Decoding ZiPS: the 2013 and 2014 Starting Rotations


Credit: Rick Osentoski-USA TODAY Sports

New York Yankees’ broadcaster John Sterling is famous for having said “you can never predict baseball.” At the granular level, this is true enough. What Colorado Rockies fan before the 2013 season could have predicted that the team would be in first place on May 25, only to look up in the standings for the remainder of the season? Baseball, however, can be projected with reasonable accuracy when taking in the big picture. So that many Colorado Rockies’ fan in March 2013 could have, in fact, felt reasonably confident about projecting the team to finish anywhere but first place in the National League West.

Aside from deflated fan predictions, there are numerous projection systems currently tasked with estimating any number of useful data, such as each player’s production and value for the upcoming season based on age, past performance, home ball-park, historically similar players, and injury history. While most have been shown to produce quality estimations, my preferred system is Dan Szymborski’s (@DSzymborski) ZiPS. FanGraphs recently published the 2014 projections for the Colorado Rockies, and Hayden Kane has already raised interesting questions about the everyday lineup and the rotation given those forecasted numbers. But before I delve into the projections with my own analysis, I want to take a moment to look back to last year. What can we learn by looking, first, at the 2013 ZiPS projections in light of how individual players and the team actually performed? Based on that comparison, how should we calibrate our expectations for 2014? I’ll first examine the starting rotation and, in a follow-up post, I will scrutinize the everyday lineup.

Before we begin, a quick recap of what we know about ZiPS (see here for more). First, it predicts playing time only insofar as it accounts for a player’s injury history and the resulting time on the disabled list. It does not predict whether or not a player will make the team on opening day based on the estimated plate appearances or innings pitched. So the forecast for Troy Tulowitzki’s plate appearances can be taken as a forecast for playing time because we know he’ll make the team, whereas Eddie Butler’s predicted innings pitched and resulting production cannot. Second, ZiPS is not a sentient being—if it is, Dan Szymborski isn’t divulging that information to the public just yet—but I’ll refer to what ZiPS “knows” and “thinks” about certain players because the system relies on information to create its projections. It knows a lot, but not everything. Finally, Szymborski does not advise that one add up the value metric WAR for each individual player to identify a team projection. But because ZiPS projections are perhaps the most interesting thing happening in the Colorado Rockies’ universe right now, I’ll risk the threatened “karate chop” and do just that, but always with the caveat that it’s a far from perfect estimation of team performance.

FanGraphs published the 2013 projections on January 17. At the time, this was the projected rotation, with the addition of Tyler Chatwood.

Jorge De La Rosa704.974.651.0
Jhoulys Chacin1364.424.422.8
Drew Pomeranz1344.964.561.9
Juan Nicasio904.483.801.8
Jeff Francis1435.344.631.3
Tyler Chatwood1295.575.180.8

The first thing that pops out to me is the minimal number of predicted innings pitched for De La Rosa. The reason ZiPS thought this is because he only pitched 59 innings in 2011 before requiring Tommy John surgery, and he pitched just ten innings in 2012. The prediction made sense because ZiPS only knew that there were two years of minimal play, and it calibrated the estimation based on that information. It did not only project about a half-season of work from De La Rosa, but in any case ZiPS was not high on him for those 70 innings. I included Chatwood here because he ended up as one of the four pitchers to toss over 100 innings; I excluded Jon Garland, who started the year with the team but had not yet been obtained when the projections were published, because as far as I can find he had no 2013 ZiPS projection on account of not having a team in 2012. The club began the 2013 season with a rotation of De La Rosa, Chacin, Nicasio, Francis, and Garland, with Pomeranz and Chatwood in AAA Colorado Springs. Now let’s look at what the rotation actually did (min 65 IP):

De La Rosa1673.493.762.9

The 2013 Colorado Rockies rotation was a mixed bag of over-performers, under-performers, and, in the case of Pomeranz and his 21 big league innings, non-performers. In all, one of the primary take-aways from this is that the rotation as a whole excelled relative to what was reasonably expected. A differential of nearly three wins above replacement is equivalent to having an extra above-average major leaguer spread out among the production value of the starting rotation. Another item to note is that De La Rosa, Chacin, and to a lesser extent Nicasio, were exactly what the team needed in 2013. Namely, they provided a lot of high quality innings. De La Rosa showed himself to be fully recovered from his surgery and pitched almost one hundred more innings than ZiPS projected. Not only that, but he outperformed his rate statistics while doing so and was worth two more wins than ZiPS thought he would be. ZiPS projected Chacin to be the team’s best pitcher, and he was. He also outperformed his projected rate statistics and was worth a win and a half more than expected. Nicasio, as anyone who paid attention last year can attest, was inconsistent. While he outperformed his expected WAR, we have to remember that WAR is a cumulative statistic, and he only was able to do that because he also pitched more innings than projected. Chatwood, who began the year in AAA, was a pleasant surprise and outperformed his expected production by quite a bit. Finally, the less said about the let’s-cross-our-fingers-experiments with Garland, Francis, and—let’s not forget—Roy Oswalt, the better.

But there is more to say about the rotation as a whole. ZiPS, in fact, has quite a bit to say about what last year’s holdovers and this year’s newcomers will do in 2014. To wit:

De La Rosa934.554.641.3
Brett Anderson644.344.051.0
Jordan Lyles1655.184.901.0

The starting rotation is one area that, as has been noted, ZiPS expects quite a bit of regression. The over-performances of De La Rosa, Chacin, and Chatwood did not set new baselines for ZiPS, but were instead read as anomalies. The oft-cited reason for Chacin’s expected regression is that his six percent home-run-to-fly-ball ratio is very unlikely to be repeated. It was, in fact, one of the very best HR/FB pitcher season’s since Baseball Information Solutions began compiling batted ball data in 2002. Tyler Chatwood’s expected regression comes from the fact that he simply does not miss very many bats. With more contact comes more hits, and with more hits come more runs. De La Rosa’s projections appear to be simply regressed toward his career averages. I, and likely many fans of the Colorado Rockies, am mostly anticipating what Brett Anderson does in 2014. Given his injury history, this forecast is clearly the 2014 analog to De La Rosa’s conservative 2013 projection. We can be relatively sure that if he pitches, he will pitch well, and because we are familiar with what De La Rosa did in 2013, there is reason for optimism.

Moving down the depth chart, we have Nicasio and Lyles. While Nicasio provided enough innings to be a quality starter in 2013, he was inconsistent, and ZiPS is not as high on him in 2014 as it was in 2013. The newly acquired Lyles is projected to be better than replacement level but below average if given 165 major league innings. They’ll be competing for a spot in the opening day rotation, but we can also add more names to the mix: Christian Friedrich, Chad Bettis, and Tyler Matzek (and, preferred-deity forbid, Franklin Morales). It’s natural to interrogate these names and read Spring Training box-scores to identify who will be the “fifth starter.” However, that’s not necessarily how the competition should be interpreted.

Jeff Sullivan recently argued that the “five man rotation” does not, in fact, exist. He showed that if we erased every start from each team’s consensus five man rotation, every team on average gave 32 starts in 2013 to guys not in the original rotation. For context, the Colorado Rockies’ workhorse of 2013, Chacin, started 31 games. When injuries occur—and they will occur—the Rockies will call on one or more of the AAA or bullpen arms to fill the void while Jonathan Gray and Eddie Butler continue cutting their teeth in the minors. That is why this year’s stable of arms looks so much better, expected regression notwithstanding. It was only a matter of time before Francis and Garland fizzled out in 2013, whereas in 2014 the team can reasonably expect solid production from at least four hopeful starters. To be sure, ZiPS is not optimistic about any of these depth options over a long period of time, but they can contribute to making depth a strength for the organization rather than a weakness. It might begin by thinking of them not as those that failed to make the team, but ready arms that can fill a temporary void for the Colorado Rockies.

The 2013 starting pitchers outperformed the total ZiPS projections, while the 2014 forecast expects regression. However, the projections are still better than the 2013 estimates, and if only Brett Anderson out-performed his projection the team would be in decent shape. Aside from the projections, the biggest difference going into this year is that the team is gambling on youth rather than the fountain of youth. While I didn’t always agree with the method of attaining said youthful pitchers (bye, Dexter Fowler!), it is a path worth pursuing—predictable or not.