Summary:
Football is one of the most popular sports in the world. In recent years, more and more companies have been associated with football depending economically on it. This led to a huge statistical interest in the sport. This thesis constitutes a review on football modeling.
Initially, theory behind bivariate analysis is developed along with properties and extensions of the bivariate distribution. Special attention is paid to the bivariate Poisson distribution which is widely used in football modeling. Regression models constitute another subject of study as they provide functions that describe the relationship between random variables. In that part, count data models are presented such as Poisson regression model and the inflated models which deal with problems with excessive outcomes. As for the parameters estimation, the EM algorithm is considered to be a rational way to find the maximum likelihood estimate when the latter cannot be calculated in straightforward way.
After presenting the theoretical framework on with football modeling is based, several bivariate predictive models are presented in terms of four main categories: naïve models, models with dependence parameter, inflated models, dynamic models.
Finally, analysis of the Greek Superleague is carried out through four bivariate models. After the comparison of the models’ fitting, prediction in a playoff match takes place.