What are each national team’s chances of winning the World Cup? We’ve built a statistical model to try to answer that question rigorously. The model works in two steps: first we measure each team’s strength using data (Elo ranking, squad market value, etc.); then we simulate the tournament thousands of times. That lets us calculate each team’s probabilities. This is our updated prediction:
After the first matches, Spain, France and Argentina are the favorites to win the 2026 World Cup according to our model. The three of them have a probability of around 14% each — and yet together they don’t win even half of the simulations. This paradox is simple statistics: we struggle to accept the uncertainty.
The model respects the nature of the game. Soccer is hard to predict and a World Cup is even harder. National teams play few matches, which obscures their level; and there will be five single-elimination rounds, where one goal can be decisive. For a strong team it remains very difficult to string together those five wins (80% × 80% × 70% × 65% × 65% = 16%). The model can identify favorites, but not a likely winner.
How the model works
Let’s start with a game. By clicking the ball button below you can simulate a World Cup with our model. Each match and tie will be “played” until a champion is found for that run…
The trick to estimating probabilities is repeating these simulations 100, 1,000 and up to 100,000 times. You can also do that in the interactive tool below. If you spend some time running simulations, you’ll see the percentages converge toward our prediction. But it will take a while! Longer than you’d think. But that’s randomness.
What do our simulations depend on? We detail the methodology below, but the essentials are simple. We trained a model to predict each match based on the strength of the two opponents. We measure that strength with two main metrics: an ELO ranking based on results and squad market value. We also account for other factors: host status (which gives an advantage), players’ age and the league they play in (to adjust market value), and a historical ELO (regression to the average).
How will each group pan out?
The advantage of having a statistical model is that you can predict many details. For example, the following table shows the probability each national team has of finishing in each position in its group.
The likely path for each national team?
Each team has a huge number of possible routes to the final, with many potential opponents in the round of 32, round of 16, quarterfinals and semifinals. But those matchups are not random and can be predicted. The next interactive shows the most likely opponents for a team on its way to the final.
For instance, Spain and Argentina have a decent chance of meeting in the round of 32.
Irreducible chance
Predicting tournaments has become a trend. Experts, academics, fans, and companies publish their models. We’ve been doing it since 2018, when we outperformed the models from Goldman Sachs and UBS. Even so, we must acknowledge that betting markets and prediction markets are hard to beat. You can get the better of them sometimes, as we did in 2022, but it’s rare to do so consistently. Some people have managed it. In the 2000s, Tony Bloom and Matthew Benham — a mathematician and a physicist — used models not very different from ours to bet against Asian bookmakers. They did it together at first and then separately, after a falling out. Both made enough money to buy the clubs of their childhood: Brighton and Brentford, both now in the Premier League.
Soccer’s uncertainty cannot be reduced much further. At some point we run up against what we call chance. That cascade of unknown factors that together produce reality. Why is rolling a dice unpredictable? Its physics are not mysterious. But it’s impossible to measure exactly the cube’s initial speed, the angle of the throw, its geometry, and the surrounding air.
Soccer is the same: the pressure on the ball, a shout from the crowd, how well (or otherwise) a player has rested, a niggling injury, a moment of doubt… Each of those elements is negligible on its own, but aggregated by the thousands they decide matches. That is why soccer cannot be solved. That is why it leaves room for the fate of the newly arrived youngster and the veteran who plays with no future. That is why we invoke the champion’s luck. That is why we tell friends to shut up before a penalty. And that is why a World Cup grips us: because we know that in the end everything will be decided in instants of irreducible chance, in those abysses of the universe where the model gives up.
* * *
Methodology
Our predictions are the result of running thousands of tournament simulations. In each match, the probability that one team or the other wins depends on their data. For example, if Spain plays Germany on neutral ground, the probability of winning might be 52% and of losing 21%.
The model has three parts:
1. Team strength metric. Here we use three metrics: recent results (measured with an ELO ranking), the quality of their players (measured by their value in euros, with data from the Transfermarkt website), and a historical ELO (to predict regression).
2. Match simulator. We trained a model on thousands of matches so that, given two teams, their strength metrics and their circumstances (for example: home advantage), it estimates how likely each outcome is. The model gives the probability of a win, draw, and loss; even the probability of each scoreline. For example, in a hypothetical Argentina vs Jordan matchup, the most likely results are 2-0 and 3-0 at about 15% each.
3. Full World Cup simulator. Finally, we simulate the tournament match by match. We repeat this 100,000 times to estimate the probability of each event.
Frequently asked questions
So, are you saying Spain will win? No. Our model says Spain is the team with the highest probability, but it also says they only have about a one-in-six chance of winning. It’s important to interpret this correctly: in reality, Spain’s victory is as unlikely as watching a missed free kick.
These figures show that a World Cup is hard to predict. And that’s no surprise. First, it’s a tournament designed to let luck influence outcomes: it’s not a regular league, it has no playoffs or two-legged ties. Second, national teams play few important matches and their performance is more uncertain than a club’s. Third, we’re talking about soccer, a thrilling sport because it’s full of surprises. Almost no one would sit to watch a match if the result were already decided.
Have you done this before? Yes. We used a similar model in the 2018 and 2022 World Cups, and in the 2024 European Championship and Copa América. The model has proven well calibrated: results to which we assigned probabilities of between 0%–15% occurred 4% of the time; and those with probabilities of 85%–100% occurred 94% of the time. In 2018 we did much better than chance, better than the FIFA ranking and better than two large banks (UBS and Goldman Sachs), even though France won and was only our sixth favorite. In 2022, the final was contested by two of the three teams we rated highest; we even beat the betting market.
Should I bet using your forecasts? No. Our model is relatively sophisticated and can perform well. But betting markets and prediction markets have proven hard to improve on. Also, to avoid losing money it’s not enough to beat the market; you must also offset the margin bookmakers reserve when setting prices.
Statistical models are useful as a reference, for transparency, and because they allow us to calculate details that betting markets do not. But to actually get predictions right, bookmakers use a hybrid approach: they combine proprietary models with expert adjustments to incorporate extra information they have (such as playing styles, current form, or injuries).
Are other forecasts published? Yes. Every year more people publish tournament predictions — academics, fans, and companies. There are bets, prediction markets, and forecasters’ platforms.
Details for nerds
What is an ELO ranking? It’s a metric that captures each team’s strength based on results. Each team has a certain number of points — its ELO points — and with each match there is an exchange. The winner takes points from the loser. If the victory is an upset (because the weaker team wins) the teams exchange more points. ELO rankings work well and are increasingly used, in sports and video games, for example, to match players of similar level. Our model uses the ELO ranking from the Eloratings website. Additionally, this year we included an historical ELO or pedigree metric (the median ranking of the team over the past 10 years). The goal is to predict regression to the average for a national team. Imagine two teams with the same current ELO of 1900 points, one with a historical ELO of 2000, and the other 1800. The first wins more matches.
Why we use squad market value. National teams play few competitive matches and that makes measuring their level before a major tournament difficult. One way to add information to the model is to use the transfer-market value of the players in each squad. We take data from the popular website Transfermarkt, adjusted for players’ age (players over 30 are discounted because they have fewer playing years left) and the country where they play (players in the Premier League, for example, are more expensive).
How the match model works. We fitted a simple model that estimates the goals each team will score based on half a dozen metrics. The main ones are differences in strength: 1) whether it’s better or worse than its opponent by ELO points, 2) whether it has been better in ELO points over the past decade, and 3) whether it is stronger by squad market value. The model also considers circumstances: tournament relevance, home or away, host status.
All these variables have some effect. For example, the chart below shows the probability that a team wins a match as a function of its advantage over the opponent in ELO points and squad market value.
We selected all of these metrics because they have been shown to improve prediction accuracy (through cross-validation, temporal validation, and out-of-sample validation). We tested additional variables — such as the distance from each venue to the host city, what is at stake for each team in the match, and tournament performance — and ruled them out: none of them improved prediction accuracy in the final stages.
The model is a GAM-Poisson. That is, it assumes goals follow a Poisson distribution, which reasonably approximates reality and has been used in statistical models and academic studies.
We also included a classic effect (Dixon & Coles, 1997): increasing the probability of draws. And why do we use a model that predicts goals and not directly wins? It helps us resolve the group stage and predict extra time.
What accuracy can we expect from the model? To train it we used a database of almost 19,000 international matches since 2004, including hundreds of World Cup and European Championship fixtures.
With those data, the model predicts the outcome — win, draw or loss — of 59% of matches: 60% when there is a home team and 55% on neutral ground, where there is less certainty. But the score that really matters is probabilistic: it does not reward “picking the winner,” but assigning 70% to events that occur 70% of the time. That metric is the Ranked Probability Score (the lower the better, as explained here or here): our model scores around 0.17 across international matches and about 0.18 in final stages of major tournaments, which are more unpredictable.
To develop this 2026 model we used artificial intelligence. This version improves on previous World Cup models in all evaluation metrics; not by a huge amount, but noticeably.
Sign up for our weekly newsletter to get more English-language news coverage from EL PAÍS USA Edition
