Introducing logic into title-race predictions

During my trip to London this week, I picked up a copy of The Times and discovered that the newspaper’s resident stats man Bill Edgar had predicted the remaining 13 matches of the Premier League season for the current top-five using a simple model. From this process, he named Manchester City as this season’s champions, ahead of Leicester City on goal difference. Comparing his projection to a number of other models, both simple and complex, reveals a number of interesting differences. Bill’s approach is more logical than most media predictions but can he do better than these models?

Times numbers man Bill Edgar has, in his own words, “had a stab at introducing logic into a title-race prediction”. He has assumed that the leading five clubs will lose away to a fellow top-five team, draw away to clubs who are sixth to 13th and win all of their other fixtures.


In order to compare Bill’s forecast to other methods, I put a request out on Twitter and quickly received projections from expected goals modeller @colinttrainor, non-expected goals modeller @goalprojection and a 56-variable neural network approach from Australia-based @teouchanalytics. I have also added the results based projection from @euroclubindex and a basic benchmark of the same results as last season for the remainder of this one (SCoRe).


The accuracy of these predictions will be assessed at the end of the season using the mean absolute errors of both positions and points. It is already interesting to note some of the differences though.

Bill Edgar’s approach is the only one to favour Manchester City as champions at this stage. The Times man’s points totals are consistently above those of the other models, even the simple SCoRe which just assumes that the rest of the season will be the same as 2014/2015. The exception is Arsenal who Edgar’s method expects to do worse than last season on the remaining fixtures. If Arsenal manage to take the same number of points in their remaining matches as they did last season, the 77 points they accrue will probably be enough to win the league.

The four more complex models all favour Leicester City to win the title now, albeit very marginally in the case of the Euro Club Index. They differ on the other three top four spots though with Arsenal, Manchester City and Tottenham Hotspur named in each position.

Across all six different approaches, Arsenal and Manchester City appear in every possible top-4 position, Leicester City are never outside the top two and Tottenham Hotspur are never named as title winners.

There is only position which everyone agrees on, a fifth place for Manchester United, so the current top four look set to play in next season’s Champions League. Hardly surprising given that, according to the Euro Club Index, only Aston Villa (20th) and Sunderland (19th) are more likely to finish in their current positions than United.


Thanks to:
Colin Trainor – his model is not public but you can google Expected goals to find plenty of information from others.
Goal Projection – you can read about his method here where there is also a link to his rating system.
Teouch Analytics – the methodology is described here.
Euro Club Index – the website is here and also contains a methodology page.

This entry was posted in Uncategorized and tagged , , , , , , , , , , , , , , . Bookmark the permalink.

3 Responses to Introducing logic into title-race predictions

  1. Pingback: Predicting the EPL top 5 end of season points | téouch analytics

  2. Pingback: Is winning all that matters in AFL or does the performance count? | Em News

  3. Pingback: New adventures! | téouch analytics

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s