Of course this is the bullpen thread and I've gone off piste. Suffice it to say my opinion is despite the importance of bullpen usage in the current evolution of the game, I believe that reliever performance year-on-year is inherently less predictable than hitting, fielding and starting pitching and so therefore should not be the focus of spending. Yes, you have to be a bit lucky to get a couple of Ryan Brasier seasons or to spend $8.25mm on the open market for 2 years of a 38-year old setup man and suddenly realise you bought Koji Freakin' Uehara, but these things seem to happen with relievers fairly often as do the sudden implosions of Eric Gagnes and Zach Brittons (projecting here - he hasn't imploded yet but merely declined).
edit: typo
That’s all very true from a cursory look at the data and it’s interesting to go further in thinking about what the strategy should be.
First, variation in reliever outcomes breaks down into variation in heath, variation in execution, and variation in small samples. You can’t control the first or third, but you can control the second. That is why we can observe that closers are much less volatile than middle relievers. This suggests that underlying talent is still the best way to limit the overall variation in relief pitcher outcomes and the key commodity you should purchase. Hence, the trend toward paying for a “second” closer (or, if you’re on the Yankees budget, 4 closers).
The other aspect of the “let’s wait and fix it later” approach to pens also can be analyzed by breaking down those three sources of relief pitcher variation. By “fix it later” teams seem to imply that within season variation is lower than cross season variation. Is that true?
If you know someone is healthy on July 31st trade deadline does that mean he’s a better bet to still be healthy in October than the guy you signed the previous winter? Depending on your assumptions of the injury distribution I guess.
Small sample size will create more volatility in a deadline acquisition than over a full season. That could go for you or against you, but in this decomposition you can’t predict it. And, the “success stories” that were really just good luck combined with confirmation biases that cause teams to attribute success to their brilliance and failure to bad luck may be biasing teams toward the fix it later approach.
That brings us to “variation in execution.” Do we think there’s a real “hot hand” component to relievers distinct from small sample size? That’s essentially what the “fox it later because relievers are volatile” argument boils down to. We know “hot hand” doesn’t hold in basketball. Has there been a study in baseball relievers?