Projecting All of Brandon Finnegan from One Appearance

On Tuesday night, during Kansas City’s improbable 9-8 wild-card defeat of Oakland (box), Royals left-hander Brandon Finnegan recorded undoubtedly the highest-leverage innings of his very brief major-league career in the highest-leverage game of his club’s season. The results were impressive: 2.1 IP, 9 TBF, 3 K, 1 BB, 1 H, 1 R, 63 xFIP-. For those unfamiliar with Finnegan previously, his performance was surely a revelation. Even for those who possessed some intimate knowledge of his college career at TCU, the outcome was likely a minor surprise, too — if for no other reason than it’s rare for any draftee to contribute meaningfully to his organization’s parent club just a few months after having become a professional.

Between his seven regular-season and (now) single postseason appearance, Finnegan is among that frustrating class of pitcher for whom (a) there exists some manner of major-league data but also (b) not so much that the fielding-independent stats which most directly inform run prevention (strikeouts, walks, ground balls) have become reliable yet.

Notably, though, there’s a collection of what one might call intermediary fielding-independent numbers — that is, metrics which (a) inform the metrics which inform run prevention but also (b) become reliable more quickly than either strikeout or walk rate (which require 70 and 170 batters faced, respectively, according to work done by Russell Carleton). Specifically, I’m thinking of these intermediary fielding-independent numbers: fastball velocity (which is useful insofar as it becomes reliable almost immediately), swinging-strike rate (which is predictive of strikeout rate), and first-pitch-strike rate (which is predictive of walk rate). Precisely how much more quickly the latter two become reliable than the stats they inform, I’m unable to say. The object of this brief exercise, however, is less about Ultimate Precision and more about attempting to extract useful information from a limited sample.

Below is a table of all the relevant correlations among players seasons from 2012 through -14 for every pitcher who faced 170-plus batters. Note that FBv denotes average fastball velocity; S-S%, swinging-strike rate; F-S%, first-pitch-strike rate; and All, each of those first three metrics combined. Ind and Dep represent the independent and dependent variables, respectively.

Ind	Dep	R^2
FBv	K%	0.17
FBv	BB%	0.05
FBv	xFIP-	0.12
S-S%	K%	0.64
S-S%	BB%	0.01
S-S%	xFIP-	0.39
F-S%	K%	0.01
F-S%	BB%	0.35
F-S%	xFIP-	0.10
All	K%	0.66
All	BB%	0.38
All	xFIP-	0.49

The results probably aren’t surprising. Swinging strikes lead to strikeouts. First-pitch strikes lead to low walk rates (considerably better than merely throwing the ball in the zone, turns out). Velocity adds some precision with the advantage that it stabilizes quickly. Altogether, they’re about 50% predictive of league- and park-adjusted xFIP.

That established, let’s apply the results from the calculations above to Brandon Finnegan himself — and not even to Finnegan’s entire major-league body of work, but specifically the 2.1 innings he recorded against Oakland. Is it possible to extract something meaningful from that limited data?

Over those nine batters on Tuesday, Finnegan produced an average fastball velocity of 94.4 mph, a 13.8% swinging-strike rate, and 66.7% first-pitch strike rate. Using those figures in conjunction with the algorithms produced from the above regressions*, one gets the following results:

Outcome	K%	BB%	K	BB	xFIP-
Projected	29.5%	6.6%	2.7	0.6	72
Actual	33.1%	11.1%	3	1	63

*Which algorithms are available at the bottom of this post.

Using just the three inputs noted above (velocity, swinging-strike rate, and first-pitch strike rate), the model suggests that Finnegan would have recorded (with rounding) three strikeouts, one walk, and a 72 xFIP-. The actual results: three strikeouts, one walk, and a 63 xFIP-.

To get a sense of how quickly such numbers can stabilize, below is a similar experiment including Finnegan’s regular-season line, during which he faced 28 batters over 7.0 innings and recorded an average fastball velocity of 92.5 mph, a 14.1% swinging-strike rate, and 64.3% first-pitch-strike rate. Projected Post denotes the numbers one would expect Finnegan to have posted using just his single postseason appearance; Projected Reg denotes the numbers one would expect given his regular season appearances; and Actual Reg is what Finnegan actual posted for the Royals in September.

Outcome	K%	BB%	K	BB	xFIP-
Projected Post	29.2%	7.2%	8.2	2.0	76
Projected Reg	29.5%	6.6%	8.3	1.9	72
Actual Reg	35.7%	3.6%	10	1	38

Finnegan appears to have recorded a slightly higher strikeout and slightly lower walk rate during his regular-season sample than the model would have predicted — leading, that combination, to just a 38 xFIP-. Insofar as only three relievers since 2002 have transcended that figure in a single season — Craig Kimbrel in 2012 (23 xFIP-), Eric Gagne in 2003 (26 xFIP-), and Aroldis Chapman this season (32 xFIP-) — it’s probably reasonable to conclude that Finnegan benefited from some variance. The general outline of the projected figures is promising, nonetheless — and, notably, the larger regular-season sample leads to almost the same exact conclusions as the smaller, nine-batter postseason sample.

In terms of Brandon Finnegan, specifically, one important question about him is his future role. While he’s recorded the majority of his professional appearances as a reliever, that’s very likely a product of Kansas City attempting to limit the innings of a pitcher who not only played a full college season, but also dealt with shoulder stiffness during same. Given Finnegan’s success out of the bullpen and his ability to deal with opposite-handed batters with relative comfort, he’s likely to assume a spot in the rotation next season. Insofar as per-batter figures improve for almost any pitcher moving from a starting to relief role, the estimate for Finnegan here oughtn’t be used to make predictions about his career as a starter. With regard to his near future as a reliever, though, they’re likely pretty reasonable.

*****

For strikeout rate (K%):

(FBv * 0.0033 + S-S% * 1.8 + F-S% * 0.098) – 0.33

For walk rate (BB%):

(FBv * 0.00103 + S-S% * 0.089 – F-S% * 0.32) + 0.17

For league- and park-adjusted xFIP (xFIP-):

(FBv * -1.2 – S-S% * 372 – F-S% * 112) + 311