club competitions link teams from different leagues, so the ratings are comparable across borders.
Second, every competition gets its own scoring level and home advantage, because a goal in one league is not worth the same as a goal in another, and playing at home helps more in some countries than in others.
Third, the model computes the probability of every scoreline for a match using a Poisson distribution for each side's goals, with the Dixon–Coles adjustment that corrects the known weakness of plain Poisson at low scores: too few 0-0 and 1-1 results, too many 1-0 and 0-1.
Fourth, the markets are read off the scorelines. Add the home-win scores for a 1, the level ones for an X, the three-goal-or-more scores for over 2.5, and so on.
No matches meet our criteria for this day yet. The list refreshes every morning and again before kick-off.
Calibration, or why a 60% should be a 60%
A model that is confident is not the same as a model that is right. Raw Poisson probabilities for goals markets are known to be over-confident, so the over/under and both teams to score probabilities on this site are corrected with a two-parameter adjustment fitted on one past season and tested on the next. The test is simple: among all the tips printed at around 60%, did about 60% land? That question is answered continuously by the track record, and it is the only kind of accuracy claim we make.
Where bookmaker odds are available for a match, the model's probability is blended half and half with the odds' implied probability after the bookmaker's margin is removed. The fair odds shown in every table are 1 divided by the final probability, which gives you a price to hold against the one you are offered.
No matches meet our criteria for this day yet. The list refreshes every morning and again before kick-off.
What the maths does not know
The model works from scores and, when they exist, from odds. It has never seen a team sheet, a weather forecast, a fixture list that puts a cup final three days after a league match, or a club in turmoil after a sacking. Its ratings move only when results move them, so a side that has just sold its best players is rated on what it did with them for several more weeks. Any of these can make a sound 70% into a real 50%, and the only person who can make that correction is the reader.
Most likely correct scores today
No matches meet our criteria for this day yet. The list refreshes every morning and again before kick-off.
Where to find the full lists
The summaries on this page are the view from above. The tables themselves live on the market and day pages: the prediction for today hub for the full slate, the correct score prediction page for scorelines, the draw prediction page for the X, and the over 2.5 predictions page for goals lines.
How we pick, in one line
A tip is printed only above a fixed threshold for its market, listed on the methodology page together with every detail above and the model's limits in full. Nothing is added by hand and nothing is removed after the match. Any bet can lose, however good the maths behind it, so stake only what you can afford to lose. The site is for adults aged 18+ only; see responsible gambling for limits and support.
Thirty days of settled tips
Every tip is logged before kick-off and settled automatically from the final score. Nothing is edited after the match.
Where to place this bet
Our recommended bookmaker for these markets is MegaPari.
Promo code: MEGAGLOBAL
Frequently asked questions
What is a mathematical football prediction?
A probability for each outcome of a match, computed from a model fitted on past results rather than from opinion. On this site the model is a Dixon–Coles adjusted Poisson, which rates every team's attack and defence and turns the ratings into a probability for every scoreline.
Is a mathematical prediction more accurate than a tipster?
It is more consistent and it is checkable. The model never has an off day and its tips are settled in public. A good tipster can know things the model does not, such as a late injury. The best use of this page is the model's number plus whatever you know.
Where are all the mathematical predictions for today?
The full tables are on the day pages. Prediction for today has 1X2, over/under 2.5 and both teams to score; the market hubs for correct score, over 1.5, draws and accumulators each hold their own list. This page explains the method and shows the day's summaries.
Is there a mathematical prediction for tomorrow?
Yes. The tomorrow prediction page carries the early probabilities for tomorrow's fixtures, produced by the same model in the same morning run and refreshed overnight.
Can a statistical model predict a match with certainty?
No model can. A probability of 75% is a statement that the outcome should happen about three times in four over many similar matches, and it says nothing certain about this one. Any bet can lose.
What does "statistical association football prediction" mean?
It is the formal name for the same thing. Association football is the full name of the sport, and statistical means the probabilities come from fitted data rather than from judgement.