Fun with KenPom

So I’m hurriedly preparing for handicapping college basketball season by revamping the model. Somehow the model gets bigger every year. Maybe by 2014 I’ll be making bets influenced by nose size and number of siblings. For ldo reasons, at the beginning of the year stats aren’t the end-all-be-all that they are most of the way through the season. For a lot of teams, last year’s stats are worthless for anything more than some general thumb in the wind guesstimate. Using the returning players isn’t much of a solution either. Individual player valuations are 1) still shitty in basketball 2) even shittier in CBB with little player history and far from static ability levels 3) not all that helpful since player turnover in CBB is ridiculous. Basically, I avoid games early in the year where most of the PT is wearing the same uniform but doesn’t resemble any incarnation of previous years’ teams.

Every year, I also go back through my old kenpom data, since it’s the dysfunctional heart of the model. Kenpom data couldn’t live on its own, but it’s a hell of resource for making your own Frankenstein. As an aside it’s the people that reply to homer message board posts with “kenpom sez, kenpom sez” that help fuel the donklash against it. Exhibit A being the “LOL WHERE IS YOUR KENPOM NOW” internet mob last March Madness season. It’s smarter than they are, but it’s not as smart as $$$ opinions either. Anyway, since I make so many adjustments to the raw kenpom data, I need to check that data for computational changes year to year. Interestingly, I am fairly sure kenpom made some not so immaterial changes to how he adjusts his efficiency stats. Think of it as something like the process of what a 50 point dunkfest over a scrub versus a 2 point squeaker over UConn would translate to in each teams’ aggregated efficiency stats after those games. It’s probably the most important part of CBB because there is such little parity.

The data in the chart below is adjusted by me in several ways which I am not going to go through, so this isn’t pure KenPom prediction & results. It is pure KenPom efficiency formulas, just adjusted to remove some bias, but adjusted in the same manner each year. It shows year by year home cover %’s for all non-neutral games. The spread ranges are at the top.



Kenpom used to have a big problem predicting blowouts. Seeing -21 online and -24 kenpom, and betting accordingly would have been a great way to light some money on fire. But it got a lot better last year. Maybe he has done something like muting the impact of UNC crushing Northwestern Quadrant of Idaho Seminary School, I don’t know. But kenpom might be getting closer to needing less and less adjusting. Just as an FYI, the high cover rates between 0-10 are mostly the result of kenpom not incorporating actual outcome distributions. Team A who (on paper) is 1 point better at home than Team B, will cover -1 more than 50% of the time.