Monday, February 18, 2008

No Matter What I Do In This Scenario, It Will Probably Not End Well

Over the weekend, Vegas Watch passed along over/unders on each major league team's total wins in 2008. I thought I wanted to check the Red Sox first, but curiosity (and size 48 font) drew my eyes immediately to the "Tampa Bay" line. Early projection on the new look Rays? 72 wins.

My first reaction was scorn. "Ha! 72 wins," I thought haughtily to myself. "Easy money. There is no way Tampa Bay wins less than 72 games. Baseball Prospectus has them winning 82. 82! That's above .500! Where do I sign up to place a sizeable and completely legal bet on this?"

Tonight, however, I was playing around with BP's PECOTA depth charts, looking to see how it did at projecting results last season, and stumbled upon something that turned me into Winnie the Pooh: "Oh, bother."

First, a little background: the Lou Piniella Experiment lasted three seasons in Tampa. In that time, PECOTA came up with the following projections for the Devil Rays:

Year PECOTA Actual
2003 57-105 63-99
2004 56-107 70-91
2005 68-94 67-95

The six win differential in 2003 seems like a reasonable margin of error (I think Vegas Watch came up with something around 5.0 or 5.1 when reviewing PECOTA at the end of last season), while the fourteen in 2004 is way off (the Yankees and Red Sox were expected to absolutely destroy the American League that year). That 2005 prediction, however...nice work, gentlemen. Nice work indeed.

Now, compare that with the way Tampa Bay's actual records have matched up with the PECOTA predictions since the Joe Maddon Era started in 2006:

Year PECOTA Actual
2008 82-80 ???
2007 78-84 66-96
2006 69-93 61-101

Two years in a row, PECOTA has swung the opposite way from previous years and drastically overrated the Devil Rays' expected performance. Even if one takes into account a three win adjustment that Nate Silver made to last year's total, we're still looking at a nine win differential. I don't know enough about computers or the way these things are figured out to try explaining why this might have happened, but it's still unnerving, especially if money is involved.

We all know where this ends. I keep my wallet in my pocket while the Rays finally break through, go 95-67 (outperforming their projected won-lost by thirteen wins), and win the American League East. I win nothing. Or, I put a few bucks on the over, Scott Kazmir and James Shields both flame out in spectacular fashion, and the team wins 53 games. It's what I like to call a no-win situation.

Of course, there's always Seattle and the under. No way they'll win 85 games...right?

3 Comments:

Vegas Watch said...

This is a very interesting analysis, and I was thinking the same thing about last year's PECOTA projection, which was obviously way high.

HOWEVA, I do not think this is all that relevant to future performance. If we were talking about PECOTA's 2008 Rays projection prior to the 2007 season, that would be one thing. But it has already taken into account where it was wrong last year by incorporating 2007's numbers into its 2008 prediction.

There are certainly no guarantees, they call it gambling for a reason. But as far as gambling goes, I think this is a pretty solid bet.

Bruce said...

PECOTA has also projected the Cardinals at 72-90, which is only six games off 2007(78-84). Now, with the news that Clement won't be ready for Opening Day, and Pujols' wonky right elbow, that might be that far off. But then again, I've always been a glass half-empty kinda guy...

stopmikelupica said...

Six games off is a large amount in predicting a baseball team's record. Basically, say 60% of teams will end up between 75 and 87 wins - you can reasonably guess which 20% will be better than 87 (Boston, NYY, NYM) and which 20% will be below 75 (KC! Baltimore!). Then just predict everyone else at around 81 wins, and booyah... you'll be within 6 games of every team pretty much.

This comment has nothing to do with the larger point of this post; I'm just pointing out that 6 games is a pretty generous margin of error, that's all...