2022 Projections Thread

[SRB]

The fact that there is never a more rigorous accounting of past performance (and this being a statistics-driven field) has always been a massive red flag to me about how accurate they actually are. Obviously lots of writers have a vested interest in trying to convince use they are useful, but who knows. I vaguely remember something from years ago suggesting that a simple three-year weighted average (is that what Marcel is/was, or am I misremembering?) was more accurate than most of the complex projection formulas.

 

Please define accurate.

 

If you are defining as accurate as being did x player get exactly within the stats? Then no it is not accurate and there really isn't an accurate system for this. If you are defining the predictions if they are within a margin of error then sure you could look at it that way. I think you would find that for the majority of the predictions they are within the margin of error or at least close enough for it to be negligible.

 

What other purpose could they have? Obviously nobody expects them to be exactly accurate, but my concern is that the margin of error is so large as to be of minimal utility compared to just looking at last season's states (or last three seasons) or subjective human guesswork.

[ClosetBrewerFan]

Basically just use the projections as a basis to judge the player on what they did and forget the accuracy on the actual projections.

 

Why not just use basic stats to get the basis to judge from if accuracy isn't relevant? I get that defining accuracy is hard, not to mention subjective. One of the articles I posted outlines the difficulty. Which begs the question. Why should we simply believe it's true if we haven't, or can't, measure it?

As far as I can tell there is a lot of articles explaining why some of these in-depth predictors are better, but nothing showing they actually are better. If one claim's it's more accurate than something else, then it stands to reason it was shown to actually be so. But I can't find anything showing it is. Again, maybe I missed it since I'm not as big on that sort of thing. If anyone can point me to something actually showing it's better that would be helpful. I'm not talking about explanations of why it should be. I've found plenty of that. I'm also appreciative of posters here explaining why they like them. But that is not the same as SHOWING its superiority. Given how long these projection systems have been around it seems like a major oversight not to show it really is better.

If these predictions were "accurate", I'm sure someone would be boasting about it. I would think someone like Stearns would expect to see proof if he was going to rely on a projection. And maybe the team's proprietary calculations are more accurate than what we are seeing for free. I get that there are a lot of variables that make determining accuracy "hard" but that's what these guys get paid for. You just pick a stat to use and then compare your projections to actual. Use WAR for example. Its not perfect but none of these are. One could easily calculate that a particular projection is on average +/- X% from actual. Accuracy is however you define it. Let's say it calculates out to +/-10% on average. That would mean if you project the player to have 3 WAR, your projection should fall within 2.7 to 3.3 WAR. I'd call that accurate enough for me. Now if it is actually +/-50% then I would not rely on those numbers all that much (1.5 to 4.5 is too wide a range for me). These mathmagicians can even calculate a confidence level on the accuracy.

 

All they can do is publish what their accuracy level is and then we could determine if that is satisfactory enough. Since they dont, and there seems to me no papers even discussing this, my guess the accuracy is closer to 50% than it is 10%.

[Eye Black]

The creator of the ATC projection system, Ariel Cohen, has published yearly performance comparisons among many of the major projection systems by using game theory (Link).

 

I usually put more stock in hitting projections than pitching projections. I’ve mentioned this in the past, but I am a big believer in Derek Carty’s The Bat X projections for hitters. The Bat X is the only projection system that currently uses Statcast batted ball data, and I trust that Carty has figured out at least a portion of the secret sauce necessary for accurate projections.

 

I do have an admittedly very much anecdotal experience from this past season. I played fantasy baseball for the first time in 15+ years by joining an NFBC Cutline league with 1,000 total participants (drafts occur among groups of 10 teams each) where you just draft a team preseason and leave it alone from there to accrue points throughout the season. I created an Excel spreadsheet for all hitters using The Bat X projections (converting each player’s projection into an expected cumulative season score using the Cutline scoring system) and one for all pitchers using ATC projections. I stuck pretty strictly to the projections for choosing players, especially hitters. I ended up finishing around #60 in the overall standings (out of 1,000), but my offensive scores were among the top 10 overall (my pitching suffered too many injuries/inconsistency and pulled my overall points total down). Maybe it was just a fluke, but after going through that process it seemed like trusting the projection system to make selections paid off. Obviously there is a massive difference between drafting using a projection system for fantasy baseball purposes, and finding real-life valuable baseball talent.

[Eye Black]

Cubs projections are up in ZiPS today. Below are the snapshots of the two NL Central teams that have been released so far…

 

 

 

 

The Reds ZiPS article is available today…

 

[monty57]

The Pirates are in obvious rebuild. The Reds put a lot of money into winning in 2020 and then Covid happened and now they're sitting with some big contracts and some talent, but not enough talent to be highly competitive and not enough payroll room to add much help.

 

The Cubs are just confusing. They should probably go into rebuild after last season's losses, and trading Contreras makes a ton of sense. But then they signed Stroman. Are they really going to try to be competitive with the mess they have, or are they just making a show to keep the fans interested and unaware that they're really rebuilding? Pretending to be competitive could cost them the opportunity to trade their biggest chip (Contreras) and could keep them in the middle of the draft so they lessen their chance of getting star talent out of the draft. But, it does keep people in the seats and money flowing to ownership.

[sveumrules]

Yeah, even after signing Stroman the Cubs would still pretty much need to ink one of Correa / Bryant / Story / Castellanos plus maybe like Carlos Rodon & Kenley Jansen to reinforce the pitching staff if they really want to compete this season.

[Thurston Fluff]

The creator of the ATC projection system, Ariel Cohen, has published yearly performance comparisons among many of the major projection systems by using game theory (Link).

 

 

Thank you for the link. It's a bit like the one I posted in that both are from a fantasy perspective. What I really want to see is each projection system evaluated from a pure baseball perspective. That would include more than the usual scoring stats fantasy baseball looks at. But your link and mine both at least try to evaluate the various systems in some manner. I'd still love to see each system show its effectiveness in some objective way. They have all the numbers so they could easily do it if they wanted to.

[sveumrules]

The creator of the ATC projection system, Ariel Cohen, has published yearly performance comparisons among many of the major projection systems by using game theory (Link).

 

Thank you for the link. It's a bit like the one I posted in that both are from a fantasy perspective. What I really want to see is each projection system evaluated from a pure baseball perspective. That would include more than the usual scoring stats fantasy baseball looks at. But your link and mine both at least try to evaluate the various systems in some manner. I'd still love to see each system show its effectiveness in some objective way. They have all the numbers so they could easily do it if they wanted to.

 

Pretty sure this year was a little wider swing than normal with the Giants essentially being thee most under-projected team of all time, but most seasons I believe the average margin of error ends up around 7-8 team wins on a macro level.

 

On an individual player level probably the simplest way would be to focus on OPS+ for hitters / ERA+ for pitchers & then set a minimum PA/IP threshold to get a large enough sample.

 

Just using roundish numbers, it looks like in 2021 there were 362 batters (about 12 per team) with at least 200 PAs & 338 pitchers (about 11 per team) with at least 50 IPs, that would essentially capture all the "regular" players. I don't have time to dig in now, but as the offseason drags on I might be able to poke around a little bit & try to put something together.

[Thurston Fluff]

The creator of the ATC projection system, Ariel Cohen, has published yearly performance comparisons among many of the major projection systems by using game theory (Link).

 

Thank you for the link. It's a bit like the one I posted in that both are from a fantasy perspective. What I really want to see is each projection system evaluated from a pure baseball perspective. That would include more than the usual scoring stats fantasy baseball looks at. But your link and mine both at least try to evaluate the various systems in some manner. I'd still love to see each system show its effectiveness in some objective way. They have all the numbers so they could easily do it if they wanted to.

 

Pretty sure this year was a little wider swing than normal with the Giants essentially being thee most under-projected team of all time, but most seasons I believe the average margin of error ends up around 7-8 team wins on a macro level.

 

On an individual player level probably the simplest way would be to focus on OPS+ for hitters / ERA+ for pitchers & then set a minimum PA/IP threshold to get a large enough sample.

 

Just using roundish numbers, it looks like in 2021 there were 362 batters (about 12 per team) with at least 200 PAs & 338 pitchers (about 11 per team) with at least 50 IPs, that would essentially capture all the "regular" players. I don't have time to dig in now, but as the offseason drags on I might be able to poke around a little bit & try to put something together.

 

That would be cool. Given the lockout maybe this is the year to do it. At least it'd be something to talk about. On the other hand, it's also probably the worst year to do it because covid may have skewed the data. It should be something each projection system does on its own every season to prove its credibility. Don't know why they don't. Even less sure why some places don't show competitors results if their own is proven to be superior.

[ClosetBrewerFan]

I did a very simple analysis of ZiPS projections to see how accurate it is. I used the brewers 2019 data as i wanted to avoid 2020 and the impact that may have had on 2021 projections. I compared both WAR and OPS+ projections vs actuals. Overall it was not too bad. On average, the WAR deviated by 1.2 points and OPS+ by 21 points. If you assume the variance is centered in the middle, WAR was accurate, on average, by +/- 0.6 and OPS+ by +/- 10. That seems pretty reasonable to me for a projection. Note: This doesn't mean that there will not be huge misses by these projections. Shaw and Aguilar were huge disappointments that year. Yelich and Hiura overperformed their projections. But some of these were surprisingly dead on (Braun, Moose, Thames).

 

I will post the pitchers WAR and ERA+ projections to actuals for the Brewers as well. Keep in mind this was a very small sample so my averages may not hold up over a larger sample.

 

[pre]2019 Pre-Season ZiPS Projections (Source Fangraphs) 2019 Actuals Deviations

Player PA BA OBP SLG OPS+ ISO BABIP RC/27 Def WAR PA OPS+ WAR OPS+ WAR

C. Yelich 673 0.298 0.379 0.522 136 0.224 0.352 7.8 3 5.2 580 179 7.0 43 1.8

Lorenzo Cain 568 0.287 0.359 0.409 104 0.123 0.333 5.9 10 3.8 623 81 2.6 23 1.2

Yasmani Grandal 475 0.238 0.349 0.462 113 0.223 0.283 5.6 2 3.2 632 119 2.4 6 0.8

Travis Shaw 560 0.254 0.34 0.472 113 0.219 0.286 5.9 2 3.2 270 45 -1.5 68 4.7

Mike Moustakas 580 0.267 0.324 0.506 116 0.239 0.271 6 0 3.2 584 114 2.6 2 0.6

Jesus Aguilar 526 0.258 0.333 0.492 115 0.234 0.303 6 4 2.4 369 87 -0.4 28 2.8

Ryan Braun 440 0.269 0.332 0.470 110 0.201 0.303 5.8 0 1.5 508 116 1.7 6 0.2

Manny Pina 334 0.247 0.303 0.385 81 0.138 0.289 4.3 5 1.3 179 86 0.8 5 0.5

Eric Thames 427 0.229 0.333 0.485 114 0.256 0.296 5.8 -4 1.2 459 117 1.3 3 0.1

Keston Hiura 534 0.259 0.315 0.409 90 0.151 0.325 4.5 -2 1.0 348 138 1.9 48 0.9

Ben Gamel 530 0.267 0.331 0.413 96 0.146 0.329 5.1 -2 0.9 356 85 0.7 11 0.2

C. Spangenberg 461 0.246 0.306 0.406 87 0.16 0.338 4.6 -4 0.7 102 64 0.6 23 0.1

Orlando Arcia 553 0.247 0.294 0.360 73 0.113 0.303 3.8 2 0.6 546 64 -0.8 9 1.4

Hernan Perez 415 0.255 0.286 0.401 80 0.145 0.301 4.3 0 0.6 246 64 -0.5 16 1.1

Trent Grisham 487 0.200 0.308 0.316 67 0.116 0.275 3.4 -3 -0.9 183 90 0.6 23 1.5[/pre]

[ClosetBrewerFan]

Here are the 2019 Brewer Pitcher ZiPS projections vs. actuals. Average ERA+ deviation is 29 and average WAR is 1.5, so ERA+ is +/- 15 points and WAR is +/- 1.8 points on average. Note how bullish ZiPS was on Peralta and not on Woodruff. Much more variance than with the hitters and greater outliers as well. And of course Burnes was ridiculously bad that year. Makes sense considering the difficulty it is to predict how pitchers will perform.

 

[pre]2019 Pre-Season ZiPS Projections (Source Fangraphs) 2019 Actuals Deviations

Player TBF K/9 BB/9 HR/9 BABIP ERA+ ERA- FIP WAR TBF ERA+ WAR ERA+ WAR

Freddy Peralta 575 12.1 5.1 1.2 0.286 109 91 4.09 2.2 382 84 -0.8 25 3.0

Zach Davies 630 6.7 2.8 1.1 0.297 103 97 4.33 2.1 672 125 2.6 22 0.5

Corbin Burnes 581 8.7 3.3 1.1 0.293 108 92 4.11 2.1 235 51 -2.2 57 4.3

Jhoulys Chacin 724 7.3 3.6 1.1 0.284 98 102 4.55 2.0 403 77 -0.6 21 2.6

Gio Gonzalez 692 7.9 3.9 1.1 0.293 97 103 4.36 1.8 366 127 2.0 30 0.2

Josh Hader 301 15.3 4.3 1.2 0.281 149 67 3.07 1.6 289 170 2.7 21 1.1

Jimmy Nelson 532 7.9 3.4 1.2 0.299 99 101 4.46 1.5 105 65 -0.5 34 2.0

B. Woodruff 511 8.6 3.6 1.1 0.298 101 99 4.14 1.5 493 123 3.0 22 1.5

Chase Anderson 609 7.3 3.2 1.7 0.278 95 105 5.12 1.4 592 106 1.8 11 0.4

Alex Claudio 315 5.8 1.8 0.6 0.292 132 76 3.43 1.2 267 110 0.6 22 0.6

Junior Guerra 518 8.1 3.9 1.4 0.284 95 105 4.7 1.2 344 126 1.4 31 0.2

Jeremy Jeffress 280 9.2 3.5 0.7 0.291 136 73 3.35 1.2 225 89 -0.4 47 1.6

Adrian Houser 397 7.1 3.8 1.6 0.304 81 123 5.38 0.1 462 120 1.9 39 1.8[/pre]

[sveumrules]

Thanks for running those numbers, CBF. My expectation was around +\- 10 on OPS+ & a little wider variance for ERA+, so cool to see it shake out in that range.

[SRB]

Would have been nice to have that extra +7.5 WAR performance if Shaw and Aguilar had not collapsed that year I guess we technically got to the playoffs anyway, and anything could have happened then.

[Thurston Fluff]

Thanks for the effort, ClosetBrewerFan. They do seem to do about what one could reasonably expect them to. Now all we need is a place that projects using simple stats and see if the more advanced metrics do a truly better job. Anyone know if something like that exist? I guess we could just use the stats from last season and see if they are more or less the same then compare that to the projection systems.

[SRB]

Thanks for the effort, ClosetBrewerFan. They do seem to do about what one could reasonably expect them to. Now all we need is a place that projects using simple stats and see if the more advanced metrics do a truly better job. Anyone know if something like that exist? I guess we could just use the stats from last season and see if they are more or less the same then compare that to the projection systems.

 

Marcel was designed as one of the simplest projection systems possible to serve as a benchmark for other projection systems, I believe it just uses a weighted average of actual stats for he last three years (3x, 2x, 1x?). No OPS+, ERA+, WAR or other league-contingent projections though.

 

https://www.baseball-reference.com/teams/MIL/2019-projections.shtml

[Eye Black]

The Brewers are now the only remaining NL Central team yet to have their team ZiPS projections released…

 

[sveumrules]

Cardinals offense looks better than ours on a spreadsheet at the moment.

 

It also looked better than ours by about 6% last year over 162 games per wRC+, yet the Brewers scored 738 runs vs 706 for the Cardinals.

 

I guess that's why they play the actual games or whatever.

[Ron Robinsons Beard]

Cardinals offense looks better than ours on a spreadsheet at the moment.

 

It also looked better than ours by about 6% last year over 162 games per wRC+, yet the Brewers scored 738 runs vs 706 for the Cardinals.

 

I guess that's why they play the actual games or whatever.

 

I know Bader is good defensively, but I just don't see a ton of upside offensively with him. There is no way he's a 4.3 WAR player. That seems crazy.

[sveumrules]

I know Bader is good defensively, but I just don't see a ton of upside offensively with him. There is no way he's a 4.3 WAR player. That seems crazy.

 

It's a defense heavy profile, but Bader posted 3.4 WAR over 427 PAs in 2018 & 3.9 WAR over 401 PAs in 2021. If he stays healthy enough & can keep his OPS+ around 105-115, he's got a shot.

 

A guy that might be getting underprojected for the Cards could be Juan Yepez. Dominated AA/AAA/AFL & should be in line for some DH action if it makes it to the NL this year. Just seems like one of those random Cardinals that kind of comes out of nowhere to be a solid MLB player.

[DuWayne Steurer]

Cardinals offense looks better than ours on a spreadsheet at the moment.

 

It also looked better than ours by about 6% last year over 162 games per wRC+, yet the Brewers scored 738 runs vs 706 for the Cardinals.

 

I guess that's why they play the actual games or whatever.

 

I know Bader is good defensively, but I just don't see a ton of upside offensively with him. There is no way he's a 4.3 WAR player. That seems crazy.

 

he's definitely a 4.3 war player if he plays 140 games.

[sveumrules]

Red Sox ZiPS dropped today at FanGraphs with a couple players of note...

 

https://blogs.fangraphs.com/2022-zips-projections-boston-red-sox/

 

Schwarber is projected for a 133 OPS+. He'll probably be too expensive for the Brewers, but would fit nicely if the DH makes its way to the NL.

 

Recently traded SS prospect David Hamilton also clocks in with an interesting projection. 0.9 WAR with a 67 OPS+, 33/5 SB/CS, and +7 defense.

[Eye Black]

Here’s the graphic, the Brewers full ZiPS article is due out tomorrow.

 

[TURBO]

Small as our #6 right out of the gate?

 

*nah*

[brewerfan82]

Small as our #6 right out of the gate?

 

*nah*

I presume that's only because Ashby's already listed in the pen (which doesn't necessarily mean the projections have him at pen-only innings) and you would expect us to dip at least into the 7th SP option at some point, so he helps account for that.

[Eye Black]

Here’s the ZiPS article on FanGraphs…

 

2022 ZiPS Projections: Milwaukee Brewers