Bayesian state-space models for the modelling and prediction of the results of English Premier League football

Ridall, Gareth and Titman, Andrew and Pettitt, Anthony (2024) Bayesian state-space models for the modelling and prediction of the results of English Premier League football. Journal of the Royal Statistical Society: Series C (Applied Statistics). ISSN 0035-9254 (In Press)

Text (RSS_Football_paper_final)
RSS_Football_paper_final.pdf - Accepted Version
Available under License Creative Commons Attribution.
Download (897kB)

Abstract

The attraction of using state space models (SSM) is their ability to efficiently and dynamically predict in the presence of change. In this paper we formulate a Bayesian SSM capable of predicting the outcomes of football matches and the associated states, which are the attacking and defensive strengths of each side and the common home goal advantage. Our filter achieves accuracy and efficiency by exploiting conjugacy in its update step and using exact expressions to describe the evolution of the states. The presence of conjugacy enables us to use a mean field approximation (MFA) to update the states given fresh observations. The method is evaluated using the full history of the English Premier League and shown to be competitive, or superior to weighted likelihood or score-driven time series based methods.