Macro Trends in Reliever Value

Back again to the bailiwick which is perhaps my least compelling at cocktail parties (and trust me, this is saying something): relief pitcher fantasy performance.

A while ago, I looked at whether or not drafting relievers was worthwhile and found, ultimately, that yes, it is worth devoting targeted draft capital to relief pitchers. But one of the outstanding questions from that analysis was whether the composition of reliever performance is changing over time, so that’s what I decided to look into next.

First, was it actually worth drafting relievers?

Before getting to the meat of this research, and because I don’t have a great place to tie this in (nor is it enough content for a standalone article), I want to briefly revisit the result of my targeted reliever recommendations from the article I linked above; with 99.4% of the regular season complete (probably more by the time you read this), I think we can reasonably leverage year-to-date statistics to see whether or not the machine-learning rules I found held up to 2026 use. In short, I think they did:

To refresh, the recommended cutoff in this tiering was a projected rank better than 140. This isn’t an absolute win (given the model we used would have predicted four hits at 68%) but the sample size here is small enough that I still feel pretty good. If you used this method to target a reliever in the middle rounds, you went from a 30% hit rate to a 50% hit rate, while still avoiding the two worst-performing relievers of the bunch.

For dart throws, the criteria were a projected WHIP better than 1.28 and at least one projected hold. Again, we don’t have an absolute win:

Out of 46 players in this pool, with a 6.8% predicted hit rate, we would have liked to see a third player hit the mark. But the two that hit really hit and we have one so close that it would round up (Morejon) and such a surfeit of arms in the general ballpark of a hit that I consider this recommendation to be successful. Seven players (15%) in this targeted sample ended up worth more than $8. Going into 2027, it will be interesting to see who from these groups stands out.

Research Questions and Data

Now that I’m done patting myself on the back, I’ll get to the meat of what I actually wanted to look at – reliever performance over time. I examined this through a combination of big-picture and small-picture lenses clustered into general themes. The first, which I’ll talk about in today’s article, is macro trends. Specifically:

• Is the number of impact relief pitching seasons changing over time?
• Has the proportion of repeat impact performers changed over time?

Next week, I’ll detail the other two – individual performance and predictions. For now, though, I’ll focus on the groundwork required to get to that point.

I pulled my data from historical FanGraphs leaderboards from 1985 to 2026 and, as has been the case in most of my recent work, generated my own dollar values for the players using a 12-team 5×5 roto, 23-man roster setup. The values might be slightly different from the Auction Calculator for two reasons: timing, and because as noted in previous articles, my code produces very slightly different (but still extremely highly correlated) results relative to the Auction Calculator.

I then needed to define how to measure the notion of an “impact” reliever season. After some experimenting, I settled on using the mean and standard deviation of dollar value of all seasons for relievers who registered at least one season of positive fantasy value (n=551 resulting in 4,272 eligible reliever-seasons). I wanted to cast a fairly wide net of players that could be remotely considered for a fantasy team, and absent prehistoric ADP data, I felt this was a decent enough proxy. Among this group, the average season was $7.56 with a standard deviation of $6.50. So, I used these to create two thresholds: above average (self-explanatory), and “elite,” i.e., $14.06 or one standard deviation above the mean.

Reasonable minds could quibble with these filtering mechanisms or definitions, but a quick glance at the Auction Calculator results for relievers this year indicates that $7 is approximately the value of the 20th-ish best reliever, right smack in the middle of the 36 relievers required to be rostered under the three-RP setting I used. Meanwhile $14 is right near the 11th/12th place threshold. So my average is in the middle of RP2s, and my “elite” mark is essentially the border for RP1. In my opinion these pass the smell test.

Macro Trends

At this point, I could start slicing the data to answer my questions. First up: Is the number of impact relief pitching seasons changing over time? Let’s see:

Because the dollar values are calibrated on a yearly basis, the year-to-year values can be compared apples-to-apples. A $15 season in 1990 is worth the same to a fantasy manager in 2026. The underlying shape of that production might be different, but the impact to your rosters isn’t. Given the slight slope in these trends, if you really squint, you might be able to say that over time, impact seasons are being gradually replaced by a greater number of above-average seasons. But these data are very noisy, and I would hesitate to attribute a causal mechanism here. To my eye, it looks like the typical number of good reliever seasons by these measures is pretty constant over time. This finding itself is interesting given the amount of ink spilled over how reliever usage is changing.

The next thing I looked at is the underlying composition of relievers in these groups, because sure, the number of good relievers might not be changing much over time, but if there is a greater churn in the actual personnel, it might still be hard to identify the good ones. I struggled to come up with a good way to measure this, but settled on the following: how many active relievers are “one-hit-wonders” versus relievers whose careers were fully enclosed by a different usage era?

I’ve defined one-hit wonders intuitively: of all relievers who’ve had an elite season, how many had only one? For active players I took data through 2026 but omitted the players who accrued their first valuable season this year. If they were included, they would unfairly look like one-hit wonders when they’ve never had a chance to accrue more good seasons.

For the comparison group, I took players whose first season was at least 1991 and whose last season was 2015. I chose this range because conceptually, I think it maps on to our understanding of pre-modern relief pitcher usage. Mechanically, I needed to go later than the earliest point in the data to ensure there was no bias in the results. I couldn’t just set the code to mark a player’s first season as the first in the data (1985) because that was every player’s first year in my data. Looking for players whose first year in my data is one year out (1986) could still bias the results because there could be players who were hurt in 1985 or had a cup of coffee in 1984 but spent 1985 in the minors. To be completely safe of this, I chose a date several years in advance. I chose my later bound (2015) because it is a point where relatively few still-active relief pitchers were playing. Anyway, here’s the wonder-wall results:

Turns out there was actually a higher percentage of one-hit wonders in the earlier data. Given the sample size here I wouldn’t make too much of this specific nuance, but it’s also worth noting that it doesn’t account for the fact that the one-hit wonder rate of active players is almost certainly biased to be higher than what the eventual “true” rate will be. A good young reliever whose first above-threshold season occurred in 2024 or 2025 might still reasonably be expected to turn another by the time their career ends but if for whatever reason they didn’t in 2026, they are counting as one-hit wonders in this framework.

All this points to a notion indicating that while teams may be more egalitarian about their save opportunities, making it harder to easily identify good fantasy relievers, the broader macro trends for relief pitcher value are not materially changing. There are just as many good relief pitcher seasons being accrued by a similarly concentrated pool of players. So, if we can’t use saves to shoot late-inning fish in a barrel, perhaps we should try a gill net.