That was what Wilton tried to do. Don’t remember Wilton? Well, at Baseball Prospectus, before there was PECOTA, there was Wilton, which tried to predict the unpredictable. Which hitter who hadn’t done much in his career was going to have the breakout year? Wilton guessed.
Which high-flying star was going to slump? Wilton guessed.
Projecting the outliers in a baseball season is like overbetting on inside straight draws. Heck, it’s like overbetting any card hand. Not only can’t you do it reliably, but every time you miss you dig yourself into a deeper hole, one from which you can never climb out, because to give this guy 10 extra homers you have to take them from someplace else. So don’t do it like Wilton did it.
A properly crafted set of player projections is a series of bets on the real talents of each player. You look at a hitter’s BABIP, his BA, his W/K ratio, his HR per fly ball rate, what scouts say about his growth potential, what you see in his bat speed, lots of other inputs, plus whatever meaning there is in his past performance, and you try to describe a typical year for a player of his talents.
There are many ways to do this. Some work mechanically off of formulas, while others work by hand the whole way. I start with a mechanical projection based on past performance, and then I adjust it to incorporate news and what I see and know that isn’t included in the original formula. These adjustments tip a player up or down, one way or the other based on his talents, changes to his ballpark, differences in his lineup, news about his personal life and just about anything you can know about him.
It isn’t like these things necessarily mean one thing or another, but that there is a whole world of cause and effect that happen outside the sphere of past performance. Capturing a piece of that, a small piece, can make a big difference.
Usually, when someone does something totally atypical, it’s because of some flukish thing. Chase Headley hit a boatload of homers, which was incredible, and 23 percent of his fly balls left the yard. Only once before had 10 percent of his fly balls gone yard, and just barely. Was that a predictable step in his development? Nope, but it is typical of where the outliers come from, each year. Going bold would mean,usually, making egregious mistakes all over the place to occasionally get lucky.
Of course, once Headley hits all those long flies the projector has to decide whether the fluke was that the balls left the yard, or that Headley had in some fundamental way changed his game. Last year, some of us projectors supported the idea that Headley had changed his game (the way Jose Bautista had a few years previous), while others judged that he’d gotten lucky. We’re not drones.
But the goal is to be righter on more players, not even more righter on one every once in a while.
The third year that Tom Tango ran his Forecasters Challenge, the three winners of the different contests were me, Nick Minnix and Mike Podhorzer (in the fourth contest the winner was Consensus, the averaging of all the others). Mike, Nick and I all makes a point of handcrafting our projections, which is our way of being “bold,” if you want to put it that way. The mechanical scoring systems didn’t perform as well in Tom’s contest, I’m pretty sure because they do not incorporate information that is hard to capture in a formula.
I’ll liken that to a card hand in which you have a great sense of what all the unexposed cards are, because of your poker experience. You put each player on a a particular hand, and if you’re right you know that your chances of drawing are enhanced.
Your poker wisdom tells you to bet aggressively because you have a read, a tell, something that gives you an edge, something that gives you more information than anyone else. This is how I feel when I’m crafting my projections.
Testing my mechanical projections against my adjusted projections, the mechanicals correlate better. The order of ranking is better than the hand-crafted projections. But the hand-crafted projections, Wiltonlike, get a closer closer fit for many players who are surprisingly good or bad. They also miss many of these by more than the mechanicals.
My difference from Wilton is that I nudge a player to hit a few more homers, or get caught stealing a few more times. The adjustment is there, but you’d have a hard time seeing it if you didn’t know. The fact is that broadly averaged past performance is a pretty good predictor of future performance, but we can beat it with savvy formulas and intelligent controlled adjustment. And that helps our fantasy game, because those little differences in the course of a fantasy season become large. Boldness has it’s place, but it isn’t in projections.