ZiPS Update: Three Year Projections!
FanGraphs now has Dan Szymborski’s Three Year ZiPS Projections on both the sortable projections pages and all of the player pages.
As Dan notes:
It’s the ZiPS you love/like/hate, now slightly less accurate! Predicting the future is foggy and the further you go, the thicker the fog gets. Every time ZiPS runs a projection, it provides a player’s rest-of-career projection, but until now, only the first-year projection has been made public on a systematic basis.
ZiPS is a non-parametric model, deriving aging curves from very large groups of similar players, so history is the main guide. After all, there’s no experimental data; it’d be nice to let Jose Altuve play out his career a million times in a million realities and see how he ages, but that’s currently impossible. Plus, the MLBPA probably would not be open to participating in this unending purgatory.
The three-year projections are start-of-season projections. There’s currently no mechanism to update future projections the same way the in-season projections are calculated. The year-to-year model that ZiPS uses is much more robust than the in-season model and I am not smart enough to have figured out an automatic workaround yet.
David Appelman is the creator of FanGraphs.
My word, those Juan Soto projections
By 2021 he is basically tied with Trout for best projection in baseball. Wow.
What’s even more amazing is that these are median projections. So, given the high uncertainty that comes with having less than a year of MLB plate appearances, that means that the odds of greater than 10 WAR in 2021 has got to be something absolutely crazy like 15% or 20%.
Wow is right.
Well ZiPS does love the biebs, so never say never! But Soto, damn!
And that’s with a penalty for defense. My word indeed!
Juan Soto is a hitting prodigy. I think sometimes we have a tendency to forget that he was 19 years old. The track record of players slugging .500+ and with an OBP near .400 is…Mel Ott? Joey Votto, the guy who suspect his career is going to resemble, was in…High-A, I think. But ZiPs has never seen a guy like this before except Mel Ott, and thinks he’s going to grow into that kind of power while keeping the OBP. I’m not so convinced he’s got 40+ homers in him regularly, but heck, even Joey Votto in left field is something to celebrate.
Its interesting how it projects Soto to become an MVP player with Acuña not improving on his rookie numbers at all(they are nearly identical and actually slightly worse OBP). I also don’t buy Soto defense to be twice as good as he was last season.
Dan’s the best.
I will take the under on all those Soto projections. I don’t deny he is a stud but we’re saying that a guy that was still in AA this time last year is going to hit 40-50 home runs a year as soon as next season?
Why is Vlad Jr projected for only 112 G in 2020 and 114 in 2021?
No DH in interleague games
(*ducks*)
Does ZiPS use height and weight? He may just be at a size where guys normally get injured.
Can you comment on some of the playing time projections? Is it just some weighted average or more sophisticated than that?
Looking at Aaron Judge for example – he had a somewhat freak injury (caused by a hit by pitch) last year and missed ~8 weeks, but otherwise has never been hurt in 2+ years. He’s projected for 128, 125, and 123 games over the next 3 years – is this just because of the weight given to his 112 games played last year?
I’d argue that there’s not really any reason to predict him to play fewer games than any other everyday player on the basis of one errant pitch (but at the same time understand that somehow incorporating into the model a measure of injury likelihood would be extremely difficult).
Aren’t the playing time figures part of the algorithm? My guess is that if you try to “beat the projection” you will be wrong as often as you are right.
On another note, Judge is definitely a player without historical comps. No one has ever been so big while also athletic. Guys like Jason Heyward (about 40 pounds lighter) and Adam Dunn (not a horrible athlete in his mid-20s) are not even that close.
The fundamental problem is historically figuring out the difference between freak injuries and serious ones. I expected that players would generally regress more towards being “mostly healthy,” but it isn’t as much as I hoped in history. Recentism does a good job establish long-term playing time baselines.
It’s something I’m hoping to deal with in the future and actively working on and my approach of ZiPS knowing the basics of injuries isn’t enough to overcome this issue. But for significant freak injuries, I’d definitely mentally add some playing time. I don’t have a model yet otherwise that I”m happy with – I make my own playing time projections for things like standings, but I’m quite strict at making sure my opinion doesn’t go into ZiPS except for the widest of things (like knowing X player is in fact playing Y position, like Ian Desmond in CF this year).
SCORE
I have been waiting for this one for a while. Thanks, everyone.
Surprised to see Albies has the best projection of any Brave for 2021, altho some of that seems to just be ZIPS projecting him to play more.
Apparently preller kidnapped Keston
Paddack projection is pessimistic
I agree. ZIPS has him with upside, but the mean projections are less enthralling than I had hoped.
Wait – are Zips mean or median estimates?
And, relatedly, are there any plans to publish variances or even distributions (maybe something like the histograms that McDongenhagen provided for the prospects)?
That would be really cool.
Also, this is already really cool. Thanks for the amazing work.
Weighted mean. I’m sometimes a bit lazy in this description because then I’ll have to explain what a weighted mean is.
I’m thinking long-term some kind of visual element.
Would it be possible to have wRC+, OPS+, ERA+, etc. on these? Without them, it’s next to impossible to get a read on some of the batting lines. For example, Garret Hampson looks like a stud with the bat, but putting Coors into context might show it’s only a 90+ line…I’m super appreciative of the 3yr projections in general, thank you!
YEEEESSSS!!!!!!!!!!!!!
Are there certain players not included in the model? I happen to check a few and noticed Pete(r) Alonso has a projection for this year but not for the next three.
Thanks, this is awesome!
Clearly, Peter Alfonso is projected to die at the end of the season.
Is there a way to show the cumulative WAR for all three seasons? Keep in mind I am stupid and lazy.
I got less lazy. Three year WAR totals: Trout (24.4); Lindor (21.1); Mookie (19.7); Ramirez (19.4); Soto (19.3). Then a drop to 16+ win players.
Poor Tim Tebow at the bottom of the list with -7.7 WAR
For the curious:
https://docs.google.com/spreadsheets/d/1YU2Eg-fetB-WxVLoLK1eq-WsdoTlptR7l-OMmxWY4bg/edit?usp=sharing