What's Better Than Poisson For Soccer Handicapping?

[Sportslover]

I can calculate the goal expectation for each team in a given soccer match with a good level of accuracy.

However, I am struggling to convert this into a win probability and fair odds for 1X2, Asian Handicap and Total markets.

I was using Poisson but looking for a more accurate method because Poisson underestimates the probability of the draw and other low scoring scorelines.

Can anyone give me a suggestion on a more accurate probability method?

[Wrecktangle]

I dunno. All the soccer freaks use Poisson, why would you think there's something better?

[JohnAnthony]

Is that a software or something?

[Sportslover]

I dunno. All the soccer freaks use Poisson, why would you think there's something better?

Hi Wrecktangle, the reason for looking for something better is because most studies have shown that Poisson tends to underestimate the probability of low scoring games such as 0-0 draw, 1-1 draw, 1-0 home win and 0-1 away win. I have also noticed this in my own testing. Therefore, I wanted to see if there was a more accurate probability distribution, for soccer, out there.

For example, would a normal distribution, with a standard deviation be more accurate long term, would this take away any error caused by the fact that Poisson's standard deviation is equal to its mean.

Or are there any other suggestions for more accurate probability distributions that would convert goal expectancy for both teams into a probability of winning or covering the spread?

[gangeriver]

Is that a software or something?

If it is a software (it seems software) this is very funny thread.Soccer is not like US sports. Spontaneous positions, everything change suddenly and lots of unknown factors....and a software. strange!

[Justin7]

Bivariate poisson might work better. There's a chapter on it in this book:

http://www.amazon.com/Statistical-Thinking-Sports-Jim-Albert/dp/1584888687/ref=cm_cr-mr-title

Overall, a great book on other gambling topics as well.

[Data]

For example, would a normal distribution, with a standard deviation be more accurate long term, would this take away any error caused by the fact that Poisson's standard deviation is equal to its mean.

Based on the above, I am pretty sure you are yet to apply Poisson properly.

[Sportslover]

Based on the above, I am pretty sure you are yet to apply Poisson properly.

Data, I apply Poisson using the Poisson function in excel.

But first, I use my own technique to figure out the goal expectancy of each team.

Once I have the goal expectancy for each team, for example, 1.78 and 1.15, then I would use those values in a Poisson distribution using the function in Excel. Note that I have already taken into account home team advantage etc.

So if I wanted to know the probability of a 2-1 home win, I would put the following into the formula bar:

=POISSON(2,1.78,FALSE)*POISSON(1,1.15,FA LSE)

Is that incorrect usage?

[Peregrine Stoop]

goals are not scored independently of each other, which is why the low 0-0 outcomes are being estimated incorrectly

[Sportslover]

goals are not scored independently of each other, which is why the low 0-0 outcomes are being estimated incorrectly

Yes, that's a definate problem.

But how do you get around this problem, how do you solve it? What other method can be used that takes into account the fact that goals are not scored independently of each other?

[That Foreign Guy]

Markov Chains - http://en.wikipedia.org/wiki/Markov_chains

Also, rethink if you need to model specific score outcomes. Total goal estimates on their own give you the relative strength of the sides. You should be able to produce win draw win probabilities from there by a couple methods (look at past results +/ odds for similar ratios, create AH+0 odds and extrapolate).

[Sportslover]

Markov Chains - http://en.wikipedia.org/wiki/Markov_chains Also, rethink if you need to model specific score outcomes. Total goal estimates on their own give you the relative strength of the sides. You should be able to produce win draw win probabilities from there by a couple methods (look at past results +/ odds for similar ratios, create AH+0 odds and extrapolate).

Thanks for your suggestion.

Could you give me an example of how I would do a Markov Chain in Excel if I knew the goal expectancy for team A was 1.78 and the goal expectancy for team B was 1.15?

I've read the Wikipedia article and watched some videos on YouTube about Markov Chains but I've still got no idea where to start if I wanted to use it in Excel for Soccer predictions.

[Sportslover]

I have another question 2, since goals are somewhat dependent on each other, how do you determine a mathematical value for their correlation?

[Sportslover]

Here's another question:

I can use the RAND() function in Excel to write the following formula which generates a random integer between 0 and 20:

INT(20*RAND())

However, how do I write a formula which generates a random integer between 0 and 20, when I expect the average to be 1.78 (the goal expectancy for a team)? How would I include or incorporate that into my formula?

Or would I need to use a completely different function?

[Data]

=POISSON(2,1.78,FALSE)*POISSON(1,1.15,FA LSE)

Is that incorrect usage?

Yes, assuming that the means are constant. In fact, the means are initially lower but increase after the first goal is scored.

[Sportslover]

Yes, assuming that the means are constant. In fact, the means are initially lower but increase after the first goal is scored.

Would it therefore be more accurate to do a Poisson distribution for each 10 minute period of the match and then add the totals together?

Do you know much about implementing Monte Carlo simulations in Excel for Soccer goals?

[Data]

Would it therefore be more accurate to do a Poisson distribution for each 10 minute period of the match and then add the totals together?

I cannot see how that would help.

Do you know much about implementing Monte Carlo simulations in Excel for Soccer goals?

Read this book:

http://books.google.com/books?id=QLI3JJ9CD5QC&printsec=frontcover

[Sportslover]

Thanks for the link Data.

You indirectly led me to an online course that I have enrolled in to learn all about modeling in Excel so thanks for that! This will be a massive help to me as I'm really just a beginner in this area.

[Pokerjoe]

Soccer games before either team scores are statistically different creatures than after the first goal goes in. 0-0 games are even different than 1-1 games. There's been some good work done on this. I probably got it by googling, so you can, too.

Because there are so many professional soccer leagues, you don't have, with DB work with them, the problems of small samples you have with US sports. So I would recommend building a DB with many thousands of games/odds/scores, and crunching the numbers.

Or, and along with it actually, deduce likelihoods from bookie estimates. The soccer markets must have been pretty well pounded into efficiency, after so many gazillions of games played and bet. If there are +EV derivative bets out there, I'd be surprised, but at the least they wouldn't be huge. This means you can almost certainly use market estimates to deduce a formula for exact scores re odds.

IOW, where the market has a line of A -.5 over B, o/u of 2.5, see the correct score odds for 0-0, 0-1, 1-0, etc. Use Pinnacle, SBO bets, I imagine.

Build a DB of such estimates, then deduce a formula that would translate your score estimates into correct scores.

Though modeling answers to such questions is attractive, when large DBs are available, as they are in soccer, crunched numbers or market-hammered estimates are better, imo. The map is not the territory; the difference between modeled worlds and the real one can be subtle and important.

[cypry_bubu]

hi Sportslover!could u share ur way of calculating goal expectation,please.anybody know how to calculate goal expectation?

[DedesSports]

After reading all this I just wonder if you all think that bookmakers go to all that work to price matches ? I would bet "no" as the answer