Predicting Results of Football Matches: Ordered Random Forest versus Bivariate Poisson Regression

Betting on football matches is a huge industry which exists already for a long time. Therefore, predicting sports matches has been the topic of a lot of researches. This paper attempts to replicate and extent the paper of Goller et al. (2018), where a new method to predict football matches is used. Various variables describing the teams in the 1. Bundesliga are used to predict the final league table of the 1. Bundesliga. The main focus of this paper is the comparison between the predictive power of the Ordered Random Forest Model (ORFM) of Goller et al. (2018) and the Bivariate Poisson Regression Model (BPRM). The predictive performance of both models is assessed by different performance measures, for example a hypothetical return on investment (ROI). The BPRM slightly outperforms the ORFM in terms of the performance measures used in this paper. Therefore, I conclude that the BPRM has a slightly higher predictive power than the ORFM. Hence, I would recommend the BPRM over the ORFM for predicting football matches.