Paired Comparison Modelling of Sport Tournaments

Abstract

Forecasting the results of sport tournaments is a topic of considerable interest. This thesis focuses on statistical modelling of football matches. The methodology discussed in the thesis stems from a cumulative logit extension of the traditional Bradley-Terry model for paired comparison data. The modelling approach includes the home-field effect and memory components that account for the current form of the players indirectly described through the fantasy football quotes. Maximum likelihood and shrinkage methods are considered for model fitting. The methodology is tested with simulation studies and illustrated with an application to the past season (2018/19) of Serie A.