Modeling basketball games by inverse Gaussian processes

Xinyu Tian, Yiran Gao, Jian Shi

Abstract: The scoring processes of home and away team in basketball games are modeled by two dependent inverse Gaussian processes with a team-specific parameter that measures the team strength. A common latent variable that measures the game pace is designed to characterize the dependence. A moment estimation method combined with maximum likelihood estimation is proposed to fit the parameters and a Bayesian method is applied to update the estimation and make in-game predictions. It is shown that the proposed model can obtain the same performance as the benchmark model, Gamma process model, in outcome prediction, point spread betting and model gambling.