What options does each team have to win the World Cup? We have set up a statistical model to try to answer that question rigorously. The model works in two steps: first we measure the strength of each team using data (Elo ranking, squad value, etc.); then we simulated the tournament thousands of times. This way we can calculate the probabilities of each selection. This is our updated prediction:
Our model says that Spain is the favorite to win the 2026 World Cup, with a probability of 16% or 17%. And yet, the Spanish lose the tournament five out of six times. The other favorites are France (12%) and Argentina (12%), but between the three they do not win even half of the simulations. This paradox is simple statistics: it is difficult for us to accept uncertainty.
The model respects the nature of the game. Football is difficult to predict and a World Cup is even more difficult. The teams play few games, which obscures their level; and there will be five single-match qualifiers, where a goal can be decisive. For a strong team it is still very difficult to chain those five successes (80% × 80% × 70% × 65% × 65% = 16%). The model can point to favorites, but not a likely winner.
How the model works
Let’s start with a game. By clicking on the ball button below you can simulate a World Cup with our model. Each match and each match will be “played” until the champion of that round is found…
The trick to calculating probabilities is to repeat these simulations 100, 1,000, and even 100,000 times. You can also do it in the following interactive. If you spend some time doing simulations, you’ll see that the percentages converge towards our prediction. But it will take a while! More than one would say. But that’s chance.
What do our simulations depend on? Below we detail the methodology, but the essentials are simple. We have trained a model to predict each match based on the strength of the two rivals. We measure this strength with two main metrics: an Elo ranking by results and the value of the squads. We also take into account more factors: host (gives an advantage), age and league of your players (to adjust their value) and a historical Elo (regression to the mean).
How will each group look?
The advantage of having a statistical model is that you can predict many details. For example, the following table shows the probability that each team has of finishing in each position in its group.
The probable path of each selection?
Each team has a huge number of possible routes to the final, with many potential rivals in the round of 32, round of 16, quarters and semi-finals. But these crossings are not random and can be predicted. The following interactive tells which are the most likely rivals of a team on its way to the final.
Chance can play tricks on you. For example, at the beginning of the tournament, Spain had a 31% chance of meeting Argentina in the round of 32, which would be a very tough rival for the first round.
The irreducible chance
Predicting tournaments has become a genre. They publish their models, , amateurs and . We have been doing it since 2018, when we beat the Goldman Sachs and UBS models. Still, we have to admit that betting and prediction markets are hard to beat. You can beat them one year, like we did in 2022, but it’s rare to do it systematically. Some people have achieved it. In the 2000s, Tony Bloom and Matthew Benham—a mathematician and physicist—used models not unlike ours to bet against Asian houses. They did it first together and then separately, fighting. And the two earned enough money to buy their childhood clubs: Brighton and Brentford, both now in the Premier League.
The uncertainty of football cannot be reduced much further. At some point we collide with what we call chance. That cascade of unknown factors that together trigger reality. Why is a die unpredictable? Its physics is not mysterious. But it is impossible to accurately measure the speed of the cube, the angle of the shot, its geometry and the air around it.
Football is the same: the pressure of the ball, a shout from the public, the rest of each player, a pain, a thought of doubt. Each of these elements is negligible on its own, but added in thousands they decide the games. That is why football cannot be solved. That is why it leaves room for the fate of the young newcomer and the veteran who plays without a future. That is why we invoke the luck of the champion. That’s why we tell our friends to shut up before a penalty. And that’s why a World Cup catches us: because we know that in the end everything will be decided in moments of irreducible chance, in those abysses of the universe where the model gives up.
* * *
Methodology
Our predictions are the result of doing thousands of simulations of the tournament. In each match, the probability that one team or the other will win depends on their data. For example, if Spain plays against Germany on a neutral field, the probability of winning is 52% and the probability of losing is 21%.
The model has three parts:
1. Team strength metric. Here we use three metrics: their recent results (measured with a ), the quality of their players (measured with their value in euros, with data from the web), and a historical Elo (to predict regression).
2. Match simulator. We have trained a model with thousands of matches to, given two teams, their strength metrics, and their circumstances (example: playing at home) estimate how likely each outcome is. The model tells the probability of victory, draw and defeat; even that of each marker. For example, in a hypothetical duel between Argentina and Jordan, the most probable results are 2-0 and 3-0 with 15% each.
3. Complete World Cup simulator. Finally, what we do is simulate the tournament match by match. We repeat this 100,000 times to be able to estimate the probability of each event.
Frequently asked questions
So, do you say that Spain is going to win? No no. Our model says that Spain is the most likely team, but also that it has only one option in six to win. It is important to interpret this correctly: in reality, your victory is as unlikely as seeing a free throw miss.
These data show that a World Cup is difficult to predict. And it’s not a surprise. First, it is a tournament designed for luck to influence: it is not a regular league, it has no playoffs, nor home and away games. Second, national teams play few important games and their performance is more uncertain than that of a club. And third, we talk about football, an exciting sport because it is full of surprises. Almost no one would sit down to watch a game if the result was decided.
Have you done this before? Yes. We used a similar model in the and , in the and of 2024. The model has been demonstrated: the results to which we gave a probability between 0%-15% occurred 4% of the time; and those with probability 85%-100% occurred 94% of the time. In 2018, we did much better than chance, better than the FIFA ranking and two big banks (UBS and Goldman Sachs), although France won, which was only our sixth favorite. In 2022, the final was played by two of the three teams that we saw as the best; We even beat the bets.
Should I bet using your predictions? No. Our model is relatively sophisticated and can work well. But betting and prediction markets have proven in the past to be very difficult to improve. Furthermore, in order not to lose money, it is not enough to beat them; you must also compensate for the margin that bookmakers reserve when setting prices.
Statistical models are useful as a reference, for transparency and because they allow us to calculate details that the bets do not respond to. But, when it comes to getting it right, the bets use a hybrid approach: they combine their own models with the fine adjustment made by their experts, to consider the extra information they have (such as details of playing styles, states of form or injuries).
Are other predictions published? Yes. Every year there are more people making tournament predictions! They publish their own academic, hobbyist and business models. There are bets, prediction markets and tipster platforms.
The details for ‘nerds’
What is an Elo ranking. It is a metric that captures the strength of each team based on its results. Each team has a certain amount of points—their Elo points—and with each match there is an exchange. The winner takes points from the loser. If the victory is by surprise (because the weak team wins) the teams exchange more points. Elo rankings are increasingly used, in sports and video games, for example, to organize games between players of similar level. Our model uses the Elo ranking of the web.
Additionally, this year we have included a historical Elo or pedigree metric (the median ranking of the team in the last 10 years) as a variable in the model. The objective is to predict the regression to the mean of a selection. Imagine two teams with the same current Elo of 1900 points, one with a historical Elo of 2000 and another of 1800. The first wins more games.
Why we use the value of templates. The teams play few competitive matches and that makes it difficult to gauge their level before a big event. One way to add information to the model is to use the transfer market value of the players that make up the team. We take the data from the popular Transfermarkt website, adjusted by the age of the players (those over thirty become cheaper due to having fewer years of career ahead of them) and the country where they play (Premier footballers, for example, are more expensive).
How the party model works. We have adjusted a simple model that estimates the goals that each team will score based on half a dozen metrics. The main ones are the differences in strength: 1) if it is better or worse than its rival by Elo points, 2) if it has been better in Elo points in the past decade, and 3) if it is better by squad value. In addition, the model considers the circumstances: relevance of the tournament, home or away match, host status.
All of these variables have some effect. For example, the following graph shows the probability of a team winning a match based on its advantage over the opponent in Elo points and squad value.
We have chosen all the metrics because they have been shown to improve prediction (in a cross-validation, temporal and out-of-the-sample process). We tested more variables – the distance of each team from the venue, what each team is playing for in the match, the pedigree in tournaments – and we discarded them: none improved the prediction in the final phases.
The model is of the GAM-Poisson type. That is, it assumes that goals follow a Poisson-type distribution, which reasonably approximates reality, and which has been used in statistical models and .
We have also included a classic effect (): increase the probability of ties. And why do we use a model that predicts goals and not directly victories? It helps us to resolve the group stage and to predict overtime.
What success can we expect from the model. To train it, we have used a database of almost 19,000 national team matches since 2004, including hundreds of matches from World Cups and European Championships.
With this data, the model correctly guesses the result – victory, draw or defeat – of 59% of the matches: 60% when there is a local team and 55% on a neutral field, where there are fewer certainties. But the note that really matters to us is probabilistic: it does not reward “hitting the winner”, but rather saying 70% in things that happen 70% of the time. That score is the Ranked Probability Score (the lower the better, as explained or ): our model scores around 0.17 in all international matches and around 0.18 in the final phases of major tournaments, which are more unpredictable.
To develop this 2026 model we have used artificial intelligence. This version improves on previous World Cups in all evaluation metrics; not very much, but noticeably.
