Scout Report – Martin Olsson

Apologies in advance for the very long post 🙂

Cracking VFM: A Fantasy Football Strategy

1. Introduction

It’s usually around this period of time that most of the Fantasy Football Managers start considering how to structure their teams.
Questions like the below are typical brain-teasers.

• How should I spend my budget?
• Should I go for a cheap defense?
• Is it feasible / effective to spend a big amount of the budget on three expensive strikers?

Some nice and interesting posts recently highlighted the need for balance. Like that from “In Tommy We Trust”: (http://www.fantasyfootballscout.co.uk/2013/07/04/scout-report-allan-mcgregor/?hc_page=3&hc_refresh=744#hc_comment_5015557).
He stated that going for an expensive defense is not a good strategy and that opting for an Uber defense is a viable option especially after considering the playing time constraint.

Daniel, on the other hand, suggests (http://www.fantasyfootballscout.co.uk/2013/07/04/scout-report-allan-mcgregor/?hc_page=2#hc_comment_5015270) that the most effective way to play in the long run is by having a less Uber-heavy team. He considers them as the most difficult to replace in case of injury/rotation and at the same time the most tricky to identify the good ones.

2. The need for Strategy

It’s common logic nowadays that in order to be successful one needs a great foundation, an effective strategy. The best strategies are those that “force” the managers to make clear choices about the allocation of the available resources (the budget in our case). Especially, when we are talking about the initial allocation of a budget, it is critical that we distribute the budget to the most “profitable” units of our team by having a long term vision.
Nevertheless, the importance of trend analysis grows when it comes to looking for alternatives to diversify our team and gain value by creatively reallocating resources and by changing our team’s composition through transfers (due to form, double game-week schedule etc.). In all those cases, a different tactical plan needs to be in place that should work in accordance and for the benefit of our long term strategy. (e.g. RVP gets injured…we should not ship him out for a cheap one and invest the money in getting the most expensive Goalkeeper).

http://s2.postimg.org/rplsk1jk9/image.gif

Strategy is the combination of those two sided interactions between the analysis done and the implementation of our strategic choice.
Similarly, for FPL, we need to come up with a strategy bases on some observations and analysis and then implement it week in week out. Hopefully there are plenty of nice posts and analysis here to make our life easier.

3. FPL Strategy

A fantasy football manager needs a solid basis in order to be successful in the long run. The main issue with that is that it is quite difficult to quantify and predict all possible variables and come up with optimal solutions especially in such a dynamic environment as that of the Premiere League. We should though take a look at the game and make some observations.

First of all, one important (probably the most important one) aspect of the game is the VFM effect (value for money). You typically get what you pay for. There are some exceptions always, but in general, and especially after considering the inflating prices of bandwagons, to win in FPL you should allocate the budget in a way that you maximize the points returned.

Every player has a price though and one simply cannot go and get even half of the good players at the same time since there is a certain budget to be spent. On top of that, there are more rules (max players per team/position etc.) that make it more complex.

So the central question that arises is what type of players should somebody invest in? Should you spend more on midfielders or get a great keeper and 4 defenders from top teams?

A way to interpret this problem, quantify it and derive to a solution is by conducting some sort of opportunity cost analysis in order to find the optimal set up of the team.

The main downside of that is the absence of available data to most of us apart from points and the market price of a player at certain stages of a season (beginning/end). Another critical issue is that we cannot incorporate easily the “form” impact into this equation for the same reason described above.

Even though, the form of a team or a player can be key, in many cases, what matters more is the strategy that one has decided to go for, simply because that drives decision making in the first place (usually via biases).

For instance, let’s suppose that an FPL manager has decided to go with only one expensive defender and picked Vidic. 4 gameweeks later, he sees that Kompany does an excellent job. So, what he does, is to get rid of Vidic and get Kompany without really affecting the balance of the team. Intentionally or not people do follow a strategy also because of the budget restriction, as the initial decisions makes it difficult to change the shape of the team completely.


4. Problem solving

So, as all of us, fantasy football managers know, we start with a budget of 100 Millions (always think of them as being millions…I bet most of you do exactly the same!) and we need to confirm a squad of 15 players. No more, no less.
The first thing one can do is to equally split the budget among 15 players.

http://s12.postimg.org/lhl1t2zvx/image.png

Each of the players will cost 6,7 M and the problem is solved. Why then nobody is doing that? What drives as into “breaking” the pattern?

The answer is the return on assets. In our case the finance term can be translated into the below (per player) formula:

FPL player ROA= Fantasy Football Points received / Price of player

Well, you don’t need to do any crazy research to realize that going like that is a suicide mission. It is almost certain that most of the 6.5 players will fail big time and we will miss all the big players of the game. In other words, we see that there is an opportunity elsewhere that is better than simply having a fully balanced team!

How then do we decide what is value for money? Is it the gut feeling that we have that this year all strikers will go crazy so we should buy 3 expensive strikers and get really cheap midfielders? Or did we just read on the forums that you have to be crazy going for 3 expensive defenders?
As we allocate our budget into a group of players for our team, in reality we do nothing other than cutting the budget from a group of assets that is not believed to have the same ROA as another. Which are those groups?

That is really up to us but and how deep we want to go in our analysis. In this analysis, I decided to go for two of them (probably the most critical ones). The position and the market price. So, one group is Defenders that are valued for 4.5 M another one is defenders that cost 6 M, another one is Goalkeepers for 5.5 etc. (you get the idea….)

Each group of those variables has an ROA (or even more than one as I will explain later). The next step is to calculate that. We could have used an average of them all but that wouldn’t work as there are many players in every group that do not play at all, something that will destroy our analysis. Another possible solution would be to use the best one or the one/s that played the most but again the figures would be biased to getting the big-guns (cherry-picking bias).

In the end, I went for 3 different ROAs per group of players that took into account the moving average of the top three players. To make it comprehensible, for the group of midfielders of 6.5 M, the first ROA is the average of the points that the top three players had. Second grade is the average of the players who got from the 2nd best up to the 4th. Similarly, the 3rd grade of this group’s ROA was calculated using the average of the 3rd to 5th best players.

Through the weighted average smoothing of ROAs, we do incorporate the fact that one will not always be able to get the top player with the best return so the average serves us well for a prediction model.

1st grade:
http://s9.postimg.org/gopx3ze9b/image.png

2nd grade:
http://s16.postimg.org/yl2elidph/image.png

3rd grade:
http://s23.postimg.org/41nph8gkb/image.png

This way also, we do take into account the fact that should somebody picks two players from the same group, he/she should expect that the second one will not be equally effective as the first one. In other words, the grades have a declining ROA which is of course logical.

An assumption we have to consider, in order to make the formula make sense, is that the FPL managers will be able to identify the good performing players from each group and actually pick those from the whole pool of players that belong in the same group. e.g. It makes no sense to get Holtby from Tottenham that costs 6.5 M and expect a return of points equal to that of Walters from Stoke. Calculating the average will give us a good enough projection of the return of points should we be able to identify one of the best for that money.

After calculating the 3-grade ROAs for all groups in the game I ended up with a matrix of 150 cells. That was just the starting point though. I did the same exercise with data from the last two seasons and run data smoothening and trend analysis in order to find all those similarities and combinations between groups and make the dataset less biased (underlying patterns among similar prices etc). So I did mixed and combined groups with other depending on seasonal trends as generally higher porices for defenders. Lastly, I incorporated last year’s final points against the initial prices to validate the projected initial value for money. There was a clear indication that defenders who performed very well during season 2011-12 were overpriced hence we should have been careful as there was a big chance of overpaying for a bubble. Actually, the results revealed that the highest danger of overpaying was by getting a 7M defender that would flop. The variance between projected and actual ROA was significant and did not justify the spending.

The next step was to develop the model that optimizes the allocation of the budget in a way that it picks the highest projected total ROA for the team. This was done by following the simple idea of opportunity cost.

Opportunity cost in simple words is: “what is the next best thing alternatively to our choice?” So let’s suppose you have a 1000 Euros and you can either go on holidays or buy a brand new bass guitar. If you buy the bass the opportunity cost is that 1000 euros spent plus the pleasure you forgo by not going on holidays. In the FPL world now, you have 2 spare millions to spend. So, should you spend it to upgrade another defender or upgrade that 5M midfielder into a 7M one?
My model takes that fundamental idea and comes up with the best spending so that the we get the maximum possible return of points, which is our desired outcome (I guess 😛 )

Of course, all restrictions that are set in the game should still be respected.

Those and their respective solution are described below:

• Total Budget : The overall budget of the team should be no more than 100 M.

• Formation : I developed optimal solutions for four main types of formation (3-4-3 , 4-3-3 , 5-3-2 & 4-5-1) but most importantly, the model is flexible enough so that the users chooses his/her favorite one and get the optimal allocation of the budget with the max total ROA. Of course, the formation picked should be among the ones allowed by FPL, so no 0-5-5s are allowed 😉

• Substitutions and rotation : The tool does take into account the positions of the players you want to leave at your bench and the cost but it considers them as inactive players. In the example below, I opted for an OK-ish 2nd goalkeeper, two cheap defenders and a cheap midfielder. That makes up a total of 18.5 M. As a result the main team should have a total budget of 81.5 M.
Again, the tool is flexible for all kind of different variations but it is not designed to optimize your rotation strategy. That is a project for another day!

http://s16.postimg.org/dn0u1638l/image.png

5. Model Results and interpretation

A 3-4-3 start-up team with a 100% balanced allocation of the budget looks as below:

http://s12.postimg.org/m7d0wxf71/image.png

The optimal solution long term is:

http://s23.postimg.org/ucrwkxj2z/image.png

How should we interpret the output of the model?

First of all, we should definitely not go and buy the players that match this scenario but instead, as discussed in the beginning, this should be only the foundation of the high level allocation of our budget especially at the beginning of the season where everybody is struggling to identify the best value for money.

The main takeaways are the below for a 3-4-3 formation:

• One expensive DF is enough
• Same goes for the Midfielder, as one expensive MF is enough
• Up front you need 2 of the best (expensive)


Doing the same exercise for multiple formations, reveal other equally interesting points which are pivotal especially after considering how dynamic the FPL environment is and how agile a winning team requires to be.

http://s16.postimg.org/agavva86d/image.png

Main observations:

- There are certain types of players that are MUST in all formations. These are:
• 2 cheap-ish DF
• 1 cheap MF
• 1 very expensive MF
• 1 very expensive FW

- It makes sense to invest in a more expensive GK when you play consistently with 3 DF as you cannot find such an ROA for 2M more elsewhere.

- Up front, the first pick should always be the top-gun (RVP type of player), followed closely by the 2nd also expensive. The third pick should be a cheap one because spending more money there will not offer an equally good return as spending it elsewhere.

6. To be done…

The analysis done above focuses mainly in creating a stable and effective team by allocating the budget where the real value is.

Once the prices are published I will run the analysis again and provide an update including this time this season’s starting prices as well. There is a chance of providing the tool so that everyone can play around as well.

Finally, looking from a different angle, it would be nice to further investigate and share some numbers on how easy it is considered to find players within the same price group that have the same return. In other words, which groups of players are easier to get a stable amount of return and which have just 1 or 2 great performers and the rest is crap.

Building the strategy for long term success together with some nice agile tactical tips would be a nice way to go!

7. Final remarks

At the end of the day, these are only numbers and projections but I would take a fact based analysis any day rather than a tip from a random guy who believes that getting three big strikers makes his team much more effective in the long run.

Carefree,
Pirate