Football Musings

[andy28]

I am in NZ but not Wexford, I am in Hawke's Bay, Wexford has a different meaning for me, not where I live.

I did get Newcastle, the match with City and Liverpool just looked wrong, my model had one part all wrong it priced City at 1.35 which thru everything out. This is why I decided to go with the Blended model as it smooths out outliers. While that result goes down as a lose I doubt I would have bet on it or would have not put as bigger bet on

[wearthefoxhat]

I am in NZ but not Wexford, I am in Hawke's Bay, Wexford has a different meaning for me, not where I live.

I did get Newcastle, the match with City and Liverpool just looked wrong, my model had one part all wrong it priced City at 1.35 which thru everything out. This is why I decided to go with the Blended model as it smooths out outliers. While that result goes down as a lose I doubt I would have bet on it or would have not put as bigger bet on

Apologies to Walshy, I thought I'd read it was Wexford, Canterbury...

New Zealand it on my travel list when I officially retire. Maybe visit a racetrack when I'm there to see how they run things and compare it with the UK scene.

[andy28]

I am in NZ but not Wexford, I am in Hawke's Bay, Wexford has a different meaning for me, not where I live.

I did get Newcastle, the match with City and Liverpool just looked wrong, my model had one part all wrong it priced City at 1.35 which thru everything out. This is why I decided to go with the Blended model as it smooths out outliers. While that result goes down as a lose I doubt I would have bet on it or would have not put as bigger bet on

Apologies to Walshy, I thought I'd read it was Wexford, Canterbury...

New Zealand it on my travel list when I officially retire. Maybe visit a racetrack when I'm there to see how they run things and compare it with the UK scene.

Cant beat a day at the races, all circuit racing here very similar to Australian tracks. Ellerslie would be a must visit, my fav would be Matamata

[walshy]

Apologies to Walshy, I thought I'd read it was Wexford, Canterbury...

New Zealand it on my travel list when I officially retire. Maybe visit a racetrack when I'm there to see how they run things and compare it with the UK scene.
[/quote]

No Hassle, Yeah New Zealand looks stunning from what I've seen.
Defo on my list of places to visit also

[walshy]

I am in NZ but not Wexford, I am in Hawke's Bay, Wexford has a different meaning for me, not where I live.

I did get Newcastle, the match with City and Liverpool just looked wrong, my model had one part all wrong it priced City at 1.35 which thru everything out. This is why I decided to go with the Blended model as it smooths out outliers. While that result goes down as a lose I doubt I would have bet on it or would have not put as bigger bet on

Apologies to Walshy, I thought I'd read it was Wexford, Canterbury...

New Zealand it on my travel list when I officially retire. Maybe visit a racetrack when I'm there to see how they run things and compare it with the UK scene.

No Hassle, Yeah New Zealand looks stunning.
Defo on my list of places to visit also

[TupleVision]

Hey WtFH, I've recently been experiment with Bivariate Poisson, not sure if you've considered this but I'm getting better results than Zero Inflated so far, just something to think about maybe... ChatGPT says this about it

Final Verdict: Why Bivariate Poisson Is Better for Your Model
Football matches are not purely independent events.

If Team A scores early, Team B may attack more, increasing the chance of a goal (or conceding more).
A Bivariate Poisson model accounts for this, while ZIP does not.
ZIP only adjusts for extra 0-0 draws, but doesn’t improve the modeling of other scorelines.

In contrast, Bivariate Poisson provides a more accurate distribution of 1-1, 2-1, 3-2, etc..
Bivariate Poisson is more versatile and can be tuned to reflect different league conditions.

The correlation parameter (λ_c) allows us to fine-tune results for different leagues and matchups.

[wearthefoxhat]

Hey WtFH, I've recently been experiment with Bivariate Poisson, not sure if you've considered this but I'm getting better results than Zero Inflated so far, just something to think about maybe... ChatGPT says this about it

Final Verdict: Why Bivariate Poisson Is Better for Your Model
Football matches are not purely independent events.

If Team A scores early, Team B may attack more, increasing the chance of a goal (or conceding more).
A Bivariate Poisson model accounts for this, while ZIP does not.
ZIP only adjusts for extra 0-0 draws, but doesn’t improve the modeling of other scorelines.

In contrast, Bivariate Poisson provides a more accurate distribution of 1-1, 2-1, 3-2, etc..
Bivariate Poisson is more versatile and can be tuned to reflect different league conditions.

The correlation parameter (λ_c) allows us to fine-tune results for different leagues and matchups.

Certainly looks worth exploring. It's an early part of the journey for me, and I'm quite aware there are some sophisticated models out there. The Bivariate one looks well suited to the task that's for sure!

I'm currently giving my ZIP model a fair test. It's performing very well across all leagues, so far, especially when you factor in the value-driven staking approach. I've recently added a lay the draw alert to all the games, which is separate to the back staking plan, and where the maria lay staking plan can be incorporated, pre-off, is necessary. Of course the data samples are relatively small, so am expecting it to level off at some time.

I can see that there is potential in taking a pre-off position backing home/away or laying the draw as long as the value criteria is met. Add in the mix the HT review strategy, on games where the strongest value is indicated, then it looks like a busy old sporting weekend ahead.

I've started to expand the excel sheets into the Italian Serie.A. (The AS Roma 4-0 was predicted with some good value for the home team pre-off) I'll look to add the U.S soccer league (MSL) when there are few games played.

Here's where I'm at for now.

WorkFlow.png

Checklist Summary.png

[TupleVision]

Certainly looks worth exploring. It's an early part of the journey for me, and I'm quite aware there are some sophisticated models out there. The Bivariate one looks well suited to the task that's for sure!

I'm currently giving my ZIP model a fair test. It's performing very well across all leagues, so far, especially when you factor in the value-driven staking approach. I've recently added a lay the draw alert to all the games, which is separate to the back staking plan, and where the maria lay staking plan can be incorporated, pre-off, is necessary. Of course the data samples are relatively small, so am expecting it to level off at some time.

I can see that there is potential in taking a pre-off position backing home/away or laying the draw as long as the value criteria is met. Add in the mix the HT review strategy, on games where the strongest value is indicated, then it looks like a busy old sporting weekend ahead.

I've started to expand the excel sheets into the Italian Serie.A. (The AS Roma 4-0 was predicted with some good value for the home team pre-off) I'll look to add the U.S soccer league (MSL) when there are few games played.

Here's where I'm at for now.

This looks good so far.

I was testing a model using a Bayesian approach for in play that worked on momentum but it allowed me to override too much and I found I was just basically gambling so I'm exploring pre-match again using the Bivariate model. I feel the less I can get involved in the process the better

[wearthefoxhat]

Certainly looks worth exploring. It's an early part of the journey for me, and I'm quite aware there are some sophisticated models out there. The Bivariate one looks well suited to the task that's for sure!

I'm currently giving my ZIP model a fair test. It's performing very well across all leagues, so far, especially when you factor in the value-driven staking approach. I've recently added a lay the draw alert to all the games, which is separate to the back staking plan, and where the maria lay staking plan can be incorporated, pre-off, is necessary. Of course the data samples are relatively small, so am expecting it to level off at some time.

I can see that there is potential in taking a pre-off position backing home/away or laying the draw as long as the value criteria is met. Add in the mix the HT review strategy, on games where the strongest value is indicated, then it looks like a busy old sporting weekend ahead.

I've started to expand the excel sheets into the Italian Serie.A. (The AS Roma 4-0 was predicted with some good value for the home team pre-off) I'll look to add the U.S soccer league (MSL) when there are few games played.

Here's where I'm at for now.

This looks good so far.

I was testing a model using a Bayesian approach for in play that worked on momentum but it allowed me to override too much and I found I was just basically gambling so I'm exploring pre-match again using the Bivariate model. I feel the less I can get involved in the process the better

My only experience with Bayesian was using it on a horse race tissue price set up.

By combining my tissue price with the live market and run it through a sort of, "Bayesian washing machine", it gave a final price that took into account the live market. It seemed to over emphasise the final price, if my tissue price was close to the market price. Maybe it was meant to happen like that by design, but I wasn't totally happy with the output.

[andy28]

Two from 2 for me today betting on the TAB they had Crystal Palace at 2.6.

I am doing the MSL this year I skipped last year, so will add them as well, however it is frustrating a lot of the time they run well behind schedule, can be 30mins sometimes more, 18sec delay & matches get left suspended for 30mins+. Also low liquidity on CS markets, I only do value betting.

I will look to add a lot more Sofascore data into this years model in preparation for next season EPL, AI has helped me a lot in this area

[wearthefoxhat]

Two from 2 for me today betting on the TAB they had Crystal Palace at 2.6.

I am doing the MSL this year I skipped last year, so will add them as well, however it is frustrating a lot of the time they run well behind schedule, can be 30mins sometimes more, 18sec delay & matches get left suspended for 30mins+. Also low liquidity on CS markets, I only do value betting.

I will look to add a lot more Sofascore data into this years model in preparation for next season EPL, AI has helped me a lot in this area

Out of the 4 games tonight, C.Palace and Fulham made it profitable, Bournemouth lost.

The Chelsea match odds were an exact match with the live market, so no alert.

PremierRes.png

---------


I wonder what A.I will look like around the start of the new football season?

[andy28]

Brentford let me down today, otherwise a clean sweep

[wearthefoxhat]

Brentford let me down today, otherwise a clean sweep

Well done, getting some consistency in the overall results.


Getting closer to the Super 6.

Super 6 Res.png

-------


Beat ole Jamie...just.

Pundit.png

[wearthefoxhat]

Hey WtFH, I've recently been experiment with Bivariate Poisson, not sure if you've considered this but I'm getting better results than Zero Inflated so far.

Took a look at Bivariate Poisson last night and earlier this morning. Of course, I leaned on GPT and DeepSeek (free versions), GPT seemed a bit more user-friendly and picked up on my input errors. It was very patient, as I kept referring it to Bavarian Poisson...I'd never heard of it before until you mentioned it!

Why Biv.png

BiviaratePoisson.png

I'll run it alongside my Zippidy-doo-da poisson that seems to be holding up well....so far

[andy28]

I couldn't get it to perform as it should, GPT recommended for using in play and I looked at a 2-0 score line to see what the odds were for 2-2 and it gave a prob of 0.8% so I didn't continue with it.

I am wondering how much longer these models will hold up in the last part of the season, teams like Forest may play different as they will be hell bent to stay in top 4, teams just outside the top 4 will be hell bent on getting in. Teams at the bottom will be playing to stay up so may settle for a draw against stronger opp or look to defend a 1-0 lead against a team toward the bottom. Plus Liverpool with a big buffer may not field their strongest team against say Southampton, plus they have a lot of home games left so they may have a few weaker sides going out at home.