Getting boring around here with nothing but the same old arguments, so…let’s start a new lightning rod and talk about “openers”, maybe something useful will fall out of the cracks..
There are many people that use opening lines (outside of line moves) as part of their handicapping. From very simple, i.e. in football the favorite is -2.5, therefore the books are looking for favorite money, to all kinds of line comparison between previous openers and other voodoo that usually makes no sense to me. There are people that swear by understanding the books’ “motives”, while others think that using the line as a variable in a model is counterintuitive as that line is one the thing you are trying to beat.
All the repetitive discussion on how everybody who builds a model is a retard and/or that the line makers have some special knowledge that can only be divined directly from the gods has me thinking. Let’s say in baseball you take all your basic stats that are easily scraped from the web (i.e. WHIP, ERA, W-L, Bullpen, ERA, Team Batting, home field advantage, etc., etc.) and stick them in a basic regression model you could probably get a pretty good opening line predictor. (Lord knows I’ve built some in the past). Now my guess is that most models that use basic regression from easily sourced data are lacking and will lose in the long run. The hard part is recognizing that the variance (i.e. picks) is probably a bad thing. If I have a simple model that predicts a line of -150 for the favorite and the Vegas line is -115 what is the likelihood of the simple model outperforming the model used by the book makers? (And yes I acknowledge that line makers take many things into account including public perception which could mean from a truly statistical perspective the simple model could be more accurate for certain games). I won’t go down the whole “fading your own model” rabbit hole, but I’m thinking of building a simple regression model as a reference point. In the above example if I liked the favorite I would be very hesitant to bet on it. If my “real” model liked the underdog, then I’d be more apt to wager on it.
At the end of the day we are really fighting over fractions of a percent. If you’ve done this long enough you know the pipe dreams of 68% winners isn’t going to happen. But the difference between 53% and 54% is actually huge. I realize we are all trying to quantify the known information with our picks, but is anyone using the “unknown” hidden information that the book makers are providing to filter picks or make decisions?. I do know some people that used a similar approach to somewhat successfully quantify inside information in pari-mutuel markets. Obviously it’s a whole different world in sports betting, so let the flaming begin, and let’s hear all the opening line voodoo you use in your handicapping!