Our big favourites, checked again: now they win more often than we say
In September we checked our own predictions and found our biggest favourites falling short: teams we gave 70% or more had won 62.0% of the time, not the 78.2% we said. We have changed how the numbers are made since. So we ran the same check on the 2,013 club matches we called from 8 to 28 October, all with today's settings. The problem has turned round. Teams we gave 60% or more won 74.8% of the time; we had said 66.6%.
What the September check got wrong
The first check pooled every prediction from 19 July onwards. That mixed two models. Until 28 August we published the model's split as it came out. On 29 August we added a correction that pulls every match's three numbers a little towards an even split, because the model had been too sure of itself. Of the 187 favourites at 70% or more in the September study, 181 came from before that correction. So the headline figure described the summer model well, and the one running when the study was published hardly at all.
Split by setting, the record for favourites at 60% or more reads like this:
| Correction | Matches | We said | They won |
|---|---|---|---|
| None to 28 Aug | 397 | 70.4% | 57.7% |
| Strong 29 Aug–7 Oct | 34 | 64.4% | 79.4% |
| Today's 8–28 Oct | 238 | 66.6% | 74.8% |
Before the correction the model was too sure. With the strong correction it went the other way: its favourites won far more often than it said, although 34 matches is a small sample. On 8 October we eased the correction to less than half its strength, and set it out on our methodology page under Corrections. In between, on 22 September, the model itself changed: each league now gets its own scoring rate and home advantage.
Said against happened, since 8 October
| We said | Average | It happened | Calls |
|---|---|---|---|
| 0–10% | 7.5% | 5.6% | 36 |
| 10–20% | 16.0% | 10.4% | 403 |
| 20–30% | 25.6% | 24.7% | 2,548 |
| 30–40% | 33.8% | 34.2% | 1,447 |
| 40–50% | 44.1% | 44.4% | 918 |
| 50–60% | 53.8% | 57.7% | 449 |
| 60–70% | 63.3% | 75.1% | 169 |
| 70–80% | 73.2% | 72.9% | 59 |
Where the numbers hold
Most of what we publish sits between 20% and 60%, and there the record is close. Across those bands the widest gap between what we said and what happened is 3.9 points. Favourites we gave 40–60% (1,366 matches) won 48.8% against the 47.3% we said. The draw, across all 2,013 matches: 26.2% said, 25.0% drawn.
Where we are now too cautious
The gap has moved to the top of the range, and it is not spread evenly.
- Favourites at 60–70%: 169 matches, we said 63.3%, they won 75.1%.
- At 70% or more: 69 matches, we said 74.8%, they won 73.9%.
- In matches with a favourite at 60% or more we gave the draw 19.8% and it came up 18.1% of the time; we gave the underdog 13.5% and it won 7.1%.
So the clear favourites of 60–70% are where we undersell, and the very biggest are about right. Away from home, which was the weak spot in September, the record has recovered: big favourites playing away won 72.7% (44 matches, 65.5% said), close to the 75.3% of those at home (194 matches).
Europe and the rest
The three UEFA club competitions, the Champions League, the Europa League and the Conference League, gave us 28 big favourites. They won 75.0% against the 69.6% we said. The 210 in every other competition won 74.8% against 66.3%. On 28 matches the margin is about ±16 points, far too wide to say whether Europe behaves differently. It will need reading again anyway: on 29 October, after this check's cut-off, we changed how clubs from different leagues are compared in those competitions.
Should we change the correction?
The obvious answer is to weaken the correction again. We tested that before saying it. We took the earliest 1,409 of these matches, found the correction that would have fitted them best, and scored it on the 604 that came after, which the fit had not seen, against the setting we use. It made no difference the data can see. On the Brier score, which measures how far each probability landed from what happened, it was 1.5 thousandths of a point worse, with a margin of ±4.6. So we are leaving it as it is. A change that does not survive a test on matches it has not seen is noise, and a model that chases noise gets worse, not better. Part of the reason is that the correction acts on every match, and most matches sit in the middle bands, where the record already holds.
What it means for reading our numbers: a team we give 60% or more outside the UEFA competitions has, so far, won 9 points more often than the number says. That is a statement about the past 2,013 matches, not a promise about the next one.
How sure are these numbers?
The bands in the middle hold hundreds or thousands of calls each, so their margins are a few points. The top is thinner: the 60–70% band has 169 calls and a margin of about ±6.5 points, and the 70–80% band 59 calls and ±11.3. Even at the edge of its margin the 60–70% band stays above what we said, so that gap is not chance. Its size is the part to read loosely.
Three weeks is also a short window. We will run this check again once today's settings have a full month of matches behind them, and the track record updates every day in the meantime.
How this was counted
Every club match our model called before kick-off between 8 and 28 October 2026 and that has a final score, across all the competitions we cover; national-team matches are left out because their model is set up differently. Results at 90 minutes. Each match gives three probabilities, one per outcome, so the chart counts 6,039 calls. A favourite is the side with the higher of the two win chances. The margins are 95% intervals. Every prediction was logged before kick-off with the correction it was published with, and none has been changed since.
Published by Fulltime Stats, an independent site run by Leo Tempo. Who we are.