Friday, August 1, 2008

In-season Marcel Projections

One of the bigger areas of research in current SABRmetrics is in the design of projection and forecasting systems to predict future performance. There are a bunch of systems popping up, but one of the most interesting is the MARCEL system, developed by Tom Tango. It's not interesting because it's complex, but rather because it's so damn simple. All it does is to take a weighted average of previous seasons, apply some regression to the mean and some age adjustment, and BAM!, there's your projection. It's actually named after Marcel the monkey because of its deliberate simplicity.

It's also interesting because it's completely open-source which means you can actually see where the results come from. And despite this lack of complexity, it still stacks up well to the other projection systems floating around out there. The projections have been typically released before the season, but Sal Baxamusa at the Hardball Times has developed a spreadsheet which you can use to calculate in-season MARCEL projections. Too lazy for that? (I'm nodding my head.) Just head over to Colin Wyers' site where he's written some code and is displaying the projections in real-time.

The best hitter right now - or at least the hitter with the best projection going forward, if you believe there's a difference - is Albert Pujols who checks in with a .333/.438/.603 line. He absolutely laps the field, coming in with a projected OPS over 50 points higher than the nearest competitors. Those competitors are David Ortiz (.293/.404/.586), who still looks good after this season's injury-marred performance, and Alex Rodriguez (.305/.404/.583).

A few Orioles' notables:

Nick Markakis (.303/.378/.481, +24 lwts) - A little pessimistic in my view. Of course I don't say that as an unbiased observer, but rather as a guy who is a huge Markakis fan. I'd still bet the over on that OPS over his next 600 PA.

Aubrey Huff (.270/.335/.456, +6 lwts) - The power of regression to the mean combined with its good friend, aging. Marcel the monkey throws cold water in our faces regarding Huff's big season, projecting him for just a .791 OPS, well in line with the .778 OPS he compiled from 2005-2007. This is why, at best, we should be cautious about Huff even approaching, let alone matching, this season's numbers in 2009.

Luke Scott (.268/.355/.488, +18) - Luke Scott, on the other hand, is a guy we might be able to count on in the future. Or at least next season. The system likes him so much that even with Scott being 30 and putting up a nice season that it thinks he's playing a few points of OPS below what he's capable of doing. Not a bad pickup in the Tejada (current projection: .300/.349/.470[!?]) trade.

Adam Jones (.270/.317/.405, -9) - That's a tick under where he's at this year, and as with Markakis, you think (hope) he can do better than that going forward. One issue with the system is that it doesn't know anything about minor league performance. If we looked at translated minor league stats, as a system like PECOTA does, we'd see why Orioles fans would call that projection pessimistic, but that's one of the issues that you have to be cognizant of with MARCEL.

The worst projection in the system goes to the immortal Corky Miller, a 32 year old journeyman backup catcher with a career .179/.269/.295 line in 378 PA over the last 8 seasons. Even Miller can't be that bad, says Marcel, as he's projected for a .182/.244/.300 line going forward.

No comments: