Need to update the model - looking at Hawthorn tips

Now that we are well and truly in the off season it is time for some reflection and tweaking of the model.  Given that it is now two years in a row of “underperforming” (compared to the slew of other models on the internet and well behind the leader on the footytips.com.au site) and the increased complexity of model 8 with its too many lookup tables and cyclic calculations I have decided to work on reworking the model before the start of next season.

I am keeping the basics of maintaining the philosophy of the model and keeping it a shot based prediction model; with additions and penalties for home ground advantage, interstate travel and number of “available” players.  The comparative rankings of defence and attack are also maintained but some more focus on the optimisation of the various weighting factors will be tackled.

I will maintain that bookies odds will not be used as an input to the model – I am in essence calculating my view of what the odds should be.  This is then used to determine the weighting of “wagers” on the footytips.com.au flexi tipping option.

A simple margin algorithm is also used to determine which games warrant tipping in the streak tipping competitions.  This aspect will also be examined in more detail over the off season as historically the tipping runs have not progressed past 15 to 20 and there is room for optimisation/ improvement.

So with those thoughts in mind I will start by examining the 2013 season team by team looking for areas to improve.  So starting with Hawthorn:

An interesting point of note which seems to have gone unannounced is that Hawthorn in the third last game of the year managed to have an all time percentage greater than 100 (that is over their entire AFL/ VFL history they now have scored more points than has been scored against them): they finished the season with 167,182 points for and 167,149 against. (47,427 shots for and 48,059 shots against)(excluding out of bounds but including rushed behinds).

The scoring shot difference I will compare with other teams as I will do with the average points per scoring shot over time as shown in this graph:

 

The following graph highlights the shots for and against over the last few seasons and is used in my modelling.  The scoring barrage of late 2012 plateaued for the whole of 2012 and I think this lower level was cause for some modellers to suggest that Hawthorn were on the slide or under performing.  The over prediction inherent in this coming off a very heavy attack scoring caused my model problems and a number of incorrect predictions particularly in the early part of 2012.

The table below shows how Hawthorn performed in my 2103 model and the second table based on my current model modification v9.  This year in games Hawthorn played I correctly tipped 19 of 25 (76%).  The early run of difficult games were the majority of the incorrect tips.  The modified model I am working on tends to favour homes teams slightly more – this is based on the complete AFL/ VLF season history.  The choice of factors so far settled on when run for Hawthorn games last year only yields a 60% accuracy with an even greater emphasis on the home team.  Comparison with the other teams is to be considered in order to weight the home ground factor.

Model v8: 2013 model results for Hawthorn games.

Model 9 preliminary factor analysis – re run of 2013 season (Hawthorn)

The tips against the odds: Collingwood, Adelaide, Port Adelaide, North Melbourne and Sydney: all of which were unsuccessful.  North Melbourne, Adelaide, Port Adelaide and Sydney all pushed the Hawks with only Collingwood being taken apart by the Hawks – a failing of Collingwood in a number of games.  The West Coast game in round two was always predicted to be tight with the models – in hindsight the good year of the Hawks and the poor year of the Weagles is brought into focus in this game.