Quote:
Originally Posted by TomG
After some thought, this might be a candidate for the "Benter Boost" (credit @Gingfacekillah for the term, Billy Benter for the concept). I think of NZ 5v5 FO as the pool of available players to choose from. And the actual opening faceoff data as the YTD utilization of that pool. We shouldn't be disregarding the observed opening faceoff probabilities altogether. This is valuable information. So why not a Benter Boost as a way to ensemble the two? Or perhaps we can do some year-over-year correlations on the opening data and regress the observed to the percentage in the pool?
I could be way off base here since I haven't read the book you mentioned. But I looked up the concept and it seems to apply to regression and market odds. I also don't know much about hockey or the model you're using for face offs. But do you think you're accurately capturing regression in your model?
You mentioned betting on a guy who has won 83% of face offs ytd vs a guy who has won 5%. I would naturally want to bet on the 5% guy and fade the 83% myself, assuming face offs are close to a coin flip, mainly because of regression. I would also like to know if 83% is a sustainable FO% in hockey and what is the long term FO% of this guy and same for the 5%.
So I'm concerned that looking at YTD face off win percentage and treating it as predictive going forward is a mistake. Those are ad hoc numbers, not necessarily predictive of the future. In fact one of the ways they could be predictive is in fading outlier positive performances YTD and tailing outlier negative ones.
To put it simply you might be betting on guys who are running hot and betting against guys who are running below EV. This seems like a losing strategy to me, but I might not be understanding your model correctly.
Last edited by donkeyballz; 01-07-2020 at 01:19 PM.