How OddsWasp calculates football predictions
Every football prediction on OddsWasp comes from the same published method: an independent Poisson model. This post explains exactly how it works, so "probabilities, not promises" isn't just a slogan — you can check our working.
Expected goals, not gut feeling
Each team has an attacking rating (goals scored vs league average) and a defensive rating (goals conceded vs league average). Multiplying a team's attack by its opponent's defence, plus a home-advantage constant, gives its expected goals for that specific match.
From expected goals to a full scoreline matrix
Poisson distributions turn those two expected-goals numbers into a full grid of scoreline probabilities — P(2-1), P(0-0), P(3-2), all the way up. 1X2, over/under 2.5 and both-teams-to-score are simply sums over the cells of that same grid, not separately-invented numbers.
A worked example
Say the home side's expected goals work out to 1.7 and the away side's to 1.1. The Poisson formula gives roughly a 18% chance of a 1-0, 15% for 1-1, and 13% for 2-1 — three of the most likely single scorelines, and already about 46% of all probability mass. Add up every cell where the home score beats the away score and you get the home-win probability for the 1X2 market; add up every cell where both teams' goals are 1 or higher and you get the both-teams-to-score probability. Same grid, different sums.
Why not just copy the bookmaker odds
Bookmaker odds already have their margin built in, so they systematically overstate every outcome's true probability by a few percentage points — that's how the house makes money on average. OddsWasp's numbers come from the model above, independently of what any bookmaker is quoting, which is the only way to tell whether a given price is actually good value rather than just familiar.
What this model will never do
It never invents an edge for a team it has no rating data for — unknown teams get the league-average prior, which just reflects venue effects, not a fabricated advantage. And a 65% favourite still loses roughly one match in three: that is what probability means, not a guarantee.
Frequently asked questions
Is this the same model bookmakers use?
No. Bookmakers price in a margin (the 'overround') on top of their own probability estimate, so their odds are never a pure probability. OddsWasp's Poisson model has no margin — it's a statistical estimate, not a price.
Why did a 70% favourite just lose?
Because 70% still means roughly 3 losses in 10. A single result never confirms or disproves a probability — only checking calibration across hundreds of predictions does.
Does the model account for injuries or lineups?
Not directly — the attack/defence ratings are built from actual results over a rolling window, so a long-term absence shows up gradually as results change, not instantly on team-news day. That's a known limitation, not a hidden one.