Can We Predict Decline Before Fastball Speed Drops?

Danny Salazar's effectiveness has declined in concert with his decline in velocity (via Erik Drost).

Danny Salazar’s effectiveness has declined in concert with his decline in velocity (via Erik Drost).

On average, pitchers begin losing velocity the minute they reach the majors. Age and injuries begin to degrade pitchers and some begin to become shells of their former selves. But some don’t. These pitchers are still productive as their fastball loses more and more zip. But not every fastball is the same. While we generally think of a pitcher’s velocity in terms of his average velocity, in each game a pitcher’s velocity on his fastball will vary by roughly 5-6 mph, and in some cases even more than that. So I wanted to investigate whether we a pitcher’s individual pitch values at different fastball speeds would help determine how a pitcher will age.

Some pitchers just can’t seem to pitch as well as their velocity declines. Tim Lincecum is one. His velocity has come off an average peak of over 94 mph in his rookie season to under 89 mph this season. His ERA has followed the decline. Here are his ERA/FIP values from 2008, which was his first full season, to 2014:

Tim Lincecum, Changes in Performance with Velocity
Season Avg MPH ERA FIP
2008 94.1 2.62 2.62
2009 92.4 2.48 2.34
2010 91.3 3.43 3.15
2011 92.3 2.74 3.17
2012 90.4 5.18 4.18
2013 90.2 4.37 3.74
2014 89.7 4.78 4.10

He was still productive when his velocity dropped in 2009. From 2011 to 2012, when his velocity dropped another two mph, he did see a huge drop in production.

On the other hand, Felix Hernandez has seen his fastball lose three mph, but he has been producing the same or better since getting to the majors. Here is his profile of these same three metrics since 2006, which was also his first full season:

Felix Hernandez, Changes in Performance with Velocity
Season Avg MPH ERA FIP
2006 95.2 4.52 3.91
2007 95.6 3.92 3.75
2008 94.6 3.45 3.80
2009 94.0 2.49 3.09
2010 94.1 2.27 3.04
2011 93.3 3.47 3.13
2012 92.1 3.06 2.84
2013 91.9 3.04 2.61
2014 92.1 3.03 2.47

Since 2010, Hernandez has lost two mph off his fastball, but is basically producing at the same level.

Carson Cistulli took a stab at trying to answer this question by using game outcomes and velocities to look at how Ubaldo Jimenez struggled with decreased velocity.

This method may have its virtues (subject to later possible study) in determining how a pitcher will perform at a decreased velocity. One issue that may arise, however, is that a pitcher might not transition slowly into his new velocity level. His velocity may take a major dive like Lincecum’s did from 2011 to 2012.

Another pitcher who had a major velocity drop this season is Danny Salazar. Last season he burst into the majors with a 96 mph fastball, and he struck out more than 30 percent of the batters he faced. This season his average fastball speed is down 2.5 mph, and both his strikeout rate (down five percentage points) and walk rate (up two percentage points) have gone in the wrong direction. His rate of home runs per nine innings has gone from 1.2 to 1.8. The combination has led to an ERA of 5.53, way up from his 3.12 of last season. The drop in performance was not lost on the Indians, who demoted him back to Triple-A Columbus on Friday after he couldn’t escape the fourth inning on Thursday.

Trying to predict the drop would have been impossible with game data, since he never averaged under 95 mph for any game in 2013. All his games this season though have averaged under 95 mph. Here is his velocity chart, per FanGraphs:


Those are averages, but just over 30 percent of his 2013 fastballs were slower than 95 mph. I wanted to know if I could use the per pitch run value on these fastballs at the lower end of the velocity spectrum to predict his belly flop of a performance in 2014. Were hitters able to make solid contact more easily with them? Could they be a sign of problems to come?

Obviously, it’s not that simple — there could be some sampling issues with the data. First, the fastball may be thrown slower, compared to other fastballs, to keep the hitter off balance. Also, a loss of speed may mean more break, which can just as beneficial as a change of pace. Furthermore, the pitcher may be looking to get a weak batted ball instead of a swing-and-miss. Besides the intent of the slower fastball, other known factors may be acting on the slower pitch. Some pitchers see their velocity decline over the course of a game (though a few, such as Justin Verlander, do increase) and the effects of the second and third time through a batting order may be over-weighting the decline. In the end, I kept the data simple and just used the run values.

The key hope behind using per pitch run values was to use a larger sample of data (every fastball) and having the ability to look at every pitch instead of games.

In a single at-bat, a pitcher could easily throw four fastballs, and each could be at a different speed. By looking at the individual pitch values, I hoped to get enough useable information. I grouped each pitcher’s fastball speeds into one mph blocks and got the pitchers’ run values for those speeds. Here are the average run values for each of the velocity intervals. A negative value is better for the pitcher:


No real surprise with the data: a faster fastball is more productive.

To get an idea of how run values correlate with ERA, FIP and xFIP, here are the correlation equations from run values to ERA and its two estimators.

  • ERA= 76*avgRV + 4.30 (r^2 of .45)
  • FIP = 49.5*avgRV + 4.23 (r^2 of .33)
  • xFIP = 28* avgRV + 4.18 (r^2 of .19)

Using the values in the above image, here are the estimate ERA/FIP/xFIP knowing the run value.

Estimated Rate Stats at Specific Velocities
Velocity (MPH) ERA FIP xFIP
>=82 and <83 4.21 4.17 4.15
>=83 and <84 4.20 4.17 4.14
>=84 and <85 4.18 4.15 4.14
>=85 and <86 4.18 4.15 4.13
>=86 and <87 4.16 4.14 4.13
>=87 and <88 4.13 4.12 4.12
>=88 and <89 4.10 4.10 4.11
>=89 and <90 4.10 4.10 4.11
>=90 and <91 4.08 4.09 4.10
>=91 and <92 4.00 4.04 4.07
>=92 and <93 3.90 3.97 4.03
>=93 and <94 3.90 3.97 4.03
>=94 and <95 3.77 3.89 3.99
>=95 and <96 3.66 3.81 3.94
>=96 and <97 3.47 3.69 3.87
>=97 and <98 3.38 3.63 3.84
>=98 3.23 3.53 3.79

Now that we’re armed with this data, let’s go back and look at Salazar. Here are his average 2013 run values at different speeds:


If you look hard enough, you see that as his velocity declines, his run values increase (which, again is bad). Could it have been possible to have an idea of his decline?

To find out, we want to see how the data correlates between seasons. The answer is, essentially, not very well, and this points to the theory being effectively moot. At the high end of the spectrum, where a pitcher has 500 fastballs in a one mph band, the data correlate with a r-squared of .10. That’s a small enough number that we can say that it basically doesn’t correlate at all. And this R-squared value is when the pitcher has the most number of pitches in a band. I was hoping the data at the low end of the spectrum could tell me something going forward, but no such luck.

Again, using Salazar as an example, he had approximately 300 pitches total less than 95 mph, and not all all of them were grouped into the one mph bands I used here, which splits the sample up even more.

So what does all the preceding work mean?

  1. Pitch values may not be the way to analyze individual pitches. The only decent other measure I can think of is swinging strike rate, though I am open to suggestions, and welcome you to share your thoughts on the matter in the comments section.
  2. To see if a pitcher may decline, we should treat velocity loss like BABIP — just use the overall values for now. A small amount of projection can be taken from the player’s past results, but basically the league average values dominate the prediction.*
  3. Maybe I could expand the data to three mph blocks, but the range may not detect a talent level decline. If the range were 90 to 93 mph, Lincecum’s average values from 2011 and 2012 would both be selected.
  4. We may not be able to determine which pitchers will be able to thrive at lower velocities, but while I am crying “Uncle!” for the moment, I am not ready to concede the idea entirely.

I expected to find some useful insight when I began using run values to look at the production loss as a pitcher sees his velocity decline, but did not. There simply does not seem to be enough data once you break up velocity into different mph bands to get a good idea of how a pitcher will decline. For now, we should just assume a league average decline until the pitcher’s other stats begin to stabilize at his new fastball velocity.

* To put some numbers behind how small the change would be, in 2013 Danny Salazar threw only 35 pitches in his current range of 93 mph to 94 mph. He would need 12 times as many pitches to just get near the 400 pitch value and a r^2 of of .1. At a .1 r^2 value, 10 percent of the player’s talent is his own and the rest is the the major league average. So using Salazar’s 35 pitches in 2013 would predict less than 1 percent of this 2014 value.

Jeff, one of the authors of the fantasy baseball guide,The Process, writes for RotoGraphs, The Hardball Times, Rotowire, Baseball America, and BaseballHQ. He has been nominated for two SABR Analytics Research Award for Contemporary Analysis and won it in 2013 in tandem with Bill Petti. He has won four FSWA Awards including on for his Mining the News series. He's won Tout Wars three times, LABR twice, and got his first NFBC Main Event win in 2021. Follow him on Twitter @jeffwzimmerman.
Newest Most Voted
Inline Feedbacks
View all comments
9 years ago

My hypothesis, based on no actual knowledge:

Divide pitchers into two cohorts – 1. Pitchers who rely on a moving fastball, 2. Pitchers who rely on location and have a relatively straight fastball

Cohort 2 will age better as fastball velocity declines, as they’ll still be able to hit their spots.

I’m too dumb to even know what I don’t know – Is this possible to measure using pitchfx?

Mr Baseball
9 years ago

The best aging pitcher with reduced velocity is one that can still locate a fastball and still generate movement. These guys exist too. It’s not one or the other, even if this is a special group.

Mr Baseball
9 years ago

Face it, the human brain still processes way more data than a spreadsheet. That is why we need scouts and will need them for a long time.

Ralph Malph
9 years ago
Reply to  Mr Baseball

A spreadsheet does not process data. It organizes data so that a human brain can process it.

Bruce Weigman
9 years ago
Reply to  Ralph Malph

A spreadsheet does indeed process data. It is a functional program with the ability to solve and execute complex equations and logical expressions.

Kevin R
9 years ago

This is just spitballing a suggestion, but could you think about using standard deviations and z scores of fastball velocity? Would standard deviations even be meaningful in terms of actual velocity differences? If Salazar’s standard deviation on fastball velocity is only 0.6mph or something like that, it may not be particularly useful. I think this concept is useful if the idea that a pitcher’s difference from his own velocity baseline is more important for performance than an absolute reduction. Idea spitballed.

Dan M
9 years ago

Might be interesting to look at all pitchers if possible to get a better sample size. I’m thinking something like this:

Rather than break down by a single pitcher and by MPH, maybe assign pitchers into groups based on “FIP on substandard fastball velocity”, defined by FIP on pitches one standard dev. (or a half of one) below their average fastball. For example, you have 100 pitchers who throw 93-96 or so, and they all have thrown 1000 fastballs one standard dev below their mean. Once you have the FIP on this subset of pitches, you can look at which pitchers were effective at reduced velocity (and of course, vice versa). Hopefully, this group shows similar measurable traits, and maybe we could get a little closer to an answer.

9 years ago

Look at BABIP as a component of functional outcome.
9 years ago

The Sport­wagon shed its hood ornament and the Vista-Cruiser got a revised front bumper because the parking lights were moved up to a spot between each pair of headlights. Power on the optional Cruiser engine was nudged up to 320 horsepower.

9 years ago

I think conceptually you are on to something, but the interactions are obviously highly complex. Pitch effectiveness is not just going to be based on its own intrinsic quality, but also a game-theoretic problem regarding a hitter’s expectations. Pitch value is probably a decent place to look, but looking at a pitcher’s fastball in isolation is not really going to capture what separates a pitcher that can survive a velocity decline from one who can’t.

For starters, you’d also want to look at the other pitches in his arsenal- how effective are they, both absolutely and relative to his fastball? If he compensates for a less effective fastball by throwing more junk, is that a viable strategy? Secondly, if you had sufficient data, can you chart effectiveness of the other pitches conditional on fastball velocity? e.g. Could you regress the pitch value of a pitcher’s slider on the average velocity of his fastball that inning, or of his most-recently thrown fastball, or (probably too hard to compile the data) his most recently thrown fastball to that particular hitter?