Computer Forecasting of Australian Rules Football for a Daily Newspaper

Abstract

An exponential smoothing technique operating on the margins of victory was used to predict the results of Australian Rules football matches for a Melbourne daily newspaper from 1981-86 and again for a competitor in 1991-92. An initial ‘quick and dirty’ program used only a factor for team ability and a common home ground advantage to predict winning margins. Probabilities of winning were accumulated to predict a final ladder, with a simulation to predict chances of teams finishing in any position. Changes to the competition forced a more complicated approach, and the current version uses several parameters which allow for ability, team/ground interaction, team interaction, and a tendency for team ability to regress towards the mean between seasons. A power method is used to place greater weight on the errors in closer matches, and errors across the win-lose boundary. While simple methods were used originally, the Hooke and Jeeves method was used in optimizing the parameters of the current model. Both the original model and the improved version performed at the level of expert tipsters.