Monday, December 24, 2012

Word of caution - long range predictions

Forensic check on the 2012 season with the new streamlined Monte Carlo full season predictor: Managed to predict six of the top eight - Fremantle came from nowhere in my model (I predicted 14th and they managed 7th: well done purple haze) and North snuck into the eight where I had them at tenth with a 43% chance of making the eight.  My big loosers/ poor performers of 2012 (apart from the Hawks loosing a seemingly straight forward statistical win in the last game of the season) were the Siants and Carlton who managed to end up 9th and 10th so I do not feel too bad about these.

The beginning of 2012 saw the Hawks facing a number of potentially tight games/ swing games that were flagged to define their season within the model, the Hawks still managed to be minor premiers despite the mid season Monte Carlo model predicting a third or fourth position - their very strong run of games in the second half of the season starting with the demolition of the Roos pushed them to the top (peaking too early?).  Both Sydney and West Coast were flagged as strong contenders right from the start which i think was borne out - the Weagles coming very clsoe to a top two position - loosing the crucial swing games late in the season

Now having said how good the model predictions were for final ladder position it must be said that the order was all over the shop - only got one of the top four correct and as shown in the last post the top four is where the premier comes from 96.5% of the time in the model.  As I have previously posted the model seems quite accurate for up to six rounds into the future.  I am focusing on using a more robust (i.e. I need to improve the) model algorithm for the six rounds into the future then switch over to the simpler streamlined Monte Carlo version for the longer range predictions. 

Model simulation: top 8 chances of winning the flag

Having run the full 2012, from round 15 and round 18: 2012 and 2013 season many hundreds of times I thought that it was appropriate to examine what chances (my model predicts) a team will have of winning the flag depending on where they finish at the end of the home and away season.  Note there will be differences from season to season as the various home ground advantages come into play depending on how may interstate teams and their relative position.  However, summary: minor premiers are predicted to win the flag 45.5% of the time, second 22.8%, third 16 and fourth 12.2% (top four accounting for 96.5% of premierships).  The second four account for 3.5% of premierships - so far the model has not thrown up an eighth place premier.


Monte Carlo runs 1000 top 4 965




top 2 683 3rd and 4th 282

5th to 8th 35

455 228 160 122 21 12 2 0

1st 2nd 3rd 4th 5th 6th 7th 8th

What do I take away from this?  Gut feel is that this proportion is about right; also should be looking at the bookies odds for premiership at about round 14 and focus on my predicted top 4 and compare the model percentage chances of winning with the bookies odds looking for some opportunities.

Thursday, December 13, 2012

Full season monte carlo (no draws) shot predictor model: final ladder predicitons

The Carlo Monty laptop has been running hot crunching numbers with a simplified iterative model predictor to sooth say (or out sooth say as best as possible the Footy Maths Institute!)

Adelaide still persist in coming out on top despite the large number of tight games.  GWS has a bit of a monopoly on the wooden spoon with Western Bulldogs giving them a run for their money.

Top eight certainties at the moment are: Adelaide, Hawthorn, Saints, Swans and Eagles.  Freo and Cats almost certain top 8.

Monte CarloScoring shot basis (no draws)
Full season100 iterations
TeamWin2ndTop 2Top 8123456789101112131415161718
Adelaide24%25%49%100%50%20%8%15%7%0%0%0%0%0%0%0%0%0%0%0%0%0%
Brisbane0%0%0%63%0%0%0%0%0%6%15%42%20%11%5%1%0%0%0%0%0%0%
Carlton0%1%1%20%0%0%0%0%0%3%7%10%28%25%17%7%2%1%0%0%0%0%
Collingwood0%0%0%0%0%0%0%0%0%0%0%0%0%0%1%4%16%72%6%1%0%0%
Essendon0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%1%8%59%32%0%0%
Fremantle2%2%4%91%0%0%1%4%6%16%42%22%8%1%0%0%0%0%0%0%0%0%
Geelong4%2%6%96%0%1%1%8%26%35%16%9%3%1%0%0%0%0%0%0%0%0%
Gold Coast0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%2%33%60%3%2%
GWS0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%4%10%86%
Hawthorn24%23%47%100%13%32%25%22%7%1%0%0%0%0%0%0%0%0%0%0%0%0%
Melbourne0%0%0%6%0%0%0%0%0%0%0%6%17%28%29%15%5%0%0%0%0%0%
North Melbourne0%0%0%0%0%0%0%0%0%0%0%0%0%2%10%34%37%13%2%2%0%0%
Port Adelaide0%0%0%1%0%0%0%0%0%0%0%1%2%13%13%32%35%4%0%0%0%0%
Richmond0%0%0%23%0%0%0%0%1%8%6%8%22%19%25%7%4%0%0%0%0%0%
Saint Kilda21%21%42%100%24%21%30%16%6%2%1%0%0%0%0%0%0%0%0%0%0%0%
Sydney4%6%10%100%1%2%4%11%41%27%12%2%0%0%0%0%0%0%0%0%0%0%
West Coast21%20%41%100%12%24%31%24%6%2%1%0%0%0%0%0%0%0%0%0%0%0%
Western Bulldogs0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%0%1%87%12%

Sydney have a big range of possible end of regular season positions but 5th or 6th is favoured at present.  North and Collingwood will not see September action (really need a big start to the season to overturn the model predicitons).

Too much number crunching for today... Will fire the laptop up again on boxing day or so