from Amazon
Monday, January 01, 2007
Forecasting Young Players
This article at Lookout Landing illustrates very nicely how Marcel can forecast a young player better than PECOTA and ZIPS.
In 2005, YuBet was .256/.296/.370 (in only 228 career PA). How to forecast 2006? The regression toward the mean equation is at the 200 PA level, which pretty much means Marcel would have forecast YuBet around half-way between his career performance and the MLB average. Plus a healthy age adjustment, because he’s on the steep ascension. Hence, a .270/.322/.407.
PECOTA on the other hand had him replicating his 2005, and ZIPS had him below his 2005. Why? Here’s the trick:
PAs count. Betancourt had 584 PA in 2006. That by itself means that it’s more likely he had a good season than not (though his stellar fielding may have forced the issue anyway, like Adam Everett). The short rule is: more PA, higher production. It’s as simple as that.
So, when you review a forecasting system at the end of the year, a dumb system, like Marcel, may turn out better, simply based on this rule! While a GM may prefer PECOTA or the others, it’s Marcel that may end up looking better.
However, I have some idea that PECOTA doesn’t handle the PA issue well. This is pretty clear in the forecasts for pitchers, and my guess is that the player sims don’t take PA into account. Because of that, YuBet’s career performance in 228 PA may have resulted in the same comp list if he has 1228 career PA. But, I don’t know.