Football Statistics

[MCLARKE]

A rare visitor to these shores and not sure if this is the right forum. I very rarely bet on the football, horse racing is my bag, but I have been offered various price boosts on BETMGM and football seems the best bet due to the lower overrounds.

As a result I have delved into the last 15 years stats for the premier league. All odds are based on BET365 prices Friday afternoon for weekend matches and Tuesday afternoon for midweek matches.

Overall the overround is 1.043 although it has increased from 1.026 in 2014 to 1.054 in 2023.

The ROI is -4.7%. The optimum bet is the home win with a loss of 1.2%. The draw is a loss of 6.8% and the away win is 5.9%.

Outsiders have the best record with a loss of 0.6%, the favourite is a loss of 2.9% and the middle odds is a loss of 9.9%.

Some statistics by teams 

Home win top 5          

Home win bottom 5   

Away win top 5            

Away win bottom 5     

Home draw top 5        

Home draw bottom 5 

Away draw top 5          

Away draw bottom 5   

Not sure how relevant these stats are but gives me a bit of a steer.

The standout bet tomorrow would be Burnley to beat Liverpool at 15/2 (Burnley +20, Liverpool -19).

 

 

 

 

 

[MMFloors]

Burnley +1.5AH is a good bet according to my model too. See other thread.

[harry_rag]

Interesting stuff. Some top of the head thoughts to stop them cluttering up my head!

19 hours ago, MCLARKE said:

football seems the best bet due to the lower overrounds

Are they that much lower in terms of overround added per runner? Obviously they're a lot lower in total if you compare a football match to a 20 runner handicap but what would a 3 runner race look like? You're only trying to win on one selection so you're only trying to beat the margin added to that runner. Worth bearing in mind before you desert a sport you know more about, if you can use your existing data to find runners with a decent AE.

19 hours ago, MCLARKE said:

Overall the overround is 1.043 although it has increased from 1.026 in 2014 to 1.054 in 2023.

It surprises me that they were apparently betting to such a low overround at one point. Must've been seriously chasing business of so. But then, as we know, they're happy to price badly and chop off winners.

19 hours ago, MCLARKE said:

The ROI is -4.7%. The optimum bet is the home win with a loss of 1.2%. The draw is a loss of 6.8% and the away win is 5.9%.

It reassures me from a logical point of view that, given a decent data sample, you wouldn't have made a profit blindly backing any of the 3 outcomes. Maybe it surprises me that home win looks best and draw looks worst as I'd have thought that home teams were more popular and would have most margin added to them while most punters ignore the draw so you might have expected least margin to be added there. But then, if business is being chased, maybe there's an element of wanting to be offering an attractive price about the outcome punters want to back. But then again, it's just one league and if you did the same thing for several divisions you might get a truer and slightly different picture! 

19 hours ago, MCLARKE said:

Outsiders have the best record with a loss of 0.6%, the favourite is a loss of 2.9% and the middle odds is a loss of 9.9%.

That chimes a bit more with my original preconception of more margin being added to the favourite. I'm not sure there's much value in considering the option of the middle odds; sometimes it will be the draw and sometimes the outsider.

19 hours ago, MCLARKE said:

Some statistics by teams 

Now you're getting into the territory of the sample size seeming to be quite decent but being too small to be of any real relevance. It would be a shock if every team closely plotted to the overall numbers, they're going to be all over the shop. I think you'd put yourself away by relying on those figures too much to make decisions on individual matches. Some teams will have just had freaky seasons or periods where they were underrated or overrated by the markets. I think what you're seeing there is expected variance and too much randomness for it to help you bet on individual games.

[MCLARKE]

3 hours ago, harry_rag said:

Are they that much lower in terms of overround added per runner? Obviously they're a lot lower in total if you compare a football match to a 20 runner handicap but what would a 3 runner race look like? You're only trying to win on one selection so you're only trying to beat the margin added to that runner. Worth bearing in mind before you desert a sport you know more about, if you can use your existing data to find runners with a decent AE.

I am struggling to make a return at the moment on the horses. The overround per runner is probably the same but most horse races are 5 runners + with an overall overround of 110% so if picking randomly I would be better picking the football.

 

3 hours ago, harry_rag said:

But then again, it's just one league and if you did the same thing for several divisions you might get a truer and slightly different picture!

I've also analysed 15 years of La Liga and the results are similar, -2% home, -9% draw and -11% away

3 hours ago, harry_rag said:

That chimes a bit more with my original preconception of more margin being added to the favourite. I'm not sure there's much value in considering the option of the middle odds; sometimes it will be the draw and sometimes the outsider.

La Liga muddies the water here with the favourite being the best bet at -2%, the outsider at -8% and the middle at -12%

 

3 hours ago, harry_rag said:

Now you're getting into the territory of the sample size seeming to be quite decent but being too small to be of any real relevance. It would be a shock if every team closely plotted to the overall numbers, they're going to be all over the shop. I think you'd put yourself away by relying on those figures too much to make decisions on individual matches. Some teams will have just had freaky seasons or periods where they were underrated or overrated by the markets. I think what you're seeing there is expected variance and too much randomness for it to help you bet on individual games.

I will have a look and see how they vary season by season and try and put together some sort of system that I can then test.

[harry_rag]

18 minutes ago, MCLARKE said:

I am struggling to make a return at the moment on the horses. The overround per runner is probably the same but most horse races are 5 runners + with an overall overround of 110% so if picking randomly I would be better picking the football.

I'll resist the temptation to say don't pick randomly as I get that you're focusing on maximising the bonuses here! There may be an argument for making sure you're playing at close to best available price where you do bet though. No point in taking a price that is the worst available where the boost only just beats what you could have got elsewhere. Apologies if you're already weighing up the merits of the price anyway. I'd have suggested looking at the anytime goalscorer market where at the top end the returns to level stakes at best price in my data are only -2.84% betting blindly but given that I'm not beating Uni this year (same prices) I'll keep my trap shut on that front.

24 minutes ago, MCLARKE said:

I will have a look and see how they vary season by season and try and put together some sort of system that I can then test.

I'll be interested to see what you come up with but you're ploughing an extremely ploughed furrow in terms of using this sort of data to look at the H/X/A market. One thing to bear in mind is that a lot of people exclude the the season(s) hit by COVID in terms of analysing historical data as I think the normal precepts around home advantage went out of the window. As I say though, I think there's a lot of randomness in the mix of the data that you'd have to be sure wasn't exerting an undue influence.

[MCLARKE]

1 minute ago, harry_rag said:

I'll be interested to see what you come up with but you're ploughing an extremely ploughed furrow in terms of using this sort of data to look at the H/X/A market. One thing to bear in mind is that a lot of people exclude the the season(s) hit by COVID in terms of analysing historical data as I think the normal precepts around home advantage went out of the window. As I say though, I think there's a lot of randomness in the mix of the data that you'd have to be sure wasn't exerting an undue influence.

You are right, I'm sure this has been analysed to death so I suspect I will not find anything. Good point about the Covid season, the results do look like an outlier.

[MCLARKE]

Seems a bit too obvious but I've analysed home wins for those 7 teams that have been in the premier league for each of the last 15 seasons (Arsenal, Chelsea, Everton, Liverpool, Man City, Man U and Spurs).

For 2009 to 2022 only Arsenal and Chelsea have not shown a profit. The 5 in profit recorded 801 wins from 1,235 matches with a profit of 75 points and an AE of 1.08. This was the base data.

I then tested it on the 2023 season and the results were even better with each of the 5 teams showing a profit, overall 64 wins from 95 runs with a profit of 23 points and an AE of 1.20.

In the current "live" season the results haven't been so good but still profitable with 27 wins from 47 runs and a profit of 2.07 points. Everton and Man City are showing losses.

I will place a bet on these 5 teams if I can equal or beat BET365 odds. I will have more opportunities than with horse racing because there is no BOG to complicate matters.

This weekend's games

Man City 1/11 (Ladbrokes)

Tottenham 1.88

Liverpool 1.48

 

[harry_rag]

12 hours ago, MCLARKE said:

Seems a bit too obvious

No surprise that I assume it must be, as my starting point is always cynicism when it comes to use of "top level" data!

12 hours ago, MCLARKE said:

I've analysed home wins for those 7 teams that have been in the premier league for each of the last 15 seasons (Arsenal, Chelsea, Everton, Liverpool, Man City, Man U and Spurs).

So we're looking at a subset of "better teams" on the basis of them never having been relegated, and would expect better than average results.

12 hours ago, MCLARKE said:

The 5 in profit recorded 801 wins from 1,235 matches with a profit of 75 point

Giving an ROI of 6.07%

12 hours ago, MCLARKE said:

I then tested it on the 2023 season and the results were even better with each of the 5 teams showing a profit, overall 64 wins from 95 runs with a profit of 23 points and an AE of 1.20.

An impressive ROI of 24.2% (be nice to have surfed that wave) but if you take that out of the overall results above you're down to an ROI of 4.56%. Profit is profit but I'd say you're into the territory where it it's inconclusive in terms of it being a strategy worth following, could just be random and have a long term expectation much closer to break even. Out of interest, what would have been the split between profitable and unprofitable seasons had you done the 5 times strategy every season for 15 years?

[MCLARKE]

6 minutes ago, harry_rag said:

So we're looking at a subset of "better teams" on the basis of them never having been relegated, and would expect better than average results.

One of the reasons I have picked these is that they will have played more matches than other profitable teams and therefore there is more data which should provide more robust results.

 

 

9 minutes ago, harry_rag said:

An impressive ROI of 24.2% (be nice to have surfed that wave) but if you take that out of the overall results above you're down to an ROI of 4.56%.

The overall results didn't include the 2023 season, therefore the ROI now stands at 7.3%

[harry_rag]

8 minutes ago, MCLARKE said:

The overall results didn't include the 2023 season, therefore the ROI now stands at 7.3%

Sorry, should have picked up on that. be interested to see how it works out, assuming you're going to give it a proper run.

I'm still suspicious of there being a degree of back-fitting (finding a subset of results that happened to be profitable and assuming they will still be going forward) but, at worst, it's unlikely to show much of a loss. I guess I'd either want a bigger sample size given the ROI or a bigger ROI given the sample size!

[MCLARKE]

26 minutes ago, harry_rag said:

I'm still suspicious of there being a degree of back-fitting (finding a subset of results that happened to be profitable and assuming they will still be going forward) but, at worst, it's unlikely to show much of a loss. I guess I'd either want a bigger sample size given the ROI or a bigger ROI given the sample size!

Always difficult to know whether a system that has been profitable will continue to be. I always leave the last year to test the system, not sure what other methods could be used to test the robustness of a system.

[harry_rag]

23 minutes ago, MCLARKE said:

Always difficult to know whether a system that has been profitable will continue to be. I always leave the last year to test the system, not sure what other methods could be used to test the robustness of a system.

True, I'm just a clumsy and enthusiastic amateur when it comes to this stuff. I could do with learning how better to validate my data. Breaking it down into quarters is always worth a quick look. If all 4 quarters are similarly profitable then that's as encouraging as it can get, if only one was then it's more likely to just be random. 

I'd still be interested to know the split between profitable and unprofitable seasons across the 15 years if that's easily ascertained.

[MCLARKE]

1 hour ago, harry_rag said:

Out of interest, what would have been the split between profitable and unprofitable seasons had you done the 5 times strategy every season for 15 years?

For the top 5 teams there have been 15 profitable seasons out of the last 16 with just 2016 showing a loss (this was the season when Leicester won the title).

The constituents would change from year to year, Chelea would have appeared for just 1 season (they then lost 11 points), Arsenal have only appeared for the last 2 seasons.

The only teams to have qualified every year are Man City, Man U and Spurs.

I am surprised at the consistency of this profitability for what appears to be such an obvious system.

[harry_rag]

8 minutes ago, MCLARKE said:

The constituents would change from year to year

I meant the specific system of the same 5 teams as you quoted the original results for, which I think is Everton, Liverpool, Man City, Man U and Spurs?

[MCLARKE]

Same results, profit in 15 out of 16 seasons. It is actually Arsenal, not Liverpool that is part of the top 5 so I should't have actually backed Liverpool today.

In reality the teams woud change from season to season.

[harry_rag]

Sorry but I'm a bit confused by the "top 5" thing now!

I thought you started out quoting figures for backing the same 5 specific teams every season, is that not the case?

17 hours ago, MCLARKE said:

The 5 in profit recorded 801 wins from 1,235 matches with a profit of 75 points and an AE of 1.08.

So what's the top 5 concept? Following the 5 teams who showed the best return in the previous season? Think I've got it! Backing the "ever present" teams for the home win provided they were in the top 5 returns for the previous season?

[MCLARKE]

1 hour ago, harry_rag said:

So what's the top 5 concept? Following the 5 teams who showed the best return in the previous season? Think I've got it! Backing the "ever present" teams for the home win provided they were in the top 5 returns for the previous season?

Almost ! It is backing the "ever present" teams so long as they have shown a profit. For the last 2 years it has been 5 teams, before that it was 4 because Arsenal had not qualified.

[harry_rag]

There's no doubt an element of me being a bit dense here but I think there's at least a small element of you not explaining it well! 

What was the initial criteria that determined the 4 teams you did back and the 3 you didn't in 2010? Whether or not they showed a profit in the previous season?

You've carried on backing the 3 teams every season because they've shown a positive total return from then to date (despite some loss making seasons)?

You started backing Everton and Arsenal when they edged into cumulative profit?

You ditched Chelsea and Liverpool because they fell into a negative cumulative return and haven't got back into the black?

I guess I'm struggling because you've not defined what "shown a profit" means in a way that explains why teams start or stop being backed.

BTW - I've just had a first look at the data and it's quite interesting so thanks for bringing that to my attention and sorry for the extent of the questioning but it is out of genuine interest!

 

 

[MCLARKE]

Yes, I might not have explained it too well.

I have backed teams that have shown a cumulative profit.

So at the end of 2010 Liverpool, Man C, Man U and Tottenham showed a total profit for 2009 / 2010 so I would have followed them for 2011. By the end of 2011 Everton had moved into cumulative profit so they would then be followed for 2012. By the end of 2012 Liverpool had moved into a cumulative loss so they would not be followed in 2013.

This is the full base data that might help (or possibly confuse even further !)

 

 

[MCLARKE]

4 hours ago, MCLARKE said:

Always difficult to know whether a system that has been profitable will continue to be.

One method is to calculate the p value, this supposedly calculates the probability that the results could have arisen by chance

For the 2023 results the p value is 0.015, which indicates there is a 1 in 67 chance that the results are purely due to luck.

If we use a confidence level of 95% (p < 0.05) then the system passes the test. If we use a confidence level of  99% (p < 0.01) then it doesn't.

 

 

[MCLARKE]

On 12/29/2023 at 11:19 PM, MCLARKE said:

Man City 1/11 (Ladbrokes)

Tottenham 1.88

2 / 2, lsp 0.95

[MCLARKE]

This week's selections

MAN UTD 5/4

EVERTON 15/8

Both with BETMGM

 

[Torque]

On 12/30/2023 at 7:52 PM, MCLARKE said:

One method is to calculate the p value, this supposedly calculates the probability that the results could have arisen by chance

For the 2023 results the p value is 0.015, which indicates there is a 1 in 67 chance that the results are purely due to luck.

If we use a confidence level of 95% (p < 0.05) then the system passes the test. If we use a confidence level of  99% (p < 0.01) then it doesn't.

 

 

I use 99%. The extra 4% makes a big difference.

[Torque]

16 minutes ago, Torque said:

I use 99%. The extra 4% makes a big difference.

You could also make the argument that a smaller sample should use 99%, whilst a bigger sample might be more able to tolerate lower confidence of 95%. I'd be looking for a bigger sample that satisfied p < 0.01 myself.

[MMFloors]

I don't have a lot of mathematics skills but my model comes up with this output for regression analysis. And I know it's based on a small sample size (120).

 

Edited January 13 by MMFloors

[MCLARKE]

On 1/13/2024 at 2:41 PM, MMFloors said:

I don't have a lot of mathematics skills but my model comes up with this output for regression analysis. And I know it's based on a small sample size (120).

 

I have a maths a level but from over 40 years ago.

What do these numbers tell us ?

[MMFloors]

1 hour ago, MCLARKE said:

I have a maths a level but from over 40 years ago.

What do these numbers tell us ?

From what I understand:

- the R-squared ("R-kwadraat") should be as close to 1 as possible, but 0,7+ is a very good start.

R-Squared is a statistical measure in a regression model that determines the proportion of variance in the dependent variable that can be explained by the independent variable.

- the Significance ("Significantie F") should be as low as possible (close to zero)

Significance is used to determine whether the relationship exists or not. For example, if the regression coefficient is significant at the .05 level, then it can be said that we can reject the null hypothesis and accept the alternative hypothesis that a relationship exists between the dependent and independent variable(s). 

- the P-value ("P-waarde") should be as low as possible but at least under 0.05.

The p-value of a linear regression model checks if there is a significant linear relationship or correlation between your predictors and the target value. If the p-value is low, this means its relationship is significant

 

[MCLARKE]

Tomorrow's selections

MAN CITY 1/10 - QUINNBET

TOTTENHAM 7/10 - LADBROKE BOOST

[MCLARKE]

This weekends selections

Everton 3.20 Betfair

Man Utd 4/6 Quinnbet

Arsenal 13/10 Quinnbet

[MCLARKE]

On 2/2/2024 at 4:56 PM, MCLARKE said:

This weekends selections

Everton 3.20 Betfair

Man Utd 4/6 Quinnbet

Arsenal 13/10 Quinnbet

6 from 9, LSP 0.72