Thursday, November 13, 2014

Model behaviour

As we enter another drab international week, it presents a well-overdue chance to dig a little deeper into this year's model and look at some of the results which don't necessarily align with what the majority of people appear to be thinking.

I don't want to get too bogged down in the merits of any particular assumption in the model, though obviously as we look back and see where the model missed reality, tweaks can and are made. I do however want to illustrate how individual forecast numbers are made up and why that leads to the situation, for example, where Alexis Sanchez has so-so forecast numbers despite leading all midfielders in points to date (by a distance).

Regressing conversion rates
The first and most obvious point to note is that Sanchez has a particularly high goals per shot on target ratio (50%) which based on history is unlikely to be sustained. The model uses a regressed conversion rate based on a combination of a player's history (where applicable) combined with the average conversion rate for a player with his particular "shot profile" which consists of his position, volume and type of shots. Players like Fabregas (25%), Di Maria (30%) and Hazard (33%) have rates we would expect from their position.

Next up, we have to remember that player forecasts should always be considered in the context of their respective teams. So, while Hazard's 19% share of his team's shots inside the box is a great number, it isn't superior enough to Di Maria's (16%) to cover for the fact that in Gameweek 13 Man Utd are forecast to have a far greater number of SiB (9.4) than Chelsea (6.5). In short, you need to consider the size of the whole pie, not just the size of a given player's slice.

Let's look at the numbers for Gameweek 13 in a bit more detail:

Fabregas's goal threat is a standout number here as with 2% of Chelsea's SiB, he isn't likely to rack up a big goal total and that should immediately set off some kind of alarm as you're going to need absolute dominance in one category if you are basically performing at a league average in another. Sanchez and Di Maria have a fairly similar profile individually but we can see that with United forecast to have a much better day against Hull than Arsenal will at West Brom, Di Maria winds up with a superior points forecast. That's really the story of the next eight weeks or so, as Arsenal have the highest forecast SiB just once (GW18), which is not coincidentally the game in which the model likes Sanchez the most:

This post isn't necessarily a knock on Fabregas or Sanchez's prospects, nor a campaign for Di Maria, just a friendly reminder that the model is at least based in logic and if a ranking looks odd, there will be a reason behind it. Now, of course, whether those reasons are correct is up for debate but it's worth pausing whenever a player is presented as "must own" or undroppable.

Okay, you say, I don't need a lesson, I need some names to look at. First up, let's plot ownership with total forecast points for the next eight gameweeks and see who the masses might be sleeping on:

This chart is a little cluttered to derive masses of information from, but I believe it does a good enough job at illustrating that, on the whole, ownership numbers are based in reality. There's no one who is owned by thousands of managers who doesn't deserve at least a large part or their attention, and we don't see anyone with potentially explosive numbers who is currently being completely ignored. As is always the case then, it's on the fringes where we find value. I don't want to get into deep dives on scores of players right now, but a couple of interesting names to highlight are:

A lot of people own him but it probably should be more. He is, quite simply, the best player in the league by a distance right now and his fantasy potential is superior to everyone by some distance. Indeed, without getting into specifics, we can probably expect some regression from the majority of those players forecast around the 4.5 point per game mark, yet Aguero pushes closer to 5.0 without any signs of slowing down. Best case scenario from a fantasy perspective is that City get knocked out of Europe and we finally get that dominant Aguero season we've waited for, where he can be slotted into a side and captained freely without fear of rotation from the likes of Dzeko or Jovetic.

With this chart and the aforementioned note about Fabregas' lack of goal threat, I am a bit spooked about his long term prospects, especially in light of the fact that the former Arsenal man hasn't missed a game yet this season and now has a slight hamstring issue, which could lead to some rotation in the coming weeks. Of course he remains very ownable and the assist potential is sky high but with his ownership rate creeping over 50%, contrarians like your humble blogger will be getting itchy trigger fingers.

It's not a flashy move, but as we noted, the value is in the fringes and so ditching Clyne who is (a) widely owned and (b) 0.7m more expensive than Alderweireld who finds himself buried in the 1% ownership group despite offering just a touch less forecast value than his teammates, makes a lot of sense.

Despite Southampton's incredible success, Tadic doesn't seem to have quite earned the same attention as some of his teammates - particularly Pelle - which leaves him as one of the best handful of midfielders in terms of forecasted points yet with an ownership number more in line with his defensively minded teammates than the league's elite midfielders. At 7.9m he isn't cheap enough to be considered an absolute lock in anyone's team but then the players he's comparable to here are generally coming in at least a million more than the young Serb.

Baines' ownership number is still something of a surprise as while 25% is high, it isn't particularly high for the oft-fancied Baines yet this is the first time I recall that he might actually deserve such a high number. Everton's shaky defensive form will probably keep me away but you cannot underestimate just how much better he is than every other defender when it comes to attacking threat (and remember that the model doesn't really give extra attention to penalty takers).

For the rest of the international break I hope to do a stock check on each position and have a look at forecasting the next third of the season. I imagine many of you will be entertaining the idea of using the wildcard this week (as I have), as it's a logical idea to try and split the season into three roughly equal slices.


nooBKiat said...
This comment has been removed by the author.
nooBKiat said...


thank you for the insights.
I am just surprised Tadic's forecasted score is higher than Chadli's, considering the latter have more SIBs in the past 3-4 games, and the fact that Tadic only probably have a "big" game this season and otherwise under-performing against your model projections.

shophome philly said...

Fоr hottest іnformаtiоn
уou have tο gο to see web and οn wоrld-ωidе-web I fοund thіѕ webѕite aѕ a
bеst ѕіte for most uр-to-date uрdаtes.Look at my website can get them vip zappos shoes