Something I’ve been asking myself this year (and the abyss, as I’ve mentioned it in posts) is whether FIP or xERA is more predictive, as xERA enters more into the…well, I would’ve said mainstream, but we’re a ways from that already, so let’s say the FanGraphs sphere. With xERA in FanGraphs, it’s more accessible than ever, and more available for my personal usage than ever, and yet I still tend to use FIP as a better indicator of a pitcher’s true performance level, not knowing whether it or xERA is the better predictor between the two.
I realized recently it would be fairly easy to check, so I checked, and while the dive I did was rather quick, and therefore not exactly comprehensive, it was helpful. From 2015, for which xERA is first available, through 2019, I took every qualifying pitcher’s xERA, ERA, and FIP, and ran some regressions between them. Specifically, I asked how well each of those three metrics predicted a pitcher’s ERA the next year, something that, because it required pitchers to hit the qualifying innings threshold two years in a row, limited the sample size to 126.
Here’s how the r-squared values came out:
Variable 1 | Variable 2 | R-Squared |
ERA | Nxt Yr ERA | 0.061569 |
xERA | Nxt Yr ERA | 0.174174 |
FIP | Nxt Yr ERA | 0.178954 |
Again, 126 isn’t the biggest sample size, but the early indicator is that FIP and xERA are comparable predictors of a pitcher’s ERA the next season, while ERA is noisy enough to not be its own best predictor (as we already knew, to an extent). Out of curiosity, I also compared FIP to next-year-FIP and xERA to next-year-xERA, reaching the following r-squareds:
Variable 1 | Variable 2 | R-Squared |
FIP | Nxt Yr FIP | 0.239168 |
xERA | Nxt Yr xERA | 0.26545 |
Again, so close that it’s hard to say, but it’s interesting that xERA at least might be more predictive of xERA than FIP is of FIP going across seasons. A deeper dive would probably involve expanding the innings thresholds or trying to compare across different time intervals than season-to-season, but those would bring with them their own flaws. For the moment, it’s probably safe to use either FIP or xERA, or to use the two in conjunction, and to look forward to getting more full-season data again at the end of September 2022, when we’re finally comparing consecutive 162-game seasons to one another again.