Calculating Edge with Poisson Distribution

[IrishTim]

Suppose I'm handicapping soccer totals and I make the fair value O/U under 3 goals with the OVER at +120 and the UNDER at -120. Now in my Poisson spreadsheet, if I specify a 5% edge that makes the payout odds on the UNDER 3 go to -105.

But the difference in the payout odds if you take the break-even win percentages of -120 (54.5%) and -105 (51.3%) is 3.2%. So what is my edge, 3.2% or 5%? And what accounts for the difference.

[durito]

edge = win prob*dec odds -1 = approx 6.5% at -105 and 54.54% win prob

careful with poisson and soccer totals unless this just a for fun exercise

[MonkeyF0cker]

Suppose I'm handicapping soccer totals and I make the fair value O/U under 3 goals with the OVER at +120 and the UNDER at -120. Now in my Poisson spreadsheet, if I specify a 5% edge that makes the payout odds on the UNDER 3 go to -105.

But the difference in the payout odds if you take the break-even win percentages of -120 (54.5%) and -105 (51.3%) is 3.2%. So what is my edge, 3.2% or 5%? And what accounts for the difference.

Without having the data in front of me, I'd guess that soccer totals aren't normally distributed. If you have a statistically significant sample in a database, there are tests that you can perform to check the (goodness of) fit of the distribution for normality. Google the Kolmogorov-Smirnov test, the Shapiro-Wilkes test, or the D’Agostino-Pearson omnibus test for more info.

[Justin7]

Poisson is not very good for soccer.

[Pancho sanza]

Poisson is not very good for soccer.

What about hockey? Any opinion?

[skrtelfan]

Poisson is less bad for hockey but still not great. Any time you have a very low median, poisson doesn't work well. If the median team goals is 2.5, 6 goals can happen but -1 goals can't.

[20Four7]

What about hockey? Any opinion?

I have found it not to be the best for hockey. I have played around with it and it is better than soccer (due to more scoring) but not great.

[durito]

Hockey is closer than soccer, but it's still not poisson. This paper takes an excellent look.

http://www.hockeyanalytics.com/Research_files/Poisson_Toolbox.pdf

[durito]

...

[IrishTim]

Thanks for the helpful answers, especially the recommendations of further reading by durito (btw - I wouldn't mind seeing that last comment deleted) and monkey. I've printed off some stuff that I'll take a look at as soon as midterms are over.

So I gather from the responses here that Poisson isn't appropriate for soccer totals? I haven't compiled the data either to look at the scoring distribution, but I was testing a few different things. My question then is why won't Poisson work? My understanding is that you can use it anytime the event is (1) counted one at a time (2) the actual probability of occurence is small, but the number of attempts is fairly large. Soccer goals seem to satisfy both of these requirements.

[IrishTim]

A corollary I guess to my question on why isn't Poisson good for soccer - is it also not appropriate for MLB season totals? If you project a team to win 85 games and the book is hanging 88.5 with the under at -125, how do you determine your edge without using Poisson?

[Justin7]

I'd use a binomial distribution for season wins. Poisson works reasonably well for mlb though, since there are so many games. In NFL (or any sport with under 40 games), Binomial works much better.