The Relationship Between FIP and Exit Velocity

One of the great things about FIP, in my estimation, is the ease with which one can understand its value. If you’re watching a pitcher for your preferred team and ask yourself “What outcome would bother me the most right here?” a home run is the clear answer. A walk is second. A single, double, or triple isn’t ideal, of course. In the case of every ball in play, though, there’s at least some chance for the defense to make a play. The walk and home run don’t allow that. They are, almost uniformly, decisively negative.

Conversely, a strikeout is generally the best outcome for a pitcher*. A batter who strikes out create no opportunity for value.

*Outside of a double play, of course. That requires a runner at first and less than two outs, though, something that happens less than 20% of the time.

What FIP does is to take those three outcomes and transform them into a pitching stat that’s consistent from year to year and better predicts future ERA than ERA itself does. One thing for which FIP doesn’t account, though, is all of those balls that are hit into play.

Or maybe it does.

We know that a pitcher exerts a decent amount of control over the types of batted balls he concedes. He might be a ground-ball pitcher, a fly-ball pitcher or a mix of both. Newer data pushes us closer to the conclusion that a pitcher has some control over how hard a ball is hit, as well — although most of the control does appear to come from the batter.

Statcast has given us the ability to help reach those conclusions. The graph below comes from the work of Sean Dolinar and Jonah Epstein — you can play around with their tool here — and illustrates the degree to which a pitcher’s observed launch angle and exit velocity represents his true-talent launch angle and exit velocity.

As you can see, there’s more hope for arriving at something like “true-talent” launch angle. And this makes sense: as noted above, we talk frequently about “ground-ball” and “fly-ball” pitchers. Grounders and flies are expressions of a pitcher’s control over launch angle. The relationship between a pitcher and his exit velocity is a bit more speculative, though.

Yesterday, I discussed how there was a detectable relationship between those two variables even looking at one year compared to the next. We also have a relationship (as discussed yesterday) between exit velocity and FIP, even if there’s also a decent bit of noise in there.

To see how the relationship with FIP works, it might be helpful to break down the components of FIP. The chart below depicts the correlation coefficient between average exit velocity and HR/9, BB/9, and K/9 for 186 single-seasons from 2015 and 2016 for the 93 pitchers who recorded more than 100 innings in both years.

While it’s possible that there’s some sort of relationship between strikeouts, walks, and exit velocity, that relationship doesn’t easily present itself in the data above. Where there does seem to be some sort of relationship is in home runs. Now let’s take a look at three groups from 2016: those with a high average exit velocity, those with a low average exit velocity, and a large group in the middle.

While the relationship between exit velocity and both strikeout and walk numbers appears to offer some promise, it might be better to consider them more deeply on another day. Not only is the coefficient lower for both those variables than for home runs, but strikeouts and walks exert less of an overall effect over FIP than homers.

The potential relationship with homers and exit velocity makes sense given that home runs are hit the hardest. That said, consider this comparison between the average exit velocity for pitchers on all batted balls compared to all batted balls without home runs.

The relationship between average exit velocity on non-homers and exit velocity overall is very strong. (That top right dot is Danny Salazar’s 2016, FYI.) If we wanted to make a leap, it would sound something like this: because (a) there’s a relationship between exit velocity and home runs, which is included in FIP, (b) exit velocity without home runs has a near-perfect relationship with exit velocity with home runs, therefore (c) there’s likely a relationship between exit velocity on balls in play and FIP.

That leap would be pretty convenient, right? Maybe the reason FIP has worked out so well over the years isn’t due merely to accounting for the three variables over which a pitcher exerts direct control, but that, at some small level, it actually accounts a tiny bit for the balls in play that pitcher concedes, as well. Thus, if true, the home-run component of FIP serves as a proxy for batted balls.

Let’s not get too ahead of ourselves, though. While we know exit velocity generally has an impact on base hits and damage, a pitcher’s average exit velocity might not have the same impact. When we compare a pitcher’s ISO against with his exit velocity, we see some relationship (r=.42), but when we look at a pitcher’s ISO against while also removing home runs and then compare that to a pitcher’s exit velocity without home runs, the relationship gets weaker (r=.31, .33 if using exit velocity with home runs). The further we try to make connections, the weaker the relationship gets. For example, the relationship between ISO and FIP is strong (r=.74), but the relationships between FIP and SLG on balls in play (r=.19) or ISO on balls in play (r=.23) are nowhere close to as strong without the home-run component.

I don’t think we have discovered any secrets connecting FIP and batted balls in play, but that doesn’t mean it was fruitless. Last season, 15 pitchers saw an increase in exit velocity by more than 2 mph relative to 2015. On average, those pitchers experience an increase in their HR/9 by 0.35, more than the league average of 0.2.

In the first graph from this post, we saw more confidence in exit velocity after 80-100 balls in play, so it might be possible to identify pitchers likely to have rough seasons sooner in the season — as opposed to those we can hope for a healthy bounce back and return to form. Teams might be able to take a slightly more proactive approach in identifying issues and trying to correct them. We failed to solve the riddle of the batted ball, but we might have gotten closer to finding useful information from exit velocity.