EdgeworthArena
Live dataSign in
Pricing a Football Match · The scoreline model

Goals are Poisson, almost

lesson 1 of 625 minnot yet done

Goals are rare events in a long match. Each side creates chances at some rate, and a chance becomes a goal with some probability, so the number of goals a side scores in ninety minutes is close to a Poisson count with a rate we call λ for the home side and μ for the away side.

Goals per side in ninety minutes: Poisson with λ = 1.55 (home) and μ = 1.15 (away)0%10%20%30%40%0123456goalsprobability21%33%25%13%5%2%0%32%36%21%8%2%1%0%home side, λ = 1.55away side, μ = 1.15
The two sides' goal distributions at the rates a mid-table home fixture usually carries. The home side's extra third of a goal moves mass from 0 and 1 onto 2 and 3.
P(X = x) = e^{-λ} λ^x / x!

The rates come from the two sides. Every team has an attack strength and a defence strength, and the home side gets a home-advantage multiplier:

λ = att_home · def_away · h, μ = att_away · def_home

Fit on thousands of results, these strengths are what the Predict page shows as expected goals for and against an average side.

Where Poisson is wrong

Two independent Poissons put too little mass on 0–0 and 1–1 and too much on 1–0 and 0–1. Dixon and Coles (1997) fixed this with a single correction factor τ that moves probability between the four low scorelines, controlled by one parameter ρ. Fitted on our store, ρ is about −0.05: a small but real tilt toward the draws.

The Dixon–Coles correction: ρ = −0.05 moves mass from 1-0 and 0-1 onto 0-0 and 1-10%4%8%12%16%0-01-00-11-1probability6.7%7.3%10.4%9.8%7.7%7.1%12.0%12.6%independent Poissonswith τ_ρ, ρ = −0.05
What ρ = −0.05 actually does: a little mass leaves 1-0 and 0-1 and lands on 0-0 and 1-1. It is a small correction, and it is the difference between a book that prices the draw properly and one that does not.
P(x, y) = τ_ρ(x, y) · Poisson(x; λ) · Poisson(y; μ)

What you can now do

With λ, μ and ρ you have a probability for every scoreline, which is the whole game: every market on the board is a sum over cells of that grid. Try it below: pick two sides and read the rates, then look at the grid.

Every scoreline at λ = 1.55, μ = 1.15, ρ = −0.05 · home 46% · draw 26% · away 28%01234560123456away goalshome goals7.3%7.1%4.4%1.7%0.5%0.1%0.0%9.8%12.6%6.9%2.6%0.8%0.2%0.0%8.1%9.3%5.3%2.0%0.6%0.1%0.0%4.2%4.8%2.8%1.1%0.3%0.1%0.0%1.6%1.9%1.1%0.4%0.1%0.0%0.0%0.5%0.6%0.3%0.1%0.0%0.0%0.0%0.1%0.1%0.1%0.0%0.0%0.0%0.0%Sum the cells45.9%home win: below the diagonal26.4%draw: the diagonal27.6%away win: above itover 2.5 goals 50.6%both teams score 54.4%
The whole model in one picture: every scoreline and its probability. Sum below the diagonal for the home win, along it for the draw, above it for the away win.
Check yourself

3 questions

explained after each answer