What a mathematical prediction model is doing
The core of almost every public football model is the same: estimate an attacking rating and a defensive rating for each club from historical results, adjust for home advantage, produce an expected goal figure for each side, then convert the pair into a probability distribution over scorelines.
From that single distribution the model reads everything else — 1X2 probabilities, over/under lines, BTTS, correct score. The "predicted score" headline is just the most likely single cell in a matrix where no cell is usually above 12%.
That last point matters enormously. A predicted 2-1 is not a forecast that the game finishes 2-1; it is the mode of a wide distribution, and it will be wrong most of the time by design.
Reading the percentages correctly
A 58% home win probability means that across many similar fixtures the home side wins about 58 times in 100. It says nothing about this specific match beyond the odds, and a loss does not falsify it.
Judge a model on calibration, not on hit rate. If fixtures rated 60% win close to 60% of the time over hundreds of matches, the model is good even when individual calls look poor.
Compare the model number to the price, always. Convert the odds with the implied probability tool and bet only where the model's number is meaningfully higher than the price implies after margin.
Probability, not prophecy
Every published percentage is a long-run frequency claim. Single results neither confirm nor refute it.
Predicted score is the mode
It is the single most likely scoreline in a flat distribution, typically carrying only a low double-digit probability.
Calibration over accuracy
A model that is right 40% of the time on 40% picks is working perfectly. Hit rate alone is a meaningless statistic without the price.
Where statistical models reliably fail
Team news is the classic blind spot. A model trained on results does not know a first-choice striker was ruled out an hour before kick-off, and public models frequently do not update after lineups are announced.
Motivation and context are the second. Final-day dead rubbers, cup rotation, and teams managing a second leg all break the assumption that both sides are trying equally, which the historical ratings quietly encode.
Small samples are the third. Newly promoted clubs, early-season fixtures and unfamiliar divisions all produce ratings built on thin data and correspondingly unreliable probabilities.
Using model output as an input
Treat the published percentage as one opinion among several. Compare it with the market price, with your own read on team news, and if you want, with your own model built from the Poisson calculator using inputs you control.
Where the three agree, you have confirmation but usually no edge — the market already knows. Where the model and the market disagree sharply, investigate the reason before assuming the model found something.
Most of the time the market is right and the model lacks information. The exceptions, found consistently, are where betting profit comes from.
Model predictions in the Nigerian market
Nigerian bettors have unusually good access to model output and unusually poor access to price comparison, which is the wrong way round. The percentage is free; the price is where the money is decided.
Operators serving Nigeria including 8xbet price a wide set of leagues, and disagreement between a model and a thinly traded market is far more common outside the top five European divisions.
Our own daily model output is published on the predictions hub with the reasoning attached, and the priced version sits on the odds page.
Live Polymarket markets
Real-time prices and 24h volume from Polymarket.
Forebet-style prediction, mathematical football models, probability output and model limitations explained for Nigerian bettors who want to use statistics properly.
Are mathematical football predictions accurate?
Well-built models are well calibrated rather than highly accurate. They estimate frequencies correctly over large samples but will be wrong on many individual matches.
Why do model predictions disagree with the odds?
Markets incorporate team news, money flow and information a results-based model cannot see. Persistent disagreement usually reflects missing information, not a hidden edge.
Can I build my own model?
Yes. A Poisson model using attack and defence estimates and a home adjustment reproduces most of what public sites publish, and our calculators run the arithmetic for you.