Building a correct score model

Football, Soccer - whatever you call it. It is the beautiful game.
Post Reply
User avatar
brimson25
Posts: 504
Joined: Sat Apr 08, 2017 11:42 am

Morning all,

I wonder if someone could help. I'd really like to sharpen up my correct score game. Is there anyway I can get correct score data without paying betfair 100s of pounds?

I downloaded the free stuff but it seemed to be in a database format and I couldn't make sense of it.

Please be gentle in responses! This is my first foray into data analysis and I'm not even sure if this is the right question.
User avatar
Euler
Posts: 24702
Joined: Wed Nov 10, 2010 1:39 pm
Location: Bet Angel HQ

There is some wonderful data right here on the forum. I'll try and find it in a bit when I get to the office.
User avatar
brimson25
Posts: 504
Joined: Sat Apr 08, 2017 11:42 am

Thank you very much. I'm starting to sift myself. This is a huge task.
User avatar
brimson25
Posts: 504
Joined: Sat Apr 08, 2017 11:42 am

I'm enjoying this more than I expected but I'm as confused as hell!
User avatar
Euler
Posts: 24702
Joined: Wed Nov 10, 2010 1:39 pm
Location: Bet Angel HQ

spreadbetting
Posts: 3140
Joined: Sun Jan 31, 2010 8:06 pm

Most of the odds are derived from people using their own tweaked versions of a poisson distribution so it'll all boil down to how good you are at setting your initial expectations of goals for each team to get that starting point.

From there a lot depends on what you're hoping to acheive from the data obviously many people will simply try to beat the odds outright whereas others may look for exploiting situations where the odds are out of line with their assessment in play, either thru bare stats or their view of how the game will pan out + stats. Understanding the maths behind the odds or at least how markets decay over time might give you an edge or at the very least tell you when not to play.
User avatar
Euler
Posts: 24702
Joined: Wed Nov 10, 2010 1:39 pm
Location: Bet Angel HQ

Poisson distribution doesn't work on football scores, so don't pursue that. Create a model independent of it.

https://www.youtube.com/watch?v=hXQoqUwiSIA
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

Euler wrote:
Wed Jul 04, 2018 12:01 pm
Poisson distribution doesn't work on football scores, so don't pursue that. Create a model independent of it.
Is the problem that Poisson over estimates the Nil score because it can't accomodate the theoretical negative scores? I've just dug out a spreadsheet I did around it but I can't quite understand what it was I was doing in it. :roll:
User avatar
brimson25
Posts: 504
Joined: Sat Apr 08, 2017 11:42 am

Euler wrote:
Wed Jul 04, 2018 11:50 am
viewtopic.php?f=6&t=14776
Thanks very much! Round 2 about to commence of me wrangling with numbers before I discover as I fear that declining odds markets like correct score offer only the illusion of money...
User avatar
brimson25
Posts: 504
Joined: Sat Apr 08, 2017 11:42 am

spreadbetting wrote:
Wed Jul 04, 2018 11:57 am
Most of the odds are derived from people using their own tweaked versions of a poisson distribution so it'll all boil down to how good you are at setting your initial expectations of goals for each team to get that starting point.

From there a lot depends on what you're hoping to acheive from the data obviously many people will simply try to beat the odds outright whereas others may look for exploiting situations where the odds are out of line with their assessment in play, either thru bare stats or their view of how the game will pan out + stats. Understanding the maths behind the odds or at least how markets decay over time might give you an edge or at the very least tell you when not to play.
Thanks very much, that's helpful.
User avatar
Kafkaesque
Posts: 886
Joined: Fri Oct 06, 2017 10:20 am

Euler wrote:
Wed Jul 04, 2018 12:01 pm
Poisson distribution doesn't work on football scores, so don't pursue that. Create a model independent of it.

https://www.youtube.com/watch?v=hXQoqUwiSIA
Like Shaun, I'm curious as why you state poisson doesn't work?

The video, you reference, explains a good alternative, but not why poisson shouldn't be pursued. Given how much you stated it as an absolute, I would suspect, you have a specific reason for it?
User avatar
ShaunWhite
Posts: 9731
Joined: Sat Sep 03, 2016 3:42 am

Kafkaesque wrote:
Thu Jul 05, 2018 1:18 pm
Euler wrote:
Wed Jul 04, 2018 12:01 pm
Poisson distribution doesn't work on football scores, so don't pursue that. Create a model independent of it.
Like Shaun, I'm curious as why you state poisson doesn't work?
I'm not curious (well not in that way) I know it's flawed. I just can't remember why....I'm sure it's something to do with the results being very close to zero. Eg great for results either side of 50, but not so good with results either side of 1.5. :?:

Poisson is afterall just a theoretical distribution formula, it takes no account of home vs away or if teams need a pasrticular result more than another.

If you already have 10 years of match results data then that's clearly a better picture of the actual distribution. For those without the stats, and ways to slice them for certain situations, they're probably left with poisson being their best hope.
spreadbetting
Posts: 3140
Joined: Sun Jan 31, 2010 8:06 pm

Simple poisson distributions have been shown to underestimate draws when compared to historical odds, there's plenty of academic articles about poisson and football. Because of that many people have tweaked versions of poisson to account for the draw deficiency using bivariate poisson models etc

I think there are so many variables when it comes to predicting football results you'll be lucky to find an edge over the market. If we look at the video Euler posted he's getting 11.8 for the draw whereas the Betfair market is sitting at 14 mid point so there's a large variation between the two and I'd be surprised if Euler is acheiving a 20% edge over the Betfair market with a simple system.

Take a look at the power of goals blog as that has quite alot of decent posts about football predicting http://thepowerofgoals.blogspot.com.

If you look at the post regarding odds decay over a match you'll see poisson distributions do tally in line with Betfair odds, now that may be due to the majority of MM's using their flawed poisson models but it does mean we have a half decent method of predicting future odds decay/distribution if not the actual scores.

http://thepowerofgoals.blogspot.com/201 ... t-ten.html

I gave up trying to get a model to predict scores a long time ago.
Post Reply

Return to “Football trading”