Next Season Forecaster
A club's seasons, one after another, are a time series. Smooth over all of them, the latest counting most, and you get a forecast for next season, with an honest idea of how far out it could be.
- Points a game, each season
- Smoothed level
- Next season's forecast
How to read it
- Each green dot is one season: the club's points a game in the Premiership. Only its unbroken run of Premiership seasons is shown.
- The gold line is the smoothed level: the model's view of how good the club really is. After each season it moves 70% of the way towards what the club did.
- The gold dot is next season's forecast, the latest level. The shaded band shows how far out such a forecast usually is.
- Points a game × 38 gives the season total, so 1.91 a game is about 73 points.
How it works
Start at the league average, 1.38 points a game. After each season, move 70% of the way from the current level towards what the club actually did. The level after the last season is the forecast. Seventy per cent was the setting that forecast best on held-back seasons; on the five test seasons since, forecasts like this were usually within about 9 points, better than assuming a repeat of last season (about 10) or the league average (about 19).
Where it goes wrong
- Results only. Signings, sales, budgets and a new manager, the things that really shape next season, aren't in it.
- Promoted clubs have no recent Premiership season, so they get what promoted sides have taken on average in their first season back: 1.15 points a game, about 44 points.
- Breaks take time to show. If a club's level jumps, the forecast catches up over a season or two.
- Frozen at the end of a season and refreshed each summer.
How it was built and tested: What happens next season? Time series and ARIMA
Results data from football-data.co.uk. These figures come from a statistical model and are for analysis and education. Football remains uncertain and model predictions will frequently be wrong.