Model v. bookmakers
How much of a Champions League knockout is luck?
Before the first leg of every knockout tie, a model worked out each team's chance of going through, using only matches played earlier. We call it the pre-match chance. Then we checked what actually happened.
Every tie since 2012-13, about to drop.
Even champions mostly got lucky
To lift the trophy, a champion has to win every tie on its way. Multiply its pre-match chances together and you get the odds of that whole run. Hover or tap a bar, or pick a champion below, to see its route.
Who'll win this season?
How the title odds have moved
| Club | Change (points) |
|---|
How this is worked out
One tie is a lottery. A season isn't.
Change one result, change the whole tournament
Pick the winner of any tie and watch every club's chance of lifting the trophy move. Team strengths stay as they were before the round of 16, so the only thing that changes is the results you pick.
Chance of winning the trophy
Comeback meter
Pick a first-leg score. How likely is your team to still go through? The model's answer sits next to what really happened in every two-legged tie that started the same way (Champions League knockouts since 2011-12, and Europa League knockouts from 2020-21 to 2024-25).
Model's chance after every first-leg score
The later the round, the closer the contest
In the round of 16, the favourite usually goes through. By the final it's not far off a coin toss. The hollow dot is how often the model expected its favourite to go through; the solid dot is how often it really did. There have only been finals, so each round is a small sample, and the line through the solid dot shows where the true rate probably lies.
Which season was the wildest?
An underdog is the side with the lower pre-match chance. For each season, the solid dot is how many underdogs actually went through and the hollow dot is how many the model expected. The number on the right is the gap. Click a season to see its ties.
Every knockout tie
Each bar splits the pre-match chances between the two teams. ✓ marks the team that went through. The scores under each bar are written in the same order as the team names. Put two clubs in the boxes below to see their whole history against each other.
Luck, or a blind spot? Real Madrid
How many ties each club won, against how many the model expected it to win (clubs with at least 8 ties). Most clubs sit close to what the model expected. One doesn't.
Real Madrid went through ties when the model expected . If the model were right, that would happen about once in tries. The model rates Real accurately in La Liga, but in Champions League knockout matches they take about a third of a point a match more than expected.
So part of Real Madrid's "luck" isn't luck. Either they lift their level for these matches, or something that matters here isn't visible in league results.
Is it just a lucky streak? Two checks
Where the extra wins came from
Each tie Real Madrid played, sorted by how it ended. Hollow dot: how many the model expected. Solid dot: what happened.
Did it last?
The same count for each half of the 14 seasons. A lucky run fades, while a real edge carries on. Hollow dot: how many ties the model expected Real Madrid to win. Solid dot: how many they won.
Is the model honest?
A luck study is only as good as its chances. So the model was tested on league matches from seasons it never saw (2017-18 onwards), against the bookmakers' final odds just before kick-off, which are the toughest benchmark there is.
When it says 60%, does it happen 60% of the time?
Home teams are grouped by the chance they were given, then we count how often they actually won. A point on the "honest" line means the chances were right.
How close does it get to the bookmakers?
0% is knowing nothing: giving every match the same average home, draw and away chances. 100% is matching the bookmakers.
The bookmakers win, as expected: their odds include team news, injuries and the betting market's money. The model gets most of the way there using nothing but past results.
The exact scores
| Model | Brier | Log loss |
|---|
Both are standard accuracy scores for probability forecasts; lower is better. Brier is the average squared gap between the chances given and what happened; log loss punishes confident mistakes harder.
Live: the strongest clubs right now
| # | Club | Would score | Would concede | Would win | Chance of winning, as a bar |
|---|
The prediction ledger
The real test: the model predicts upcoming league matches, and every prediction is saved before kick-off and never changed afterwards. Once a match is played, the prediction is scored against the result and against the bookmakers.
Upsets watch
Upcoming games where the model rates a result well above the bookmakers do. If the model is right, these are the results to watch for; if the bookmakers are, the model is about to learn something.
Coming up
The model's chances for each result: home win, draw, away win.