What happens if you simulate the Premier League season 10,000 times

A league table looks like a fact. Thirty-eight games each, points banked, teams sorted top to bottom, and there it is, the season, settled. But that final table is only one version of events. Replay the same campaign with the same teams and the deflections, penalties and red cards fall differently, and so does the order. The table we frame at the end is a single sample from a much wider set of seasons that could have been.

Playing the season, not reading it

Here is the trick. Instead of predicting the table, you play it. Give every side an attacking rate and a defensive rate, drawn from what they have actually done, and then simulate all 380 fixtures one scoreline at a time, each game a random draw from those rates. Add up the points, sort the standings, and you have a finished season. Not a forecast, a full imagined campaign with a champion, a top four, and three teams down.

Then you do it again. And again. Ten thousand seasons, each with a fresh set of dice. No single run means anything on its own. What matters is the tally across all of them: how often each team finished first, how often it slipped into the bottom three, how often the whole thing broke its way. Counting those outcomes turns one table into a spread of probabilities.

The title is rarely a lock

Run it and the first thing that goes is the idea of a nailed-on champion. Even a clearly strong side tends to win the league in only a share of the simulated seasons, not all of them. A team that comes out on top in, say, four runs in ten is the rightful favourite and still does not win the majority of the time. The title is a probability, never a name you can bank in July.

The top four is messier still. It is not four fixed slots, it is six or seven teams with overlapping chances, trading places run to run. A side can be firmly “in the race” and miss out more often than it qualifies, purely on the bounce. Relegation is the same picture flipped: a variance zone where three or four clubs share most of the risk, and the drop lands on different names depending on which season you happen to be watching.

That is the shape a season simulation hands back. Not a rank for each team but a band of plausible finishes, wide for the mid-table clubs and only a little narrower at the very top. Read it as a distribution and the table stops looking like fate.

Early on, it is mostly luck

The spread is at its widest in August. A handful of games is nearly all noise: a team can sit top after four wins and still be, on its underlying rates, a mid-table side that caught a soft run and some kind bounces. That is not a flaw in the fixtures, it is what a small sample looks like. It takes a couple of dozen matches before results start to pin down what a team really is, and the simulated spread narrows to match.

This is why an honest model resists the early table. The standings in September are a snapshot of variance as much as quality. The simulation’s job is not to chase that snapshot but to ask, given what we know so far, how the remaining games are likely to fall, and how wide the honest range around them still is.

Reading the spread honestly

The temptation is to collapse all of this back into one predicted table, because a single order is easier to argue about down the pub. Resist it. The value is in the spread itself: a side the model rates at 70% to survive still goes down three times in ten, and a season where it does is not a broken model, it is one of the seasons that was always on the cards. Probabilities, kept honest, tell you more than a confident finishing order ever could.

So the next time a table lands and someone calls it inevitable, remember it is one draw from a hat with thousands of tickets in it. Simulate the season and the certainty dissolves into something more honest and, oddly, more interesting: a champion who was only ever probable, a top four that was a scramble, and a drop zone that could have claimed almost anyone.