Picking an FPL team is hard. There are so many factors to consider and many of these become overwhelming or are downright impossible. My approach is to try to maximise the points per match of my squad. Using this approach, the first couple of players are fairly easy, but it gets increasingly difficult as you are start running out of budget, and factoring in fixtures, and rotation, and form, and impacts of transfers, and manager changes, and new stadiums, and recent international competitions, and preseason fixtures, and . . .
To make things simple, I outsourced the complicated calculations to a computer. While this resulted in an “ideal” team, the process provided some incredible insight into value, team structure, and where to spend your money. Join me on this quest where the real treasure is the journey and not the destination.
I built an optimiser in Excel using the ‘Solver’ add-in. I fed in the list of players with team, position, price (2019/20), and points per match data (2018/19). The only modification to the points I made was to adjust the points for the reclassified players to reflect their new positions (I did not make any changes to their bonus points). I cleaned the list of player to remove anyone with less than 10 GP and also thinned players from the lower price brackets with low PPM (because the 8th best 4.5 defender would never be picked by the model). I set the optimizer to pick 11 players (following the FPL starting 11 restrictions) using 82m as the budget (allows four 4.5m bench players). I tried to include captaincy points, but it proved to be too complex. In the end, each team the optimiser generated had at least one of the top 5 PPM players in it, so I doubt the results would differ much if it was included.
The “Perfect” Team
So once everything was entered, this is the team that was spit out.
Five at the back was no surprise to me, but I found it interesting that Zinchecko was selected over Laporte. With no room for a LIV attacker, Sterling with a handful of sub premium attackers makes sense given the remaining budget. A premium keeper also makes sense as the optimiser does not look at rotating keepers.
My next questions was how much value do the premium defenders bring? Are they still worth it if their returns drop off a bit this year? To test this, I applied a flat reduction on all defender’s PPM (in 5% increments). The results are below:
It wasn’t until an 80% reduction in returns for defenders that the optimizer moved away from five at the back. To put this in context, the LIV defenders scored around 200 points last season. 5% of 200 is 10 points, 20% of 200 is 40 points. 40 points is 8-10 less returns (cleans sheets, goals, or assists) and would drop their ~30 returns to ~20 returns over the season (a nearly 35% drop). Laporte was around 180 points last season so a 20% drop puts him around 145 points (4.1 PPM). This means that the threshold for five at the back is in the order of 4.1 to 4.3 PPM for the premium assets (Zinchenko is on the same PPM as Laporte, but is only 5.5m).
The keeper also switched between a budget Gunn (he may not play, but is within 0.1 to 0.3 PPM of a couple other 4.5m keepers) and a premium Lloris. This must have come down to a couple of break points for other players. Also, Wilson (8.0) is a mainstay in the teams with either Vardy (a sub premium forward) or Jiminez (upper mid-priced) only once the midfield is maxed out with five players.
Second Tier Defensive Choices
Only seven unique defenders have shown up to this point and these include Zinchecko (a twitter favourite) and Alonso, but with some doubts to the extent of minutes they will get. If we remove these players, the resulting team we get is:
Removal of these two defenders leaves us with only four at the back, and also two strikers. Sterling remains as the only premium player with Jimenez preferred over Vardy. If we take this one step further and remove all seven of the top defenders from optimizer, we end up with:
A five at the back with a power midfield. The only issue is there are some fringe defender picks in here with several playing very few games. I ran the optimizer again, but only with defenders that have played at least 20 games last season:
A 4-5-1 with a premium keeper and not too far off some ‘twitter template’ teams. The defenders all seem like fairly constant players and LEI options are prefered in the midfield.
Looking at these ‘second tier’ defender teams, it shows the value that is available in the defense (4-5 at the back) beyond the first choice premium defenders and that loading up on premium attackers doesn’t seem worth it. The price of the premium defenders seems to limit teams to one premium option.
Removal of Wilson
Wilson has been a mainstay in the drafts and so I had a look at a version without him, the result looks like this:
Interestingly enough, Aguero slots in as the striker without Wilson. Going back to the original team, Vardy and Pogba became Tielemans and Zaha. The removal of Wilson looks like a good way to get a second premium asset with 5 at the back. The optimizer suggests a premium forward, but you could look at a mid with a 6.0 to 6.5m striker. Mind you, with triple LIV defense, Mane and Salah are off the table.
This analysis has been done assuming 4.5£ players on the bench. Outside of the defence and maybe some mids, this price point will likely result in a non-playing bench. This is worsened when playing only 1 striker in your starting 11. I tried one more scenario with only 80£ in the starting 11 to allow 1 or 2 playing bench players. This resulted in Vardy becoming Zaha. At 78£, Pogba also become Perez. No new names or restructures showing up with this exercise, indicating that there is some flexibility in the team to beef up the bench a bit to enable rotation or cover absentees.
Points per Match
Looking at all the teams generated above, their points per match are provided below:
|95% Defender Points||58.9|
|90% Defender Points||57.5|
|85% Defender Points||55.9|
|80% Defender Points||55.1|
|75% Defender Points||54.2|
|No Alonso & Zinchenko||58.1|
|No top 7 Defenders||56.6|
|No top 7, 20 GP minimum||56.1|
The points per match for these teams peak around 60 (remember, these do not include captaincy points). Avoiding the top 7 defenders results in the biggest point drop down to around 56 PPM. Removing Wilson or reducing the budget only drops the PPM by 0.5 to 1. If you factor in the potential benefit from a playing bench through rotation (based on fixtures) and missed games, this margin may be even less.
To provide some context of the PPM of each player in these teams, below is the PPM vs price plot for each position. These plots include all players (even those not in the optimizer) and have not been cleaned to remove outliers and one game wonders. Players identified by the optimizer are noted with red labels.
In the context of the other players, it is easy to see why the optimizer preferred the players it did. It also provides you with information on a) players who are very similar to the selected players, and b) the kind of output that would be required to make the team in each price point.
While the information above is valuable, there are limitations to the model, a couple things to consider:
Accuracy of data
With nearly 200 players in the optimizer, the task to go through and create a realistic projection of points based on past performances, managerial/structural changes, transfers, and any other number of things is to onerous. Using the PPM of up to 38 games from last season is fairly recent and appropriate, but it isn’t 100%. Redmond for example got all of his attacking returns following a managerial change, Jota significantly increased his output after he started playing up front with Jimenez. Even if the player was consistent across the 38 gameweeks, their stats could be different this season for any number of reasons.
The optimizer creates a set-and-forget team with no consideration for rotation. There are ways to adjust for this (boosting defenders PPM, for example), but I did not account for this. Outside of the reduction of the team budget, the optimizer assumes your 11 will play gameweek in and gameweek out.
The optimizer is using PPM data based on all of last season. Players’ PPM is impacted by the venue and opposition. A player with a favourable run of fixtures to start the season may be better than a higher PPM player with a tough start.
The optimizer does not include captaincy points. This is not a huge factor as all drafts had Sterling who had the most PPM last season and would be a great set and forget captain choice.
New players to the game and promoted teams were not included in the analysis. They may offer value in their price brackets, so keep an eye on them.
Some players go through hot and cold steaks while others (Vardy, KDB, and Sigurdsson) can be very consistent over the season. These streaky players can outscore all the other players for a portion of the season so get them in if you can.
I doubt that the optimum team generated will be the ideal team for the coming season, but we can use some of the findings to shape our successful teams.
- There is value in the defence. Only with nearly 65% of returns did five at the back become “un-optimum”. There is serious value back there and you should look to capitalize on it, even if you want to shy away from the most expensive defenders.
- Find your forward(s). Sub 6.5£ forwards didn’t make the cut. They get outscored by their defensive and midfield counterparts and offer poor value. But the cheapest options are worse. One or two sub-premium/mid priced forwards could do wonders for your team and get you returns without breaking the bank.
- Midfield value. Midfielders offer great value at several price points. If you can find 2-3 mid priced forwards in form, that can be huge boost for your team. Research by others has shown that swapping a couple of these players throughout the season can get you big points (if done correctly). (Article)
- Captaincy picks. Make sure you have a high PPM player (or two) to slap the armband on. However, filling your team with expensive premiums and heap budget players is not the best option. Forgoing an additional premium player for a couple mid priced ones can get you a bit more value. (Article)
- Finally, trust your own projections. Set your team based on the points you think the players will score. Look into the short and long term history of the player and factor in all the things you want.