Promoted clubs: what our model knows and when it guesses
Published
“Promoted club” sounds like one category. To a prediction model it is two — and the difference is not how good the team is, but whether the model has ever seen it play. Four fixtures between Friday and Monday put both kinds on display, and the model’s behaviour changes completely from one to the other.
Two kinds of new
Our dataset covers the Championship, England’s second tier, with the same depth as the Premier League — Coventry City and Hull City each arrive in the top flight carrying 230 finished matches of their own history in our database. The model has watched them, rated them and re-rated them for five seasons. It simply hasn’t watched them here.
Spain and France are different. We don’t model the Segunda División or Ligue 2, so Málaga, Deportivo La Coruña and Le Mans crossed into our coverage with essentially nothing behind them: Málaga and Deportivo have one finished match each in our dataset (their opening rounds), Le Mans has none. For these clubs the model isn’t translating a rating from a lower tier — it has no rating to translate.
When the model knows the newcomer
Knowing a promoted club makes the model opinionated, sometimes strikingly so.
At Hull City vs Manchester United, our 16 August model run makes the promoted side very nearly even money in the 1X2: 36.3% for Hull at home, 37.0% for United, 26.7% the draw. That is not a shrug — it’s a specific claim, built on five seasons of Hull data, that this squad at home is roughly a match for a side it has never faced in our dataset.
Arsenal vs Coventry City shows the same conviction colliding with the market. Our 16 August run prices Arsenal at 61.4% and Coventry at 15.6%. Betfair Sportsbook’s odds, fetched on 19 August (1.18 / 7.00 / 19.00), imply roughly 81% for Arsenal and 5% for Coventry once the bookmaker’s ~4% margin is stripped out. The model is giving the promoted side about three times the chance the market does — a wider gap than the Atlético vs Málaga divergence we wrote up last week, and for the opposite reason. There, the model knew too little about the newcomer. Here, it arguably knows a lot — and the open question is whether a rating earned in the Championship survives contact with the Premier League, or whether the market is right that the tiers are further apart than results-based ratings suggest. We genuinely don’t know yet. That’s not false modesty; it’s the first season in our data where we can test it.
When it’s guessing
Now watch the model when the history is missing.
Málaga vs Deportivo, two promoted clubs with one match each in our dataset, comes out of the 19 August run at 45.6% / 25.9% / 28.5%. Le Mans vs Brest, from the 16 August run, is nearly identical: 45.7% / 25.5% / 28.9%. Two different countries, four different clubs, one shape.
Compare those home-win numbers with the long-run base rates in our dataset: home teams have won 45.7% of 1,922 finished La Liga matches and 43.3% of 1,688 in Ligue 1. When the model has little or nothing to distinguish two teams, its output collapses toward the one thing it does know — home advantage. These probabilities aren’t insight into Málaga or Le Mans; they are the sound a model makes when it has no information. An honest site should say so rather than dress the number up as analysis.
How to read the difference
Neither kind of call is automatically wrong. But they deserve different levels of trust, and the embeds on each match page will keep re-rendering the current numbers while the figures quoted above stay frozen at their model runs — dated so you can hold us to them. Worth adding: the model sees results, not headlines. Summer signings, new managers and pre-season form are invisible to it, for the well-documented newcomers just as much as the blank ones.
Which kind of guess ages better — the confident Championship-informed call or the prior-driven placeholder — is exactly the sort of question a public track record exists to answer. Every one of these four fixtures will be graded there after full time, hits and misses alike.
