Forecast Models — wearefamilAI · World Cup 2026
FORECAST MODELS  ·  FIFA WORLD CUP 2026

See our Predictive Models here. World Cup 2026.

We publicly show and disclose the predictive model methodology that we use, which informs any of the content we generate. It gets regularly updated — and will always be public. We present the data and the opinions; players decide which way to act.

Last refreshed 23 April 2026
Simulations run 100,000
Teams modelled 48
Model benchmark Brier 0.172 · beats bookmakers
01 · Power Rankings

Who's actually the best team in the world, right now?

A strength index blending long-run Elo rating with the attack/defence profile from the match model. Higher is better. Argentina still lead the board — but the gap to Spain and Brazil is tight enough that one tournament shakes it.

01 Argentina 2.39 02 Spain 2.27 03 Brazil 2.12 04 Colombia 2.02 05 England 1.92 06 France 1.93 07 Uruguay 1.90 08 Portugal 1.86 09 Japan 1.83 10 Germany 1.77 11 Ecuador 1.71 12 Morocco 1.66 13 Netherlands 1.61 14 Italy 1.49 15 Belgium 1.48 16 Mexico 1.40 17 USA 1.38 18 Denmark 1.39 19 Australia 1.38 20 Croatia 1.38 0 1.2 2.4 STRENGTH INDEX (HIGHER = STRONGER)
The top cluster
The CONMEBOL trio plus Spain form a clear top four. Argentina (2.39) edge it, but Spain and Brazil are within the margin where draws and penalties decide things.
Hero team check
Portugal sit 8th in the world by the model — elite, but not elite enough to close the 14/1 price. See the Value section below.
Who's missing from the top 10?
No Netherlands, no Italy, no Belgium. The model rewards recent form and international fixtures, and the Oranje are paying for a patchy 24 months.
Dark horses
Ecuador (11) and Morocco (12) sit just outside the top 10 — both priced far longer than that by bookmakers.
02 · Match Predictions

One match at a time.

The model converts each team's attack and defence into a goal-expectancy for the 90 minutes, then simulates the scoreline. Here's how it currently reads a handful of the fixtures that matter.

Portugal vs Spain
Heavyweight · neutral
0–1
24.1%
Portugal
29.6%
Draw
46.2%
Spain
Over 2.5
37.6%
BTTS
43.5%
xG
0.88 / 1.32
Argentina vs Brazil
South America's big one · Argentina home
1–0
50.9%
Argentina
28.0%
Draw
21.1%
Brazil
Over 2.5
40.4%
BTTS
44.4%
xG
1.46 / 0.84
England vs Germany
Classic rivalry · neutral
1–1
40.6%
England
27.8%
Draw
31.5%
Germany
Over 2.5
47.3%
BTTS
52.6%
xG
1.38 / 1.19
Japan vs South Korea
AFC top clash · neutral
1–1
54.2%
Japan
25.2%
Draw
20.5%
South Korea
Over 2.5
49.8%
BTTS
51.1%
xG
1.70 / 0.96
How to read this
The bar above each match is the model's three-way probability (home · draw · away). xG is the expected goals each side creates across 90 minutes. BTTS means both teams to score. The scoreline badge is the single most-likely scoreline — not the most-likely outcome.
03 · Tournament Outlook

100,000 simulated World Cups. Here's how they finished.

We played out the full 2026 bracket a hundred thousand times — every group match, every knockout, every penalty shootout. The percentages below are the share of those tournaments each team actually lifted the trophy.

Top 10 · Probability to win

out of 100,000
Argentina 18.5% Spain 15.0% Brazil 9.9% Colombia 7.4% England 5.8% France 5.8% Uruguay 5.3% Portugal 5.1% Japan 4.2% Germany 3.5% Other 38 teams: 19.5% combined

Confederation split

CONMEBOL has only 6 teams at this World Cup. UEFA has 16. The model still gives them roughly equal odds of producing the winner.

CONMEBOL
44.3%
UEFA
43.3%
AFC
6.9%
CAF
4.4%
CONCACAF
1.1%
OFC
0.3%
Hero team check

Portugal's real path through the bracket

Ranked #8 by the model
5.1%
To lift the trophy
19.6%
To reach the semis
19.6
Fair odds (model)

Bookmakers currently price Portugal around 14/1 (6.7% implied). The model sits lower — closer to 19/1. Not far off, but the wrong side of the line. More on this below.

04 · The Value Board

Where the model sees edge.

The question that actually matters: where does the model think bookmakers have mispriced the tournament? Positive expected value (EV) means the model believes the odds are generous versus its own win probability. Three names stand out.

10
Teams with positive EV
2.8%
Suggested portfolio stake (quarter-Kelly, 5% cap)
+3.4%
Expected return on bankroll
01 · HIGHEST EV
Ecuador
+259%
Expected value per unit staked
Model
2.85% [1.2–5.3%]
Market
0.79% · 125/1
02
Australia
+207%
Expected value per unit staked
Model
1.02% [0.2–2.2%]
Market
0.33% · 300/1
03
Japan
+136%
Expected value per unit staked
Model
4.63% [1.5–7.8%]
Market
1.96% · 50/1
AVOID
Portugal
The model likes Portugal — just not at the current price. Bookmakers are quoting 14/1 (5.88% implied); the model puts them at 5.12%. That's a small gap, but it's the wrong way round. Nothing to bet here.
Model 5.12%
Market 5.88%
Edge −0.8pp
F
A note on staking. The 2.8% portfolio figure uses quarter-Kelly with a 5% cap per bet — deliberately conservative. Models have uncertainty; quarter-Kelly halves the damage when the model's wrong and still captures most of the upside when it's right. Nothing on this page is a recommendation to bet; the figures are honest estimates of what the model thinks is priced well and poorly. Gamble responsibly.
05 · How it works

Four models, stacked.

Every number on this page comes out of the same four-stage pipeline. It's not magic. It's a ratings system, a goals model, a tournament simulator, and a value filter — each doing one job, in plain sight.

STEP 01
Elo Power Ratings
We rate every team by the history of their results against each other — wins, losses, margin, venue, and how long ago the match was. It's the same family of ratings used in chess, adapted for international football.
Brier 0.172 · 7% better than the bookmaker benchmark
Calibration on held-out matches
STEP 02
Dixon-Coles Goal Model
The ratings tell us who's better, but not how many goals. This stage turns each team's attack and defence into an expected-goals number for a given match, and uses it to build the full scoreline probability grid — which gives us over/under, BTTS, and fair odds.
Draw calibration 23.6% model · 22.9% actual
Sharper on draws than the Elo baseline
STEP 03
Tournament Simulator
For the World Cup outlook, we play the whole tournament out a hundred thousand times — every group match, every knockout round, every penalty shootout. The "probability to win" for each team is simply the share of those simulations they finished holding the trophy.
100,000 trials · 42 seconds
Full bracket with uncertainty propagated
STEP 04
Bayesian Value Filter
The model knows it might be wrong. We resample the ratings 40 times and re-run the tournament, giving every probability a credible range instead of a single number. Kelly staking then tells us how much (if anything) a probability edge is actually worth at current market prices.
Ensemble of DC + Elo · honest uncertainty
Picks disappear when the edge isn't real
See what Bob, Franklin and Shirley make of the numbers.
The model gives you probabilities. The Wright family gives you the argument. Cold data, warm takes — published alongside every refresh.
Meet the family →