iTranslated by AI
Japan Has a 1.4% Chance to Win: Insights from 100,000 Simulations of the 2026 World Cup
What I built
A web app that simulates the 2026 World Cup tournament. You can choose countries of interest from the 48 participants and either run a single simulation to view a bracket with scores for each match or calculate the probability of winning by running 100,000 Monte Carlo simulations.
Main Features
- Toggle between two types of ratings: Elo and Opta
- Match score generation using Poisson distribution
- Realistic simulation including extra time and penalty shootouts
- Official FIFA 2026 bracket structure
- Available in 10 languages
- Displays win probability for all 48 countries
Tech Stack
- Vanilla HTML/CSS/JS (no frameworks used)
- No build steps, 100% client-side
- Deployed on Netlify
Design Decisions
Decoupling atk/def (Attacking/Defensive strength) from Win/Loss calculation
In my initial implementation, the goal/conceded data (atk/def) from the last 20 matches was fed into λ (expected goals), and the winner was determined by the magnitude of the scores generated by the Poisson distribution.
However, this meant that high-scoring games in the qualifiers skewed the win probability predictions for the final tournament. This caused teams like Norway or New Zealand, who scored heavily in qualifying, to be significantly overvalued in the main event.
After the revision:
- Outcomes are determined solely by ratings (Elo or Opta).
- Once the result is decided, atk/def is used to generate a realistic-looking score.
This separation allowed for both a "rating-pure win probability" and "realistic-feeling scores."
Tuning Draw, AET, and Penalty Rates
Initially, the draw rate was only around 11% (whereas in real World Cups, it is 20-25%). I adjusted the thresholds and probabilities to align with the following values:
- Group stage draw rate: 20%
- Knockout stage AET (Extra Time) rate: 15%
- Knockout stage Penalty shootout rate: 7%
Replicating the FIFA 2026 Bracket
The tricky part was that the Quarter-Final matchups are not symmetrical. Since the winners of the Round of 16 cross paths in the Quarter-Finals, the bracket display cannot be a simple left-right tree.
I implemented this after reviewing the official documents multiple times.
Results of 100,000 Simulations
Win probabilities in Elo mode (top teams):
- Spain: 28%
- Argentina: 18%
- France: 14%
- England: 7%
- Brazil: 4%
Ranking changes in Opta mode:
- Spain: 24%
- Argentina: 18%
- France: 17%
- England: 9%
- Brazil: 6%
In Opta mode, Spain's lead narrows, while France and England show a tendency to rise.
Japan's Figures
Japan's probability of winning is 1.35% (Elo mode, 13th out of 48 countries). This drops to 0.75% in Opta mode.
Breakdown of reaching stages (Elo mode):
- Round of 16 or better: 49%
- Quarter-finals or better: 27%
- Semi-finals or better: 12%
- Winner: 1.35%
Since Japan has reached the Round of 16 in the past, as a fan of the Japanese national team, I am curious about the probability of breaking through the barrier of the Quarter-finals. Whether a 27% chance of reaching the Quarter-finals or better feels high or low is subjective.
The near twofold difference between Elo and Opta stems from their different evaluation criteria. Opta tends to reflect more recent strength, while Japan generally receives a higher evaluation in Elo.
Give it a Try
Please try setting your country to "Focus." With more than 50 days until the World Cup kicks off, it is a fun way to make predictions before the tournament starts.
Sequel: Revision Records up to v2.6
After publishing this article, I implemented 6 stages of revisions to statistically address parts that felt "off" after actually running the simulations. I have compiled a record of how I reached v2.6, including a Poisson mixture model, λ damping, and handling rejection sampling for AET, in a separate article:
Discussion