Skip to main content
top

Bibliography

Conference Paper (international conference)

Modelling the score of the sport match via dynamic generalized linear models

Volf Petr, Hejdušek M.

: Bulletin of the International Statistical Institute 56th Session. Proceedings, p. 1-4

: ISI 2007. Session of the International Statistical Institute /56./, (Lisboa, PT, 22.08.2007-29.08.2007)

: CEZ:AV0Z10750506

: 1M06047, GA MŠk

: generalized linear model, Poisson distribution, sport statistics

(eng): In the contribution, several Poisson models (independent, inflated, bivariate) are used for prediction of the result of sport match. All these models are treated in the framework of generalized linear models. Furthermore, we show also two possibilities how a time development of parameters during the season can be incorporated. First, by updating the parameters, giving more weight to recent results than to the older ones, as in Dixon and Coles (1997). Further, we shall introduce a dynamic autoregressive model for parameters innovation through the season (described in a Bayes way, similarly as Crowder et al., 2002)), and consider an approximation leading to a variant of Kalman filtering method. Finally, a set of models is compared on the analysis of large data sets from several European football leagues and also from ice-hockey NHL.

(cze): Několik variant Poissonova modelu (nezávislého, smíšeného, dvoj-rozměrného) je užito k predikci skore sportovních zápasů. Modely jsou formulovány jako zobecněné lineární, a je v nich zavedena dynamika vývoje parametrů v čase. Parametry se vyvíjejí autoregresně, metoda jejich odhadu používá aproximaci, která vede na lineární dynamický systém typu Kalmanův filtr. Metoda je ilustrována na rozsáhlých datech z několika evropských fotbalových soutěží a také z hokejové NHL.

: BB