A Basic Cost Projection For a Dozen 2015 Free Agents by Dave Cameron March 11, 2014 Yesterday, I created a model of agent pricing using projected 2014 WAR as the only variable. As I noted in the post, I would classify this model as more of a toy than a rigorous attempt at analyzing the market, as we know there are plenty of factors beyond next season’s performance that influence the size of a free agent contract. Ignoring age and health are obvious flaws, as both are significant factors in contract size, especially when forecasting how many years a player might sign for. And we know that the market pays a different price for offense than it does for defense, reflecting some of the uncertainty that the defensive evaluations have relative to our ability to measure offensive production. So, before I say any more about this model, let’s remember that this was essentially designed as something like the Marcel projections for salaries. Tom Tango built the Marcel projections as something of a baseline forecast, using very simplistic adjustments and ignoring things like park factors in order to show the bare minimum that a projection system should be able to accomplish. Likewise, I’d say that any serious attempt at evaluating the free agent market should be able to do better than what this basic model can do. This system is as simple as it gets — without being completely wrong, anyway — and we can almost certainly do better, but doing better comes with added complexity, and it’s still nice to have a simple, easily explained baseline to measure other forecasts against. That said, let’s see what this basic pricing model — naming suggestions welcome, by the way — spits out for a dozen of the more notable 2015 free agents. For now, let’s ignore the future and pretend that these 12 players were free agents this past off-season, and see what the model would have forecast for them as 2014 free agents. As a reminder, we’re using a hybrid ZIPS/Steamer WAR forecasts, and the model multiplies projected WAR by five to estimate annual average value and projected WAR by 2.0 (for 3+ WAR players), 1.5 (for +2.0 to +2.9 WAR players), or 1.1 (for 1.0-1.9 WAR players) to estimate the number of years a player will sign for. Because I selected 12 of the more notable pending free agents, we’re only looking at players with either a 2.0 or 1.5 multiple, as they’re all projected as above average performers for 2014. On to the forecasts: Player 2014 WAR ProjYears ProjAmount ProjAAV Max Scherzer 4.8 10 240 24 James Shields 4.1 8 164 21 Jon Lester 3.9 8 156 20 Pablo Sandoval 3.8 8 152 19 Hanley Ramirez 3.7 7 130 19 Chase Headley 3.5 7 123 18 Justin Masterson 3.2 6 96 16 JJ Hardy 3.1 6 93 16 Jake Peavy 3.0 6 90 15 Colby Rasmus 2.7 4 54 14 Asdrubal Cabrera 2.1 3 32 11 Jed Lowrie 2.1 3 32 11 I think we can see right away that the model probably has problems with over-projecting years for free agent pitchers. We know that MLB teams are comfortable going to 10 years for position players, so that length got built into the model, but because we’re not separating out pitchers and hitters, the model is overestimating a team’s willingness to pay Scherzer the same as the Mariners paid Robinson Cano, because it sees them as equally valuable free agents. I don’t see any scenario under which Scherzer would land a 10 year contract as a free agent, and my initial reaction to all of these numbers is that they’re just way too high. No one would really give Pablo Sandoval $152 million coming off a +2.3 WAR season, right? Probably not, no. But it’s worth keeping in mind that Sandoval’s career averages put him at +3.7 WAR per 600 PA, and he’s entering his age-27 season. If a team thought they could keep Kung Fu Panda in shape and healthy, betting on him wouldn’t be all that different than betting on Jacoby Ellsbury, whose price is heavily influencing this projection for Sandoval. We live in a world where Hunter Pence got $90 million for his mid-30s, so maybe pricing Sandoval’s peak years — with the expectation that he’ll be healthy, as these forecasts do — at $150 million isn’t entirely insane. And, to be honest, I’d probably take the over on the Hanley Ramirez forecast. Like Sandoval and Ellsbury, the questions about him revolve around health more than performance, and I think Hanley would appeal to a wider base of teams than Shin-Soo Choo did. And I think both Asdrubal Cabrera and Jed Lowrie could have done better than the 3/$32M that the model is suggesting, especially given how badly it missed on Jhonny Peralta. So it’s not like the model is just overstating the value of all 12 players. It just seems systematically high on the numbers of years for high-end pitchers, because it doesn’t know that teams don’t really go beyond seven year deals for hurlers. That’s why it projected nine years for Tanaka too. It’s a flaw that stems from the simplicity of the model, but it’s also a pretty easy one to fix. For instance, if we just cap the pitcher multiplier at 1.7 instead of 2.0, Scherzer goes from 10/240 to 8/$192M, a believable figure given his pedigree and the recent contracts signed by guys like Zack Greinke and Masahiro Tanaka. Instead of 8/$164M and 8/$156M for Shields and Lester respectively, they come in at 7/$144M and 7/$137M, putting them in the same range as Cole Hamels, which seems about right. And Justin Masterson goes from 6/$96M down to 5/$80M, which seems a little bit more reasonable given the deals signed by the Garza/Jimenez/Nolasco trio this winter. Whether you want the model to make positional differentiations is a question of trading simplicity for accuracy. The more you break down specific players into different buckets, the less the model can work as a catch-all baseline that is easy to explain in English, but of course, I already flagged catchers as being an exception to the rule, so maybe it’s better to flag pitchers too. I’m not married to the uniform multiplier model, and since this is mostly a toy, I’m happy to hear feedback on which model would be preferable. But at the same time, let’s also keep in mind that these players weren’t free agents this winter, and we don’t want to use their 2014 WAR estimates to calculate expected 2015 prices. For one thing, players generally get worse as they get older, and these players were selected as the “most notable upcoming free agents” because they generally have a track record of success, so we’d expect some regression to the mean just because of how I picked the dozen players for the list. Odds are pretty good that these 12 players will be forecast for a lower 2015 WAR than they are for 2014, and that’s the projection that the model would use to actually determine what they’d get as free agents. So, let’s age everyone by one year and assume that their 2015 forecast will be 90% of their 2014 WAR forecast, and re-run the calculations based on that number instead. Player 2015 WAR ProjYears ProjAmount ProjAAV Max Scherzer 4.3 9 194 22 James Shields 3.7 7 129 18 Jon Lester 3.5 7 123 18 Pablo Sandoval 3.4 7 120 17 Hanley Ramirez 3.3 7 117 17 Chase Headley 3.2 6 95 16 Justin Masterson 2.9 4 58 14 JJ Hardy 2.8 4 56 14 Jake Peavy 2.7 4 54 14 Colby Rasmus 2.4 4 49 12 Asdrubal Cabrera 1.9 2 19 9 Jed Lowrie 1.9 2 19 9 Now, these numbers look more reasonable, right? Free agents are almost always going to look a little rosier before they actually get to free agency, since age often chips away at their skills and health, and applying some expected decline to this group gives us a set of projections that look pretty solid at first glance. Scherzer gets the same forecast as Tanaka did this year, with Shields, Lester, Sandoval, and Ramirez landing in the same range as Choo’s deal. And remember, this is with the 2.0 multiplier for pitchers, which is pretty clearly wrong. What if we re-ran the numbers with the the 1.7 multiplier for 3+ WAR pitchers? Then the table would look like this. Player 2015 WAR ProjYears ProjAmount ProjAAV Max Scherzer 4.3 7 151 22 James Shields 3.7 6 111 18 Jon Lester 3.5 6 105 18 Pablo Sandoval 3.4 7 120 17 Hanley Ramirez 3.3 7 117 17 Chase Headley 3.2 6 95 16 Justin Masterson 2.9 4 58 14 JJ Hardy 2.8 4 56 14 Jake Peavy 2.7 4 54 14 Colby Rasmus 2.4 4 49 12 Asdrubal Cabrera 1.9 2 19 9 Jed Lowrie 1.9 2 19 9 These “feel” like the best estimates yet — with the exception of the Scherzer forecast, which seems entirely too low — for whatever my gut feelings are worth. In general, though, tweaking a model to make the output match your initial expectation does more harm than good; what’s the point of modeling something if you’re just going to force it to conform to what you already believe? We could keep manipulating the numbers all day until we got a set of projections that we all agreed seemed reasonable, but then the model wouldn’t add anything beyond basic crowdsourcing. Because this is a toy in progress, I’m certainly open to suggestions on tweaks or improvements that retain the simplicity that was the original goal. Since we already have the more complex models of free agent pricing that include aging curves and the like, I think the value in this tool is that it can be calculated using a minimalistic number of variables and done without the aid of a calculator or spreadsheet. If there’s a way to make the model better while retaining that strength, then I’m all for it. But that last table does feel pretty close to a decent forecast, and gun to my head, I’d probably be okay using those as a barometer of what kinds of deals the impending free agents should set as baseline expectations if they decide to test the market. Scherzer, Cabrera, and Lowrie could easily land bigger deals, while maybe Sandoval’s health and weight problems mean he’d have to settle for less, but I think the numbers listed in that final table are fairly okay estimates of what these players should be looking for in an extension if they want to bypass free agency and sign in the next couple of weeks. Maybe they’ll raise their stock in 2014, and maybe inflation — another important variable we didn’t include — will render these forecasts obsolete, but if any of these 12 sign extensions in the next few weeks, I won’t be shocked if the numbers look pretty similar to the ones listed in that last table.