WeBall
Model probability
The model probability is what the engine thinks the chance of an outcome is, derived from historical results rather than from the odds. It is one half of every edge calculation; the other half is the market's fair probability.
How it is produced depends on the shape of the quantity. Yes/no outcomes are modelled as binary events, counts like totals and receptions with a Poisson or negative binomial fit chosen by dispersion, and continuous quantities like margins and passing yards with a normal approximation.
Crucially, the raw model number is shrunk toward the market before it is used. The weight of that shrinkage depends on sample size: a model with deep history moves the number a long way, a thin one barely moves it at all. The market is treated as a strong prior that a model has to earn the right to disagree with.
Want the full version? Read the in-depth guide, with worked examples and the math.
See also
- Fair probability — The market's honest opinion — implied probability with the bookmaker's margin removed.
- Edge — The gap between your estimated probability and the probability the price implies. Positive means value.
- Confidence score — How much data sits behind a grade — sample depth, book coverage and consistency, on a 0 to 1 scale.