Estimação de rating no futebol: campeonato brasileiro de 2017

Abstract

Observing games from 2017 Brazilian Footbal Championship League series A, we adjusted Poisson models with linear predictors for the expectation of the number of goals from home and away teams. We estimated attacking (g ) and defense (d ) strength parameters for each team. A common home advantage effect (m ) was used for all teams. The first leg of the double roundrobin was used to estimate initial ratings. Those parameters were then re-estimated at each round of the second leg. We intend to present a descriptive model that could estimate as well as predict scores and results from each game. Additionial features of the final table could also be predicted as probability of playing South American Champions League or being demoted to series B of the competition. Model has been shown flexible in the pure likelihood analysis version, but some of desirable features of parametric Bayesian inference could enhance some of its properties and will be object of future research.