Been a while since i posted an update. Who wants to remember round 2? Not me.Ugh, what a shocking week. Unlike last week, 3 out of 4 50/50 games went against me. And melbourne. Bloody hell.
Model predicted 3/9.
118 invested for 51 returned.
Profit -59%
Lessons learnned:
Will need to incorporate gamestyle and home ground advantage into the model somehow. Will require some thinking.
Not going to knee jerk into chasing value at the expense of accuracy just yet. If i had been chasing value these past two weeks i'd be in the same spot as i am now as my win this week would have been balanced by a big loss in week 1.
Dealing with loss. I lost 60% this week. That's bad. Some tweaks to the method required in the future however that's gambling amd i started small so i could learn and limit losses. I committed my money for the season and so far have gotten value out of enjoyment of the excercise alone. Win win. I think i'll refer back to this for some perspective later in the season.
First, the numbers:
Rnd 3:
9/9 for 51% profit
Rnd 4:
5/9 for -35% profit
Rnd 5:
6/9 for 3% profit
Rnd 6:
7/9 for 28% profit
Rnd 7:
8/9 for 38% profit
After that terrible round 2, i am back to square (almost - overall RoI is -0.1%).
So far i have blatantly ignored the model for two games and my gut has been right twice (melb/richmond and geelong/west coast - it's really hard to account for mental fragility and homesickness using statistics)
I have also subtly adjusted 'judgement factors' in my model to suit my intuition 4 times from memory:
-Hawks/WCE (loss - had hawks at 53% initially)
-GCS/Bris (loss - had brisbane at 52% initially)
-Stk/Melb (loss - had saints at 51% initially)
-Melb/Coll (loss - had coll at 63% no change no matter what ratings i adjusted. Learned to ignore model occaisionally from this case and have since adjusted some methodology to account for collingwoods stacked defence which was skewing results)
The lesson has been believe the model for the close ones unless i have a strong feeling (richmond are terrible, wce always forget to pack their appetite for the contest).
Having watched quite a few games, i've also managed to fill the gaps in my kbowledge of other teams somewhat. The model is now fairly well balanced and is giving more accurate results most of the time.