A Dive into the Astros’ Rotation Projections

The Astros don’t have many questions, but one obvious question is “how much will the rotation miss Verlander?” That is a key issue for projections of 2023 performance.

A recent Fangraphs article, “Spotlight on the Most- and Least- Improved Rotations,” (Jay Jaffe), provides an interesting (but perhaps alarming) perspective. The author makes a few adjustments to the Fangraphs depth chart projections to rank the most to least improvement starting rotations. The most improved rotation, Texas Rangers, has an increase of 8.6 WAR, and the least improved (No. 30) is the Houston Astros, with a decline of -8.2 WAR. This sounds bad, but it’s not disastrous. The projections still show the Astros’ rotation to be firmly above average, and, let’s face it, the Astros won so many games in 2022 that a decrease of 8 - 10 wins still leaves the Astros in the playoff driver's seat. Still, here are a couple of Astros-related conclusions from the article:

• Two division rivals, the Mariners and the Rangers are projected to have a rotation better than the Astros. The Astros’ starters’ ERA is projected at 3.85, compared to 3.76 and 3.78 for the Rangers and Mariners. It’s worth noting that this projection glosses over uncertainty regarding playing time. For instance, one can question whether the Rangers’ DeGrom can pitch 172 innings in 2023. Likewise, injuries have dogged the Astros’ Lance McCullers, Jr., and one can also question whether he will pitch more than 100 innings. So, pitching depth will be important. And the Rangers improved their depth during the off-season, and the Astros suffered a decrease in their rotation depth.
• Regression is not the Astros’ friend. As the article states: “on the regression front, note that Valdez and Javier, who combined for 7.5 WAR, are projected for just 5.6 in about 25 more innings.” When pitchers like Valdez and Javier enjoy their best seasons ever, it leads to good things—like a World Series championship. But there is a downside for 2023—-the superlative 2022 performances mean that the potential for regression is much larger.

“Regression to the mean” is the statistical tendency for performance over time to move in the direction of the mean or average. Almost every reasonable projection is based on past seasons’ performance and will include regression to the mean. The more sophisticated projection systems include their own “secret sauce” to address other factors. But you can see a basic projection that includes only regression and aging effect on the Baseball-Reference.com player pages. This is a baseline projection (called “Marcel”) that utilizes three years of data and gives three times and two times weight to the first and second year (respectively) performance, compared to the initial year. Looking at the Marcel projection, we can get a sense of the degree of regression. For Framber Valdez and Cristian Javier, regression accounts for the projection's 23% and 30% (respectively). For a player like Hunter Brown, whose career has relatively little actual data for the projection, regression accounts for 83% of the projection.

Eric Hartline-USA TODAY Sports

The WAR projection is scary (-8.2 wins), but remember that Fangraphs WAR for pitchers is based on Fielding Independent Pitching (FIP), which does not include the impact of defense. FIP is an important pitching metric because numerous studies have shown that FIP is better than this year’s ERA at projecting next year’s ERA. But if you are projecting a team’s record, it is not ideal to ignore defense. The Astros were known for their above-average defense last year, and as a result, most of the Astros pitchers had an ERA below their FIP. For example, the defensive component of Baseball-Reference’s pitcher valuation would suggest that Framber Valdez’s 2022 FIP should be reduced by 15% if you want to account for the defense behind him. Of course, there are questions about the impact of the Astros’ defense in 2023. For example, will eliminating the shift hurt groundball pitchers like Framber Valdez?

In addition, projections will not reflect any changes, mechanical or otherwise that a pitcher makes to improve his performance. For example, will Luis Garcia’s new delivery help or hurt his performance? Early returns seem to indicate the answer may be positive, but...it’s early.

Comparison to Stuff+

Projections are based on high-level outcomes, like ERA, FIP, or BABIP. Efforts to project future performance based on more granular measures like velocity, pitch type, pitching break, location, etc., have been less successful. Well, perhaps until now.

`Fangraphs recently added two new pitch modeling metrics to its leaderboard: PitchBot and Stuff+. Tango’s blog reviewed the new metrics and concluded that Stuff+ was a clear winner. According to Fangraphs, Stuff+ includes but is not limited to: release point, velocity, vertical and horizontal movement, and spin rate. Although Tango said he is normally skeptical of this approach, he found that Stuff+ has an impressive correlation with the succeeding year’s ERA. His test suggests that Stuff+ may be better than FIP at predicting the next year’s ERA. He said of Stuff+: “I don’t know what it’s doing, but it is allaying my skepticism somewhat.” Tango also helpfully sets out a factor to convert Stuff+ to an ERA equivalent—which I have used to develop an ERA based on Stuff+. [ The conversion: 7-(Stuff/30) ]

So, let’s compare how Stuff+ compares to ZIPS 2023 projections for the Astros rotation. Also for comparative purposes, the table includes the 2022 expected ERA (x-ERA).

2022 x- ERA / Stuff ERA / ZIPS ERA

Valdez 3.31 / 3.30 / 3.26

Javier 2.43 / 3.13 / 3.74

McCullers, Jr. 3.57 / 3.43 / 3.36

Garcia 3.59 / 3.57 / 4.09

Urquidy 4.56 / 3.50 / 4.04

Brown 2.96 / 3.27 / 3.78

I should note that the sample size for Brown is very small (less than 20 IP), which likely affects the reliability of the Stuff ERA and x-ERA for him. That said, Stuff+ provides a more favorable ERA projection for him, compared to ZIPS.

In addition, the ERA based on Stuff+ is much more favorable to Garcia and Urquidy than ZIPS. In particular, the Stuff based ERA for Urquidy is much better than I would have expected based on his FIP. Perhaps Urquidy’s stuff explains why he tends to outperform his x-ERA and FIP.

Stuff+ would suggest that Javier is the Astros’ best starting pitcher. For those of us who watched him in 2022, that’s not hard to believe. In any event, ZIPS projects a notably higher ERA than both x-ERA and Stuff+ for Javier.

All three of the methods (x-ERA, Stuff+, ZIPS) produce very similar results for Valdez. Although the result is good (3.3-ish ERA), this is significantly higher than Framber’s 2.82 ERA in 2022. Possibly defense accounts for some of that difference. But it’s also possible that he will experience regression from the 2022 result.

So far, I have not addressed the weak nature of pitching depth beyond the six players above. So, to complete the comparison, I calculated the Stuff-based ERA for Brian Bielak and Seth Martinez. The Stuff+ ERA for Bielak is 4.00, compared to a ZIPS projection of 4.08. The Stuff+ ERA for Martinez is 3.50, compared to the ZIPS projection of 3.86.

Any thoughts or questions?