Building a winning squad using data from the 2012-2013 season

It’s that time of year again. The time when every single one of the millions of participants in the Fantasy Premier League (FPL) are standing on equal ground. The time when hopes are high, bets are hedged, and punts are made. The time when fantasy managers are actually considering starting Fernando Torres, because he can’t continue this poor run of form forever, can he?

It is also the time of tinkering. With transfer rumours swirling and new managers settling in, everyone is attempting to predict who will be leaving for greener pastures and who will enjoy greater success under new management. For most fantasy squads, the excruciating month leading up to the first kick-off of the season leads to personnel turnover in quantities that would distress even Roman Abramovich.

Amidst all this conjecture and double-guessing, every fantasy manager is trying to figure out the best method of assembling their squad. It wouldn’t be prudent to pick all our players from top clubs even if we could afford it, but for which positions should we choose players from clubs lower down in the table? Where do we find value in the player market? How important is our goalkeeper decision? (We only have one of them after all.)

Here, we take a shot at answering these questions, amongst others, using data from the 2012-2013 FPL season.

When building a fantasy squad, the largest conundrum with which most of us internally quarrel is ‘which positions should I choose first’? Or, perhaps more importantly, ‘which positions should I fill last’? The answers to these questions are easy if we think about them objectively. The positions we should fill first are those which give us the largest differential if we choose the right players and accordingly, the positions we should choose to fill last are those which give us the smallest differential.

Luckily, assessing position-specific differential is fairly easy. Using defenders as an example, we can take the average point total of the top 15 defenders in the 2012-2013 season and subtract the average point total of all the defenders that played at least 1000 minutes. If we perform the same calculation for the top 15 midfielders, top 10 forwards, and top 5 goalkeepers (to take into account the proportional representation on a fantasy squad), we arrive at the following numbers:

Position Differential (pts)
Goalkeeper 48
Defender 59
Midfielder 86
Forward 67

It’s pretty conclusive, if unsurprising: choosing the best-performing midfielders and forwards rewards you more than picking top class defenders.

So we choose our midfielders and forwards first. From which clubs should we choose them? Well, of the top 15 players in each of these positions last year, about 50% were from clubs in the top quarter of the table:

Position Proportion from top 5 clubs
Goalkeeper 33%
Defender 47%
Midfielder 53%
Forward 47%

Clearly, if we’re hedging our bets, we should be choosing a good portion of our midfielders and forwards from top 5 clubs. But what happens when we run out of money for the big-club players who typically cost more? At that point, we need to choose players based on their value.

If we plot the number of points each player accumulated over the course of last season against that player’s average price, we can get a sense of that player’s overall value. For example, below we can see the value graph for each midfielder who played at least 1000 minutes in the Premier League last year.


Any player above the regression line outperformed their expected value. That is, they scored more points than what could be expected based on the rest of the midfielders who played at least 1000 minutes. Even just eye-balling this graph, we can see that a much higher proportion of midfielders in the bottom 15 clubs were more valuable than expected compared to midfielders in the top 5 clubs.

In fact, midfielders in the bottom 15 clubs were 1.65 times as likely to exceed their expected value. But that’s not even the largest discrepancy. No goalkeepers in the top 5 clubs were more valuable than expected. In contrast, forwards and defenders in the top 5 clubs were just as likely to exceed their expected value as those in the bottom 15 clubs. See the exact proportions of players that exceeded their expected values below.

Position Top 5 clubs Bottom 15 clubs
Goalkeeper 0% 37%
Defender 50% 51%
Midfielder 35% 58%
Forward 50% 50%

The results are pretty clear: when looking for value, we should pick midfielders and goalkeepers from the bottom 15 clubs. For defenders and forwards, it doesn’t matter which clubs you choose from when you’re looking for value.

Take home messages – building your squad

Forwards and midfielders give you the largest differential if you choose them wisely, so you ought to pick them first. Last year, half of the top 15 fantasy producers in both of these positions came from the top quarter of the table. Therefore, you should probably ensure that a decent portion of your midfielders and forwards are from top 5 clubs. If you’re strapped for cash and trying to choose an attacking player with good value, pick a midfielder from a bottom 15 club rather than a forward.

Choose your defenders next, as they provide the next largest differential if you choose them well. A defender from a top 5 club is just as likely to reach his expected value as a player from a bottom 15 club. So if you have cash leftover, choose defenders from top 5 clubs because they will get you a few more points. However, consider spending that extra cash on an upgrade for an attacking option rather than a more costly defender, as that will likely give you more points in the long run.

Goalkeepers don’t gain you as many points and a keeper from a top 5 club will not reach their expected value. Take a keeper from a bottom 15 club.

The wonderful (and frustrating) part of Fantasy Premier League is that no two seasons are alike and despite all our best efforts, we cannot predict what will happen any more than Arsène Wenger can predict when he’ll sign a transfer target. There is no right answer. All we can do is play the probability game.

Do these findings line up with your start-of-season methods for assembling your squad?