Thursday, December 14, 2017

Forward prediction model – historical perspective


This blog examines the historical performance of the Carlo Monty forward prediction model since the final eight system in 1994.  The model predicts the team rankings 22 games into the future from round one each season.  The model is based on a simple regression model of weighted averages of the previous 20, 10 and 5 games scoring shot differences.  A positive number equals more wins than losses while a negative number indicates more losses.  Within the model a forward shot difference prediction greater than 5 is a good indicator of grand final form.  A value more negative than minus 5 and the team is in contention for the wooden spoon.  Zero to five is roughly a final’s appearance and zero to minus five is the duldrums.



In the 24 seasons from 1994 to 2017 the team predicted to finish atop the ladder appeared in the finals 23 times (Brisbane did not appear in the finals in 2005).  However, on seven occasions the premier was correctly forecast.  The premier came from the top four predictions 71% of the time – the most recent exceptions being last years Tigers and the previous years Bulldogs with the Tigers being more of a surprise, surprisingly.  Notably both were correctly tipped on Grand Final day. (This time last year I was predicting a Cats: Crows grand final).

Looking at the ladder at the end of the home and away season the proportions are:




So the predictions for the end of 2018 are shown below:



As noted in previous posts when compared to the simulated seasons the models are looking for possible upward movements of Hawthorn and Essendon at the expense of Geelong and Western Bulldogs; and cause much media attention running into the finals.  Richmond and GWS may swap spots higher up the ladder. The battle for the wooden spoon is for Brisbane to get off the bottom and possibly force Gold Coast to take the "prize."









No comments:

Post a Comment