Official Poisson Discussion Thread

[SpodoKomodo]

So there are hundreds of Poisson betting articles out there, I thought it would be a nice idea to collate the information and tips we all have on using Poisson for sports betting, to see where the advantages/disadvantages and possible values lies. What I have done so far:

• For testing purposes, I am using this season's goals scored and goals conceded stats to create my expected numeric value (mean). I am doing the conventional: (Home team goals scored at home + Away team goals conceded away)/2. And vice versa.
• I am using these numbers to create the probability for each correct score (multiplying chances of each team scoring each number of goals).
• I then add the data of all the home win correct score odds to create the home win odds, and then the same for draw and away win.

My queries/worries so far:

• I am only calculating chances for scorelines up to 4 goals for each team, meaning that the odds for home win, draw and away win add up to around 98%, instead of 100%. Do other people calculate chances of scorelines up to, say, 10-9, just so the data is completely clean and not skewed?
• Is there a simple way of importing the "best odds available" for each match to allow easy comparison between "my odds" and bookmakers odds? (I'm aware I've asked this, and that is probably requires hard-coding, but I'm guessing it's something other people would like to do too!)
• And the question that is bothering me most: I am aware that by just using this season's scores as data, or the previous 20 matches etc, that I will not gain any edge over the bookmaker, as I am not taking into consideration any other factor. How would one go about incorporating a ratings system into the Poisson means? I cannot get my head around how you can use anything other than PURE statistical data in Poisson. For example, if I'm using elo ratings, and Chelsea and have a rating of 18 higher than Burnley, what would I do with this data to alter my mean goal predictions? One idea I had was that, for every point difference in rating between the 2, I could add 0.05 to the goal prediction for the favourite, and subtract 0.05 from the goal prediction of the outsider. Does this make sense?

I am hopeful of finding something that could work - at the moment I am paper trailing a CS value system with Poisson (using basic goal stats) and although my yield is crap, the scorelines have nearly ALL finished within 1 goal of the CS prediction. Obviously this requires more testing, but I still think this is a promising sign. For the foreseeable future, I am going to continue to experiment with Poisson and keep this thread updated with anything I learn/new systems etc. If anyone has any opinion, stories, facts, anything to do with Poisson, I hope you can contribute!

[SpodoKomodo]

Re: Official Poisson Discussion Thread

Match Odds		My odds	Odds
Home		2.331607	2.88
Away		3.414011	2.75
Draw		4.112086	3.40

These are my odds for Scunthorpe v Ipswich. The odds suggest my sums are correct. I just need to concentrate on deciding what data should be used for the Poisson variables.

[relf]

Re: Official Poisson Discussion Thread

My queries/worries so far:

• I am only calculating chances for scorelines up to 4 goals for each team, meaning that the odds for home win, draw and away win add up to around 98%, instead of 100%. Do other people calculate chances of scorelines up to, say, 10-9, just so the data is completely clean and not skewed?

I would say it's fine for calculating up to 4 or 5 goals. After all, high scorelines would be classified as 'Others' by the bookies. The Poisson distribution has a slight tendency to veer towards 0 rapidly as the number increases, so I would be very careful to use a Poisson distribution to predict high scorelines. It might not be accurate or reflective of the high scorelines proportions in real life. (I'm just suspecting here, I have not done any back-testing to prove.)

• Is there a simple way of importing the "best odds available" for each match to allow easy comparison between "my odds" and bookmakers odds? (I'm aware I've asked this, and that is probably requires hard-coding, but I'm guessing it's something other people would like to do too!)

How do you define best odds? I remembered someone once mentioned that the bookies actually set their odds such that they would make a profit no matter if it's a home win, draw or away win. Thus, their odds might be skewed. As long as your odds produce value and yield in betting, it simply shows that your odds are better and more realistic.

• And the question that is bothering me most: I am aware that by just using this season's scores as data, or the previous 20 matches etc, that I will not gain any edge over the bookmaker, as I am not taking into consideration any other factor. How would one go about incorporating a ratings system into the Poisson means? I cannot get my head around how you can use anything other than PURE statistical data in Poisson. For example, if I'm using elo ratings, and Chelsea and have a rating of 18 higher than Burnley, what would I do with this data to alter my mean goal predictions? One idea I had was that, for every point difference in rating between the 2, I could add 0.05 to the goal prediction for the favourite, and subtract 0.05 from the goal prediction of the outsider. Does this make sense?

I would say this is by pure experiment. You can try changing the lumda value, e.g. lumda = mean + 0.01, mean - 0.05. There's really an infinite amount of tweaks you can make. You could try a few to see which ones produce better results.

[SpodoKomodo]

Re: Official Poisson Discussion Thread

I would say it's fine for calculating up to 4 or 5 goals. After all, high scorelines would be classified as 'Others' by the bookies. The Poisson distribution has a slight tendency to veer towards 0 rapidly as the number increases, so I would be very careful to use a Poisson distribution to predict high scorelines. It might not be accurate or reflective of the high scorelines proportions in real life. (I'm just suspecting here, I have not done any back-testing to prove.)

Ok yep that's whart I've done, I've just hidden the cells that show scores like 5-4 etc, but I've still used the data to help work out the match odds. I think, rather than Poisson not being accurate of high scorelines, bookies just don't reflect the true odds of the higher scorelines because the odds would be too high. For example, you get odds of 500/1 for Barcelona to win 8-1, but in reality the odds should be a lot higher.

How do you define best odds? I remembered someone once mentioned that the bookies actually set their odds such that they would make a profit no matter if it's a home win' date=' draw or away win. Thus, their odds might be skewed. As long as your odds produce value and yield in betting, it simply shows that your odds are better and more realistic.[/quote'] Yes of course, bookies have an overround of 120% or so, so I wouldn't use the odds of 1 single bookmaker. What I mean is, I would like to import the data (in a web query?) from a website like oddschecker.com, to show the best available odds for each correct score in my spreadsheet. Does this make sense?

[relf]

Re: Official Poisson Discussion Thread 1) As for the probability of high scores, I'm unsure about the true probabilities. It happens rarely, thus making the sample size really small to judge. So whether it's 500/1 or 5000/1, personally I have no idea. 2) It makes sense. Sorry I have no idea how to do this, so I can't help you on it. 3) As long as it makes sense, you could try. Whether or not it will work, I can't give you an answer. The best way is to try it out with a paper trail.

[SpodoKomodo]

Re: Official Poisson Discussion Thread

1) As for the probability of high scores, I'm unsure about the true probabilities. It happens rarely, thus making the sample size really small to judge. So whether it's 500/1 or 5000/1, personally I have no idea. 2) It makes sense. Sorry I have no idea how to do this, so I can't help you on it. 3) As long as it makes sense, you could try. Whether or not it will work, I can't give you an answer. The best way is to try it out with a paper trail.

1) Yes I see whast you mean - personally I would probably not back any scoreline involving over 4 goals. Although it does happen occasionally, the spreadsheets have used have nearly always thrown up scorelines like 5-4 as value. Whether this is down to the data I have used, or the fact that bookies offer similar odds for all high scorelines, I'm not sure. 2) Ok fair enough, hopefully lunatic will be able to help with this! ;) 3) Ok, I've read some threads on ratings, I don't think my mathematics skills are high enough for some of the more complicated ones. Maybe I'll just experiment with some basic ones.

[bagaspuds]

Re: Official Poisson Discussion Thread

What I have done so far:

• For testing purposes, I am using this season's goals scored and goals conceded stats to create my expected numeric value (mean). I am doing the conventional: (Home team goals scored at home + Away team goals conceded away)/2. And vice versa.

SpodoKomodo, Great to read your post. I have been using Poisson also to predict the outcome of football matches. Incidentally I have used up to a score of 7-7, this covers 99.46% of the outcome at scoring rates of 60mins (A) and 60mins(B). So that's fine with me. Just add up all the probabilites for all the scores and see how far short of the 100% mark you are. What concerns me is the above quote. I too have been taking the average of 'Team A Home score rate' & 'Team B away Concede rate'. Is this correct? You say it is conventional. I know it is better than taking the scoring rates alone but is there a better way? Would love to hear from others. These are values I got from my calculator. Do they agree with yours. A-60, B-60 score rates Odds A-2.63 B-2.63 Draw - 4.19 A-45, B-70 A-1.84 B-4.11 Draw-4.72 As with the above reply, I believe the bookies move their odds to reflect the flow of money. So I would not be overconcerned if they differ from yours.:dude

[Vilamoura04]

Re: Official Poisson Discussion Thread

And the question that is bothering me most: I am aware that by just using this season's scores as data, or the previous 20 matches etc, that I will not gain any edge over the bookmaker, as I am not taking into consideration any other factor. How would one go about incorporating a ratings system into the Poisson means? I cannot get my head around how you can use anything other than PURE statistical data in Poisson.

• For example, if I'm using elo ratings, and Chelsea and have a rating of 18 higher than Burnley, what would I do with this data to alter my mean goal predictions? One idea I had was that, for every point difference in rating between the 2, I could add 0.05 to the goal prediction for the favourite, and subtract 0.05 from the goal prediction of the outsider. Does this make sense?

I suggest you read Kevin Pulleins book "Definitive Guide to Betting on Football". In it there is a ratings system you can use, which once youve stuck it into excel and automated it is relatively quick and easy to calculate success rates for teams for any given match. There are also tables of data showing how goal expectation increases/decreases with the above success rates, based on past english league data. Basically, a game between two equally matched sides (actually away side marginally stronger to cover home advantage) has the lowest goal expectation at 2.45 goals/game. A game between two very uneven sides would have a goal expectation of 3.03 goals/game.

[Kumquat Tree]

Re: Official Poisson Discussion Thread

What concerns me is the above quote. I too have been taking the average of 'Team A Home score rate' & 'Team B away Concede rate'. Is this correct? You say it is conventional. I know it is better than taking the scoring rates alone but is there a better way?

Of course there is a better way. Assume Team B1 have played ManU, Arsenal, Villa and Tottenham in their last 4 away games and conceded 12 goals. Then assume Team B2 have played Fulham, Birmingham, West Ham and West Brom in their last 4 away games and have conceded 8 goals. You really think that Team B1 are more likely to concede in their next away match than Team B2 just because 12 is greater than 8? So to improve your formulas, you have to consider the opponents where goals conceded (and scored) came from too. ;)

[bagaspuds]

Re: Official Poisson Discussion Thread Guys, I have now looked at over 300 soccer games, randomly across different european leagues. I have compared my starting odds with Betfair's. If I was to bet on the highest Value of either Home, Away or draw on ALL of these matches then the return would be 13% over Betfair 5% commision.:dude How does this seem to you? How many games is "a good sample" to have confidence in?

[relf]

Re: Official Poisson Discussion Thread

Guys, I have now looked at over 300 soccer games, randomly across different european leagues. I have compared my starting odds with Betfair's. If I was to bet on the highest Value of either Home, Away or draw on ALL of these matches then the return would be 13% over Betfair 5% commision.:dude How does this seem to you? How many games is "a good sample" to have confidence in?

Sample size can only help if you are calculating the strike rate of your selections. If you want to analyse your returns, you need to take in consideration the statistics of your odds, the mean, the variance, etc.

[bagaspuds]

Re: Official Poisson Discussion Thread Guys, I am now up to 600 games and the margin is still +12% over Betfair's 5% commission.:cow I calculate the scoring rate from the start of the season. I'm too lazy to add weight to each score based on the oponent. My question is, what level of confidence can I have in this? Can I quantify it in any way?

[Kumquat Tree]

Re: Official Poisson Discussion Thread Superb result. You should be richer than Uncle Scrooge before Christmas then. :cash No idea how to compute a confidence level based on that. I would not recommend to calculate goal averages on results long ago like September. At least not weighted equal to recent results. But if you say your results are profitable, then hey, why not. Regarding your concern of not weighing the opponents goal averages, it should be the lesser problem the more the season moves on. I'd say after 10 or 15 weeks or so, there will be no more significant differences between the strenght of schedule among teams. Every team will usually have played a couple of highscoring and lowscoring opponents by then on a somewhat similiar level.

[bagaspuds]

Re: Official Poisson Discussion Thread Cheers "Kumquat Tree". I agree with all you say. I think poisson is accurate over a large number of games so you really have to roll with the punches and not worry over about the run of losses. If you are looking for value then you will find that it is the long odds (though not always so) that offer the best value. Soccer by it's nature is very low-scoring so therfore very random, after all a game can literally be decided on a lucky deflection or dubious referee decision. Over a large number of games, this randomness is smoothed out. I'm still undecided though how best to compute the scoring rate, ie recent form with weightings versus average over high number of games. Less work in the last approach though.:ok

[KThom]

Re: Official Poisson Discussion Thread can someone please explain to me why the formula in example 1 at www.stattrek.com/lesson2/poisson.asps shows the last figure as x! (where x equals 3) and in the formula the 3 turns to a 6? I don't understand what x! Means

[relf]

Re: Official Poisson Discussion Thread

can someone please explain to me why the formula in example 1 at www.stattrek.com/lesson2/poisson.asps shows the last figure as x! (where x equals 3) and in the formula the 3 turns to a 6? I don't understand what x! Means

I can't see the link. As for x!, it's means factorial. 1! = 1 2! = 2 x 1 = 2 3! = 3 x 2 x 1 = 6 ... 10! = 10 x 9 x 8 x 7 x 6 x 5 x 4 x 3 x 2 x 1 = 3628800 It's just multiplication of the number (x) all the way down to 1. Hope this helps.

[KThom]

Re: Official Poisson Discussion Thread Thanks relif. Understand

[Jimaldo]

Re: Official Poisson Discussion Thread Spodo are you calculating your results by the poission distribution or the bivariate poisson distribution?

[SniffTheGlove]

Re: Official Poisson Discussion Thread I tried poission distribution in 2006-2007 season using the same criteria of HomeTeamGoalFor+AwayTeamGoalAgainst / 2 but it never worked out for me unless of course I was doing the maths wrong but if other can get good results from it then I am all for it being used again.