I haven't made much effort to promote this, but last fall I developed a new projection system. The occasion was John Burnson's book The Graphical Player, for which he commissioned the forecasts.
The reason I'm writing about it now is that David Appelman has been kind enough to include my projections along with some other better-known, better-established (and maybe just better) systems on FanGraphs. Thanks John and David!
Here's a quick overview of what MINER is, and what makes it different:
- It's a regression-based system, and it makes no effort to identify player types, or similar players. In fact, it's based almost entirely on top of Marcel.
- I make extensive use of batted-ball data, so that BABIP depends on the player's distribution of batted balls, and pitcher HR allowed is based on fly balls allowed. (Among other things.)
- I also make extensive use of level- and park-adjusted minor league data, including batted ball data.
- There is some effort to project playing time: in addition to basing projections on past playing time, I've docked players PAs and IPs if they are projected to perform well below replacement for their position, and the amount they are docked depends on just how bad they are projected to be. This is very approximate, especially since I am only using offensive numbers to determine their production relative to replacement.
- In The Graphical Player, there are three years of projections and three tiers--20th, 50th, and 80th percentile. FanGraphs is publishing just the one. For anything beyond the median 2008 forecasts, well...you'll have to buy the book.
For now, I've also published the spreadsheets as google docs. As David was including them on FanGraphs, an incredibly obvious bug popped out; when that is resolved, I'll publish Excel files that you can play with. In the meantime, you'll find the Google Docs here and here.