Holding up the tipsters ladder

With one more game to go in round 10 and after “enjoying” the twin live coverage of Bulldogs v Port (thanks doggies for stuffing up my streak tipping) and the Hawks walking all over the Dees: all courtesy of live coverage on Australia TV in the comfort of a Hanoi hotel  (after travelling for a month surely I could get a decent game to watch!) it is time to fire up the waybeyondredemption.com blog

Modeller                        Correct tips

Swinburne                      67

MAFL margin average       66

Bookies                         65

Footy Maths                   63

Footy Forecaster             62

Carlo Monty                   58

Approaching the halfway mark of the season and I am sadly adrift and a long way from the tried and true form of the Swinburne “Tin Head”. 

I am not surprised that I am behind the statistical rigour of the MAFL (I have taken the liberty of comparing myself against the average of the margin predictors amongst the multitude of tipping analysis that is embedded in the MAFL blog). 

Hats off to Swinburne and MAFL for being in front of the ever reliable Bookies (who I set as my benchmark to beat).

Then comes Footy Maths (who as they refer to me as the “mysterious” Carlo Monty I shall refer to as my Nemesis (although not as mean and nasty as Brick Top in Snatch: more like Moriarty in The Goons)).  The Footy Maths Institute at least can dine out on a free burger from Footytips.com.au.  Given the western suburbs flavour of the footy maths blog I presume the free burger will be eaten at the Flemington Hungry Jacks?!

The Footy Forecaster has had a mediocre year so far but is still out performing me.  Historically it has been neck and neck between our models which probably comes down to a similar model structure.  I admit I do not know the Footy forecasters technique (ladder position ranking I gather), or the Nemesis’ thought process (team ranking), Swinburne have made comments on the web in the past which implies a level of sophistication but in my interpretation erring on the side of simplicity, MAFL does describe his technique but I need to refer to the textbooks to keep up, and the bookies – well they just shuffle money around depending on public opinion.

As other bloggers have shared their modelling philosophy and made veiled requests to know my structure...

The Carlo Monty system is a simplistic prediction model based on scoring shots primarily.  Scoring shots are allocated based on:

·   *      Data from 2000 onwards.  During this time I consider that all teams have roughly similar advantages and disadvantages based on draft, salary cap, travel commitments and general level of player fitness/ professionalism etc.

·   *      Home ground advantage – if both teams are from the same state the advantage is minimal.  The advantage stems from vocal supporters rather than knowledge of the wind in the Rising Sun end of the ground or whether the sprinklers were turned on on Friday at Moorabin.

·   *      Winning streak over the previous month (“a month is a long time in footy”).  If you have won the last three your confidence is up compared to only one win in three etc.

·   *      Defensive and Attack rank comparison between the two teams.  I include a number of games from the previous season.  The early models examined using the previous 22 games down to the previous couple.  Again I settled on the month is a long time in footy methodology using the last three games of the home and away season from the previous year.

·   *      A moving average of a team’s previous shots on goal difference.  Again the model has used a range of different moving averages over the course of the previous 7 versions and a range of ladder percentage.

·   *      Finally a small correction is made based on team “health”: a proportion of the team’s best 22 who are available for the game.  Altering this factor has had little effect on the model over the years and has a small weighting.

The two teams predicted scoring shots are compared and a “fair odds” estimate is made.  This fair odds is compared to the bookies odds on Tuesday and are used to determine the footytipping.com.au flexi tip options.

The shot predictions are also run through a monte carlo algorithm which in the complicated form has set distributions for goals to points.  After 1000 runs for each game the percentage chance to win is again compared to the bookies odds.

Streak tipping is based on games predicted to be won by 3 goals or more.

Because the model is history based forward predictions can be done by assuming that the predictions for a round are correct (this of course gets less accurate the further out the predictions go and I have found that 6 rounds is about the limit for reasonable robustness of the model).  Of course with a disastrous tipping in the early rounds this year long term predictions become a mute point as will be displayed in the next blog.

Although historically the model is running at 75% accuracy.  The round 9 totals from previous years are:

2001  54

2002  57

2003  52

2004  57

2005  53

2006  49

2007  56

2008  46

2009  46

2010  47

2011  60

2012  57

2013  53

(note for Nemesis: I count draws as “not a loss” so therefore a win for everyoneJ: the mysterious Carlo Monty is an optimist)

The model definitely suffered from too many one sided contests in 2012 which bolstered a team’s expected performance the following one or two weeks.  It is a little harder to pinpoint the problems with the model this year but they revolve around Saint Kilda (Gone), West Coast (underperforming) and Essendon (doing better than expected early: and over rated) (and a little on Port).  More of this in the blog for the flight home...