Is PSG the Champions League favourite? Our two models rank it 1st and 10th

Two models replayed the same Champions League, 50,000 times each, league phase and knockout bracket alike, from the same settled ground: the 18 matches played taken as they stand, only the 126 remaining ones projected, and not a single point separating them on what has been played. Any disagreement therefore comes from the strengths, never from the data.

Across 36 clubs they agree almost everywhere. On exactly one, the first ranks Paris Saint Germain 1st with a 24.5% title chance, and the second 10th with 2.2%.

The cause is not "one reads squads, the other reads results". It is more precise, and it is the whole subject: both correct for the level of the league a club plays in, but one does it club by club and the other player by player — and only the second can tell an exception apart from the average around it.

The problem both models have to solve

Comparing a French club to an English one requires a common scale: an index measured at home is worth nothing elsewhere until it has been brought onto one.

The first reads squads: it rates players one by one and brings them onto the scale through each league's median Elo. The second, Dixon-Coles, knows nothing about players and reads results — goals scored, goals conceded, quality of opposition — from which it derives an attack index and a defence index per club, brought onto the European scale as well.

Both produce the same thing: the probability of reaching each round, not of winning it. For the league phase, the top eight places, the play-off and elimination sum to 1 — PSG's 81.2% for the quarter-finals in the player model is a probability of getting there, not of coming out on top.

First, they agree

It has to be established first: two models that diverge everywhere diverge nowhere in particular. As of 26 September 2026, their two rankings correlate at 0.944. The mean absolute gap between ranks is 2.1 places, and 22 of the 36 clubs are ranked within three places of each other. On each side, three clubs are enough to hold half of the title chances — and two of those three are the same, Arsenal and FC Bayern München.

Paris Saint Germain		24.5%
Arsenal		20.0%
FC Bayern München		19.5%
Manchester City		11.2%
FC Barcelona		9.5%
Liverpool		5.5%

Paris Saint Germain		2.2%
Arsenal		25.3%
FC Bayern München		18.0%
Manchester City		20.3%
FC Barcelona		9.4%
Liverpool		2.9%

The two charts look alike, except on their first line: FC Barcelona moves from 9.5% to 9.4%, Manchester City from 11.2% to 20.3% — the noise of two methods measuring the same thing differently. PSG, for its part, changes sides.

Except on one club

On the probability of lifting the trophy, the player model gives PSG eleven times more than Dixon-Coles. And this is not the ordinary spread of two methods: the second largest disagreement on the top eight places is worth 25.5 points, on Liverpool — half of the Paris gap. PSG is not the peak of a distribution, it has left it.

A squad judged by its league's average

The mechanism does not concern one line of the team but the whole squad.

To bring a club onto the European scale, Dixon-Coles applies one single coefficient per league — a constant, not a measurement: identical to the sixth decimal place for the five English clubs entered as for the three French ones, it carries no information about the club it is applied to and moves every club of a country with the same gesture.

Here, side by side, are the two measures of league level our two models actually use, for the eleven leagues an entered club represents. The median Elo is the player model's, the very quantity that enters its correction; the coefficients are Dixon-Coles's. Two distinct scales — an Elo is an absolute level, a coefficient a multiplier around 1 — which do not divide into one another and can only be compared by the order they produce. The higher the attack coefficient, the stronger the league.

The two measures agree broadly: the same league on top, the same border between the five big ones and the six others, and no league more than 2 places apart from one ranking to the other.

Except where it counts. On median Elo, Serie A, Ligue 1 and La Liga sit within five points — indistinguishable. The attack coefficient, meanwhile, lifts La Liga 11 % more than Ligue 1. Two leagues the player model's measure judges equivalent, Dixon-Coles's constant separates by a tenth: the mechanism of the disagreement over PSG is right there, on two lines of a table.

And it is the bottom of the table that gives the measure of the top: between the Premier League and the Austrian Bundesliga there is a factor of 2.09 on attack. The same goal is not worth the same thing depending on the country it was scored in — that is what "paying your league's average" means.

Three notes. In defence a coefficient below 1 improves the club: the lower, the better. The coefficient does not depend on how many clubs are entered — the Pro League's, with a single one in the competition, is neither better nor worse estimated than the Premier League's, which has five. And the columns do not rank the eleven in the same order: Eliteserien is ninth in attack and last in defence, so "a strong league" is not a single quantity and we compute no overall level score. (Of the 36 entrants, 5 clubs are measured on no domestic league at all.)

Ligue 1 carries the lowest attack coefficient of the five big leagues, 1.124, within a thousandth of Serie A. It penalises PSG in both directions:

• its attack is lifted 1.25 times less than that of an English club with the same raw attack;
• every defensive gap between a French club and an English one is enlarged by 40%.

It is the same constant acting twice: an exceptional squad in a less lifted league is undervalued throughout, not only where it shows best.

The defence, because that is where it shows best

Let us start with the figure that suits us least. On the raw measurement, before any correction, the PSG defence stands at 0.8235 and Arsenal's at 0.6246: Arsenal really is better, by a factor of 1.318. We are not arguing with that.

After correction, those same two defences stand at 0.7287 and 0.3955 — a factor of 1.842 where the raw measurement gave 1.318.

Arsenal's defence is better, and that is measured on the pitch; but the size of the gap you read has been enlarged by two fifths by the league, not by how the two sides played.

Arsenal		0.40
Manchester City		0.46
Liverpool		0.69
Manchester United		0.69
Paris Saint Germain		0.73

The four English clubs shown here all come ahead of PSG, and the last two only narrowly. What that order owes to the coefficient and what it owes to the football cannot be read off the chart: between clubs from two different countries, it is in part a modelling choice.

The player model does exactly the opposite

It does not estimate its correction on everyone, but on players who have changed league and only those, through within-player variation — the same man before and after his transfer — and then applies it player by player, at every appearance.

The reason is written in the code: estimating that slope on all players would be a mistake, because the best players really do play in the best leagues — you would be subtracting their talent while believing you were subtracting a bias.

The consequence is the one at hand: an exceptional squad in a weaker league keeps its exceptional character, player by player, because the correction bears on individuals and not on the badge. A club constant cannot tell the difference.

Both assumptions come at a price, and not at the same place: the constant is robust and blind to exceptions, the slope sees exceptions but assumes that transferred players are comparable to those who stay. And since the player model builds its eleven on the domestic league, it projects a European cup with a domestic starting side.

Why PSG, and nowhere else

Because it is the one place in this competition where an extreme squad meets an average league. The other clubs both models place at the top play where the coefficient is generous: Arsenal and City in the Premier League, Bayern in the Bundesliga at 1.269. Elsewhere the two methods have little disagreement available to them; on PSG, every bit they can produce is produced.

Which makes nobody a comfortable favourite: even on the more generous of the two models, PSG does not win in three of four draws, and reaches the final in only 39.0% of cases.

What the disagreement teaches us

A single model would have settled the matter while telling us nothing. Two models that agree almost everywhere and part ways on one club have named the exact place where the question lies: should the PSG squad be judged by the Ligue 1 average?

Dixon-Coles answers yes, by construction, for every French club at once. The player model answers no, player by player. The pitch will settle it by the end of the league phase, and we will ask the question again with the same two models.