Season Simulation
A season simulation plays every remaining fixture 100,000 times over from a fitted match model, then reports the range of points and finishing positions each club landed in rather than a single predicted table.
Why it matters. A projected table published as one line per club hides how much of the order is guesswork. The range tells you which positions are genuinely settled and which are a coin flip.
One predicted table is a claim the model cannot support
Most projected tables give you twenty clubs in an order and nothing else. That presentation says the model knows where each club will finish. It does not.
What a match model actually knows is how likely each scoreline is in each fixture. Turning that into a final table means playing the fixtures out, and football played out twice from identical starting conditions gives two different tables.
So the projection plays them out 100,000 times and reports the spread.
How a simulated season runs
Each of the 100,000 runs does the same four things:
- Draw the ratings. Each club's attacking and defensive strength is nudged up or down, because those ratings are estimates from a few dozen matches, not measurements.
- Score every remaining fixture. A scoreline is drawn at random from the probabilities the drawn ratings imply for that specific matchup.
- Add it to the real table. Points already earned this season are the starting point; only the future is simulated.
- Rank and record. The finished table is sorted on points, goal difference then goals scored, and every club's finishing position is tallied.
After 100,000 runs, a club that finished first in 27,000 of them has a 27% title probability. A club that finished between 51 and 80 points in 90,000 of them gets a 51 to 80 range.
The range is the useful part
A narrow band means the model is confident. A club whose 90% range runs from 78 to 88 points is going to be near the top whatever happens.
A wide band means the position is not settled. Ranges around 30 points wide are normal early in a season, and they overlap almost everyone in mid-table.
Overlapping bands mean the order between those clubs is noise. If three clubs all project between 51 and 70 points, printing them 9th, 10th and 11th is an artefact of having to print something. Treat them as tied.
The bands narrow as the season goes on, because fewer fixtures remain to be simulated and the ratings behind them have more evidence.
Promoted clubs are an assumption, not a measurement
A newly promoted club has played no matches in this division. There is nothing to fit its ratings on from top-flight results, so the model starts from a squad-level market-value estimate instead of a league-average placeholder.
Those rows are marked and their bands are drawn as a dashed outline, and the bands are deliberately wider than everyone else's. That extra width is the model saying the prior is information, but not match evidence. Once the club has played, its own results take over and the band narrows like any other.
What the simulation does not model
The ranges cover the variance of football, not the variance of a football club. Every run holds each club's strength fixed from now until May, so none of the following is anywhere in the numbers:
- Injuries, suspensions and managerial changes
- Points deductions
- European and cup fixture congestion
- Clubs easing off once safe, or pushing once there is something to play for
Read the projection as "if the clubs stay roughly what they have been". When that assumption breaks, so does the projection, and the range will not warn you.
Where the ratings come from
The attacking and defensive ratings are the same fit that produces fixture difficulty. It is a Dixon-Coles model, a standard approach to football scorelines, fitted on non-penalty expected goals rather than goals so that a lucky finishing run does not get treated as attacking strength.
Before matches are played, projections are additionally informed by squad-level market-value estimates derived from publicly available crowd-sourced data. This influence fades to zero as the season progresses.
The projection is recalculated once a day, immediately after that model is refitted.
Common questions
- How is the projected table calculated?
- Every result this season and last feeds a Dixon-Coles model that gives each club an attacking and a defensive rating. Every fixture still to be played is then given a random scoreline drawn from the probabilities those ratings imply, and the finished table is recorded. That happens 100,000 times, and the published table summarises all 100,000 outcomes.
- What does the points range mean?
- It is the 90% interval: the club finished inside that span in 9 out of every 10 simulated seasons. A 30 point range is not a hedge, it is the honest width of what 38 matches of football can produce from the same starting ratings.
- Why are two clubs shown in a specific order when their ranges overlap?
- Because something has to be printed first. Where two clubs' ranges overlap heavily, the order between them carries almost no information and should be read as a tie. The order is only meaningful when the bands barely touch.
- What does "Default" mean next to a promoted club?
- That club has not played a match in this division, so there is nothing to measure it on. The model substitutes an average promoted side and widens the range to say so. Read those rows as an assumption about promoted clubs in general, not a read on that club.
- Why do the numbers not change after every match?
- The projection is recalculated once a day, straight after the model is refitted. A single result rarely moves a 38 match projection enough to matter, and recomputing it hourly would show Monte Carlo noise moving rather than the model changing its mind.
- What is not included in the projection?
- Anything the model holds no information about: injuries, managerial changes, mid-season transfers, points deductions, European fixture congestion, and clubs easing off once they are safe. Every simulated season holds each club's strength fixed from now to May.
Related terms
See Projected Table across the Premier League
Every player, team and match measured on the chances created rather than the scoreline.
Last updated