Momus
Blog

How Momus predicts football matches: the Dixon-Coles model, explained

Momus prices every football match with a self-fitted Dixon-Coles model. Here's how it turns team ratings into fair 1X2 odds — and why it blends the result with the sharp book.

August 1, 2026 · 6 min read

Before Momus can find value in a football market, it needs its own honest answer to one question: how likely is each result? It gets that from a self-fitted Dixon-Coles model — the same class of statistical model professional quants use to price football.

What the Dixon-Coles model is

At its core it models the goals each team scores as a Poisson process driven by two strengths per team — an attack rating and a defence rating — plus a home-advantage term. Dixon and Coles' refinement adds a correction for low-scoring games (0-0, 1-0, 1-1), where the naive model is slightly off. The output is a probability for every possible scoreline.

Momus fits those attack and defence ratings for every team from years of results, weighting recent matches more heavily (a time decay), so a team's rating reflects how good it is now, not three seasons ago.

From ratings to a fair line

Given two teams' ratings and home advantage, the model produces the expected goals for each side, then the probability of every scoreline. Sum the scorelines where the home team wins, draws, or loses and you get a clean fair 1X2 line — home / draw / away. The same score grid also gives over/under and both-teams-to-score probabilities for free.

Why Momus blends it with the book

Here's the honest part most 'AI tipster' pages skip: a goals model on its own does not reliably beat a sharp bookmaker. The market prices in team news, motivation and money the model can't see. So Momus de-vigs the sharp book and blends it with the model, and measures its edge against that blended fair value — never against the model alone. The model's job is to be an independent second opinion that catches what the market misprices, not to replace it.

Where the model is weak — and how Momus handles it

Thin data breaks any football model: a just-promoted side or the opening weeks of a season give unreliable ratings. When the model isn't confident, Momus leans on the book and its read instead, and treats the call as higher-variance. That restraint is the point — a model that knows when to defer is worth more than one that's always sure.

See the method turned into bets on the track record, or read how Momus finds value on Polymarket.