Croatian First Football League: Teams' performance in the championship

The goal of our research was to use simulation modelling for prediction of the Croatian First Football League seasonal ranking and analyse variation in teams’ performance during a season. We have developed a model of the number of goals scored by a team in a match based on the Poisson distribution. Parameters of the model were estimated from the data on consecutive matches in a season. Variation in a team’s performance was modelled as a moving parameter estimate. The final rankings were predicted from 1000 simulation runs of the second part of the season based on parameter estimates from the first part of the season. For each team the most frequent outcome of the simulation defined the team’s rank. The method was tested on seasons 2014/15 and 2015/16. Prediction was correct for six teams in the season 2014/15 and five teams in the season 2015/16. Proposed methods enable dynamic monitoring of a team’s performance and prediction of final rankings during the season. An advantage of the prediction method is that in addition to predicting the final ranking it also estimates probabilities of alternative positions.