Colin Wyers ran a fantastic piece on predicting ERA a few days ago. Read it here. The basic summary:
- As a predictive tool, ERA is bloody horrible.*
- FIP is better than ERA by almost half a run.
- tRA is slightly better than FIP.
- xFIP is slightly better than tRA.
And after consulting with Graham, he hinted that tRA* is slightly better than xFIP by about the same margin between tRA/FIP. I wanted to expand on this, partially because I think we all need a refresher on how poor ERA is, but also because we need to get an idea on the true talent levels of our pitchers to date.
The metrics listed are pretty self-explanatory. I've thrown in run average (RA) which is ERA, only with unearned runs tacked on. Believe it or not, RA is a better predictor of future performance than ERA. Here are some thoughts:
- Dan Wheeler is the closest thing the Rays have to Akinori Iwamura the pitcher. Not in terms of talent, but rather predictability. The four good metrics possess a standard deviation of 0.13, by far the lowest on staff. If you like consistency, he's your guy; if you like good performances, he's probably in the middle.
- Home run regression is the tale of Jeff Niemann/Andy Sonnanstine. xFIP projects Sonny to regress better, while tRA* essentially says it's a tie. You have to wonder how many teams, like Seattle, will attempt and buy low on Sonnanstine. Perhaps the real question is whether any teams will bite Niemann's apple.
- David Price is up and down. Owner of the highest deviation, Price is liked by the regressed metrics, albeit nowhere close to the RA/ERA level of infatuation.
- Joe Nelson hasn't turned out quite as planned. His RA is nearly an entire run higher than his ERA. The good metrics say he ranges from replacement level to "not that awful". tRA* shows the most forgiveness and regresses Nelson's career high HR/FB% by quite a bit.
- How to know you're good, part one: J.P. Howell's xFIP is 3.
- How to know you're good, part two: James Shields' xFIP is 4.
- None of the metrics really agree on how good or bad Balfour is. FIP loves him, xFIP cautions against the homer rate, tRA is meh, and tRA* is a slightly more enthusiastic meh.
A lot of people are probably reading and wondering why we'd turn to regressed models. All of these players belong to a common fraternity of major leaguers. David Price has fantastic stuff, but at the end of the day, he's still a major league starting pitcher. That means we should regress towards that, if only to be conservative in our estimations. That's why we can look at a player like Joe Nelson and say, "Well, he's been unlucky," without resorting to clichés or only anecdotal evidence.
When in doubt: regress, regress, regress.