The Guttridge-Wang trade model

The most important single central fact about a free market is that no exchange takes place unless both parties believe they will benefit.
–Milton Freidman

Both as a fan and a professional, I’ve seen trades assembled and evaluated from an a subjective and all-too-imprecise platform.

Take this hypothetical quote for an example. It may be what you’d read in a newspaper or what you’d hear if you had an executive on the phone:

We’re trading this veteran because he’s getting too expensive and we’re not contending. He’s still a good player, but we’re getting a decent pitcher who’s under control for four more years, and their #2 prospect who was a first-rounder. We forfeit the draft picks we’d get for him next June, but I think it’s worth it. Besides, we’ve got a good replacement waiting in the wings.

Obviously, nobody is making a trade based on the above statement alone—it’s an oversimplification, though it’s at least partially representative of the way many decisions are made. But what you have above is a multitude of disparate information. When considering such a deal it will be impossible to precisely gauge the value of each bit of information, balance it, and compare it to alternative strategies without a framework to evaluate and organize the risks and benefits.

Even holding a spreadsheet with the most accurate five-year total value forecasts of the players involved, I think it would still be virtually impossible to balance the relative value of the divergent aspects of such a scheme (unless the deal was just comically lopsided, and those cases are very rare). Don’t believe that?

Then tell me, what’s the value of the fact that the pitcher in the above deal is in arbitration for four more years? Is the discount he will receive by virtue of being in the arbitration process a more valuable or less valuable than receiving Type A compensation picks? And that’s just a single and relatively minor point of a much more complex deal.

What’s being advocated is basically managerial (in other words, internal) accounting. In industry, we make a decision according to the bottom line. “I have to pay $x million for this piece of equipment, which depreciates at a rate of $n per year. I could sell the old unit for $y. I expect to get $r per year in additional revenue from the superior productivity, which is a better return on investment than I get elsewhere. But we’d have a month of downtime, which costs us $p, and we’d have to add an employee to operate it at the cost of $q.”

Again, can you imagine how muddled this decision would get if the manager(s) in charge simply evaluated the information more or less in their head, without the aid of the balance sheet?

It is clear that one person, or a group of people, will not be able to precisely evaluate and balance all the information contained in a trade (let alone multiple offers) without a prevailing framework to organize, sort, and balance the inputs.

I created such a framework, which became part of the ‘toolbox’ I showed to prospective employers (MLB executives) during the Winter Meetings in Las Vegas. I had been initially flattered by their response. That is until one executive responded “Oh, cool… kind of like they do at The Hardball Times.” Damn—here I was thinking I had been original.

Victor Wang was the author of the work the exec had been referencing. Incidentally, Victor contacted me after my Winter Meetings article as a fellow job-seeker soliciting advice. He was glad to hear the news that his work had caught the attention of a prominent executive, and we gradually began comparing notes on our respective models.

Victor and I agreed on the major points:

A) That virtually all pertinent aspects of a trade can be converted into economic values. This is obvious in the major cases; a 3 WAR player has an inherent market value (currently around $13m per year). But we can also derive economic values for potential compensation picks, contract risk, prospects in A-ball, and more.

B) That surplus value is the currency of exchange. For example, suppose teams are paying $4.5m per win on the open market. You have a 2 WAR third baseman who is under contract for $6m. Thus, he carries $3m in surplus value; he is a $9m value that you only have to pay $6m for. In a vacuum, you should not trade him unless you receive $3m surplus value in return. If you were to trade him straight-up for another third baseman worth $6m and earning $6m (again, in a vacuum, and assuming these players are both on one-year deals) you have unquestionably decreased your assets.

C) Trades are not zero-sum equations. Primarily due to the fact that teams have divergent near-term strategies (and also varying ability to replace a player) it is entirely possible—even common—that both teams in a trade benefit.

There were some areas of disagreement.

1) For prospects, I had simply been running a current projection and projecting forth using a standard aging curve. This is problematic, due to the fact that (when using proper figures) virtually no 22-year-old will project as an All Star if he simply follows the standard aging curve from the present to his peak. Virtually all top-caliber performers have ‘jumped’ that curve at some point in their careers.

I first thought this impediment to be a fair penalty based on the inherent risk of prospects, but Victor had come up with a better solution. By studying past prospect lists of major publications, Victor was able to assign economic values to different ‘slots’ of prospects. As in, if Baseball America, Baseball Prospectus, and John Sickels seem to agree that Jhoulys Chacin is between the 15th and 25th best pitching prospect in the game, we can use Victor’s research (and a few basic financial calculations) to determine the present economic value of that grade. You can see Victor’s work on these valuations here.

2) I’ve convinced Victor to adopt my future value discount and inflation figures. The way he had been calculating things was incorrectly inflating the net present surplus value of future receipts, because he had inflation set higher than his discount rate. To convert that into plain English; what is more valuable—2.5 WAR projected for 2010 or 2.5 WAR projected for 2013? 2010, of course, for two reasons:

&8226; You get to realize the value more quickly,
&8226; There’s far less risk in a projection one year away than four years away.

The model now makes a more accurate account of this, and it makes a huge difference.

&8226; As far as organization (which is the whole point), we’re combining models. The “process tree” framework Victor used to illustrate potential outcomes will be used, with some tweaking of the “branch”-level coefficients. For the projected values along the timeline, we will use using my spreadsheet model, now with Victor’s prospect valuation.

Using the Matt Holliday trade as an example, here’s how the Guttridge-Wang model looks:

image

For some explanation: The change in contention odds and its value is sort of a “dumb” calculation, in that it’s not intended to be precise. Baseball Prospectus estimated that making the playoffs is worth, all told, $40m to an organization. For the purposes of this calculation, teams are just subjectively slotted as contenders or non-contenders.

This is because…
{exp:list_maker}It’s not practical, nor desirable, to use a current projection (like 84.6 wins) as a starting point. Especially during the offseason, so much can change in terms of roster construction and competition between the time of a deal and the season that it doesn’t make sense to fixate on a particular number, and
I’m confident we can do a decent job of subjectively slotting a team as contenders or not. {/exp:list_maker}
Thus, to calculate the change in contention odds, we’re going with a simple premise. A team who wins 83 games has only the slightest chance of making the playoffs. A team who wins 95 has only the slightest chance of not making it. In that sense, you could say each marginal win added to a contending team increases their playoff chances by about 8%. (Yes, I’m aware it’s not actually that simple. But it’s a pretty good estimate).

Holliday improves the A’s, whom I slotted as a contender (I had them right around 81 wins pre-trade, but the Angels were looking quite weak) by about 2.4 wins. Thus, they’re about 19 percent more likely to make the playoffs than before the deal—quite a large figure. So, being 19.1 percent more likely to earn $40m means he has added an expected surplus value of around $7.5m to the A’s.

Now, onto my spreadsheet model:

image

As you can see, this mock-up is from the Rockies’ perspective. Remember that because of divergent goals and varying ability to replace players, the deals aren’t zero-sum equations; the fact that I assess the Rockies at a loss of $4.25m does not mean the A’s gained the same amount. I presented Victor’s process-tree from the A’s perspective and my spreadsheet model from the Rockies perspective, just to present a taste of each—to fully evaluate a trade, you need to assess each side.

Now, as far as that $4.25m loss for the Rockies… is that the real value of this model—being able to say that a team “lost” $4.25m (as opposed to, say, $3.81m), or gained $4.1m in their franchise value?

Well, only sort of. It is valuable for that reason. Even if we accept that the projected values and salaries are only 90 percent accurate to their optimal baseline (which would be a very low threshold that I’m positive we’ve met), there’s a great deal of value in being able to establish that bottom line.

However, the real beauty of this model is how it allows an organization to examine all the possibilities. What if the Rockies’ scouts are big believers in Carlos Gonzalez, and believe there’s a good chance he’ll be an All-Star caliber player by 2011 (a highly implausible scenario in my opinion, but that’s beside the point)? Well, we can plug that scenario into the model, and see how the trade would look if it panned out that way:

image

If things work out that way, it’s a great deal for the Rockies. Now, what if the “stall” experienced by Gonzalez during ’07-’08 is indicative of a very early peak, and he’s really just a toolsy 5th-outfielder, a la Reggie Taylor (which I regard as equally implausible as the All-Star scenario)? Plugging it into the model, we would see that it would be a horrible deal for the Rockies.

What if murmurs of Huston Street having some shoulder issues are true? What if he misses half a season at some point prior to free agency? Plug that into the model, and see how it affects the bottom line. What if Greg Smith develops into a passable 3rd starter instead of a 5th starter? Plug it in. What if we feel Seth Smith is far more valuable than the projection is giving him credit for? Plug it in.

This is the real value of this model for an organization; you must take the baseline as a starting point, sure—but plugging all the reasonable or anticipated deviations will allow you to get a clear picture of the range of possibilities, put a quantifiable value on the subjective parts of the evaluation, and let you pick your spots in terms of risk most effectively.

The above models don’t contain anything any team isn’t already considering in a trade; no new information is really being introduced. It’s just that the information, instead of being disparate, non-quantified, and unorganized, is now tied into a single platform with a single unit of value and presented in a now-intelligible manner. Where once it was too overwhelming to accurately account for, it is now clear, flexible, and actionable.

The hypothetical trade I quoted in the opening represents the way our minds distill these disparate inputs, probabilities, and values. General managers, unlike their counterparts in industry, never really used the balance sheet, primarily because we’ve never viewed major league players, minor league players, draft picks, and playoff value within a single plane of tangible value (both present and future). Those who learn to do so effectively will gain a massive leg up in the information wars.

Print Friendly
 Share on Facebook0Tweet about this on Twitter0Share on Google+0Share on Reddit0Email this to someone
« Previous: The 10 worst lineups ever no-hit
Next: Clone Wars: Jacoby Ellsbury and Juan Pierre »

Comments

  1. Mass Cap said...

    Both are interesting theories and models. One has to presume that the people working the model aren’t swayed but their over valuing a particular player they want in return or the in house replacement.  I think it is a tool that could be used to evaluate trades 1 or 2 years after the fact to maybe evaluate the thinking of those making the trades.

  2. Adam Guttridge said...

    Mass Cap…

    Are you suggesting that if someone had a man-crush on say, Matt Holliday, they’d manipulate the numbers? A) I don’t see why anyone would have a motive to do that… if you work with the Cardinals, and thinking about trading for Holliday… why would you want it to look like he’s a 5.8 WAR player instead of a 5.1? B) It really couldn’t be done, as WAR is verifiable, so you’d be caught awfully fast C) If they wanted to do that, they wouldn’t manipulate/hide the numbers… they’d just say ‘I think Matt Holliday is still progressing… what if he’s a 5.8 player net year?’ and plug that into the model in contrast to the baseline.

    Also, using this model in a retrospective sense seems to me… weird and pointless? The whole point here is the value projections. If you used it in a retrospective sense, you’ve got… pretty much nada.

  3. Adam Guttridge said...

    Thanks, DB.

    Yeah, I think it’s always been around to some degree or another. The point of this framework isn’t necessarily that we’re doing anything new here. It’s just that we’re quantifying and organizing the data in a much more clear way than doing it in one’s head.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

Current day month ye@r *