Every offseason, GMs, fans, and sportswriters alike put their powers of prognostication to the test and try to predict the performance of a number of free agents. There are a multitude of ways to approach the problem, from the scientific (projection systems), to the informed scout ("He's a big guy and I've seen plenty of big guys break down in their early 30's"), to the crazy fanboy ("Coco Crisp will hit 30+ homers in Fenway", although to be fair that seems to have originated with a Yankees fan).
In this post, we'll try to predict the long term offensive performance of arguably the top 3 FA batter from this offseason by finding comparable players and looking at their career trends.
So how does this whole thing work?
Basically I'm finding the 10 most similar players to our free agent in question at his 2008 age. From there, I'm charting their performance over the rest of their careers using wOBA. The hope is that the shape of the performance can suggest something about how the free agent might perform.
Why use similarity scores and wOBA?
Of course projection systems like PECOTA, Chone, ZiPS and even Marcel are going to be a lot more accurate than simply taking the weighted average of the performance of the comparable players. But most projection systems are based on aggregate performance trends. While those are definitely going to improve the accuracy across the entire sample, they may miss an unexpeced fluctuation for an individual player that could be picked up by looking at his comps. Also, projection systems tend to be poor at predicting playing time (or don't even attempt to do so). Since one of our questions in evaluating a free agent is how much playing time we expect him to rack up, looking at comparable players may be a way to answer that question. Finally, very few of those systems project out more than a couple of seasons at a time.
Why not use similarity scores and wOBA?
First and foremost, sim scores and wOBA only look at the offensive side of the equation - and sim scores don't even really do that very well. I like to think my similarity scores are a little more accurate than the ones on Baseball Reference because I've park and era adjusted the values - but they're still really not much more than a junk stat. So why use them at all? Well, one of the first things fans do when trying to predict how a player is going to turn out is figure out who he reminds them of. Then they apply the doppelganger's career arc to the original player to get an idea of a possible shape to the career. Using sim scores does much the same thing, but with a little more stringency to the whole thing. As I mention above, it's clearly no replacement for a real projection, but I like to think it's an interesting alternative look at things.
Mark Teixiera
Entering this offseason, the 28 year old switch-hitting first baseman of the Angels was clearly the prize offensive player on the market. Coming off a season where he had a 151 OPS+ with excellent defense in 157 games, his future was the subject of a bidding war between a handful of teams, including the Yankees, Red Sox and Nationals. In the end, the Yankees won the right to pay him $180 million over 8 years to solidfy a position that has frankly been a weakness over the past 5 or 6 years (since Giambi turned into a statue).
What can the Yankees expect over the length of the contract? Dan Symborski used his ZiPS system to predict the following performance:
ZiPS Projection - Mark Teixeira
------------------------------------------------------------------------------
Year AB R H 2B 3B HR RBI BB SO SB BA OBP SLG OPS+
------------------------------------------------------------------------------
2009 545 101 159 37 1 28 115 83 103 2 .292 .392 .517 137
2010 552 99 165 37 1 31 114 90 104 1 .299 .402 .538 145
2011 536 95 157 34 1 29 110 90 99 0 .293 .399 .522 141
2012 537 92 154 34 1 26 102 90 102 0 .287 .394 .499 134
2013 533 87 148 34 1 24 97 87 104 0 .278 .384 .480 126
2014 540 87 150 34 1 24 95 88 106 0 .278 .384 .478 126
2015 529 82 144 33 1 21 88 84 103 0 .272 .377 .457 119
2016 529 80 143 32 1 20 84 83 102 0 .270 .375 .448 116
------------------------------------------------------------------------------
And what does our similarity score method suggest?
First, let's identify Teixeira's top 10 most comparable players at age 28.
Player | Sim Score |
Boog Powell | 966 |
Nate Colbert | 942 |
Greg Luzinski | 942 |
Eddie Murray | 939 |
Willie McCovey | 930 |
Carlos Delgado | 927 |
Jack Clark | 925 |
Vic Wertz | 923 |
Glenn Davis | 922 |
Kent Hrbek | 921 |
A few Hall of Famers and a lot of very good players with some great years. Now let's see how they performed in their careers after 28.
Age | Players | PA | PA/Player | wOBA |
28 | 10 | 5507 | 551 | 0.400 |
29 | 10 | 4796 | 480 | 0.377 |
30 | 10 | 4120 | 412 | 0.380 |
31 | 9 | 5142 | 571 | 0.394 |
32 | 9 | 4577 | 509 | 0.375 |
33 | 8 | 3698 | 462 | 0.372 |
34 | 7 | 2878 | 411 | 0.370 |
35 | 6 | 2807 | 468 | 0.358 |
36 | 5 | 2400 | 480 | 0.357 |
37 | 3 | 1234 | 411 | 0.356 |
38 | 3 | 757 | 252 | 0.300 |
39 | 2 | 1007 | 504 | 0.381 |
40 | 2 | 1012 | 506 | 0.318 |
41 | 2 | 576 | 288 | 0.308 |
42 | 1 | 128 | 128 | 0.264 |
And now in graphical form:
As a point of reference, Teixeira's wOBA this season was 0.410, or slightly higher than the weighted average of his comparables. In general, the biggest concern with Teixeira's comparables is durability. Only half of his comps even made it to age 36, which would be the end of his current contract. And they only had two seasons in that time frame that averaged over 550 PAs. Performance-wise, his comparables maintained their ability to be above average hitters until around the age of 37, although the sample size is so small at that point it's hard to draw any conclusions on how Teixeira would do at the same age. Considering how athletic Teixeira appears to be, I tend to think he'll last longer and be more durable than many of his comparables. Lasting as long as Murray or McCovey (the only two comps who are Hall of Famers, and the only two who made it past 38 - something Carlos Delgado could still do), is harder to predict. But if Teixeira remains a productive player until age 40, he likely will have amassed Hall of Fame numbers.
Manny Ramirez
After Ramirez was traded from the Red Sox to the Dodgers last season, he caught fire and was given much of the credit for LA's late season surge to the playoffs. Despite being rewarded with a 4th place finish in last year's NL MVP race, Manny finds himself without a home for 2009. The relative lack of interest probably stems from at least three aspects. First, Manny was perceived to have quit on the Red Sox last season, which might make some teams wary of bringing him into the clubhouse. Second, Manny's not the best defensive player out there. While he's likely better than a lot of the advanced metrics make him look (the Green Monster has caused some issues in evaluation in the past) it's certainly fair to claim Manny's below average, if not substantially below. Finally, all reports are that Manny's been asking for a 4 year contract in the $80 to $100 million range. To many GMs, that's an awfully big risk to take on a 36 year old below average outfielder who can be a bit of a headcase.
What can we learn from Manny's comps? Is 4 years too long?
Here's the list of comparable players used for the analysis.
Player | Sim Score |
Reggie Jackson | 926 |
Ken Griffey | 923 |
Frank Robinson | 914 |
Willie Stargell | 911 |
Frank Howard | 897 |
Willie McCovey | 893 |
Gary Sheffield | 883 |
Jeff Bagwell | 877 |
Johnny Mize | 876 |
Frank Thomas | 865 |
Three of the players on this list are still theoretically active so any analysis is somewhat incomplete (unless Griffey, Sheffield and Thomas all retire). Not surprisingly, there's a lot of Hall of Famers or likely Hall of Famers.
And here's how they've performed by age.
Age | Players | PA | PA/Player | wOBA |
36 | 10 | 4759 | 476 | 0.379 |
37 | 9 | 3073 | 341 | 0.364 |
38 | 8 | 3903 | 488 | 0.361 |
39 | 7 | 2924 | 418 | 0.368 |
40 | 6 | 1594 | 266 | 0.336 |
41 | 3 | 830 | 277 | 0.317 |
42 | 2 | 212 | 106 | 0.287 |
Apparently, if you've reached 36 at a high level of performance, (Manny's age 36 wOBA was .423, substantially higher than his comps), you can generally hold on pretty well until 40. I think it's fair to expect declining playing time and some decline offensively, but he'll probably continue to be a well above average offensive player for at least another 4-5 years. Defense is obviously a different story. He's probably already in DH territory, which means he's probably only a good fit for American League teams. That said, he'd be a pretty good fit at DH if a team were willing to maintain a full-time DH. I don't think he's worth $20 million a year, but if he wants a 4 year deal, he might be worth $12-15 million (without doing any real WAR calculations).
Adam Dunn
To me, Dunn is the biggest mystery of the off-season. After 5 straight seasons of over 40 home runs, he's still unsigned, and rumor has it that he might sign for around $5 million a year. Unfortunately for Dunn, he's run into the perfect storm of factors that are depressing his value. Obviously the economy plays a part, but all free agents are having to deal with that. Perhaps more importantly, there's been a glut of similar players on the market this season - the aforementioned Manny, Pat Burrell, Raul Ibanez and Milton Bradley. Manny and Ibanez are substantially older, and Bradley, despite being a better defender, is much more injury prone. And there just aren't that many opening for a LF/1B/DH right now. Finally, Dunn just doesn't look like a valuable baseball player. Sure, he hits his home runs, but he strikes out way too much and hits for a low average. And he's purported to not really like baseball or care how he does. Think Dave Kingman or Rob Deer.
Of course those of us in the sabermetric community recognize his value at the plate (as well as his deficiencies in the OF), and most would rank him near the top of the list of players above - especially as all of the listed players save Bradley should probably be DH's. So what do Dunn's comparables say about his likely performance?
His most similar players at the age of 28 are:
Player | Sim Score |
Dale Murphy | 930 |
Roger Maris | 918 |
Harmon Killebrew | 912 |
Rudy York | 912 |
Jeff Burroughs | 911 |
Mike Schmidt | 907 |
Dave Kingman | 905 |
Mark McGwire | 905 |
Sammy Sosa | 900 |
Tom Brunansky | 900 |
A couple of Hall of Famers (who probably don't really compare, considering Schmidt was a 3B), and a few players who notably cratered in their early 30's. After 28, these players put up the following composite performance:
Age | Players | PA | PA/Player | wOBA |
28 | 10 | 5754 | 575 | 0.370 |
29 | 10 | 5212 | 521 | 0.399 |
30 | 10 | 5153 | 515 | 0.387 |
31 | 10 | 5039 | 504 | 0.406 |
32 | 10 | 5075 | 508 | 0.384 |
33 | 10 | 5256 | 526 | 0.371 |
34 | 8 | 3663 | 458 | 0.392 |
35 | 6 | 3623 | 604 | 0.380 |
36 | 6 | 2603 | 434 | 0.363 |
37 | 5 | 1896 | 379 | 0.339 |
38 | 3 | 1268 | 423 | 0.324 |
39 | 2 | 533 | 267 | 0.315 |
And graphically:
Most of Dunn's comparable players kept their value as elite players through their early 30s. At that point, there appears to be a substantial downturn in their performance. Dunn, who had a .385 wOBA in 2008, can probably be counted on for another 4 years of performance close to that mark, but then his output might begin to slip. Combined with his lack of defensive prowess and the concerns about his approach, I might be reticent to give him a 4 year deal. Perhaps a 3 year deal with a team option for $10-12 million a year would make sense. It's a lot lower than he's looking for (supposedly 4/$15) but it's a lot more than Burrell's deal.