Lille vs PSG: a twelve-point gap between our model and the market
Published
Friday night at the Decathlon Arena is the largest model-versus-market divergence on our board this week, and it is worth walking through slowly, because the size of the gap has a single identifiable cause rather than a mysterious one. Our 26 August model run prices Lille vs Paris Saint-Germain at 31.6% Lille, 26.7% draw, 41.7% PSG. The market has PSG closer to 54%. Twelve percentage points is a lot of daylight on a fixture both sides have plenty of data for.
Where the market actually sits
ESPN’s odds page for the fixture, read on 26 August, lists Lille at +340, the draw at +285 and PSG at −135 — decimals of roughly 4.40, 3.85 and 1.74. Those imply 22.7%, 26.0% and 57.5%, which sum to 106.2% because the overround is baked in. Stripping that margin out proportionally leaves a fair-price read of about 21.4% Lille, 24.5% draw, 54.1% PSG.
Line the two up:
| Outcome | Our model (26 Aug run) | Market, de-margined |
|---|---|---|
| Lille | 31.6% | 21.4% |
| Draw | 26.7% | 24.5% |
| PSG | 41.7% | 54.1% |
The draw is close to agreed. The result market is not: we give Lille half again the chance the bookmaker does, and PSG twelve and a half points less.
The parameter that explains most of it
Our model applies one home-advantage term for the whole league. In the Dixon-Coles fit that is a single parameter estimated across every Ligue 1 match in our database; in the Elo ledger that runs alongside it, home advantage is a flat 65 rating points, identical for Le Havre at home to Angers and for Lille at home to PSG.
That is a defensible modelling choice and a real limitation, and this fixture is where the limitation is most expensive. A single global home term means the model applies exactly the same Pierre-Mauroy discount to Paris Saint-Germain that it applies to every other visitor — and elite clubs are precisely the teams whose away performance least resembles the league-average away performance the parameter was fitted on. The market is under no such constraint. It can, and clearly does, price PSG’s specific record on the road.
The supporting evidence for the market’s side of that argument is not subtle: PSG can extend their competitive unbeaten run against Lille to eleven games on Friday, having scored three or more in six of the previous ten, as Goal’s preview notes. Our model sees those matches only as scorelines feeding attack and defence parameters, with no separate channel for “this visitor travels unusually well”.
What the league table is not telling you
There is a tempting counter-argument sitting in the standings, and it should be resisted. Lille go into matchday 2 top of Ligue 1 on three points after a 2-0 win at Angers, Olivier Giroud scoring on 11 minutes and Tiago Santos doubling it around the half hour; PSG sit eighth on one point after a 2-2 draw with Rennes, per the same preview.
Neither our model nor the market gives that table any weight, and both are right not to. One round is one match per club. The gap between 31.6% and 21.4% has nothing to do with Lille being first and PSG eighth — it existed before either result and would have existed had the two been reversed.
Team news, which neither number contains
Two absences are confirmed and neither is priced by us. Hamza Igamane is out for Lille with a cruciate ligament tear, and Nuno Mendes is suspended for PSG following a dismissal, both reported in Sports Mole’s preview.
Our model does not read that page. It never has. A left-back suspension for the visitors is the kind of thing a market moves on within minutes and a goals-based fit notices only if it eventually shows up in scorelines — so if you want to argue the 41.7% is closer to fair than it looks, the Mendes absence is the argument, and it is one we have not made inside the number.
Which side we would expect to be wrong
Honesty is cheaper before the match than after it, so: on a fixture like this one, the market is the better-informed party more often than we are. It prices squad information, travel, rotation ahead of European weeks and the specific away profile of a club that has dominated this fixture for a decade. We price goals.
What we do not do is nudge our output toward the bookmaker to make the gap look smaller. A model that quietly corrects itself towards the closing line stops being a model and becomes a slower copy of one, and the whole value of publishing a number that disagrees is that it can be scored later. Every settled call we have made lives on the track record, which is where performance claims belong — not in a preview, and not in a paragraph that cannot be verified.
The rest of Friday’s card and the current version of these probabilities are on the Ligue 1 page; the numbers above are frozen to the 26 August run and the match page will have moved on by kick-off. Twelve points of disagreement, one parameter, and ninety minutes to find out who was closer.