Community

A practical guide to finding undervalued players

Introduction

Playing FPL is simple – you just need to spend your budget wisely. However, in order to do that you need to compare the value offered by different players, which is not so straightforward an exercise. Traditional method of dividing player’s points by his price does not help much, as you will always find that cheap players offer greater value than expensive ones. Therefore, in order to circumvent this bias you need to find a systematic approach to compare players in different price brackets.

Methodology

This analysis tries to find overpriced/underpriced players based on their last season’s performances and this year’s prices. I consider average point per match (PPM) rather than total points over the season to be a more valuable metric. Fantasy managers are usually able to transfer out injured or otherwise unavailable players, so a player should not penalised for him being temporary unavailable (however, you should avoid injury prone players as transfers are limited).

I utilized regression analysis of 2018/2019 average points per match and 2019/2020 prices. I excluded players who played less than 1000 min. (for data reliability) and are currently owned by less than 4% (I consider these players not to be “fantasy material”). For players that were reclassified from one position to another, I used their recalculated PPM based on this excellent analysis by Ludo.

Results

I came up with the regression of PPM=1.08+0.46*Price. To put it in simple language, while buying a player you get 1.08 PPM free of charge and 0.46 PPM for every million spent. Therefore, I derived the player value as Value=(PPM-1.08)/Price. Traditional value metric would use a basic formula of Value=PPM/Price, which is biased towards cheaper players (it does not account for a “free” 1.08 PPM).

Based on my methodology the most valuable players in each position would be as follows:

 

 

The way to interpret the results is simple – the higher the value score, the more underpriced the player is. The hurdle rate to decide if the player is underpriced/overpriced is 0.46. If the value is higher than 0.46, the player is underpriced and vice versa.

We can see from the results that the hype about defenders are justified by my analysis. Despite the increase in their prices Liverpool defenders still offer exceptional value, while only two forwards are underpriced (value score above 0.46).

How to use my approach making daily decisions

The tables above indicate the underpriced/overpriced players based on solely their historic PPM. However, if you believe that a certain player may underperform/overperform his historic PPM, you may recalculate his value.

For example, if you believe that Andrew Robertson this season will average 5 PPM (rather than 5.9 last year), you can recalculate his value using the same formula Value=(PPM-1.06)/Price resulting to value of 0.56 (which is still impressive).

I have also built the inferred average PPM for each price point. If you believe that a certain player can overachieve the PPM for his price bracket – go for it!

There is no need to memorise this table as you can do mental calculations for any price point by yourself. You can use simplified formula of Price/2+1 (the table uses formula Price x 0.46 + 1.06) to arrive at required PPM to make the transfer.

Let’s take Robertson again as an example. Using simplified formula you will arrive at 7/2+1=4.5 PPM to justify the transfer. It is 0.2 more than in the table above, which is not a problem as you want to transfer in players who offer above average value.

Other implications – cost of making transfers late

Another widely discussed topic is should we make transfers early or incur price increase/decrease and wait as close to deadline as possible. My regression can calculate the cost of waiting. The value of any additional 0.1 in team value diminishes from 1.7 points at the start of the season to 0 points at the end.

As an example, if there are 37 Gameweeks left, the cost of 0.1 increase in price can be calculated as 0.1 multiplied by incremental return of 1m (0.46) and multiplied by remaining Gameweeks (37). That would give us 0.1 x 0.46 x 37 = 1.7 points for every 0.1 increase in price. The cost of missing 0.2 price increase (3.4 points) is just below the cost of a hit (4 points).

Conclusions

The main takeaways:

1. Use the formula Price/2+1 to arrive at required PPM for any player that you want to transfer in. If you believe he can overachieve it, transfer him in!

2. Use formula Value=(PPM-1.06)/Price to compare the value offered by any two players. If the result is above 0.46, he offers above market average value.

3. The value of any additional 0.1 in team value diminishes from 1.7 points at the start of the season to 0 points at the end.

Fantasy football is not only about data, there are certainly more factors that fantasy managers need to navigate, i.e. changes in forms of teams/players, injuries, captaincy, rotation risk and many more. However, I believe that my approach could help the managers to make better informed decisions and find undervalued/overvalued players.

55 Comments Post a Comment
  1. Rotation's Alter Ego
    • Fantasy Football Scout Member
    • Has Moderation Rights
    • 12 Years
    4 years, 8 months ago

    Cheers Paulius, some good research and working here. No surprise to see TAA and Robertson so high, but to see Zinchenko right at the top is interesting - if only I could convince myself about his minutes!

    1. Loftus cheeky nandos
      • 5 Years
      4 years, 8 months ago

      Surely higher now mendy injured

    2. FPLTrader
      • Fantasy Football Scout Member
      • 6 Years
      4 years, 8 months ago

      Thanks for posting! I will try to follow up on my methodology during the season using this season`s average PPMs. You may find me on twitter (FPL_Paulius) if anyone is interested on further analysis.

  2. Zebrahimović
    • 8 Years
    4 years, 8 months ago

    Bilva team - what you recon out of 10? Any thoughts appreciated m8s

    Ryan (Button)
    TAA(v) Robbo Zinc Coleman (Lund)
    Sterling Salah(c) Bilva Siggy (Donker)
    Jota King (Greenwood)

    1. Zebrahimović
      • 8 Years
      4 years, 8 months ago

      The other option was have Fraser and Barkley for Bilva & Siggy and have Vardy instead of King

  3. KujaliaFC
    • 11 Years
    4 years, 8 months ago

    Given that Ravel Morrison has signed for Sheffield United, is Lundstram still a viable OOP 4.0 D?

    1. Chandler Bing
      • 7 Years
      4 years, 8 months ago

      He's not nailed on but will get his chances early on as it stands.

      1. KujaliaFC
        • 11 Years
        4 years, 8 months ago

        I'd imagine that Morisson would have a more attacking role anyway perhaps?

  4. Mullered in Maenam
    • Fantasy Football Scout Member
    • 9 Years
    4 years, 8 months ago

    Thanks Paulius, very interesting read.

    Is there another way to access the results tables, I can't read them very well in this article, thanks.

    1. KujaliaFC
      • 11 Years
      4 years, 8 months ago

      Right click and select "Open in New Tab".

      The image itself is legible. It's resized down in the article.

      1. Mullered in Maenam
        • Fantasy Football Scout Member
        • 9 Years
        4 years, 8 months ago

        Thanks Kujaliafc, very helpful and thanks for taking the time.

        Cheers

  5. Chandler Bing
    • 7 Years
    4 years, 8 months ago

    If Matip starts on Sunday:

    A. Matip and Coleman
    B. VvD and Diop (rotation with Rico)

    Have Robbo. 4-4-2 formation either way.

  6. Rupert The Horse
    • Fantasy Football Scout Member
    • 11 Years
    4 years, 8 months ago

    Some great work here.
    So glad TAA and Robbo aren't 8m+

  7. Rossi3254
    • 5 Years
    4 years, 8 months ago

    Really interesting read thanks for this!
    Was just wondering in the case of Zinchenko did you recalculate his ppm for last season as if he were a defender?

    1. FPLTrader
      • Fantasy Football Scout Member
      • 6 Years
      4 years, 8 months ago

      Yes, I used recalculated 5.7 PPM for him instead of actual 3.1 PPM.

  8. Pompel
    • 10 Years
    4 years, 8 months ago

    Excellent work!
    Lking the simple way that the regression formula shows that simple PPM(PPG) metric is flawed as skewed towards cheaper players, as every starting player will also get points regardless.

  9. CL4PTRAP
    • Fantasy Football Scout Member
    • 9 Years
    4 years, 8 months ago

    Cheers FPL Paulius. I enjoyed the article and will keep the formulas in mind for the season. Could you run the value calculator based on projected scores over the season in the members area with 2019/2020 prices? Does it differ from the values obtained when based on 2018/2019 total scores?

    1. FPLTrader
      • Fantasy Football Scout Member
      • 6 Years
      4 years, 8 months ago

      My methodology would not work for total scores, as total scores account for injuries. I may do some analysis for projected average scores over next 6 GWs. You may follow me on twitter (FPL_Paulius) if you are interested in more analysis.

      1. CL4PTRAP
        • Fantasy Football Scout Member
        • 9 Years
        4 years, 8 months ago

        Perfect. Not on twitter but I'm sure you'll drop something here for us to follow. Cheers.

  10. Annie
    • Fantasy Football Scout Member
    • 11 Years
    4 years, 8 months ago

    Nice read - such a welcome break from pre season reports and Villa talk.

  11. Ask Yourself
    • 7 Years
    4 years, 8 months ago

    Wow nice article thanks

  12. pingissimus
    • 5 Years
    4 years, 8 months ago

    Fascinating thank you

    It’s made me stop and think given that 3 of my certain avoids (King Zaha and Zinch) are good value. Still believe other factors make them dicey choices but there we go.

    Also intrigued by the early transfer info.

    1. FPLTrader
      • Fantasy Football Scout Member
      • 6 Years
      4 years, 8 months ago

      I currently own King and Zinch. Waiting where Zaha will end up to decide on him, as his stats are pretty good.

      1. pingissimus
        • 5 Years
        4 years, 8 months ago

        The thing with Wilf is his stats are boosted by the time he spent with Bats. Before that he was not much ahead of Townsend. Statistically he’s much better with someone who isn’t Benteke! FWIW fans eye view says the same thing.

        The Palace article the other day backs this up in some detail. It’s also worth considering Palace are in their traditional transfer mess and will only bring people in on deadline day - hard to integrate.

        Will he move on? Possible of course. But it’s going to take a monster offer and there’s a very finite number of clubs who can put the cash down now - that’s the killer it’s an absurd amount of money to pay up up front.

      2. pingissimus
        • 5 Years
        4 years, 8 months ago

        Oh and my King aversion is based on the same principle.

        Season stats are misleading as they are boosted by the time he spent alone up top. If you strip out the matches when both he and Wilson were on the field together his stats and heat nails look far worse. This really just stands to reason - he’s good as a central striker but average at best when in support of very often as a fairly withdrawn winger.

        1. pingissimus
          • 5 Years
          4 years, 8 months ago

          *maps

  13. Markus
    • 14 Years
    4 years, 8 months ago

    Great contribution. The one build I would make is to factor that base cost of defenders if 0.5 lower, so for defenders expected ppg=price-0.5x0.46+1.06.

    I've always gone for 20 points per million ballpark but this is slightly less at 17.5, but arguably more accurate given methodology. I do think you are more likely to get overachievers low down the scale though even if 'average' brings it down hence why the expensive options have to justify even higher returns to be the 'best' options, but I guess it works both ways of having more budget to play with too then.

    1. Markus
      • 14 Years
      4 years, 8 months ago

      Sorry that should be +0.5 shouldn't it

    2. Dead Balls
      • Fantasy Football Scout Member
      • 6 Years
      4 years, 8 months ago

      Did you have Doherty last season?

    3. FPLTrader
      • Fantasy Football Scout Member
      • 6 Years
      4 years, 8 months ago

      Defenders should not be treated differently. The only difference is that they offer 4.0 price point, which is not available for other groups. However, it still makes sense to treat 6.0 assets across diferrent positions.

      1. Markus
        • 14 Years
        4 years, 8 months ago

        Fair enough. It's just if I reduce Digne to the cheapest player as a 2nd or 3rd sub, I can afford a 6.5m mid in his place my midfielder sub. So in my case comparing apples with apples is a 6.5m mid and a 6m defender and this will always be the case, outside of the mandatory 3 defenders 2 mids and 1 striker, which are generally the comparative calls you're making...put another way, if all the defender prices were shifted to go from 7-10m and everyone had an extra 15m it would be literally exactly the same game and decisions as now but psychologically and with the charts defenders would all come out as awful value.

        1. FPLTrader
          • Fantasy Football Scout Member
          • 6 Years
          4 years, 7 months ago

          I get your point. However, the impact of being able to buy 4.0 defender is pretty low, as you usually have no more than one 4.0m asset. Last year we were blessed with van Bissaka, this year we may not have any playing 4.0 defender to fill our bench. Having 4.0 defender on the bench as opposed to 4.5 mid would save you 0.5 which is around 0.5x0.46=0.23 PPM you could get elsewhere, which is not enough to sacrifice defender spot. Another way to rectyfy that woul run a regression assuming that price of defenders are price-4, while other positions would be price-4.5, that would account for different starting points of each position.

  14. abaalan
    • 7 Years
    4 years, 8 months ago

    Excellent analysis. Will be sure to use some of this data in my decision making.

    Also Confirmed a number of thoughts I had in regards to value ... Eg. Coleman better value than digne.

    Interesting to see a player dropping 0.1 and getting in a player rising 0.1 is nearly worth a hit on its own at the start of the season! Seems crazy

    1. advdborg
      • 4 Years
      4 years, 8 months ago

      This analysis is shaky as it does not take into account the 50% "tax" on resale of the player

  15. Whits
    • Fantasy Football Scout Member
    • 7 Years
    4 years, 8 months ago

    First few GWs?

    A. Coleman
    B. Wan B

  16. pingissimus
    • 5 Years
    4 years, 8 months ago

    Just spotted that only one of the six 10 million plus premiums isn't best value for position but only makes the significant others list. Strange that he's the one almost no one is considering going without.

    1. Dead Balls
      • Fantasy Football Scout Member
      • 6 Years
      4 years, 8 months ago

      Who blue?

  17. sirmorbach
    • 7 Years
    4 years, 8 months ago

    Amazing work. Thank you very very much for this.

  18. advdborg
    • 4 Years
    4 years, 8 months ago

    Could you make the same analysis with a regression per position. It would give much more valuable insights in my opinion

    1. FPLTrader
      • Fantasy Football Scout Member
      • 6 Years
      4 years, 8 months ago

      I do not see much value in doing seperate regression, as you can move money around between the positions, therefore forwards compete directly with defenders and so on.

  19. Veggente85
    • Fantasy Football Scout Member
    • 8 Years
    4 years, 8 months ago

    Hello,

    I have a question about the methodology. Have you tried to calculate your regression line with players prices without baselines (4.5 for gk, defenders and midfielders, and 5.5 for forwards - or 4.5 for coherency with other positions)? The rationale is that the first 4.5m are probably providing a significantly different amount of points per million than all the successive ones. If you don't remove that, you probably add some noise to your estimates.

    1. Bagheri Arce
      • Fantasy Football Scout Member
      • 8 Years
      4 years, 8 months ago

      Great article - really interesting and a bit different.

      I too am keen to know the answer to this question. As ever the trick in fantasy football is spending your disposable income wisely. I class disposable income as excess above a base price...so let's say base price for all positions is 4.5m (and 5.5m for strikers) to get legitimate playing assets, less £17m as a minimum you can leave on the bench, leaves you with about £31.5m to play with.

      It can be inferred by eye-balling the spread of value across the positions in your tables above which still indicates defence is where the 'true' value is at and someone like Wilson or Jiminez could also be crucial to own.

      Obviously in reality there are other factors that are probably too complicated to build into the model like the fact that Zinchenko scores highly on PPG but is a rotation risk...meaning you need to cover some of this risk with your bench budget...hence reducing his value to your team. To what cost I don't know.....

      All truly engaging stuff though - thanks

    2. FPLTrader
      • Fantasy Football Scout Member
      • 6 Years
      4 years, 7 months ago

      The first 4.5m indeed provide more points than successive millions spent. However, that is accounted for by regression intercept 1.06, which I regarded as "free points" in the article.

      1. Veggente85
        • Fantasy Football Scout Member
        • 8 Years
        4 years, 7 months ago

        The regression intercept doesn't capture it. The 1.06 the refer to PPM of an hypothetical 0M player. For a 4.5 players, your model predicts 3.13 PPM (1.06 + 4.5 x 0.46). What I suggest you, is to remove 4.5 from the price of every player. This way, your intercept will refer to the points of a 4.5 player. And the estimate will be more accurate of the value of each additional million.

        1. Andy_Social
          • 11 Years
          4 years, 7 months ago

          Can you do that with the players in his table?

          1. Veggente85
            • Fantasy Football Scout Member
            • 8 Years
            4 years, 7 months ago

            No, it is necessary to have the complete dataset and to run the regression again. If Paulius publish this dataset somewhere I can try to do it. Anyway his analysis is nice, removing the 4.5 will improve the estimate, but I think that the 0.46 that he obtained isn't very far from reality. I suppose that removing 4.5 from prices will improve the value of low cost players and decrease the value of high cost ones.

            1. Veggente85
              • Fantasy Football Scout Member
              • 8 Years
              4 years, 7 months ago

              Btw, someone know where I can find last season points per player (or directly the PPM stat)?

          2. Veggente85
            • Fantasy Football Scout Member
            • 8 Years
            4 years, 7 months ago

            OK, I replicated the study with -4.5 to all prices. The new equation is 2.5087 + 0.5356 * adjusted Price. It means that 4.5 players are expected to score 2.5 points per appearence and that, on the mean, you can expect 0.5356 extra points per match for each extra million invested.

            With 0.5356 as the new "treshold" for value, the players of the tables comes out like that:

            Zinchenko 0.63
            Alexander-Arnold 1.55
            Robertson 1.36
            Van Dijk 1.48
            Alonso 1.34
            Laporte 1.27
            David Luiz 1.36
            Coleman 1.80
            Digne 1.34
            Walker 1.36

            Tielemans 1.17
            Zaha 0.68
            Fraser 0.75
            Sterling 0.58
            Perez 0.65
            Milivojevic 0.74
            Pogba 0.65
            Sigurdsson 0.65
            Brooks 0.80
            Mane 0.56

            Wilson 0.88
            Jimenez 0.75
            King 0.76
            Vardy 0.58
            Deulofeu 0.96 (midfielder points)
            Kane 0.49
            Aubameyang 0.49
            Jota 0.85 (midfielder stats)
            Aguero 0.48
            Murray 0.59

            Lloris 1.89
            Alisson 1.42
            Pickford 1.73
            Ederson 1.29
            Fabianski 2.51
            Kepa 1.44
            Ryan -
            Heaton -
            Leno 1.61
            Schmeichel 1.30

            Pereira 1.11
            Van Anholt 1.38
            Felipe Anderson 0.72
            Salah 0.54
            Doherty 0.85
            Richarlison 0.53
            Bernardo Silva 0.51
            Maddison 0.52
            Lacazette 0.45
            Rashford 0.34

            Probably there is a difference in the way we calculated PPM. I took total points and divided it by games played (starts + subs on).

            1. FPLTrader
              • Fantasy Football Scout Member
              • 6 Years
              4 years, 7 months ago

              I tried running regressions with suggested -4.5 on all prices with my data set, I get outcome with the same incremental value of 1m. Regression would be 3.14+0.46x(Price-4.5). However, going down with this approach would require to subtract 4 from defenders and 4.5 from other asstets (base cost for defenders is lower). I will contemplate using this approach in future analysis.

              1. Veggente85
                • Fantasy Football Scout Member
                • 8 Years
                4 years, 7 months ago

                Nice. How do you compute PPM? Total points / (starts + subs on)? I think I'm doing something wrong on my side...

                1. FPLTrader
                  • Fantasy Football Scout Member
                  • 6 Years
                  4 years, 7 months ago

                  You can find PPM statistics in official fantasy page: https://fantasy.premierleague.com/statistics. For reclassified players I used recalculated PPM (link in the article)

                  1. pingissimus
                    • 5 Years
                    4 years, 7 months ago

                    Where is Guaita in all this? You miss him for some reason?

                    1. pingissimus
                      • 5 Years
                      4 years, 7 months ago

                      Oh you exclude him by ownership I suppose. Still a first choice keeper and in last season’s form the value pick or very nearly.

  20. zinaks
    • Fantasy Football Scout Member
    • 4 Years
    4 years, 8 months ago

    it would be interesting if u made the value of 2017/2018 players based on 2018/2019 prices and chose the perfect team and see how many points would u have scored in 2018/2019 season to test its reliability.

  21. Shatner's Bassoon
    • 5 Years
    4 years, 7 months ago

    Currently on 3-4-3 with Coleman & Bilva.

    Should I knock one of them down by 0.5 to improve my bench from the Dendo-Kelly-Lundstram fodder trio?