Dr Peter Woods
← Work
Prediction Matrix · World Cup 2026

Match Model

A prediction model for the 2026 World Cup that earns its keep only if it can out-read the market. Every match-day, its probabilities are placed side by side with bookmaker odds — and the gap is tracked honestly, win or lose.

2026
World Cup
Poisson
Core model
1X2 · O/U
Markets
Match-day
Benchmark
Data pipeline
01
Ingest
Fixtures, results, xG feeds
02
Feature build
Team-strength & form vectors
03
Poisson core
Expected goals → scoreline grid
04
Benchmark
Model vs implied book odds
Model vs market

Decimal odds for the modelled outcome against the bookmaker line. A positive edge means the model rates the outcome more likely than the market implies.

StageFixtureModelBookEdgeResult
Group AMexico v Croatia2.052.20+0.15
Group DUSA v Wales1.921.85−0.07
Group ESpain v Japan1.581.66+0.08
Group GBrazil v Serbia1.491.55+0.06
Round of 16England v Senegal1.741.80+0.06
Quarter-finalFrance v Argentina2.382.30−0.08

Illustrative sample · full tournament ledger tracked live.

Feature engineering

Poisson goal model

Each side's expected goals are modelled as independent Poisson processes; the outer product gives a full scoreline probability grid, from which 1X2, over/under and correct-score markets are derived.

Team-strength vectors

Rolling attack and defence strengths are encoded as decayed form vectors, weighting recent matches more heavily and adjusting for opponent quality rather than raw results.

Tournament effects

Group and knockout structure, rest days and host advantage are modelled explicitly — a neutral-venue tournament behaves differently from a league season.

Calibration

Predicted probabilities are calibrated against historical outcomes, so a stated 60% genuinely resolves true roughly 60% of the time.