The St. Louis Cardinals got a great performance this year from Joe Kelly. The right-hander put up a 2.69 ERA in 124 innings, including 15 starts. In his starts he had a 2.28 ERA, while he posted a 3.65 ERA in 37 innings in relief. Against the Pittsburgh Pirates, Kelly had a 2.53 ERA in 21.1 innings, with a 16:10 K/BB ratio. Today Kelly will go up against the Pirates in Game 3 of the NLDS.
The ERAs look great, but when you dig deeper into the advanced metrics, you’ll see that Joe Kelly is a regression candidate.
On the season, Kelly has a 4.19 xFIP, which is 1.5 runs above his ERA. Out of all pitchers in the majors this year with 120+ innings pitched, Kelly has the second biggest difference between his ERA and his xFIP.
Kelly has a dominant ERA, but he hasn’t been dominant on the mound. As a starter he has a 4.76 K/9 and a 3.52 BB/9 ratio. A 1.35 K/BB ratio doesn’t usually lead to a 2.28 ERA. First you’ve got the low strikeouts. That means you’re relying too much on your defense for outs, which gets into batting average per balls in play territory.
Normally a pitcher will post a BABIP of .290-.300, or another way of looking at it is that 29-30% of balls hit into play will fall in for hits. Joe Kelly had a .271 BABIP, which gets into luck range? Why can we chalk that up to luck? There are only two ways to argue that a low BABIP is legit, and both are false for Kelly.
1. He’s getting help from the team defense. This isn’t true, because the Cardinals as a whole have a .297 BABIP. If Kelly’s low BABIP was a result of the team, then the entire pitching staff would have the same low BABIP. By comparison, the Pirates have been employing some extreme defensive shifts, leading to a team BABIP of .285. So if a Pirates starter had a lower BABIP, you could argue that the team defense was playing an impact in that low number.
2. He is doing something to keep runners off base. This is a common argument, but it’s absurd. It assumes that a pitcher can control where the ball is going. But Kelly has never shown this ability. He had a .306 BABIP last year in the majors. He had a .322 BABIP in Triple-A. The only other time he’s been under .300 has been in high-A.
We can safely say that Kelly’s BABIP is luck, and it’s not something that is the result of the team or the player.
Then you’ve got his strand rate. He’s stranding 82% of batters this year as a starter. The average for starters is 70%. Once again, this is something that assumes the player has some advantage here. But Kelly has never been above 74.5% in his career (his time in high-A was the high point), and he was 73.7% last year. The 82% strand rate means he was getting lucky, and should have allowed more runners to score.
As a starter, Kelly actually had a 4.43 xFIP. People are looking at today’s matchup as a potential pitcher’s duel, but Kelly isn’t in Liriano’s league at all. He’s a number four starter or a number five in a good rotation, but he has lucked into numbers that make him look much better than he actually is. Going forward, he will give up hits at close to a 30% rate, rather than 27%. He will strand runners at closer to a 70-73% rate, rather than 82%. Combined that means he’ll be giving up more hits, and allowing more people to score, which means his 2.28 ERA as a starter probably won’t last.
This might not happen today. The thing about regression is that it doesn’t happen in one game. Regression just looks at what a player might do going forward. In this case, Kelly should be expected to be closer to a 4.43 ERA pitcher going forward, than a 2.28 ERA pitcher. This means that while the two pitchers today both have strong season numbers, only the Pirates will be throwing a top of the rotation pitcher in game three.