|Advanced metrics: SPs to buy/sell|
|Theory and Strategy - Platinum|
|Written by Todd Zola|
|Wednesday, 11 July 2012 19:18|
Is Jon Lester the key to your season? Is it time to cut bait on Cliff Lee? And what in the world should you do with Tim Lincecum? The answers to these questions and more can be learned from understanding how to apply some of the "newfangled" analysis developed by the sabermetric community.
A few weeks ago, we examined some of the more common pitching metrics and elucidated what's in a pitcher's control and what's left to fate. To review briefly, a pitcher's strikeout and walk rates best identify a pitcher's skill. Home runs and hits allowed entail both skill and luck elements. With home runs, the skill involves inducing more ground balls, while the luck deals with the percentage of fly balls that leave the yard. When examining hits on balls in play, ground ball pitchers surrender more "safeties" than fly ball pitchers. Meanwhile, the ability to consistently and repeatedly induce weaker contact is being investigated and cannot yet be conclusively deemed a skill.
The disconnect between a pitcher's skills and outcome, either good or bad, can result in an ERA that is not indicative of how the hurler truly performed. If he exhibits the same skills going forward, the outcome is apt to change. It is this particular line of thinking that is going to be put under the microscope and embellished via the review of some advanced pitching metrics. We'll conclude by putting theory into action and identifying pitchers to target and avoid when the games resume after the All-Star break.
Regression, a practical definition
In order to efficiently broach this subject, the connotation of regression must be clarified. The word is being bandied about by fantasy pundits in a rather cavalier manner, which has led to some misunderstandings and misapplications of the principle.
By definition, regression is the act of going back to a previous place or state. For what ensues, regression will specifically refer to those metrics that are out of a pitcher's control and normally hover near a global average.
It's important to note that regression can occur in either direction. The perception for some is only better-than-expected results regress in a negative manner. But the reality is worse-than-expected results also can regress, but they do so in a positive manner.
When considering regression, it's important not to fall victim to the "gambler's fallacy," or the notion that luck always evens out. Gambler's fallacy is the incorrect belief that if multiple consecutive coin flips come up heads, chances are the next will be tails. The fate of the next flip is independent of those previous. If a pitcher has been the victim of bad luck, one should not expect good luck going forward, but rather, neutral luck. The same is true if the pitcher has enjoyed good luck; bad luck isn't necessarily around the corner. As time progresses, if the luck is indeed neutral, the quantification of the metric approaches the league mean. This movement is regression.
Thinking about our coin-flipping example, while the odds for each successive flip are 50/50, over a large enough sample, tails should catch up. The problem is, if we relate this to the luck associated with pitching, a season is most often not long enough for luck to completely even out. In some cases, it does. In others, the luck (good or bad) continues. But in most cases, the effect is regression, where the metric moves toward, but never quite reaches, the league mean.
Which metrics regress?
The two metrics that are subject to regression are batting average on balls in play (BABIP) and HR/FB. Soon a third will be introduced, but for now let's focus on this pair.
The notion of BABIP regression is fairly commonplace. And for the purpose of this exercise, continuing to paint BABIP with a broad brush will suffice. Given that, in general, fly ball pitchers should sport a lower BABIP than ground ball types, all pitchers should cluster around the global BABIP average, presently .296. A fly ball guy might be a tick below, while a ground ball pitcher could be a smidge above.
However, it won't be long before this analysis is as archaic as Myspace. Baseball has evolved to the level that BABIP is being broken into its hit trajectory components of line drives, ground balls and fly balls, and each batted ball is classified as hard, medium and weak. Then corresponding BABIPs are determined. Infield popups can be broken out from fly balls, and bunts from grounders. Baseball Info Solutions, one of the leaders in data collection, has even created a classification called "fliner," which is in between a fly ball and liner. BABIP can be parsed into over 30 pieces. As more data is collected, a pitcher's skill will be further delineated from happenstance. It's the happenstance that regresses, but as we learn more, not everyone will be expected to regress to the league mean, but rather their own baseline, much like hitters. Keep in mind these will still not stray too far from the league mean, but every little bit helps.
HR/9 rate is a function of the number of fly balls a pitcher induces and the percentage of flies that leave the yard (HR/FB), with HR/FB metric subject to regression. The global HR/FB is usually about 10 percent. Presently, it sits at 11.2 percent. Time will tell if the league average will in fact regress, but individual pitchers who are significantly above or below that mark (after correcting for their home park) should regress toward the league norm. Again,should being the key word there. Some HR/FB rates will continue to be lucky or unlucky, some will in fact reverse, while most will regress to the league mean but never quite reach it.
The next level: LOB percentage
Left-on-base percentage, LOB% for short, has been developed to help differentiate a pitcher's effectiveness from luck. LOB% measures the percentage of allowed baserunners who do not score. The formula is (H+BB+HBP-R)/(H+BB+HBP-1.4xHR). LOB% is a bit like BABIP for hitters, in that better pitchers can sustain a better LOB%, though it will not typically stray too far from the league average, which usually sits between 71 and 73 percent. The top starter's LOB% can be as high as 78 percent, while the poorer ones can dip below 70 percent.
More baserunners do not automatically result in a worse LOB%, and the reverse it also true (fewer baserunners, better LOB%). Two pitchers can have the same LOB% but have different ERAs by allowing differing numbers of men on base. However, since the LOB% formula does not differentiate the type of hit, the more extra-base hits the pitcher allows, the worse his LOB% will be. However, more strikeouts can yield a higher LOB%, as minimal events leading to a run occur during a punchout. Ground ball pitchers can sport a higher than average LOB% since they are more likely to induce a double play, while fly ball pitchers tend to have a slightly lower than average LOB% since they allow more extra-base hits and sacrifice flies. Finally, LOB% formula does not distinguish between runs charged to the pitcher after he has been lifted. Since better pitchers are usually supported by better relievers and better relievers allow fewer inherited runners to score, the better starters can have their LOB% aided by a strong bullpen.
The best utility of LOB% is as another quick eyeball test, along with BABIP, to see if a pitcher has been helped by Lady Luck or he has been snakebit. A lower than expected LOB% leads to an artificially inflated ERA, while a higher than expected LOB% renders an ERA below what should be expected.
Pitchers with a low LOB% make for nice buy-low targets, as their LOB% going forward can be expected to normalize, resulting in a better ERA. As the same time, be careful when acquiring pitchers with a high LOB%, as their ERA should rise as their LOB% regresses. Just like with BABIP and HR/FB, the season is not enough of a sample for every LOB% to actually reach its baseline, but most will regress toward it.
Expected ERA: FIP, xFIP and tERA
Be it due to BABIP, HR/FB or LOB%, all analysis ends with "an ERA higher or lower than expected," so it's no surprise that there are a variety of formulas that aim to generate an expected ERA based on the elements most in a pitcher's control: his skills. What follows is a thumbnail sketch of three such entities.
The most elementary determination of an expected ERA is the brainchild of Tom Tango and is called FIP, short for Fielding Independent Pitching ERA. The formula is ((HR*13+(BB+HBP-IBB)x3-Kx2)/IP) + X, where X is a league-specific factor in the neighborhood of 3.2. All that is incorporated is HR, K, BB, HBP and IP. The exclusion of hits is because they are a fielding-dependent metric. The idea is that all pitchers should give up about the same number of hits relative to innings pitched, so only homers, walks and whiffs influence ERA.
Dave Studeman refined FIP to eliminate the luck and introduced xFIP, which normalizes the home run factor according to the following algorithm: xFIP = HRx13xFB%x(HR/FB)+(BB+HBP-IBB)x3-Kx2)/IP) + X. FB% is that of the pitcher in question, while HR/FB is the league average. xFIP includes the element of homers allowed that is within the pitcher's control, the number of fly balls allowed and assumes the league mean HR/FB for everyone.
There are some analysts who prefer FIP, though most feel xFIP is better since it predicts future ERA a bit more accurately. Those who favor FIP take issue with assuming all pitchers regress toward the league mean HR/FB. They may be right, but FIP ignores that ground ball pitchers serve up fewer homers than their fly ball brethren.
The final form of expected ERA that we will review is tERA, short for true ERA. The formula extends beyond the scope of this discussion. The Cliff’s Notes version is that some components of batted-ball data are within the pitcher's control, such as hit trajectory. The tERA category assigns an expected out and run value to the various batted-ball types plus strikeouts and walks. The end result is a true ERA, or tERA based on each pitcher's distribution of these events. Adjustments are made for park factors and the run-scoring environment of each season. However, like FIP, there is no adjustment for luck with respect to home runs. Some consider this a shortcoming; others are not so sure HR/FB should be regressed as much as was once believed.
Comparing tERA to FIP and xFIP, tERA and xFIP are slightly better indicators of future ERA than FIP, while all three are superior to ERA itself. That is, a pitcher is more likely to sport an ERA akin to the expected ERAs than the actual ERA turned in this season. The actionable fantasy application is given the standard sample-size alert that there might not be ample time for everything to flesh out, a pitcher's post All-Star break ERA is better approximated via his expected ERA than his first-half ERA.
Unfortunately, there is no correct answer to this inevitable question: Which expected ERA is best? You need to keep in mind the individual nuances, most notably the manner each treats homers. For what it's worth, Baseball Prospectus has a proprietary iteration called SIERA that deals with the home run issue. No doubt, there will be additional refinements as the collection of batted-ball data improves, primarily via electronic classification of batted-ball types.
We'll wrap up today's festivities with an examination of pitchers to target and avoid by focusing on their LOB% as well as expected ERAs.
Guys to avoid
Ryan Vogelsong, San Francisco Giants: On the surface, Vogelsong has built upon his somewhat remarkable career resurrection from last season, as his 2.36 ERA is even lower than his 2.71 mark in 2011. The problem is his already-pedestrian skills have taken a step back and do not come close to supporting such a sparkling outcome. His K/9 rate has declined from 7.0 to 6.3, though his BB/9 has improved a tick, going from 3.1 last season to its present 2.9. Lady Luck has shined on Vogelsong, as evidenced by a .250 BABIP and an LOB% of 84.2 percent. The low BABIP means fewer men on base and the high LOB% means a lower than average number of allowed baserunners has scored, which explains his sparkling ERA. However, a 3.72 FIP, 4.47 xFIP and 3.60 tERA scream that a correction is coming. While one can argue that an ERA in the high 3s is not horrible, if you acquire Vogelsong on the hopes that he'll significantly lower your present ERA mark, chances are you'll be disappointed.
Ryan Dempster, Chicago Cubs: How does a pitcher drop his ERA from 4.80 to 1.99? In Dempster's case, it's not because he has developed a new skill set, it's simply dumb luck. To his credit, fewer walks have helped, as he's sporting a career-best 2.30 BB/9 rate, but better control can only partly explain a .241 BABIP, 7.0 percent HR/FB and 84.2 percent LOB%. His 3.20 FIP, 3.70 xFIP and 3.97 tERA portend a significantly higher ERA in the second half. Shrewd owners will try to sell you on the belief that Dempster will be dealt to a contender, where his wins will flourish, but buyer beware that his ERA should rise.
Jose Quintana, Chicago White Sox: The former Yankees farmhand has been a godsend for his owners, but here's a word to the wise: Sell! Quintana's spotless control has certainly contributed to his success, as good things will happen when you carry a 1.6 BB/9 rate. But not this good, especially considering his lower than average 5.8 K/9 rate. His .261 BABIP, 5.8 percent HR/FB and 83.3 percent LOB% are prime candidates for regression. His 3.03 FIP, 3.72 xFIP and 3.83 tERA suggest his current 2.04 ERA is largely smoke and mirrors. Ultimately, how much control Quintana actually has over his miniscule HR/FB rate will dictate the extent of his correction. If he can indeed limit HR/FB, his post-break ERA should be closer to his FIP than xFIP, but considering U.S. Cellular Field is a launching pad, I'm not optimistic.
Jarrod Parker, Oakland Athletics: The rookie has certainly impressed thus far, checking in with a 2.86 ERA at the break. Parker's pedigree suggested he'd be good, and he is. Just not quite this good, at least not yet. His 7.1 K/9 is league average, but a 4.3 BB/9 is worse than the norm. Parker has been buoyed by a .260 BABIP and 5.1 percent HR/FB. The low HR/FB is especially relevant, as Parker is a fly ball pitcher (though the spacious O.co Coliseum certainly helps keep the ball in the yard). That said, a HR/FB of 5.1% is likely unsustainable, so expect Parker's ERA to approach his 3.62 FIP, 4.36 xFIP and 4.59 tERA. The more his homers correct, the closer he will come to his XFIP, as opposed to his FIP.
Ryan Cook, Athletics: I'm going to inject a little poetic license here and dissect a reliever. Normally this sort of investigation is not as useful for relievers since their paucity of innings can lead to a sample size anomaly that clouds the analysis. However, Cook's numbers are so extreme, his inclusion is warranted, especially since fantasy enthusiasts love to chase saves. Cook's 9.2 K/9 rate is closer-worthy, but it isn't eye-popping, whereas his 4.93 BB/9 rate is quite worrisome. Simply put, his .149 BABIP is not sustainable, and the zero in the home-run-allowed column will change. Cook's present 1.41 ERA is a mirage. His still-low 2.83 FIP and 2.35 tERA are a reflection of allowing no homers. Cook's post-break ERA is more likely to tend more toward his 4.27 xFIP. If a competitor is asking me to include a closer in a deal, I'll gift wrap Cook and put a green and yellow bow on top.
Guys to target
Jon Lester, Boston Red Sox: Lester's whiffs are down this season, as he's sporting a rather disappointing 7.5 K/9. However, after fanning only 48 in 77 2/3 April and May innings (5.6 K/9), the southpaw has ratcheted that up, punching out 46 in his 44 2/3 innings in June and July (9.3 K/9). His BB/9 rate is a career-low 2.4. Lester's unsightly 4.49 ERA heading into the break is a result of an uncharacteristically low 66.9 percent LOB% and slightly high .326 BABIP. Lester's 4.52 tERA is not particularly optimistic, but his 3.57 FIP and 3.60 xFIP suggest we'll see the guy we are used to after the break. If Lester can come close to sustaining his June and July K/9 while continuing to limit the free passes, he has an excellent chance to better his FIP and xFIP. In fact, this analyst is taking the under.
Adam Wainwright, St. Louis Cardinals: Coming into this season, many pundits expected Wainwright to improve as the season progressed and he shook off the rust from his Tommy John procedure. Well, not only is that possible, he is also due a correction from some first-half misfortune. Wainwright's 8.6 K/9 is actually better than his presurgery rate, while his 2.5 BB/9 is where it usually is. However, his .333 BABIP, 13.9 percent HR/FB and 67.7 percent LOB% are all out of whack. Wainwright's present 4.56 ERA is significantly higher than his 3.33 FIP, 3.08 xFIP and 4.12 tERA say it should be. Personally, considering Wainwright's home park should suppress homers and he's a ground ball pitcher on top of that, I expect his post-break ERA to most resemble his xFIP, which corrects for the first-half gopheritis.
Alex Cobb, Tampa Bay Rays: Cobb is intriguing, as his rotation spot seems a little insecure, especially if you focus on his 4.89 ERA. But Cobb has pitched better than his ERA indicates, as evidenced by a 3.40 FIP, 3.56 xFIP and 3.80 tERA. The main culprit is an inexplicably low 60.2 percent LOB%. Granted, Cobb's pedestrian 6.3 K/9 does not help matters, but to be this much lower than league norm is simply bad luck, especially since Cobb has been quite fortunate with respect to homers allowed. If Cobb's owner is concerned that Wade Davis will return to the rotation or that Jeff Niemann will come back and reclaim his starting gig, do them a favor and make Cobb your problem. The odds are favorable that Cobb is part of the solution, not the problem.
Cliff Lee, Philadelphia Phillies: Admittedly, this one is almost too easy, but sometimes it helps when the numbers support intuition, and Lee has scuffled lately, so his owners might be getting even more antsy. Lee's peripherals are down a speck from last season, but a 9.1 K/9 and 1.9 BB/9 are still elite. A .330 BABIP and 71.6 percent LOB% have inflated Lee's ERA to an uncharacteristic 3.98. His 3.00 FIP, 3.06 xFIP and 3.24 tERA suggest you should hang in there, because better days are in store.
And then there's Tim Lincecum …
There is no bigger enigma than the Giants' Lincecum. I'll apologize in advance, as I don't have the definitive answer on what's wrong with him or what you should do with him. At the end of the day, though, you can forget the numbers; it comes down to risk and reward. If I am wallowing at the bottom of the pitching categories, perhaps due to Lincecum himself, I have no choice but to hope for a return of The Freak. As high as I am on Lester and Wainwright going forward, they can do only so much. In order to save my season, I need Lincecum to get it together. If I'm in the thick of things, I am avoiding the two-time Cy Young Award winner like the plague. Give me Cobb; I just want to stay the course. The problem is for those of you in between. Do you put your season on the shoulder of a guy who can take you from fifth to ninth as easily as he can vault you into first?
As implied, the numbers do not offer much guidance. A 4.7 BB/9 and an elevated fly ball percentage indicate some bad pitching. However, a .333 BABIP and 12.4 percent HR/FB also suggest at least some degree of bad luck. Historically, Lincecum carries a LOB% between 76 and 79 percent; this season, it's 59.2 percent. Granted, a 12.4 percent HR/FB contributes to this, but he is still sporting a 9.7 K/9, which should help LOB%. So not only is Lincecum allowing more runners, a greater percentage of those runners are scoring, hence his 6.42 ERA. Yet his 3.98 FIP, 3.84 xFIP and 4.70 tERA say it's not all bad pitching.
Usually you can reverse-engineer an outlying ERA, but in Lincecum's case, you run into too many contradictions. He has been both bad and unlucky. Unless I am truly desperate and have no choice but to hope that Lincecum returns to form, I'm not touching him.
Thus concludes today's discussion. As time permits, I will be glad to address follow-up questions in Conversation. Next week, we will shift our focus from the bump to the batter's box and use advanced metrics to pinpoint hitters to target and to avoid as you make your second-half run.