Football modelling using Bivariate Poisson

[scruffbag]

Hi, I'm new to this forum and have only recently begun thinking about developing my own models for football. My background is in statistics and I've been directed to some of the published literature in this area (e.g. Dixon & Coles, 1997). I was wondering whether anyone else has tried using Bivariate Poisson models, as described in the literature, and whether anyone has any advice on applying them. I'm using the statistical package "R", and as I haven't already had much experience of bivariate modelling I'm initially trying to acclimatise myself with the "bivpois" library (developed by Karlis and Ntzoufras). I'd be very interested in hearing about anyone else's experiences of modelling football scores similarly. Cheers

[salford_student]

Re: Football modelling using Bivariate Poisson

Hi, I'm new to this forum and have only recently begun thinking about developing my own models for football. My background is in statistics and I've been directed to some of the published literature in this area (e.g. Dixon & Coles, 1997). I was wondering whether anyone else has tried using Bivariate Poisson models, as described in the literature, and whether anyone has any advice on applying them. I'm using the statistical package "R", and as I haven't already had much experience of bivariate modelling I'm initially trying to acclimatise myself with the "bivpois" library (developed by Karlis and Ntzoufras). I'd be very interested in hearing about anyone else's experiences of modelling football scores similarly. Cheers

Ive just written up the Karlis and Ntzoufras model in excel and have replicated the results in the paper. I'm in testing phase now, and it's working quite well. I've not got round to modelling the means and deriving the correlation coefficent yet, so I've back engineered the prices from from sporting index and arrived at the following parameters for tommorows match. theta(0)-correlation coefficient:

0.052788

theta(1)-wigan

0.591656

theta(2)- man utd

2.293633

are these about right ? they generate the following output from the model.

spread prices			error
	Pred	Sp. Indx
Total Gls	2.95	2.9	0.02
Gls Supr.	-1.55	-1.6	0.03
H Win	3.730985	3.75	0.01
A Win	20.37	20.25	0.01
H Goals	0.70	0.7	0.00
A Goals	2.25	2.3	0.02
Handicap	15.00	15.25	0.02

the sporting index prices are the midpoints of the margins. when generating fixed prices it gives me predicted odds of

	Pred
Home	12.15
Draw	5.95
Away	1.33

are my parameters about right ? anyone any idea if the correlation co-efficient varys much between matches ?