THE BOOK--Playing The Percentages In Baseball

Saturday, June 14, 2008

How do you build a bullpen?

Mark Shapiro, the GM of the Indians, was chatting in the booth today during the CLE/SD game.  He seems like a real smart guy.  The Indians are a very good organization, so that is no surprise.  He was talking about bullpens, and the TV announcers asked him why bullpens were so unpredictable.  His answer was two-fold.  One, he said, that there might be a bounceback effect, whereby relievers who pitch well are overworked and may be less effective the next year, and vice versa for relievers who pitch poorly.  I did a study on that a while ago, which suggested that that is not the case, although, to be honest, I don’t think my conclusions were definitive.  Shapiro did not seem to sure about that either.

The second reason he opined, was right on the money.  He said that because relievers only pitch 70 IP at the most per year, that teams make a lot of mistakes in terms of evaluating them.  While he was on the right track, he also should have mentioned that when we compare bullpen performance from one year to the other, we are comparing two sets of performances both of which have a lot of noise.  By definition, that would create lots of unpredictability.

I have mentioned this concept before, and I think it is an important one:  There is a big difference between 550 IP (about how many IP relievers pitch per team per season) thrown by 10 or 15 guys (the number of relievers per team) and 550 IP thrown by one pitcher, in terms of the variance of performance.  A huge difference actually.  When people think of variance, they tend to focus on the total sample size rather than the sample sizes of each player, where the performance of each of those players is pretty much independent of one another.

Anyway, I am going to answer the question of, “How do you build a bullpen?” which seems to be such a difficult and illusive answer, at least according to baseball insiders.  It is not.  And of course, I am not really sure what they mean by “build a bullpen.”

1) Find halfway decent starting pitchers that you think are also suited or even better suited for relief, and turn them into relievers.

2) Stop rotating relievers in and out of the major leagues, based on short-term performance.  Keep evaluating your reliever projections, including your pitchers in the minors, using sabermetric techniques, and your scouting reports, and use them to determine who stays and who goes.

3) Sort your relievers according to their projections, and make sure that your better relievers get more high leverage situations and that your worse relievers get fewer high leverage and more low leverage situations.

4) Resist the temptation to use your better relievers when you are losing by 4 or 5 runs (or more), on order to “stay in the game” and resist the temptation to use your better relievers when you are up by 4 or 5 runs (or more) for fear of losing the game.  Use your worse relievers in low leverage situations, period.

5) Use you righties and lefties wisely.  In fact, each reliever should NOT have just one projection.  They should have one projection versus RHB and another versus LHB.  Use those to determine who comes in when.  Also, rather than having just an “ERA” type projection for each pitcher, classify pitchers according to high or low K, BB, HR, and GDP (basically G/F ratio).  Use those also to determine who comes in when, depending on the game situation.

6) Of course, us your ace in more IP per season, and use him in multiple innings when you can, and use him any time the leverage is high, regardless of whether it is a save situation or not.

7) Bring in relievers as early and as often as possible, especially when your non-ace starter is on the mound.

That is how you “build” (and use) a bullpen.  It is not that hard.