kenpom predictive accuracy
Admittedly, kenpom is the leader in college basketball metrics. It is not, however, the end-all-be-all.
On kenpom.com: "The first thing you should know about this system is that it is designed to be purely predictive. If you’re looking for a system that rates teams on how 'good' their season has been, you’ve come to the wrong place."
I recently tried to express interest in how well kenpom does with its ratings, especially how well it takes strength-of-schedule into account. Some of you disapproved.
So, let's take a look at how well kenpom has done with Michigan this season. Comparing the predictions that mgoblog has published and comparing them to actual results, kenpom has accurately predicted the winning team in Michigan's games 65.2% of the time, with an average error of 8.4 points in those contests. During B1G play, the kenpom accuracy in Michigan games has dropped to 60% with an error of 9.8 points.
Couple of interesting articles that might help:
http://www.lasvegassun.com/news/2008/jan/18/s-spread/
http://www.reviewjournal.com/matt-youmans/lvsc-new-path-without-white
There's confirmation here, too, that they're trying to draw equal sums on both sides of the bet. That's what they say they're doing, and it's how I'd expect a financial firm like Cantor to make a market. Why take on risk when you can pull 11:10 without it?
Will read, thanks.
I don't see the factor for offense wins games and defense wins championships. Hogwash the lot of it!
It would be interesting to see a basic analysis about game winning probabilities. Kenpom predicts "Team A" has a 65% chance of beating "Team B". Why not take a huge set of predicted games and sort them into bins by percentage chance of winning (50-100%), and see what percent of each of those games were actually won by the predicted winner. Theoretically it should be 50% for the 50% bin and 80% for the 80% bin, etc.