Friday, January 8, 2016

Back from hibernation

The WayBeyondRedemption tag line has historically been "too many airport lounges, excel spreadsheets and bottles of red to try" was very heavily weighted to the former of the triad during 2015 to the detriment of the later two.  In an attempt to readdress this lamentable situation (without loosing full time employment) the Chief Alchemist at WayBeyondRedemption has decided to update the motto to the more traditional Solve et Coagula (dissolve and coagulate) with the implied intention of focusing on the number crunching (distill the stats and crystallize the wining margins) a bit more in 2016.



During the off season the WBR super computer has crunched the numbers and come up with the "22 rounds into the future shot predictor" (22Future); the "monte carlo season simulator ranking" (Monte Rank); and the "patent pending end of season weighted guess" (Ladder Guess).  This blog will focus on the 22 Future model.  The next two will focus on the other two predictions then four blogs examining the teams as ranked (bottom four; chance for the eight; hopefully in the eight and top four).





22 Future Model

The basis of the model is based on the primary analytical data within the Carlo Monty model (weighted shots on goal differential): the difference between:

* the weighted average shots on goal for (of a particular team):

(a combination of the 20 point, 10 point and 5 point moving average of the shots on goal)

* the weighted average shots on goal by the opposition (in the same manner as above)



The historical data set for Hawthorn was then analyzed and the following simple linear regression model was chosen:

The shot on goal differential 22 rounds into the future is equal to a function of the current 20 point moving average differential and the 5 games and 10 games prior weighted shot differential plus a constant.



Hawthorn was chosen for a number of reasons but mainly because they have enjoyed a number of premierships and endured a number of wooden spoons since joining the AFL.  They also avoid the statistically variable early years.  Thirdly, I follow the hawks.



I have looked at using the same thinking for the modern era (1980 onwards roughly) but have not found a great improvement in the end results.  New teams entering the competition tend to skew the data.  Similarly analysis of the data from 2000 onwards for all teams has not generated a noticeable improvement over the initial Hawthorn based model.



The headline data from the model is  if a teams 22Future number is greater than 5 they are in contention for a Premiership.  If it is less than -5 then a Wooden spoon is a distinct possibility.  With the current 18 team roster a value of 0 to 5 indicate top 8 and 0 to -5 indicates bottom eight.  A more detailed correlation requires more red wine and free time - both near term goals.



The 22Future predictions shown below clearly show Hawthorn and West Coast as the only two serious contenders for the 2016 Grand Final - with a lot of teams with a lot of catching up to do.







Looking at the predictions at the beginning of 2015 and the actual results:









Shows that using the same thinking Hawthorn was the team to beat.  By the end of the 2015 Home and Away Wet Coast were the form team on a big roll; Sydney and Adelaide were contenders while the rest were making up the numbers.



22Future has obvious issues with the accuracy of individual teams scores (such as the massive jump in West Coasts values during 2015) however, the form indicator that monitoring this model provides is interesting.



So going into Round 1: 2016 the 22Fut ladder is (there are some minor differences between the above and the below  - lookup tables out by a round: something to check which is correct within the loops within loops : ) ):







Next blog: the subtle effects of the "fixture" as examined by the Monte Carlo model (examining the end of round 6 and the crystal ball of the end of the 2016 Home and Away season)...