I’ve written before about how we do our fantasy baseball projections. Being that I’m neck-deep in projections for the 2013 fantasy baseball season, I thought I’d share a few lot more thoughts on our methodology.
At the core of our projections is the Marcel system, which was developed by Tom Tango, who in my opinion has done as much for the fantasy baseball community as anyone, including BIll James or Daniel Okrent. Marcels is shorthand for “Marcels the Monkey“, a nod that it’s a fantasy baseball projection method that is so simple a monkey could do it.
There are a dozen or more baseball projection systems out there, many offered for free, many behind a paywall, and one (PECOTA) made famous because Nate Silver, the genius behind it, took his skills to the presidential election. Marcels has stood strong against them all, yet it is far simpler than it’s more complicated and proprietary prognosticating peers.
Crackerjacks uses a modified version of Marcels and I’m going to walk you through a step-by-step of how we do it. The steps are weighting, regression, and aging. We add a 4th, which is conditions.
Weighting
Some systems use comparables as their foundation, but most use multiple years of back data whenever possible. Marcels adds weights to the data, which makes sense because it allows you to place more emphasis on the player’s most recent season, yet still take account for possible random variation.
For example, let’s consider a player who has a home run per plate appearance of .037 in his most recent season, .047 the year before, and .053 the year before that. Simply averaging these gives us a .046.
But in order to look toward the years ahead it’s becomes more accurate to place more weight on the most recent season and Marcels does that. The weights are 5/4/3, with 5 going to the most recent season.
You multiply the component figure by the weights, add them, then divide the total by the weight total. Continuing our example, the weighted sum would be (.037*5) + (.047*4) + (.053*3) = .532. The sum of the weights is 5+4+3 = 12, making our weighted average .532/12 = .044.
Multiply this figure with your projected number of plate appearances and there is your number for home runs hit. .044 X 575 PA = 25.3 HRs.
Our example was Carlos Gonzalez. You might remember that Cargo had a breakout in 2010 and hit 34 homers, but had just 22 last season. We’ve given each season a weight so you might assume that we’re saying Cargo will hit 25.3 home runs next season, but not so fast. We still have 3 more steps. Continue Reading…















