7th
Ratings, markets, and filling out your NCAA tournament bracket
So the men’s NCAA championship is tonight, and I’ll be watching nervously, not because I have any long-standing emotional stake in either team, but because I’ll win a relatively large NCAA pool if Kansas beats Memphis. (I’ll actually win both 1st and 2nd if Kansas wins - I’ve got four entries in the bracket - and 2nd if Memphis wins, but glory goes to the winner, yes?)
This brings up one of those fundamental philosophical debates (right up there with natural law and trees-falling-in-the-forest): how to fill out an NCAA tournament bracket.
I’ve always been a conservative bracket-filler, which is to say that I rarely pick upsets. This isn’t a dominant strategy for everyone; if you’re in a very large, winner-take-all pool, and/or you aren’t terribly confident in your ability to forecast college basketball games, there’s a lot to be said for picking a longshot to go all the way. You’ll probably lose badly, but it maximizes your chance of winning. (The wrong way to pick upsets is to take a lot of 13, 14, and 15 seeds to win, which is simply a bad strategy.)
Since I watch a lot of college basketball, and flatter myself that I understand the game better than most bracket-fillers, I generally go with favorites. I spend my time worrying about 8-9 and 7-10 games in the first round, 4-5 games in the 2nd round, and the games in the Elite 8 and beyond, and then I hope that the tournament is upset-light and trust in my ability to out-pick others on the close games. It’s a strategy that’s served me well this year - in addition to the pool that I’m hoping to win tonight, I’m currently in 1553rd place in the ESPN Tournament Challenge, which doesn’t sound very good until I point out that there are probably about 3 or 4 million entries.
… except that I can’t give myself credit for great basketball savviness this year. That’s because my bracket picks were almost completely guided by the Pomeroy Ratings (http://kenpom.com/rate.php). I pretty much went with the teams that these ratings predicted to have the best chances to win each round, and in cases where those chances were pretty even (e.g., Memphis or UCLA to reach the final), I split the difference and submitted team A on one entry and team B on another.
In previous years, I’ve paid increasing interest to other rating services - first Sagarin, then Massey - but since last year, Pomeroy has been my favorite. What I’ve thought about doing (but have been too lazy to do, so far) has been to track Pomeroy’s predictions across an entire season - and not just with regard to wins vs. losses: since he predicts the percentage change that a given team will win, we should be able to see whether his predicted odds match the distribution of results. E.g., take all the games where the favorite is given 50-55% odds to win, 55-60% odds, 60-65% odds, and so on, and see if the results match those proportions.
Data point #1 could be tonight: we have an interesting case where the Pomeroy ratings show Kansas as a small but clear favorite - 62% to win - and the Vegas oddsmakers and sports-futures trading sites prefer Memphis (1.5-2 point favorites and 52-55% odds, respectively). I have great respect for Pomeroy, but I also believe in the wisdom of markets. Obviously tonight won’t prove anything either way, but maybe a comparison could be part of a master’s or Ph.D. thesis. :)
