← Back to analysis

Before a ball was kicked: what Golmetria's model projected for the 2026 World Cup

We published our forecast on June 6, before kickoff. The four teams at the top of our estimate turned out to be exactly the four semifinalists — a reminder that a model doesn't predict the future, it quantifies it.

Every statistical model is, at heart, an estimate wrapped in uncertainty. It doesn't promise the future — it assigns probabilities to every possible outcome from the information available. So the honest question after a tournament isn't "did you call the champion?" but "were your probabilities well calibrated?".

On June 6, five days before kickoff, we published Golmetria's projection for the 2026 World Cup. These were the four teams at the top of our estimate:

What happened on the pitch: Spain were champions, Argentina took second, England finished third and France fourth.

In other words: the four teams our estimate placed at the top were exactly the four semifinalists of the tournament. We called the champion and the runner-up in the right order. The only difference was between third and fourth — the model had France a shade ahead of England (9.5% vs 6.9% to win the title), and it was England who prevailed in the third-place match. Two teams effectively tied in our estimate, separated by a single game: that is precisely the kind of outcome a well-calibrated uncertainty anticipates. We didn't miss the teams; the coin landed the other way in a contest we already projected as a toss-up.

How we get there

The projection comes from an econometric model with a few layers. First, a measure of each team's strength built from real results and expected goals (xG) — not just who won, but how they played — anchored also to squad quality. From that strength we derive the probability of every possible scoreline in each fixture, with a Dixon-Coles adjustment that corrects the correlation typical of football's low scores. Finally, we simulate the entire tournament 20,000 times, respecting the bracket, to turn per-match probabilities into probabilities of the title, the final and each round. The model is calibrated against historical seasons, so that an "18%" actually means something close to 18%.

None of this is fortune-telling. It's estimation — with an explicit margin of error. The value isn't in nailing a single result; it's in assigning honest probabilities and letting them be checked afterwards. This time, they held up well. Getting the top four teams right before a ball is kicked isn't common, precisely because evenly matched sides are decided by fine margins — which is why we're pleased, but cautious: next time, the margin may fall the other way.

What's next

The same model, with the same methodology, is already running on the Brazilian Championship, which resumes now. The Série A projections live in our Índice, and we'll track them round by round with the same yardstick: transparent probabilities, verifiable afterwards. Stay tuned.