Behind the Monkey

With the debut of our Fan Projections this week, the Marcels have been a topic of discussion among our analysts. The Marcels are a very appealing system to me because of their simplicity. For those of you that are unfamiliar with the Marcel projection system, it’s not a total black box like PECOTA and it doesn’t use special aging curves or player comps like other systems such as CHONE or ZiPS. Here’s the rather simple basis behind the monkey, and these can certainly be expanded to our fan projections here. For a more detailed (and better) explanation of what Marcel is, check this out.

– Recent performance matters the most

Nobody is going to go onto these projections and project Pedro Martinez to have a 1.95 FIP season just because he did it in 1999, and nobody’s going to project Andruw Jones as the +20 run centerfielder that he was earlier in his career. We’ve seen both of these guys perform at much worse levels. Conversely, it’s a much better bet for Joe Mauer to hit 20 HRs now that we’ve seen him hit 28 in 2009. Marcel handles this by using 3 years of performance, weighting them 5/4/3 respectively. That is, the most recent season gets 1.25 times the weight of two years prior, and 1.66 times the weight of three years prior.

– Players regress

This is the point of using player comps in other projection systems – different players regress to different means. When you are entering your projections, consider that Albert Pujols is probably not going to regress to a league average, .260/.330/.425 type line. He’s going to regress to his superhuman .333/.425/.630 line. Still, most players either don’t have enough experience to define their own mean to regress to or are close enough to said mean that a regression towards the overall league mean will do quite nicely. Along with the 5/4/3 weighting mentioned above for recent seasons, Marcel weights average performance as a “2” to handle regression.

-Age matters

Players peak around 29. They probably won’t improve in their 30s, and the decline phase comes soon after. We can also expect improvement out of players who have yet to reach their physical peaks, although this doesn’t necessarily apply with defense. Marcel doesn’t use any fancy, academically researched aging curves, but it does use an aging factor – younger than 29, the player’s projection gets bumped up, and older, it slides down, and the magnitude is higher based on how far away the player is from 29. It’s not perfect, and certain body types and skillsets age differently, but the goal is to hit on the whole population, not on individual data points.

-Marcel doesn’t take a close look

Marcel misses a lot of the little things that we can use to predict breakouts or attrition. It doesn’t know injury histories. It doesn’t know about BABIP or K% or HR/FB or any of the other stats that can be used to explain anomalous years. It doesn’t know anything about the minor leagues. It doesn’t know anything about the lineups surrounding a player or the defense surrounding a pitcher. It doesn’t adjust for parks. It can’t come anywhere near projecting playing time. These are the things that the various projection systems should attempt to solve, and what we hope to accomplish with our fan projections.