Estimating ratings in football: brazilian championship 2017

Abstract

Observing games from 2017 Brazilian Football Championship League series A, we adjusted models with linear predictors for the expected number of goals for home and away teams. We estimated values for attacking (γ) and defense (δ) strength parameters for each team. A common home advantage effect ( µ ) was used for all teams and conditionally estimated. The first leg of the double round-robin was used to produce initial estimates. Those were then re-estimated at each round of the second leg. We intend to present a model that could describe past performance and predict results (and scores) from each game. Additional features of the final classification of the tournament could also be predicted, such as the probability of playing South American Champions League or being demoted to series B of the competition. The proposed model can be considered flexible in the pure likelihood analysis version, but some desirable features of parametric Bayesian inference could enhance it’s capabilities.