Ahead of the 2023-24 Premier League season, football statisticians at BetVictor have built a predictive supercomputer to simulate the upcoming campaign. The model expects Man City to be crowned champions for a fourth consecutive year.

Key Predictions:

Manchester United to finish outside of the top four.
Burnley to avoid relegation.
Newcastle United and Liverpool to finish inside the top four.

The Predicted Premier League Table:

Probabilities of Finishing in Each Position:

Predicted Season Summary:

Last Premier League season saw Man United net 58 goals with a goal difference of +15, the lowest figures for a top-four finishing side in the past decade. They also fell short of their Expected Goals (xG) by -11.7, highlighting their underperformance in front of goal. Without a new striker, our model predicts a 25.2% chance of them repeating last season’s feat and achieving Champions League qualification.

Brighton, achieving the second highest xG and fourth-highest Expected Goal Difference (xGD) last season, are among the league’s statistical best. If current performance levels can be maintained, our model projects 67 points for them.

With an xGA of 39.6 last season, Newcastle‘s defence was second only to Manchester City. Our model foresees a solid performance this season with 71 goals scored, 36 conceded, a +35 goal difference, and a potential third-place finish.

Following a surge in form during the latter part of the season, Liverpool are predicted a 49.4% chance of Champions League qualification. Meanwhile, Chelsea, coming off their lowest finish in a decade, have a mere 0.6% chance of UCL qualification based on their current performance data.

As for the relegation battle, our model forecasts Luton Town, Bournemouth, and Sheffield United as the probable teams to face the drop. Both Wolves and Nottingham Forest are teetering on the brink with respective relegation probabilities of 37.1% and 38.2%. In contrast, newly promoted Burnley have a more optimistic outlook with a relegation probability of 18.5%.

The Science Behind the Supercomputer:

The supercomputer adopts the Monte Carlo method, at its heart is a Python-based match simulator that uses two Poisson distributions – one for the home team and one for the away team – to anticipate the number of goals each team could score in a match.

A Poisson distribution is a powerful mathematical concept that predicts the probability of a given number of events (in this case, goals) happening in a fixed interval of time. The key input to a Poisson distribution is the ‘lambda’ (λ) value, which represents the average rate of an event’s occurrence.

Let’s break this down using an example of Man City playing at home against Arsenal:

In our simulator, we first compute the λ for Man City, using both xG (Expected Goals) and actual goals scored per match based on their last 19 home games. This gives us an ‘attacking lambda’. For instance, with an xG of 2.24 and actual goals per match of 3.16 whilst playing at the Etihad, Man City’s attacking λ is: 2.7 – both xG and goals are used to measure both underlying and actual performance.

Next, we calculate Arsenal’s ‘defensive lambda’ based on their last 19 away games from their average goals conceded per game (0.95) and xGA (1.2) whilst on the road, giving a value of 1.08.

To tailor these λ values to the specific match-up, we adjust Man City’s attacking λ by factoring in the strength of Arsenal’s defence. We do this by calculating a multiplier, which is the ratio of Arsenal’s defensive λ to the league average defensive λ (1.54).

Multiplying Man City’s λ (2.7) by this factor adjusts it for Arsenal’s specific defensive strength. This adjusted λ better represents Man City’s goal-scoring likelihood against Arsenal. The same process is replicated for the away team.

In formulaic terms:

Home team λ = home team attacking λ * (away team defensive λ / league average away defensive λ)

Away team λ = away team attacking λ * (home team defensive λ / league average home defensive λ)

Our supercomputer uses the match simulator to predict the outcomes of a full season’s fixtures. The simulation is run 10,000 times, following which we calculate average standings and probabilities.

Do check out: Top 10 youngest goalscorers in Premier League history