Dynasty Rankings Update: August 24

I’m pleased with how the first release of the RotoGraphs Dynasty Model went, both in the results it produced as well as the reception here on the site. I’m excited for its future use as well, because the methodology is flexible meaning there is no reason it couldn’t be expanded to reflect a deeper league or a different settings scheme. For now, though, I’m satisfied with the 12-team, 5×5 proof of concept that this run provided.

As mentioned in the release article, I wanted to take some time to highlight both what I think the model does well in addition to some blind spots. Broadly, I’m quite satisfied with how it turned out but there are a couple areas that I think warrant a bit of caution.

I’ve also provided a fresh run (as of Thursday, August 20). Long-term, I will probably settle on more of a monthly cadence, but there were a couple oversights pointed out in the comments which were easy to correct and re-release.

Likes: Overall Results

First of all, I think the model produces results that are basically reasonable. This might sound reductive, given that it wouldn’t hit the press if it didn’t, but I think it’s valuable to still point out why. When considering player value in a dynasty or keeper context, you have to weight age (and related, prospect status) and current performance differently than you would in a redraft setting. Here is the breakdown of the Top 50 ranks by age, contrasted with the top 50 aggregated ADP per FantasyPros, which I’m considering as a reasonable proxy for player value in a redraft context.

In the dynasty rankings, you will notice only seven players in their 30s. Other than Ohtani and Judge, they are all ranked in the back half of that bucket:

There is a much older skew, with 16 players in their 30s of the ADP top 50. The distribution is consistent even with the top 100 ranks for each list:

There are nearly twice as many players in their 30s (39 vs. 20) in the top 100 ADP ranks than they are in the model’s ranks. This helps show that the model systematically prefers younger players, and isn’t just playing musical chairs within the same general rankings bucket.

Circling back to prospects, you also want a situation where they are valued more than they would be in a strict redraft, but not dramatically so. To this end, I looked at where a few top prospects were being selected in drafts and compared them against where the model has similar players from the midseason cycle:

This is admittedly cherry picked somewhat, but I think the rankings do a nice job of raising the value of prospects without discounting current performance too much.

Likes: (Relatively) quick adjustments on players

Another thing that I wanted to be sure the model was good at is not being stuck in the mud when it came to players whose values are (or should be) changing rapidly throughout the season. This is where using the year-to-date stats in the blend is useful. I made a few tweaks to the code to run the model on a pre-season only basis, and here are some noteworthy fallers:

It doesn’t take too much detective work to figure out why; each of these players has significantly underperformed their preseason projections to the point that rest-of-season projections are depressed.

Meanwhile, the list of significant risers is primarily attributable to the flip side of that dynamic:

One of the things I think will add value in the model rankings over time is an ability to depict arrow-up / arrow-down players. Rank in a vacuum is helpful but misses something significant for a guy like Garrett Crochet. Absent context of this season to date, an owner rostering a 27-year-old pitcher ranked 34^th would be thrilled. However, the fact that he’s fallen 25-odd spots is an extremely useful signal that his ranking is tenuous. Were his season to officially end due to his shoulder injury, the model would strike him significantly down to 125^th.

Speaking of, I also like that the model handles injuries in a manner that is consistent player-to-player. My methodology and weightings might need to be fine-tuned over time, but injuries are one of the hardest aspects of player valuations to get correct. Relying too much on gut feel and intuition while making a ranking list will inevitably lead to biases creeping into the model. With a systematic methodology, biases still might find their way into the ranks, but they’ll be easier to diagnose and mitigate over time.

Dislikes: Over-ranking of young players with middling projections

My only real gripe with the results is how much it tends to juice the values of young players without major prospect pedigrees whose rest-of-season projected values are in the low single-digits. Case in point:

Nothing against these players, but I’m not sure they belong around the top 100, let alone in the upper crust where Lile is intruding. But to be fair, each of these players does have something going for them.

Lile’s underlying batted ball data do not look stellar, but he’s nonetheless put of $11 of value this season, and is only one spot below Julio Rodriguez on the outfield leaderboard. As one commenter in the initial rankings post pointed out, he seems to be in a bit of a BABIP dry spell. If luck were to bounce his way a bit better over the latter quarter of the season, we might feel differently about him by October. Wilson is only one season removed from starting an All-Star game and has 5.1 WAR to his name in just under 1000 career plate appearances. This is clearly a very good real life player who may yet have latent fantasy value lurking below the surface as he ages into his peak years.

As someone who doesn’t have Antonacci anywhere, I was shocked to see the season he’s having: 121 wRC+, wOBA and xwOBA to match, good plate discipline, speed, and a bit of power. If the player authoring those stats’ name rhymed with Shmoner Shmiffen we’d be going nuts for him. The picture for Bolte is similar – bottom half of his org’s top 10 prospects going into the year, very slightly worse prospect ranking, and stats that are just a bit worse than Antonacci’s across the board.

I don’t think there is any action that needs to be taken – it’s possible that the model is picking up on something that we silly humans aren’t or can’t. As such, I’m happy to take a wait-and-see approach.

Observations from the comments: Back-end catchers

While it’s a good general rule to avoid the comment section, our readers are generally pretty astute and pointed out a few quirks in the model results. I replied back to a few and wanted to discuss some of those here as well because I think they point to key ways in which the model works.

One reader mentioned that catchers like Alejandro Kirk (459) and Salvador Perez (475) ended up on the list while neither Francisco Alvarez nor Endy Rodriguez cracked the ranks (though in the latest run, Alvarez does sneak in at 494). This specific situation highlights a couple of key aspects in which the model behaves: injury reversion to pre-season projections, extra weight on rest-of-season projections relative to year-to-date performance, and clustered tiering. Kirk has been hurt most of the year, so the model primarily relies on his pre-season projections. Meanwhile, the robots still like Perez a good bit rest-of-season which is keeping him afloat in the ranks despite all signs that father time has finally caught him. Meanwhile, Alvarez has likewise missed about a quarter of the season so far, and does not have a particularly good track record of health and consistent playing time throughout his career. Rodriguez is already 26 and has only played well for 45 games at the MLB level.

Even though they differ by 35 spots in the ranks, the dollar value difference between Alvarez and Kirk (with Perez in between) was just $8, which translates to the model preferring Kirk by ~$1 at most in any given year. As the model goes further down the ranks, the tiers get bigger and it won’t make sense to seriously sweat differences in ordinal rank in the back quarter of the list.

Observations from the comments: Pitcher weirdness

One sharp reader clocked that a few SP-eligible relief pitchers seemed way too high. Turns out, they were right! While I think that having SP designation does afford these guys some extra value, it wasn’t enough to take priority in teasing out for the model. For these purposes, these arms should be treated as relievers and therefore discounted heavily. Doing so dropped most of them from the rankings.

Likewise, another reader spotted that Cole Ragans was not on the list at all. In my earlier drafts of the model run, both he and Brandon Woodruff were showing up in the top half of the lists despite what, in my opinion, are injuries which could ultimately prove catastrophic to their careers. Woodruff is down for another procedure on his shoulder, while Ragans is undergoing the third UCL repair of his short career. There was no good way to systematically account for these edge cases, so I manually changed their positions to RP, causing them both to fall off the list. I think this is defensible – sustained MLB success following a third TJ and second major anterior capsule repair surgery (with a scope in the interim) would be unprecedented. Additionally, the extent to which they may come back, if they do, may well be in relief. That will likely also be their eligibility status upon return in some leagues after missing the 2027 season as well. So this isn’t perfect, but I think their ranking is closer to NA than it is 150ish and I wanted that reflected accordingly.

On a similar note, one other commenter had a question about the model’s pessimism on Gerrit Cole relative to other noteworthy oft-injured arms like Jacob deGrom, Chris Sale, Blake Snell, and Drew Rasmussen. The answer here is straightforward: each player is simply expected to be worth more in the rest-of-season projections. In this run, Cole’s rank has improved to 153, but all the other arms are still ranked in the top 100. Sale, deGrom, and Rasmussen are all projected to be worth at least 2x what Cole is for the remainder of the year, and have generally been healthy and good for the past two seasons. Snell meanwhile is a more complicated case, but sits closer to his peak on the aging curve and projects better for the remainder of the year. Cole may well end up the better dynasty player for the remainder of their respective careers, but it’s hard to fault the model for preferring a younger player whose recent injuries have been more of the grab-bag oddball variety (and is superb when he does pitch) than a 35-year-old coming off of a major elbow procedure whose road back to success has been a bit bumpy at time.

Other observations: Sal Stewart

I don’t want to leave without mentioning the single thing which surprised me the most. Lurking in our ranks as he is in the dugout of this article’s art, Sal Stewart slots in at ninth. Not to say he sticks out like a sore thumb, but going into the season, he was being treated an order of magnitude differently by drafters than any of the other players surrounding him on the list.

I looked at his player card and saw good-not-great peripherals (0.347 wOBA matching a 0.348 xwOBA), acceptable plate discipline, and good counting stats. Surely, I thought, this was another case like those elaborated above of the model overhyping a young player who happens to be performing well.

Then I started poking around some other player pages. Consider Stewart’s career to date, and a mystery player’s performance over a nearly identical sample of plate appearances to begin his career:

Who is our mystery player? Hint: he’s a dead ringer for Stewart physically: 6’2”, 225 pounds compared to Stewart’s 6’1” and 224. He’s a right-right first-baseman, still playing, though in the twilight of his career.

…

…

…

Got your guy? If you guessed Paul Goldschmidt, you’d be correct. I’m not saying that Stewart will eventually win an MVP and probably make the Hall of Fame. Goldschmidt was clearly a better hitter to begin his career than Stewart has been. But! Goldschmidt also started two years later, debuting in his age-23 season compared to Stewart who broke in at 21 last year. Look, maybe Stewart is ranked too high. But I’d like the record to show, if Stewart ends up accepting a bid to Cooperstown in 2050, that the RotoGraphs Dynasty Model called it first.

August 24 Update:

Please note again that these results are calibrated as of Thursday, August 20.