FPL - A series of posts about formations - Part 3: Points per million as a way of valuating players

 There’s a well of different ways to assess the values of the players within the game. Form, fixtures, points per game, points per 90min and so on. What most of these factors don’t consider is the price of the player you’re trying to valuate. That’s why a lot of managers also value players by the metric “points per game per million”.   

                                                                             Example:
                              Salah has scored 179 points while playing 22 games this season.
                  This means that, based on his starting price (9.0M), he has offered a value of :
                                                         179/22/9= 0,90pts per million.

This way of valuating players has been discussed in older articles, such as “REDARROWS”s http://www.fantasyfootballscout.co.uk/2017/08/08/points-per-game-per-million-in-search-of-value/.
 

There’s several reasons why I wanted to explore this theory further:
1: Possibly defining differences in value based on positions.  
2: Figuring out if historical numbers of points and matches, coupled with season startingprices, can be used to predict a players value ahead of the season.
3: If findings suggest an optimal set-up/formation,  - A total of 278 players has been studied in search of answers.

What the study showed:
1: Differences in value
There is very noticeable differences in value when comparing averages in each position.     
                                         Goalies   Defenders   Midfielders   Strikers     
                History avg   0,773        0,602            0,533          0,5141     
                     17/18 avg   0,76          0,59               0,5378       0,4    
Based on their startingprice, the table shows that the history of the players suggested that, this season, the most valuable players would be goalies, then defenders, then mids then last attackers. When comparing the numbers with the season averages thus far (GW23), it’s clear that this has also been the case. The different positions have delivered very similar numbers to what the history suggested, with only the strikers failing to match their expected value.     

                                   Goalies   Defenders   Midfielders   Strikers     
No criteria             0,76          0,5378          0,5378          0,4007     
At least   15gms    0,77          0,556            0,556             0,4456     
At least   20gms   0,77          0,581            0,581             0,4727    
This table shows how amounts of games played affects the value in each position. It seems that on top of not getting those pesky points of the bench, having players who are “nailed on” also increase the value of their returns.    
                         Goalies                               Defenders                     Midfielders                                         Strikers     
               History   17/18                History     17/18                       History     17/18                  History   17/18  
Average     0,773   0,76   Average      0,602    0,59    Avg             0,533   0,5378                         0,5141   0,4
Avg 4-5 M 0,79     0,76  Avg 4-5 M    0,6         0,58    4,5-7 M     0,537   0,5111    4.5-7 M      0,5106   0,41
Avg 5.5 M 0,72    0,77  Avg 5.5-7 M 0,604    0,632  7.5-9.5 M 0,513   0,5641   7.5-9.5 M   0,52       0,39
                                                                                                   10M-->   0,5214   0,5036    10M-->    0,523     0,46    
This table shows what history suggests according to pricebrackets, and how its played out this season. The most notable things are:
- expensive goalies, expensive defenders and mid-priced mids are all currently overperforming compared to what history suggests.
- All strikers are underperforming, but the ones who are underperforming the most are the cheap and midpriced ones. 

 2: predicting value
The way the players have performed thus far this season does suggest that one can use historical data of points and matches to fairly accurately identify which positions are most valuable.  It should however not be taken as gospel in every individual players case, because different circumstance might alter what the history can’t account for. New role, new club, age, new manager, new players contending for a spot in the lineup. Of course one has to assess all these factors, and many more.  This of course doesn’t mean that it’s not useful in the individual cases. Out of the current 11 in the dreamteam, all players except DDG, Kane and Salah had FPL-history which predicted that they would offer greater value than average this season.    

3: Set-up / Formation 

If one was to use the findings when building the team ahead of the season, it would seem clear that the best way to invest your money would be in defence. Investing elsewhere, lets say in attackers, would be the equivalent of choosing to place a bet with the bookie offering the lowest odds on the outcome of your choosing.  The history of the players in the study suggested that 5-4-1 would be the ideal setup this year. This has been pretty much on the money so far. Only 2 strikers have offered a value greater than 0,7pts per million, 12 midfielders have managed the same, and a whooping 27 defenders have broken the mark. This basically means that 5-4-1 has offered the easiest way to steady returns, giving loads of decent options.  

Conclusion 

I believe REDARROWS said it well in his article “When we play Fantasy Premier League we are essentially trying to spend 100m as efficiently as possible, so the calculation in the backs of our minds is always point per game per million (ppg/m) even if we don’t think about it when we are setting up our teams.”  

Allthough I believe this to be true, I also believe there’s a lot to gain by using points per million in a systematical fashion. If we did, perhaps 3-4-3 would’nt be the template almost every single season. Perhaps a lot of us would have scrapped the idea of “killing” several positions in search of getting our hands on premium attackers at the start of the season. Perhaps a lot of us would have opted for the “safer choice” instead of the “risky one”. Perhaps. Playing the perhaps game is allways gonna be hindsight bias. Just because this study shows that a lot of the trends this year was predictable ahead of the season, doesn’t mean that’s gonna be the case next season, or going forward. However, going forward, I’ll make sure to use “points per million”, as a metric, more than I’ve done in the past.
 

The complete spreadsheet can be found here: https://docs.google.com/spreadsheets/d/1o_a8mLKtW2uV5z2mbwAwX5BnC3xRNnRx0XnRobZirQI/edit#gid=96237
(All numbers should be correct as of GW 23.)