2026 FIFA World Cup Prediction
Introduction
The 2026 FIFA World Cup is special for a number of reasons. It is newly expanded to 48 teams, making it the largest tournament ever. It is the first time three nations are jointly hosting the event. For African viewers, it features the most African teams to ever play at a World Cup (although it is disappointing that my nation, Nigeria, failed to qualify and we will miss seeing their iconic and stylish jerseys as we usually do).
So, I wanted to do something challenging yet fun by building a model and simulating the tournament to make projections. This is something I have never done before, and given how complex it was, I definitely learned a lot through the process.
Tournament Projections
Before getting into the data and the model, here is what my model projects based on 10,000 simulations.
Here are a few major points from the simulation results:
- My model gives Spain a 23% chance to win the tournament, which is significantly higher than other models I have seen online. England sits in second at 12.2% , followed by France at 11% , Germany at 6.6% , and Argentina at 5.8%.
- Looking at the host nations, Canada advances from the group stage the most at 87% of the time. Mexico advances roughly 82.2% of the time , while the United States advances in 65% of the simulations.
- Morocco is the highest projecting African team with a 2.5% chance to win the tournament. Senegal and Algeria also project well with a 2.2% and 2% chance to win, respectively.
- Group D appears to be one of the most even groups in the tournament. Below Turkey, the odds to win the group are nearly flat between the United States at 21.4% , Paraguay at 20.4% , and Australia at 18.4%.
- Brazil shows up much lower than expected, sitting at just a 1.7% chance to win the tournament. The model actually ranks three African nations above them.
Data
The men’s World Cup tournament began in 1930, but for this model I decided to use historical data only as far back as 1994 for two reasons:
- The farther back I went, I noticed the data quality kept getting worse.
- A simple Google search shows that the modern era of football (soccer) is generally agreed to have begun in 1992. We can all agree that football played in the 1930s is not the same as football played nowadays.
I pulled data from multiple sources, including Wikipedia pages, World Football Elo Ratings, The Fjelstul World Cup Database, and Football datasets.
I wanted to go beyond just using Elo ratings. While Elo gives a great holistic view of a team’s performance history, I wanted to account for the actual player quality in the selected 2026 squads. I engineered several features to capture this, including the number of squad players in Europe’s top 5 leagues, average age, average caps, number of players in Ballon d’Or rankings, average World Cup appearances, etc.
One important engineered feature was “Elo-adjusted goals,” which served as a makeshift offensive and defensive rating. It tracks form over a team’s last seven games, rewarding weaker teams for scoring against stronger opponents and punishing stronger teams for conceding to weaker opponents. Using 1500 as an “average” opponent Elo, the math looks like this:
\text{Offensive Rating} = \text{Goals Scored} \times \left( \frac{\text{Opponent Elo}}{1500} \right)
\text{Defensive Rating} = \text{Goals Against} \times \left( \frac{1500}{\text{Opponent Elo}} \right)
For example, England (Elo: 2031) beat San Marino (Elo: 793) 10-0 in November 2021. England’s offensive rating would be calculated as 5.29. In another example, Ivory Coast (Elo: 1676) upset France (Elo: 2081) 2-1 in June 2026. Ivory Coast’s offensive rating for that match would be 2.77, while France’s defensive rating would be 0.89. A higher offensive rating is better, and a lower defensive rating is better.
After testing, the only features that proved statistically significant and made the final model were Elo, average squad age, average squad caps, offensive rating, and defensive rating.
Model
I used a Poisson Regression model for the predictions. Instead of simply predicting a Win/Draw/Loss outcome, the Poisson model calculates the expected goals (xG) for both teams based on their matchup.
Using the features, the model calculates a goal probability to each team. I then ran 10,000 Monte Carlo simulations of the entire tournament structure, updating Elo ratings after every match and applying FIFA’s group stage tiebreaker rules, to arrive at the final projections.
Limitations & External Factors
While the model uses historical data and current form, it cannot perfectly account for sudden player injuries or a star missing the tournament. It also does not factor in specific tactical matchups, mid-game managerial adjustments, or squad chemistry.
Also, here are several unique external factors impacting the 2026 tournament. First, it will be interesting to see how players adapt to the extremely hot temperatures across North America, as teams have already reported having a hard time training in the heat.
Additionally, U.S. visa policies may negatively affect certain squads. There has already been news of hostility towards specific nations as they land in the U.S., including airport detentions and visa denials. One major example is the Iranian national team. According to ESPN, they had to move their training base from Arizona to Tijuana, Mexico. All three of Iran’s group stage games are held in the U.S., meaning they face increased travel distance and a depleted staff, as 14 staff members were denied entry.
Finally, reports suggest that the high altitude in Mexico City could cause physical problems for teams playing matches there, which is a massive variable that historical data cannot easily predict.
Contact Information
I’m open questions, and would appreciate any comments or feedback. Feel free to reach out.
Code Availability
Code and data available on GitHub