Football Musings

Football, Soccer - whatever you call it. It is the beautiful game.
Post Reply
User avatar
wearthefoxhat
Posts: 3411
Joined: Sun Feb 18, 2018 9:55 am

andy28 wrote:
Fri Jan 10, 2025 11:19 pm
I use this site https://fbref.com/en/. When I say scrape I mean I just load the pages I want via Data, From Web into excel. Then just refresh each time you want it updated.

I am still not sure about using Xg tho as I looked at Arsenal Spurs match and it's saying the XG will be around 1.6, I am not sure that will be the case. I know you have to look at other data for the full picture. I am thinking of trying a binary method into Xg as that will factor out outlier 6-3 score lines

I think I will concentrate on 1 league as well, I think confidence is important, I see teams that are level with a few minutes to go but manage to pull off a win. I watched the A league match last night and Western Utd scored 2 goals in extra time to win 3-4, They did they same thing at Perth down 2-1 they got up 2-3, 2 goals in extra time. You could see the momentum shifting and at 3-3 I knew they would score again, but more importantly so did they players.

A few years back I had a confidence meter for the NRL that worked quite well. So might have to dust it off and adjust for football
Have looked at FBref before. Quite a few out there when you take a deeper dive.

----

I just use xG as an another indicator, but with this current project, it seems to show better accuracy than the normal league form tables everyone else seems to use. The guys I mentioned before touched on ZIP (zero inflated poisson model) as a way to counter the outliers.

ZIP.png

Ahem...I'll be leaning on good ole GPT to solve it. (won't take long, it's solved pretty much solved everything else I've done in excel)

----------

The advantage of specialising, as you say, are the nuances of what certain teams do and any other patterns/trends a team might show that some models may not factor in.

-------

A confidence meter sounds like a good idea, definitely worth dusting that one off.

-------

One decision I would need to make when choosing a league to specialise in, is, do I choose one where the average league goals are high? (Premier/Bundesliga) or ones that are low? (Serie A/League 1/League 2).

I'm leaning towards the low scoring leagues as they get over looked and it's possible to create out an edge in the various markets.
You do not have the required permissions to view the files attached to this post.
sionascaig
Posts: 1514
Joined: Fri Nov 20, 2015 9:38 am

wearthefoxhat wrote:
Sat Jan 11, 2025 9:01 am

----------

The advantage of specialising, as you say, are the nuances of what certain teams do and any other patterns/trends a team might show that some models may not factor in.

-------
One thing that I started looking at is the ELO implied probability of a team wining and the actual market price (before team news comes out) e.g.

http://clubelo.com/Fixtures

It can provide a remarkably good indication of where the market will be (before the market has properly formed) and if there is an significant difference there is usually an obvious reason for that, e.g. a lot of injuries.. And sometimes the reason isn't obvious )

Just messing about with it atm but suspect it can be used as some sort of indicator / flag for specific games. Maybe some mileage in that for you?

(mind once the team news comes out it is another ball game entirely)

Keep up the posts. Enjoying your journey )
User avatar
wearthefoxhat
Posts: 3411
Joined: Sun Feb 18, 2018 9:55 am

sionascaig wrote:
Sat Jan 11, 2025 9:17 am
wearthefoxhat wrote:
Sat Jan 11, 2025 9:01 am

----------

The advantage of specialising, as you say, are the nuances of what certain teams do and any other patterns/trends a team might show that some models may not factor in.

-------
One thing that I started looking at is the ELO implied probability of a team wining and the actual market price (before team news comes out) e.g.

http://clubelo.com/Fixtures

It can provide a remarkably good indication of where the market will be (before the market has properly formed) and if there is an significant difference there is usually an obvious reason for that, e.g. a lot of injuries.. And sometimes the reason isn't obvious )

Just messing about with it atm but suspect it can be used as some sort of indicator / flag for specific games. Maybe some mileage in that for you?

(mind once the team news comes out it is another ball game entirely)

Keep up the posts. Enjoying your journey )

Thanks, yep, have seen that site. Good point about a guide before the team news comes out, or a new manager installed.

Can see that being worthwhile including as an indicator, bit like using a betting forecast before all the other factors knock the market into shape.

Thanks for kind words. Posting sometimes helps the thought process.
sionascaig
Posts: 1514
Joined: Fri Nov 20, 2015 9:38 am

wearthefoxhat wrote:
Sat Jan 11, 2025 10:09 am

Can see that being worthwhile including as an indicator, bit like using a betting forecast before all the other factors knock the market into shape.
And the corollary is, I suppose, using the markets to adjust your model for a specific game, e.g. there can be significant changes in goal expectations once team news comes out, and that will help your model be more in line with game expectations & hence identify value.

Sorry if pointing out the obvious )

Onwards....
User avatar
wearthefoxhat
Posts: 3411
Joined: Sun Feb 18, 2018 9:55 am

sionascaig wrote:
Sat Jan 11, 2025 10:30 am
wearthefoxhat wrote:
Sat Jan 11, 2025 10:09 am

Can see that being worthwhile including as an indicator, bit like using a betting forecast before all the other factors knock the market into shape.
And the corollary is, I suppose, using the markets to adjust your model for a specific game, e.g. there can be significant changes in goal expectations once team news comes out, and that will help your model be more in line with game expectations & hence identify value.

Sorry if pointing out the obvious )

Onwards....

With me you sometimes need to point out the obvious!

What you say about adjusting the model may apply more in the main leagues where the syndicates have more salient data to mould together and snap up value prices and where the liquidity is plentiful.

You never know, I might stumble across a piece of data that the syndicates don't weight correctly in their algorithms.

Knocked together some comparisons using Poisson/Betfair/Club Elo, early days so format will likely change.

In this example, The Hibs v Motherwell game catches the eye. The Poisson model indicates some value in the Home team (Hibs), where as Club Elo has it more even.

Compare.png
-----

The Correct scores for 2-0, 3-0 & Any Other Home Win look value when comparing v Betfair and the top 6 predicted by the Poisson model.

Hibs v Motherwell.png
------
You do not have the required permissions to view the files attached to this post.
sionascaig
Posts: 1514
Joined: Fri Nov 20, 2015 9:38 am

Nice !
User avatar
wearthefoxhat
Posts: 3411
Joined: Sun Feb 18, 2018 9:55 am

One observation as I set up the sheets for the midweek games.

Do Pinnacle follow Club Elo?
Do they have their own independent approach?
Do the betfair market makers follow last of all?

Club Elo's website has all the info up long before everyone else, so I think I can guess.

Premier.png
You do not have the required permissions to view the files attached to this post.
User avatar
wearthefoxhat
Posts: 3411
Joined: Sun Feb 18, 2018 9:55 am

Anyone that can't be arsed with poisson spreadsheets, but want to have Poisson on their side, here are two websites that look interesting. (Free to use)

All you need to do is come up with the "Expected Goals" for Home/Away. (Can't do it all for you ;) )

Both sites have some other interesting stuff on there. (Free to use)

-------

Since a win


https://sinceawin.com/data/tools/poisson

SinceAWin.png
-------

xGscore.io


https://xgscore.io/apps/odds-calculator


xGscore.io.png
You do not have the required permissions to view the files attached to this post.
User avatar
wearthefoxhat
Posts: 3411
Joined: Sun Feb 18, 2018 9:55 am

This is where I'm at for now.

The adjustment is combining my poisson model data with ClubElo data. As sionascaig said, maybe something is needed to normalise the stats, with club-elo, it integrates the teams form and their match ups.

Also, I'm aware that Bill Benter had to include the horse race market data into his binary/merge sort algorithm before it became profitable. Using the club-elo data, is one way. as it seems that Pinnacle/Betfair follow their lead somewhat before making their own adjustments, with their own algorithms.

Premiership.png

If the Algorithm indicates a Lay2Back, then the price is considered value and becomes an entry point for the trade.

With the West Ham trade, new manager Graham Potter may have a positive influence, so will need to tread carefully.

Club-Elo doesn't cover the lower leagues or Euro II leagues, so I reckon that's where the real edge could exist.
You do not have the required permissions to view the files attached to this post.
User avatar
wearthefoxhat
Posts: 3411
Joined: Sun Feb 18, 2018 9:55 am

Got the trade wrong with the West Ham game, but the Poisson scores for all the games held up well overall.

My adjusted xG calculation indicated more of a 2-2 than the 3-2 score-line and the actual xG supported that too.

WestH v Fulh.png

Watched the highlights and Fulham's defence was a tad shabby, at this level, you can't mess around at the back.
You do not have the required permissions to view the files attached to this post.
sionascaig
Posts: 1514
Joined: Fri Nov 20, 2015 9:38 am

That is a big difference between old & new xG's for West Ham!

Did you look at some previous matches to see how new model performs against old?
User avatar
wearthefoxhat
Posts: 3411
Joined: Sun Feb 18, 2018 9:55 am

sionascaig wrote:
Wed Jan 15, 2025 9:28 am
That is a big difference between old & new xG's for West Ham!

Did you look at some previous matches to see how new model performs against old?

I'm of the belief that you should be factoring in the actual score of the game when reviewing the expected goal process. So, I just average my own xG with the final score. Even though West Ham didn't deserve the win on the stats, Fulham can't be rewarded with a win even though the xG stats indicate it. That way, the stats normalise when using the form tables.

Paul Steele, in his old book, picked up on this. He found by adding 1 goal to the winning team, it improved the accuracy of his systems when they rolled over game by game. His approach back then, was to use penetration stats based on SOT, SoffT and corners, which, for 30 years ago, may have been okay, nowadays, there are more stats freely available and can give a more accurate perspective especially if you're not watching the game live.
User avatar
wearthefoxhat
Posts: 3411
Joined: Sun Feb 18, 2018 9:55 am

Here's an example of how I'd maintain the Expected Goals Averages information.

Ev v AV.png

This example uses the basic score-lines, and their averages for Home v Away games. (I would use xG).
A decision can be made to use Last 3 (L3) right up to Last 10 (L10) games.

It can also show how the team is performing with a cheeky graphic. (Maybe can see why Sean Dyche was under pressure)
You do not have the required permissions to view the files attached to this post.
andy28
Posts: 544
Joined: Sat Jan 30, 2021 12:06 am
Location: Wexford

Everton, Villa match I am having issues with my model, It is telling me Everton should be 2.26 with recent home matches. Also I am thinking they will be more confident that Villa. How ever my racing side says Villa will win based on recent form. The last 5 away losses were to Newcastle, Forest, Chelsea, Liverpool, Spurs, which is a tough run, so I think they will be up for it as they will see this as a must win. So when I frank that form Villa come out on top

Another match today Arsenal, Spurs one model shows a lack of goals however another a goal fest. A North London Derby maybe action packed, however it all indicators I use say Arsenal are too short. I think these are sit and watch games for me
Michael5482
Posts: 1676
Joined: Fri Jan 14, 2022 8:11 pm

andy28 wrote:
Wed Jan 15, 2025 7:25 pm
Everton, Villa match I am having issues with my model, It is telling me Everton should be 2.26 with recent home matches. Also I am thinking they will be more confident that Villa. How ever my racing side says Villa will win based on recent form. The last 5 away losses were to Newcastle, Forest, Chelsea, Liverpool, Spurs, which is a tough run, so I think they will be up for it as they will see this as a must win. So when I frank that form Villa come out on top

Another match today Arsenal, Spurs one model shows a lack of goals however another a goal fest. A North London Derby maybe action packed, however it all indicators I use say Arsenal are too short. I think these are sit and watch games for me
Bet delay is on 8 seconds, can sup a full can of lager before you get in and out the market :lol:
Post Reply

Return to “Football trading”