The Saber's Edge: Second-Half Pitching Predictions

Chris Archer is having a breakout season. He has a 2.74 ERA that is supported by his ERA estimators. This performance is a huge improvement from 2014 when his ERA and estimators were about 3.50. He is making the turnaround in the simplest of ways by striking out more batters (21.0 K% to 30.0 K%) and walking fewer (9.0 BB% to 6.0 BB%). So, for the second half of the season, should his owners believe more in his phenomenal first half or his previous season results? Maybe it is somewhere in between.

At this point in the season, pitchers can be completely different than they were last year. The reasons can be many, but most of the time it comes down to two causes, they are striking out or walking a different number of batters. On the other hand, a pitcher's previous achievements can't be ignored. The pitcher was previously great or horrible and he may just morph back into that pitcher. Here is a look at which numbers to trust more when looking to second-half pitcher performance -- the previous-season data or the first-half data.

Math Alert

For this study, I looked at starting pitchers from 2002 to 2014 who threw at least 120 innings in the previous season, at least 50 innings in the first half of the next and 30 innings in the second half. I limited the innings pitched in the second half to try to add some struggling pitchers. I was trying to eliminate the survivor bias. Struggling pitchers, no matter their true talent level, might get demoted or benched while pitchers exceeding expectations likely will get a longer leash.

For the stats, I wanted to stay with just the core pitcher talent measurements, strikeouts and walks. Additionally, I included my favorite overall pitcher talent stat, K%-BB%. For each of the stats, I compared the previous-season and the first-half data to the second-half data to get the R-squared (how well the values correlate). Here are the results (a value of 1.0 is perfect correlation):

Now, I will move to the next obvious step and look at combining both the previous-season and first-half values. The predictive results are quite impressive.

Previous Season plus 1st Half vs. 2nd Half R-squared:

K%: .97
BB%: .94
K%-BB%: .91

Now those are some numbers to write home about with 1.0 being perfect correlation. Here are the equations for the two values to use to calculate second half values:

K% = .38 * Prev K% + .62 * 1H K%

BB% = .44 * Prev BB% + .53 * 1H BB%

K%-BB% = .44 * Prev K%-BB% + .56 * 1H K%-BB%

While the first-half numbers have more weight, a person can create a decent projection on the fly by just averaging the first half and previous season values.

Math Alert Over

Conclusion for those who ignored the math section: First-half strikeout and walk values are more predictive of second-half values, but if the two values are combined, an extremely predictive value can be calculated for the second half of the season.

With the previous formulas and the starters' stats so far, here are the second-half projections. For K%-BB%, I included a value created from the formula and another by subtracting the projected K% and BB% values.

From these values, I will look at pitchers who are expected to see the largest change in their K%-BB% from the first half to the second half.

Chris Archer (24% 1H K%-BB%, 19% projected 2H K%-BB%)

With Archer, the biggest projected change isn't with his walks (~1 percent change), it is with his strikeouts coming down out of the stratosphere.  While is swinging-strike rate (SwStr%) has increased from 9 to 13 percent, a 13 percent SwgStr% should only support a 26 percent K% (27 percent projected). Archer's change hasn't come from his fastball, but his slider. In 2014, he threw his slider 29 percent of the time and his 2015 usage is up to 41 percent.  The pitch has a 21 percent swinging-strike rate and is in the strike zone (Zone%) 47 percent of the time. With the high Zone%, hitters can't just layoff the pitch for a called ball. I expect hitters to adjust a bit to Archer's heavy slider usage, and his number to regress a bit.

Phil Hughes (12% 1H K%-BB%, 15.5K% projected 2H K%-BB%)

Hughes's 2014 season will probably be his career best. The real Phil Hughes is now pitching.  Like Archer, Hughes is not experiencing a change in BB%, but in K%. His K% has dropped from 22 percent in 2014 to 14 percent this season. The main reason behind the drop in strikeouts is his 1.3 mph drop in fastball velocity. His 2015 average fastball velocity (90.8 mph) is most similar to his 2011 value (91.3 mph). His K% in 2011 and 2015 are 14 percent. While Phil's K% may tick up with some regression, I don't expect a ton of improvement unless he can regain some fastball velocity.

Garrett Richards (10% 1H K%-BB%, 13% projected 2H K%-BB%)

Richards broke out in 2014 until a knee injury ended his season. This season, the results just have not been the same. Like the previous two pitchers, his decline in strikeouts (28 to 23 percent) is the main reason his K%-BB% is down in the first half. I could see the K% jump as his swinging-strike rate is just up from 10.8 to 11.1 percent, but up nonetheless. In the meantime, his strikeout rate is down with hitters swinging-and-missing at his pitches.

Stephen Strasburg (16% 1H K%-BB%, 19% projected 2H K%-BB%)

The main issue with Strasburg this season is health. He has been on the disabled list twice. The drop in strikeouts is because he just isn't missing enough bats. His career swinging-strike rate is 11 percent while it is only 8 percent this season. I could see his strikeout rate bounce back up once/if he gets 100 percent healthy.

Also, Strasburg's 2015 season is a perfect example of why I like to use K% over K/9. In 2011, he had a 9.0 K/9 and a 27.0 K%. In 2015, his K/9 is up to 9.3, but his K% is only at 23 percent. He is not dominating as many hitters, and they are putting the ball in play against him. With more batters faced, he's able to face a few extra free swingers and eventually get a strikeout per inning.