Football trading : xG - What's it telling me?

Football, Soccer - whatever you call it. It is the beautiful game.
User avatar
Kafkaesque
Posts: 265
Joined: Fri Oct 06, 2017 10:20 am

Thu Jan 04, 2018 12:22 am

Wolf1877 wrote:
Wed Jan 03, 2018 11:55 am
There are some glaring failings in xG in my view. So I'm sort of half in Stelling's camp.

For a start off there is the accuracy of the xG model's measure of the goal attempt probability. I think Peter pointed out the extreme long range Wayne Rooney goal for Everton v West Ham failed to take account of the fact that the keeper was 30 yards away from his goal. The Mahrez goal for Leicester v Spurs in November from just outside the penalty area was assigned a 0.04 probability by understat (i.e. odds of 25) which in my opinion vastly understated the probability of a player of Mahrez's quality scoring from that position with the Spurs defenders all standing a couple of yards off him. I'd estimate the probability in that case as somewhere between 0.1 and 0.2 rather than the 0.04 calculated by the understat algorithm.


I think, this (the Mahrez bit) is common misunderstanding. One of the reasons xG came to was that someone realized that there's very little difference between the outcome based on who takes the shot. So Mahrez's ability has little impact; at least so little that it's statistically irrelevant. Sure there'd be a difference between Gary down from the pub and Mahrez. But Mahrez, Kane, Aguero, Shane Long, Rondon, or Vokes matters little. The difference is the better players play in better teams, that move the ball faster and gets it into more promising positions, whereas say Rondon will just whack it at goal at first sight. Look at someone like Messi. He very rarely takes on difficult shots, for the simple reason that he doesn't have to. Kane use to shot at everyone that moved a few years ago, but has also become much smarter with his finishing as the team has gotten better.

As for the Rooney thing, that's such a rare instance that it matters little.
Wolf1877 wrote:
Wed Jan 03, 2018 11:55 am
Another factor that xG fails to recognise that teams tend to up the tempo and attack when they need a goal and tend to defend when they dont need a goal (unless they are completely dominant like Man City). So the timing of goals will greatly affect the xG stats for most teams IMO. So taking last nights Southampton v Palace game. Southampton went ahead on 19 minutes. After that point Southampton's xG barely rose for the rest of the game. I suspect that Southamptons xG would have finished far higher if they had not scored so early as on 19 minutes as they would have likely carried on attacking Palace and accumulating a bigger xG for longer until they actully scored.
Not really a flaw in the system imho and it evens out over a full season in most cases. Say there was a 1 in 10 chance that Soton took an early lead, it should be factored in for future calcs, as there'll be a 1 in 10 of something similar happening in future matches of about the same character.
ShaunWhite wrote:
Wed Jan 03, 2018 3:56 pm
Is there even a standard was to calculate xG ?
Who's decided that when Sterling passed the ball last week he'd created a 30.2764% chance of a goal being the outcome ? To even suggest you can calcuate that from even a dozen matches is slightly sillly gven the huge number of variables, pitch condition, opponent strength, his current 'form' etc etc etc. I know these things aren't about 1 game, it's about averages, but sample size and variance have been conveniently ignored.
I don't know where you're "a dozen matches" from? The standard to calculate it is by looking at the thousands upon thousands of matches since Opta starting recording shot time, shot position etc. and then finding out how big a percentage of shots goes in, when the position, condition, shot type, pass type etc. are the same.

Incidently, given that you're a trader in the NBA, you might be interested to know that NBA teams are to varying degrees using a Basket version of xG. My understanding is that a lot of teams have toned it down from being the whole foundation of their set-up, but it's still significant. Most prominently in how the numbers of 3-pointers have risen dramatically. A team like the Cavs will take 3's all day long with noone to rebound, if it's the percentage-wise best shot. And someone like James Harden has spots, where he'll never ever take a shot from because the xG is poor.
ShaunWhite wrote:
Wed Jan 03, 2018 3:56 pm
xG also seems to be designed to produce inherently plausible results. A scale has been invented to say a team should have scored between say 0.5 and 3.5 goals (i don't see many xGs much higher) in a sport where teams usually score between 0 and 4 goals. It's a figure that's been reverse engineered. It's always going to be quite close but give the impression that there's something to be learned from the modest variations.
Reverse engineered is simply wrong and, frankly, a gross misunderstanding of how it's compiled and works. I fail to see how it would at all be uncomprehensable that most teams weren't likely to score more than 3.5 goals (as calculated after the fact). I've watched a ton of football in my time, and most matches were a team gets 4+ goals, they were getting the rub of the green with at least some of their goals, and scoring a higher percentage than what I would from a purely logical standpoint (as opposed to after the fact statistical). It's a huge rarity that I'll watch a match where one team gets 4+ and I'm left thinking, they could or should have had more than that based on the match.

A quick check of number of goals by each team in PL this season for 0, 1, 2, 3, 4+ respectively is 32, 30, 21, 9 and 8 % with similar numbers last season, so lumbing 0-4 goals together as what teams score is a bit simplified imo.

User avatar
ShaunWhite
Posts: 2933
Joined: Sat Sep 03, 2016 3:42 am

Thu Jan 04, 2018 2:39 am

Thanks for your views, it was interesting.
Kafkaesque wrote:
Thu Jan 04, 2018 12:22 am
A quick check of number of goals by each team in PL this season for 0, 1, 2, 3, 4+ respectively is 32, 30, 21, 9 and 8 % with similar numbers last season, so lumbing 0-4 goals together as what teams score is a bit simplified imo.
I just mean't that the normal range of goals was between 0 and 4, obviously not that each outcome was equally likely.
32% + 30% + 21% + 9% + 8% = ?
Kafkaesque wrote:
Thu Jan 04, 2018 12:22 am
I don't know where you're "a dozen matches" from?
It's starting to look like xG isn't as complex as I thought, the reason I'd referred to the stats over a dozen games was that I thought expectation was calculated for each player in context, not that it was an average over hundereds of random players and games.
Kafkaesque wrote:
Thu Jan 04, 2018 12:22 am
One of the reasons xG came to was that someone realized that there's very little difference between the outcome based on who takes the shot.
That's the killer one that's hard to believe and impossible to prove.

Do you have the stats on the personal xGs of all the strikers in the the PL for this season? I'd love to know what their xG varience is from this mythical Standardo Generico fella who's played in 10,000 games.

I get that top flight players are all of a high standard and probably don't deviate from the average that much, but if one guy is worth £75m and the guy next to him is worth £40m, somebody thinks that very small difference is worth having. Maybe it's just 2 or 3 extra goals from a 100 opportunities, but depending when they occur that could win a championship or serious spoil a 'system'. And that variation exists in every team.

I do accept though that this arguement doesn't prove a link between value and conversion rate as his additional value might be in other less direct ways, tackle success, pass completion, leadership or many others etc.
Kafkaesque wrote:
Thu Jan 04, 2018 12:22 am
Incidently, given that you're a trader in the NBA, you might be interested to know that NBA teams are to varying degrees using a Basket version of xG.
It's interesting you say that because I've been looking at basketball from this angle since I heard about xG in football.

The original premise being that,
-If only 3 in 10,000 are drafted into the NBA then surely they must all be within a gnats whisker of each other.
-With only 5 on the court not 11 the variables are massively reduced.
-Players have many more games/shots/goals so you can obtain a better sample without going back years
-In basketball where you're pretty much either assisting or shooting rather than being generally being more unqualifiably 'useful'
-No goalie
-Smaller playing area
-Only 30 teams in the whole country
...and one or two others, a piece of cake compared to football so it seems?

So far I've found almost nothing in data or on the net. Different players deviate from the norm enough to make a significant difference to the outcome. Again justifying certain players being worth a lot more. You've only got to look at free-shots, a man, a basket, no goalie or even surface, 100's taken in a season. Certain players outperform others week in week out.

Now I admit I'm a million miles from being a statistician but when you say this....
Kafkaesque wrote:
Thu Jan 04, 2018 12:22 am
My understanding is that a lot of teams have toned it down from being the whole foundation of their set-up,
....then it's probably something I won't find if the entire might of NBA are downgrading it's importance. And basketball is much much more simple than football.
Kafkaesque wrote:
Thu Jan 04, 2018 12:22 am
And someone like James Harden has spots, where he'll never ever take a shot from because the xG is poor.
No way, James Harden has spots where he'll never ever take a shot from because his xG is poor. Basketball has so few variables, not even weather, that players know their strong areas from taking 1000s of sample shots in practice. Different players have different throwing styles and faults that make certain angles more appealing. They don't just shoot from where the nationwide hotspot is.

If this general notion of 'average' xG was poor, nobody would shoot from those spots, not just Harden.
Kafkaesque wrote:
Thu Jan 04, 2018 12:22 am
Most prominently in how the numbers of 3-pointers have risen dramatically.
The are factors other than guessing xB caused that change.

The rise in 3-pointers is partly driven by the success of teams who's coaches have favoured 3-point attempts. Favouring 2-point or 3-point attempts makes little statistical difference (again or they'd all play the same way) There's also a case of crap teams trying to copy the style of the big boys, but the big boys favour 3s because they have the guys who can score them. The gameflow psychology is different too with bigger swings and coaching styles go in and out of fashion.

There's also issues around how refereeing has changed slightly, more offensive fouls, less defensive ones, which has made short 2s less appealing, and if you're going to try long 2s, it might as well be a 3.

I can't back that up because I only heard it in commentary and can't easy find the stats to confirm it.


..sorry all for writting such a stupidly huge message and straying off football, but I think the parallels with basketball xB are worth consideration.

Is anyone looking at darts or snooker from this angle? I don't follow them so I don't know.
Football seems the most complicated sport you could ever apply it to, has it been proven in the much more simple sports?

Wolf1877
Posts: 120
Joined: Fri Sep 08, 2017 10:59 am

Thu Jan 04, 2018 9:56 am

Kafkaesque wrote:
Thu Jan 04, 2018 12:22 am
Wolf1877 wrote:
Wed Jan 03, 2018 11:55 am
There are some glaring failings in xG in my view. So I'm sort of half in Stelling's camp.

For a start off there is the accuracy of the xG model's measure of the goal attempt probability. I think Peter pointed out the extreme long range Wayne Rooney goal for Everton v West Ham failed to take account of the fact that the keeper was 30 yards away from his goal. The Mahrez goal for Leicester v Spurs in November from just outside the penalty area was assigned a 0.04 probability by understat (i.e. odds of 25) which in my opinion vastly understated the probability of a player of Mahrez's quality scoring from that position with the Spurs defenders all standing a couple of yards off him. I'd estimate the probability in that case as somewhere between 0.1 and 0.2 rather than the 0.04 calculated by the understat algorithm.


I think, this (the Mahrez bit) is common misunderstanding. One of the reasons xG came to was that someone realized that there's very little difference between the outcome based on who takes the shot. So Mahrez's ability has little impact; at least so little that it's statistically irrelevant. Sure there'd be a difference between Gary down from the pub and Mahrez. But Mahrez, Kane, Aguero, Shane Long, Rondon, or Vokes matters little. The difference is the better players play in better teams, that move the ball faster and gets it into more promising positions, whereas say Rondon will just whack it at goal at first sight. Look at someone like Messi. He very rarely takes on difficult shots, for the simple reason that he doesn't have to. Kane use to shot at everyone that moved a few years ago, but has also become much smarter with his finishing as the team has gotten better.

As for the Rooney thing, that's such a rare instance that it matters little.
Kafkaesque, I agree with some of your input here. Given the shooting opportunity that opened up for Mahrez v Spurs then someone like Rooney or another player with good ability to pick his spot out from distance whilst under relatively little pressure from defenders would have been just as likely as Mahrez to score. I dont however agree that it is statistically irrelevant who takes the shot. Shooting ability varies widely even amongst top class players and if xG models are to be improved then they need to take this into account. My point is that when I review a video of the play that the probability of Mahrez scoring from that shot was not 0.04 but in my estimation is around 0.15. I believe that the xG model assigning a probability of 0.04 is probably accurate for a typical shot from that position because usually any shot from that position would be rushed and under intense defensive pressure. Mahrez had time to pick his shot out due to defenders standing off him and the Understat xG model is not reflecting that in my opinion. I agree that the Rooney v West Ham situation is statistically irrelevant.
https://understat.com/match/7249

I believe that the current xG models are still in their infancy which could be hugely improved in the years to come. I also believe that xG can also potentially be a useful metric however the underlying princile of xG is based on the accuracy of its probability estimations for individual goal attempts. At the moment that probability is very rough and ready and currently appears to be little more than a heatmap.
Kafkaesque wrote:
Thu Jan 04, 2018 12:22 am
Wolf1877 wrote:
Wed Jan 03, 2018 11:55 am
Another factor that xG fails to recognise that teams tend to up the tempo and attack when they need a goal and tend to defend when they dont need a goal (unless they are completely dominant like Man City). So the timing of goals will greatly affect the xG stats for most teams IMO. So taking last nights Southampton v Palace game. Southampton went ahead on 19 minutes. After that point Southampton's xG barely rose for the rest of the game. I suspect that Southamptons xG would have finished far higher if they had not scored so early as on 19 minutes as they would have likely carried on attacking Palace and accumulating a bigger xG for longer until they actully scored.
Not really a flaw in the system imho and it evens out over a full season in most cases. Say there was a 1 in 10 chance that Soton took an early lead, it should be factored in for future calcs, as there'll be a 1 in 10 of something similar happening in future matches of about the same character.
I personally dont think there is very much to be gained by averaging xG stats out over the course of a entire season as it would just become meaningless noise that may still appear to be credible. The way that headline xG stats are currently digested by most people is probably fundmentally wrong though. I'd love to have xG stats for last nights Benfica v Sporting Lisbon game. Two quality and evenly matched teams for that league. Benficas home record was W7 D0 L0 and Sportings away record was W6 D1 L0. Sporting took the lead on 19 minutes and Benfica finally equalised on 90 minutes. Benfica had 24 shots with 4 on target. Sporting 9 shots with 3 on target so I can only assume the xG stats would probably be somewhere around 2.5 - 1. My assertion is that had Benfica scored on 19 instead of Sporting that the xG totals would very likely have been the other way around. Consequently the xG headline figures from last nights game are not very useful to project forward IMO because the headline stats fail to take into account the overall context of how the the game panned out. I'd suggest that individual goal attempt xG stats could potentially be used more effectively by collating the individual stats depending on game scenario that they occur in. i.e. what were Benfica's or Sporting's xG stats for periods where they are losing by a one goal margin to a similarly ranked team. Surely for inplay exchange traders that is far more relevant than the seasons xG league table.

spreadbetting
Posts: 1450
Joined: Sun Jan 31, 2010 8:06 pm

Thu Jan 04, 2018 1:04 pm

Wolf1877 wrote:
Thu Jan 04, 2018 9:56 am

I personally dont think there is very much to be gained by averaging xG stats out over the course of a entire season as it would just become meaningless noise that may still appear to be credible.
In theory surely they should just average out alongside the match goal totals/averages. It's the underlying data on teams/individuals that may be of value, certainly will be value to the data providers as I'd guess all of the big teams will throw cash at it in the hope of some slight edge over the opposition.

User avatar
ShaunWhite
Posts: 2933
Joined: Sat Sep 03, 2016 3:42 am

Thu Jan 04, 2018 3:18 pm

Football seems to be a very complex sport to develop a method on, are the principals proven on the easy stuff like darts, snooker or tennis?

I don't trade it myself but when people trade tennis, do they look at indiviual player stats or average stats for everyone on the tour ?

Orixian
Posts: 74
Joined: Sun Sep 06, 2015 12:36 am

Wed Jan 17, 2018 2:41 pm

Apart from understat where else offers xG. I cant see it on who scored.

Orixian
Posts: 74
Joined: Sun Sep 06, 2015 12:36 am

Wed Jan 17, 2018 6:42 pm

I emailed Understat and asked them if their model accounts for the uniqueness of each player and they said itdidn'tbut they are working on a model that does.

chris2785
Posts: 30
Joined: Sun Dec 11, 2011 1:32 am

Sun Mar 18, 2018 6:09 pm

Hi, I wondered if someone could point me in the direction of a good website that shows xG for teams as matches are inplay etc so I can look at a game at half time to see how its going - apologies if someone has answered this already.

Thanks in advance
Chris

Post Reply

Return to “Football trading”

  • Information
  • Who is online

    Users browsing this forum: No registered users and 3 guests