I didn’t have a ton of time to put into this years predictions, and I’m working on a couple of things I might use to come back and make predictions at a later date. In short, here are three dumb models (i.e. ones that require no input on my behalf) that I’m using this season to predict how many points each team will score this season, based on their performance last season. There are obvious flaws to this approach – it doesn’t account for changes in managers, players, the strength of competition in the league, the ageing of a squad, or injuries. However they are exceptionally simple to understand, and do a pretty damn good job even compared to much more complex and expensive products (the EuroClub Index and a bookmakers estimates).
As a) I haven’t calculated all of the relevant numbers from Chammpionship data and b) predicting how well a given team promoted from the Championship will perform in the Premiership is basically pointless, I’ve just assumed that each of the three newly promoted teams will score 38 points under each model, which is the average of the newly promoted clubs during the past 13 seasons.
The first two models are based regressing each teams shot ratios (total shot ratio (TSR) and shots on target ratio (STR)), a given distance towards the mean and converting that number to an expected points total. These models have variously been used previously to predict the ’10-11, ’11-12, and ’12-13 seasons (in fact every season since ’00-01 is looked at here).
So first up, TSR:
Finally I’m going to throw a third, more experimental, version into the mix. Last year I had a play with a comparables model, which did admirably, but this year I wanted something more automated for me to compute. As such I started messing around with a combination of TSR and PDO, to see if there was a combination of exponents that performed better historically than either simple TSR or STR. And it appears that TSR2 multiplied by PDO1 is marginally better. So here’s how it would predict next seasons standings:
So where do the models agree? Well Liverpool should be in for some improvement this time around, Newcastle should bounce back and finish roughly mid table, Southampton are maybe the best of the non-top-seven teams, whilst Stoke and Sunderland (and to a lesser extent Villa) are in trouble.
There are some discrepancies – United’s extraordinarily high PDO from last season gives them a solid boost in the final model, Chelsea also see an improvement, whilst Newcastle and Southampton take solid hits.
I actually like the look of this new model moreso than that of the comparables last time around. Whether it can better the other models shown here, or those devised elsewhere (a nice roundup is provided here by Zach Slaton), remains to be seen.
Finally: These models imply that the cutoff for relegation will be roughly 36 points – and that the league is basically split into four tiers when it comes to a relegation battle:
<1% chance of relegation – Spurs, Liverpool, Man City, Arsenal, Chelsea, Everton, and United
1-6% chance of relegation – Southampton
9-24% chance of relegation – Newcastle, Swansea, West Brom, West Ham, Norwich, Fulham
20-37% chance of relegation – Stoke, Sunderland, Villa, Cardiff, Palace, Hull
(I'm aware there's an overlap in the last two groups – it's because Vila look much better via STR than the other two models)
THat's enough for now – I'll work on converting these into title and top four odds at a later date.
I’ve had a quick blitz and converted these models to top four, and title odds.
They suggest that the title will be won with ~78 points (which is undoubtedly low – the average of Premiership champions since ’00-01 is 87 points) and the range of title odds for each team are:
Man City 19-25%
Man United 1-3%
No other team is in excess of 1%
I’ll come back another time and look at the odds for a top four spot.