THE BOOK--Playing The Percentages In Baseball

I mention it in the replies, but for some reason I don’t talk about (or use) regressing toward the mean in the study itself.  For example, for the players with banner years (first part of the study), if you regress the 3 year weighed or unweighted average around 25%, you get a much better projection.

There is a lot of work to be done in terms of discernings “patterns” or “red flags” in a player’s career path that might be helpful in terms of projecting his performance.  As I have said many times, a similarity type of system (like Pecota) might pick that up assuming that it gets that granular (for example, what other similar players have had 2 improvement years in a row in BB rate and are age 27)?  Again, I suspect that if you try and get that granular in such a system you run into severe sample size issues, and if you don’t, you don’t get to pick up patterns that MIGHT have some peredictive value.