Building a bullpen in real life is difficult, but so is trying to put together a reliable one for fantasy purposes. Not only do you have to deal with guessing which relievers are going to be able to replicate or sustain their performances going forward, but unlike actual major league clubs, you need to make educated guesses about usage patterns and the overall quality of the teams in order to make better informed decisions.

Anyway, that’s the idea that most people seem to have. In reality, however, wins and save opportunities are nowhere nearly as closely related as many think they are. Saves are one of those stats that are considered somewhat irrelevant to sabermetrics and performance analysts, but there has still been some heavy lifting done in their service, mostly by fantasy-oriented analysts.

Prior to the 2008 season, Derek Carty ran a number of tests to find out if saves and save opportunities could be predicted, based on a variety of factors (including, but not limited to, team wins, runs scored, and runs allowed). What he found was intriguing, as none of those factors correlated with save opportunities in any significant way. It seems counterintuitive to say that the things that win ballgames-namely, preventing and scoring runs-and the wins themselves, have little to no bearing on the number of times that a team will find itself in a save situation, but the numbers don’t lie. Saves may weigh heavily in the minds of many teams, as well as in fantasy baseball, but it’s an arbitrary statistic that means far less in the grand scheme of things than runs scored or prevented-factors which actively contribute to winning or losing ballgames. It would be a different story if we had a team losing 120-130 games per season, since save opportunities only come when your team is ahead; if a team is in a position to win a game just 45 times all year, then you’d probably want to avoid that team’s closer, but that’s rarely the case.

Carty’s study came out before 2008, but the lesson holds true. Even though the Angels were the outlier who won 100 games and had 90 save opportunities (opportunities for middle relievers are also counted; just because the last pitcher gets credit doesn’t mean that the opportunity didn’t exist beforehand), the rankings fit together snugly:

Rank Team         W-L   SaveOpps
 1.  Angels     100-62    90
 2.  Mets        89-73    75
 3.  Cardinals   86-76    74
 4.  Brewers     90-72    71
 5.  Red Sox     95-67    70
 6.  Cubs        97-64    69
T7.  Rays        97-65    68
T7.  Mariners    61-101   68
T9.  Twins       88-75    66
T9.  Rangers     79-83    66

The first few teams on the list make it look as if I’m ignoring the data, as it’s made up exclusively of quality teams. But once you get down to Seattle, who had the same number of save opportunities as the American League’s World Series representative, things begin to look a little strange. Texas is another head scratcher, and the order from there seems completely arbitrary:

Rank Team        W-L   SaveOpps
T11. Astros     86-75    65
T11. Phillies   92-70    65
 13. Tigers     74-88    64
T14. D'backs    82-80    62
T14. Orioles    68-93    62
T16. Marlins    84-77    61
T16. Rockies    74-88    61
T18. Royals     75-87    60
T18. Giants     72-90    60
T18. Nationals  59-102   60
 21. Padres     63-99    59
T22. Blue Jays  86-76    58
T22. Dodgers    84-78    58
 24. Pirates    67-95    57
 25. Yankees    89-73    56
 26. Reds       74-88    55
 27. A's        75-86    53
T28. White Sox  89-74    52
T28. Indians    81-81    52
 30. Braves     72-90    49

Yes, there are more bad teams at the bottom end of this list than at the top, but what are clubs like the Yankees, Blue Jays, White Sox, Indians, and Dodgers doing in the bottom 10? There is some correlation between wins and save opportunities, but it’s very slight, and the season isn’t long enough for the teams on the list to end up being ranked in any relevant order.

While you can’t pick which teams are going to finish where in the save opportunity rankings, is it possible to determine which relievers are going to end up with the most opportunities? The short answer is again “no,” as the number of opportunities that a given team had did not necessarily affect their closer’s output. Jose Valverde was second in save opportunities with 51, but his team tied for 11th in overall opportunities. Brian Wilson was third with 47, but the Giants were tied for 18th. Francisco Cordero had 40 save opportunities (ninth) pitching for the team ranked 26th in that statistic. There are plenty of other examples floating around that show that your team’s save opportunities don’t necessarily translate to more saves or opportunities, and since the standings don’t affect save opportunities as much as we wish they would, predicting them becomes nearly impossible.

The best strategy is to focus on the best pitchers at the position, rather than trying to use their team’s projected standings or ability as a benchmark. With few exceptions, the pitchers with the most opportunities to close last year were the ones that consistently excel at what they do: Francisco Rodriguez, Jonathan Papelbon, Joakim Soria, Joe Nathan, Mariano Rivera… pitchers like this should be your targets, regardless of how their team is performing. If you can’t predict saves or save opportunities, you still have a leg up on all of your league mates who will continue to do so, and that counts for something.

A version of this story originally appeared on ESPN Insider Insider.

You need to be logged in to comment. Login or Subscribe
wouldn't it be a good idea to go for K-Rod above maybe any other for save opportunities? from what i can tell about Citi, its not going to be a hitter's ballpark, and with the Mets being a good team and having lots of close games, it would be ideal for fantasy.
One could make a similar argument about Heath Bell in Petco. The Pods aren't gonna win 95 games or blow anyone out, but when they DO have those opportunites, Bell is pretty good.
..the whole point to the article was that save opportunities can't be predicted accurately by looking at a specific team and seeing how good they or are how likely they are to play in close games. I especially don't see how a hitter's ballpark would be a good thing when targeting a closer.
Considering the somewhat random distribution of save opportunities, and the amount of saves that enter the league during the season, wouldn't it be better to just not draft for Saves highly and instead get a productive player? You aren't really drafting for ratios in your closers considering how greatly their innings are dwarfed by the performance of your starting pitchers.
I have to agree.. Is the delta between an elite closer and an OK closer really worth the premium it would cost to draft/buy one?
In general I agree with not drafting highly for saves; however, I have found that more and more owners are adopting the stratedgy of getting cheap closers late. When multiple owners take that approach, you can find yourself over paying for the higher risk closer. There are 17 pitchers that have 3 or more saves so far this season. Going back to the February 24th Player Forecast Manager, 10 of those were ranked in the top 15 for projected saves. Only 3 of the 17 (Morrow, Ziegler, and Rodney) were not projected to lead their team in saves. While you may get lucky and land three quality closers from the bottom third of the projected closers pool (such as Bell, Sherrill, and Francisco) those are also the most likely closers to lose their job. The three closers mentioned above who were not projected to lead their clubs in saves all replaced closers in the bottom third of the pool.
I wonder if adding Run differential and park factors to the above lists (and removing real win/loss record) would yield a pattern or some sort of grouping in the middle?
My initial thought is that run differential would tell us more than Runs Scored or Runs Allowed by themselves, but still probably less than wins. How would you involve park factors for something like this? Serious question.
Low-run environments may correlate with save opportunities. Ergo, are there indeed more save opportunities in pitchers' parks? Did Derek test for this?
Who is Derek Carty, and is his work either available or citeable? I would like to know all what he tested for, and how weak the correlations were. 'Statistically insignificant' and 'totally irrelevant' are not the same things. (As well I'd like to know just how evidentiary strong his findings were)
I like Carty and I respect his work so maybe he was just using a different data set or only using RA and RS in conjunction, but I did a study on the same thing and found a relationship between runs allowed and saves, though not runs scored. It makes sense that low scoring games would have more saves because there is a smaller distribution of runs. For that reason the ballpark can actually have an effect on saves as well.
That's a good question. Who is Derek Carty?
Richie, Derek Carty writes over at Here's his article on save opportunities: As far as the strength of his evidence, I'll leave that up to you to evaluate (I'm no mathematician).
Thanks, hammneggz. The article itself, leastwise, doesn't show his work. I don't feel like digging around the site to see if it is posted somewhere. What I see doesn't impress me much. Never mind that deductively I'm inclined to agree with him entirely. For fantasy purposes, 7th inning and most all 8th inning save opportunities of course ought to be discounted entirely. For roto auction leagues, if saves are 98% unpredictable, that just makes that 2% of knowledge all the more valuable, especially since it too is up in the air. The article confirms my opinion that it is just 2% that we're searching for, but provides little help with regard to where that 2% lies.
Why would you count all "save opportunities," including seventh and eighth inning ones? Those only get counted as opportunities if they get blown -- if Arredondo comes in w/ a one-run lead in the seventh and leaves with a one-run lead, then it doesn't get recorded as a save opp at all. It's not really a save opportunity because it can't possibly become a save.
It does get recorded as a hold which is essentially a save opportunity that did not result in a save.
Seeing the Cardinals and Mets near the top of the list seems a bit alarming to me: these are two teams that made a lot of late-inning pitching changes (the Cardinals because of a Tony La Russa fetish, and the Mets because of poor bullpen performance). They reliably turned a lead into more save opps than other teams, but the bulk of saves still end up going to the closer. Counting the LOOGY coming on in the 7th inning as a "save opportunity" seems like it would really muddy the waters here, since we all know he's not going to get the save, none of us drafted him, and some teams use such pitchers extensively, whereas others don't have them at all. I'm fairly convinced that conflating 7th-and-8th-inning save opportunities with 9th-inning save opportunities isn't the way to go here; very few 8th-inning guys (and no situational lefties) are actually drafted, so their performance might as well be the starter's for the purposes of predicting save opportunities. Do you have data for 9th-inning-save-opps by any chance?
I think I'm in almost complete agreement with Carty. His evidence isn't perfect, but it is good. Once you add that teams wins (and other performance indicators) are difficult to predict as well, I think it becomes pretty clear that predicting saves is unimportant to picking a closer. Here's my job advertisement for the position: WANTED: CLOSER. An ideal candidate has a history of strong baseball performance as a reliever including recognized fantasy skills: Ks, ERA and WHIP. Lack of competition from his own bullpen a plus. Closing experience not necessary, but candidates without the full support of their manager need not apply.
I don't understand why people pay top dollar for closers. Get a cheap closer at draft/auction, then pick up one of the 10-15 guys who become closers over the course of the season.
For competitive roto leagues, punting saves likely won't work. (for your neighborhood leagues, I'm sure it's a valid option) For challenge games, it's not an option at all if you're ambitious. Knowing what correlates with saves, however loosely, while your roto competitors have no idea, would be a very large competitive advantage.
I don't know if this helps "predict" saves, but I think the more a team is overshooting it's own pythagorean record, the more they will be playing (and winning) close games, while of course losing the big ones (leading to of course, overshooting their own pythag. record)
Great article, Marc. This bears out my thinking on closers. If you can get a top closer for good value in the draft, you do it, but you do not overpay and you do not discount the closers on poor teams. Joakim Soria is a perfect example.
I notice that the 2 worst teams in the league had more save opportunities than wins. I don't know if that means lousy pen or complete inability to win a game late or both.
Marc et al: Interesting take on things, and I'd love to see Derek Carty's actual work (someone mentioned the details aren't really available). Any idea how big a sample he used? Because I recalled doing just this analysis a few years ago and getting exactly the opposite result. I dug back through my files and found it - turns out it was based on the '98 MLB season (my how time gets away from us!). I found that for the 30 teams (df = 28), SvOpps correlated highly with team wins (r = .62, p << .005 one-sided, as this hypothesis would be). The slope was .41, meaning that for every extra win a team had in the standings, 4/10 of a save chance materialized. Surely the relationship isn't linear at the extremes, as the intercept was 28 (28 SvOpps for an 0-162 team of course is impossible). It's possible that this analysis should be done after aggregating many years together. As long as the save rule remained consistent and run environment was more or less constant, it seems like that should be a reasonable thing to do. It also would be interesting to put many of the parameters people above have expressed curiosity about (possibly Wins, RS-RA, PkEffect against SvOpps) into a multiple regression and see how the structure shakes out. Of course, the bottom line for me is that I try hard not to pay for saves in the auction (unless someone just seems grossly undervalued or I miscalculate and end up with units sitting there with no other home for them). The turnover for closers is so high that I've found you get better bang for your buck loading up on great setup men behind shaky closers, and watching one or two of them take over. Worst case, you end up with quietly solid pitching numbers for just a few units.
Just to nitpick, 28 SvOpps is possible, though improbable, for an 0-162, since all 28 opportunities could be blown in losses.