from Amazon
Thursday, November 09, 2006
Baseball Prospectus readers know what they are doing
Chalk up yet another win to the Wisdom of the Crowds, this time courtesy of Baseball Prospectus.
If we…
...count a “win” as any pick that was within 3 wins of actual and a loss as any pick that was at least 8 wins away from actual, BP readers are 10-10, PECOTA is 13-12, Vegas is 10-9, ESPN is 13-13, and DMB only 9-12.
The actual standard deviation of wins was 10.1, and all forecasts were (correctly) below that. PECOTA at 7.4, DMB at 8.1, BP readers at 8.5, Vegas at 8.7, and ESPN at 9.5. When you guess, you are guessing on talent, and you are not trying to guess the noise as well (unless of course you are competing with a group, where a wild guess actually helps).
In this case, 10.1/162 = .062. ObservedSD^2 = TrueSD^2 + LuckSD ^ 2. LuckSD = .039, making our TrueSD = .048 wins per game, or 7.8 wins per 162 games. PECOTA and DMB were the only ones that were really close to this, though BP readers weren’t that bad.
(Weirdly, the Vegas odds predicted 81.4 wins per team. Seems to my uneducated gambling mind that if you picked the under for every single team, you’d have won more than you lost. In fact, 16 teams won less, and 14 won more than the Vegas line.)
***
I also recommend this study by me a few years ago:
http://www.tangotiger.net/forecastFinal2.html
As well as this one by David Cameron and Greg Spira:
http://www.baseballprospectus.com/article.php?articleid=229