2026 World Cup Forecast

Who Wins The 2026 World Cup?

Phew, we're back. In this post we'll present a model that uses soccer strength estimates and Monte Carlo simulations to predict the chances of each team winning the tournament.

Stylized illustration of a young footballer with a ball standing before a giant opponent

As mentioned before, I'm a huge fan of FiveThirtyEight and took massive inspiration from their work. For years they have been my source of truth when it comes to sports analytics. As a final tribute to their work, we've built a model that uses a similar methodology to predict the chances of each team winning the tournament.

As part of the model planning process we dug a bit deeper into Elo, squad market values, predicted lineups, and venue conditions. We decided to stick with a strength based model and combined the modified strength with a Poisson scoreline model to predict the final outcome of each match and by virtue a path to victory for each contender. The methodology and some of the math used is explained below.

Standings

Here are the current standings according to our model using the strength estimates, Monte Carlo simulations and a Poisson scoreline model. To get our numbers we ran the tournament 100,000 times, summed up the outcomes for each game and followed each team through the tournament.

Forecast from before the tournament 50,000 simulations · ratings as of June 9, 2026
Group
Team rating Chance of finishing
group stage in …		 Knockout stage chances

Design inspiration from FiveThirtyEight. We miss you 🖤.SPI is the share of points our model expects a team to take against a field-average opponent (100 = take all of them). OFF and DEF are expected goals scored and conceded per match against that same average team. In this model they are two views of one strength number (see how this works). 3rd✪ is the chance of finishing third and advancing as one of the eight best third-placed teams. ★ marks tournament hosts. Sort any column by clicking its header.

Match by match

Every fixture, every scoreline

The same Poisson model that drives the simulations gives an exact probability for every scoreline. Pick a group-stage fixture (or any hypothetical pairing at any venue) and read the full grid. Probabilities are for the 90-minute result.

vs
At a glance

The most likely group stage, match by match

The most likely scorelines for all 72 group games. Hover any line for its probability, and see the full scoreline matrix for every match in the Matches tab.

Bracket builder

Build your own road to the final

Pick the group finishers, then click teams through the knockout rounds, finally explore the most likely path for your team to win the tournament. The model fills in everything you leave open and recomputes the whole bracket as you change your mind, with ratings that move on your picks, just like our hot simulations.

1 Pick the group finishers: first, second and third in each group

Prefilled with the model’s favorites. The small numbers are each team’s model chance of that finish (for third place: finishing third and ranking among the eight best thirds).

2 Click a team to advance it to the next match

Percentages are each side’s chance of winning that specific matchup (90 minutes plus penalties) at its real venue and date. Just like our hot simulations, every result moves ratings: an upset winner carries extra Elo into its next round (the expected eloratings.net update for that result, K = 60), so beating France makes you stronger against England. Each time you change a pick, the model recomputes the whole bracket under that new reality. Only the picks you clicked yourself stay pinned.

3 The most likely road to the title

Pick a team and we filter the 100,000 simulated tournaments down to the ones it actually won, then bundle those title runs into storylines: a way out of the group plus the knockout matchup that defines the road (the highlighted card). The other cards show the most common opponent at each remaining round.

Features

Candidate features for which we had historic data had to beat a leave-one-tournament-out no-harm gate before shipping. Attack/defense style splits failed that gate, which is why OFF and DEF in the standings are mirror images of one rating rather than independent dimensions, and we say so rather than fake it.

On top of fitted coefficients we also wanted to have some fun with contextual information that might impact odds. Four context effects enter as stated priors, because no historical summer World Cup in this region exists to fit them on:

From team strength to a match forecast

Okay, this is going to be a bit mathy. For anyone who wants to dig a bit deeper, I'm trying to spill all the beans here. Every match is scored by a Poisson model fitted on games of the 2014–2022 World Cups: goals ~ Poisson(exp(μ + β1·ΔElo/400 + β2·ΔMV)). Team strength starts from eloratings.net ratings on the eve of the tournament and adds feature-based adjustments:

There are roughly three steps to map two teams that are playing against each other to a potential scoreline:

  1. Compute the strength of each team on that day, given the opponent.
  2. Turn each strength value into expected goals (based on historic context).
  3. Calculate the probability of scoring 0, 1, 2, 3, ... goals.
Step 1: Computing the strength of each team

We’ll make this as illustrative as possible by calculating every step with the opening game, Mexico vs. South Africa. We define strength from these components:

Step 2: Turning strength into goals

All of those components combine into a single line. First, the generic shape of the formula:

…and then the same line, filled with the opener’s numbers from step 1:

Step 3: Calculate the probability of scoring 0, 1, 2, 3, 4, 5, 6… goals

Goal scoring in soccer is well described by a Poisson process, essentially a way to model random events that occur at a known rate. Each team’s expected-goals number from step 2 becomes a full distribution over how many goals it might actually score. Unlike FiveThirtyEight, we don’t bump the draw cells by hand: our simulated draw rate already comes out in line with World Cup history (slightly above it, if anything), so an artificial boost would make things worse.

Where SPI, OFF and DEF come from

The three rating columns in the standings table are not extra inputs, they are the formula above run one more time against a fixed reference opponent: the average team of this World Cup field. Picture every side playing that average team once, on neutral ground, venue effects off. OFF is the goals it would score in that game, DEF the goals it would concede (lower is better), and SPI converts the two into expected points per match, scaled to 0–100. One number for attack, one for defense, one power ranking, all in goal units instead of abstract rating points.

Because both numbers are read off the same strength rating, OFF and DEF mirror each other by construction: multiply them and you get the same constant for every team. One strength dial, read out twice, as the feature-gate note above explains.

References

As part of this process I went way to deep into the rabbit hole of soccer analytics. Just a quick shoutout to some of the sources that helped me get here.

Data

Methods & literature