Serie A matchday 1: the round that punished narrow home favourites
Published
A prediction model does two different things and they are easy to confuse. Sometimes it has an opinion. Sometimes it merely has an ordering — a top number in the thirties, which is less a forecast than a note saying “if forced to pick, this one, barely”. Serie A’s opening round separated those two cases about as cleanly as a round of football can, and the separation, not the scorelines, is what matchday 1 was actually about.
Our model went into the weekend with three fixtures in which the home side led without conviction: a top probability under 45%, with the draw close behind and the away side within single figures of the lead. Those three were Parma vs Cagliari at 37.7% home, Bologna vs Lazio at 39.6%, and Venezia vs Lecce at 42.0%, all from the 20 August model run. They finished 0-1, 0-1 and 0-2. Every one of them went to the visiting side, and in every one of them the home team failed to score.
The three fixtures the model shrugged at
Take them in order, because the way each was lost differs and the difference matters.
Parma were the weakest of the three leads at 37.7%, in a fixture our model also priced near the foot of the round for goals — 38.7% for over 2.5. It got the goals right and the winner wrong: a single strike from G. Romano on 79 minutes settled it for Cagliari, per Ahram Online’s matchday 1 results and scorers.
Bologna at 39.6% is the one that should sting least and probably teaches most. Lazio were a 32.7% away side in our numbers — not an outsider, just second in a three-way spread whose leader was under seven points clear of them. A 0-1 in a fixture priced 39.6 / 27.7 / 32.7 is not a model failure. It is the second-likeliest outcome landing, which happens roughly a third of the time by construction.
Venezia at 42.0% is the one with an actual argument attached. Newly promoted — Wikipedia’s season page records that “Venezia, Frosinone and Monza return to the top tier after one-, two- and one-year absences, respectively” — they carried a rating built substantially on second-tier scoring rates into a top-flight opener, and our model still made them favourites at home. Lecce won 0-2 through O. Gorter just after half-time and Gabriel on 80. Of the three, this is the one where the 42.0% looks least defensible after the fact, and the reason is a known and documented weakness rather than bad luck.
The one away lean that did not land
Honesty about a round means the misses that do not fit the thesis, too. Udinese vs Como was not a narrow home favourite — our model had it at 26.9 / 26.7 / 46.4, a firm lean to the visitors, one of the clearest positions it took all weekend. It finished 1-1: H. Kamara for Udinese on 29 minutes, Douvikas levelling for Como on 54.
That is a real miss, and it is the kind that is more instructive than the near-coin-flips. A 46.4% call is the model saying something. When it is wrong, the question is whether the underlying rating was off or whether a 46.4% simply lost the 53.6% of the time it was always going to. One match cannot answer that; a season of them, scored properly on the track record, can.
Where a firm number was a firm number
The round’s other side is quieter and worth stating plainly. Inter vs Monza was our model’s strongest position of the weekend at 72.5% and finished 4-1. Atalanta vs Sassuolo at 53.6% finished 2-1, with Raspadori inside six minutes, Odenthal levelling on 55 and Krstovic settling it on 65. Roma vs Fiorentina at 50.1% — the one fixture in the round repriced on 22 August rather than the 20th — finished 4-0, D. Malen scoring on 27, 52 and 60.
The away leans in the mid-forties largely did the same. Genoa vs Napoli at 46.1% away ended 0-2, Frosinone vs Juventus at 47.5% ended 0-1, and Torino vs Milan at 45.7% ended 1-2 with A. Cisse and Rabiot scoring three minutes apart and C. Adams pulling one back in stoppage time.
So the shape of the round is not “the model read Serie A badly”. It is that the fixtures it read with a number worth having behaved, and the fixtures where the number was a shrug went uniformly against the home side. Uniformity across three is not a large enough sample to be a finding — but it is large enough to be a question for matchday 2.
A round decided after the 79th minute
There is a second pattern running underneath, and our model has no way of seeing it at all. Cagliari’s winner at Parma came on 79. Lecce’s second at Venezia on 80. Napoli’s two at Genoa on 82 and 87. Torino’s consolation on 90+3. Five of the round’s goals, including two that decided matches outright, arrived in the last eleven minutes plus stoppage time.
A Dixon-Coles model has no minute dimension whatsoever. It produces a distribution over final scorelines and is indifferent to whether a 0-1 was settled in the ninth minute or the eighty-ninth. That is fine for scoring a forecast — the result is the result — but it means the model cannot distinguish a home side comfortably held from one that pressed for eighty minutes and was caught. Those look identical in its input and are not identical in what they imply about the next match. If Serie A’s opening weekend really was defined by late goals rather than by control, the ratings our model carries into matchday 2 have absorbed the wrong lesson from at least a couple of these games.
The part of this round that carries forward
Not much, and that is the point worth ending on. Ten matches is a fraction of a percent of the history a Dixon-Coles fit is built from, and a promoted club’s rating does not become trustworthy because it lost one game. The three narrow home favourites will not have moved far; Venezia’s number will have moved most, because a 0-2 defeat at home carries a two-goal margin into a margin-sensitive fit.
What should change more than the ratings is how a reader treats a top probability in the thirties. It is not a tip with a slightly lower success rate. It is the model declining to have a view, and this round is a reminder that declining to have a view is information in itself. The full set of Serie A prices, including matchday 2, sits on the Serie A page, and the ten fixtures above each carry the numbers exactly as they were published before kick-off.
The matches, at a glance
Serie AFT
Udinese1
Como 19071Serie AFT
Inter4
Monza1Serie AFT
Genoa0
Napoli2Serie AFT
Parma0
Cagliari1Serie AFT
Frosinone0
Juventus1Serie AFT
Venezia FC0
Lecce2Serie AFT
Atalanta2
Sassuolo1Serie AFT
Torino1
Milan2Serie AFT
Bologna0
Lazio1Serie AFT
Roma4
Fiorentina0