Joshua Iversen contributed to this article.

As many of you know, former GM Jim Bowden often writes about the baseball trade market in The Athletic. Much of that coverage involves Bowden creating his own trade proposals, but he also evaluates proposals from others – either from fans or local beat writers. As far as we know, he is the only national baseball writer doing this type of thing on a regular basis.

As part of his proposals and evaluations, he weighs in on whether or not he believes them to be fair. But how fair are they? Has anyone checked? We couldn’t find any evidence of that.

So we thought we’d take a stab at it. We do this only in the spirit of auditing, with no malice intended. We believe everyone, including ourselves, should be cross-checked for accuracy.

So we reviewed every article Bowden has published at The Athletic since August 2019 (the same time period we use to audit our model) that contains trade proposals, either originated by Bowden or evaluated by him. How closely do the proposed trades correlate to the market?

Gauging the accuracy of trades is a bit of a slippery eel, because there is no official, objective way to do so. There is no industry standard. So Bowden’s judgment may not be contestable.

But let’s try anyway, using the model we’ve built. The measure of success for both BTV and Bowden should be the highest possible correlation to the market.

Here at BTV, we’ve established that correlation at 95.3%, with a variance of 1.7. We show this on our history page, and continue to update it with each real trade. We’ve counted a total of 278 trades since August 2019, and 265 of those have been accepted as fair by our model. In table form:

Number of trades tracked by BTV	Number accepted by our model	Acceptance rate	Variance/Degree of loss
278	265	95.3%	1.7

The way we’ve done this is to document every real trade on our site and on Twitter, applying our estimated numbers at the time the trade occurs. 

For example, when Blake Snell was traded by the Rays to the Padres for four young players on December 28, 2020, we immediately published it on our trade board page, then tweeted about it, then added it to our trade log spreadsheet behind the scenes, using the estimated values we had for each player in the trade on that day. We then compare those estimated values to the return to see how close we got. (You are welcome to audit us by searching for each trade on either our trade boards or on Twitter.) 

We keep a log not only to track our performance, but to see if any patterns emerge that would indicate we need to make a tweak to the model. This allows us to be honest with ourselves and with our audience, and to continue to improve the accuracy of the model over time.

In order to test Bowden’s performance, then, we should apply the same standards and methods over the same time period. 

So that’s what we did. We copied and pasted every trade proposal from all of Bowden’s articles (except the most recent ones, which can’t be measured yet against the market), logged them into a spreadsheet, looked up the median trade value (according to our model) at the time of each proposal, applied them, and did the math. To be generous, we considered any trade within +/- 40% to be acceptable as fair (we typically use +/- 20% as fair, and go up to +/- 40% for major overpays).

Given that we’ve established that BTV has a 95.3% correlation to the market, perhaps we can use it as a proxy. In other words, what is the correlation of Bowden's judgment to BTV, as market proxy? Here are those results:

Number of Bowden trade proposals	Number accepted by our model	Acceptance rate	Variance/Degree of loss
115	42	45.2%	20.4

Bowden’s trade proposals correlated to the BTV model at a rate of only 45.2%. In other words, the majority of the trades he published as “fair” would have been rejected by our simulator, as they were beyond the +/- 40% limit.

If you buy that BTV is an effective proxy for the market, Bowden’s market correlation is lower than a coin flip. 

Further, his variance is quite high. What that means is, any GM making a trade that was deemed fair in Bowden’s articles is likely to lose $20.4M in value with each one.

Okay, but what if we took it a step further? It turns out that the majority of those 115 proposals did not actually occur – that is, on most occasions, the biggest-name player in the trade was not actually traded, making this exercise somewhat theoretical.

So let’s narrow it down to only the proposals that included players who were traded – Betts, Lindor, Arenado, Scherzer, etc. 

The benefit of that is that we know the return packages of each transaction, which represent real market values. (We still have to use BTV numbers for each package, because that’s all we have – but it’s still better than nothing.) 

Note that on several occasions, Bowden proposed multiple different trades with the same player as the lead piece, so we’re measuring each of those against what they were actually moved for. (This is important, as there is a wide range of outcomes in many of his proposals.) We’re correlating, as close we can, Bowden to the market. 

Here are those results:

Number of Bowden proposals that matched players actually traded	Number accepted by the model using market numbers	Acceptance rate	Variance/Degree of loss
47	24	51.1%	17.5

By this measure, tested more closely by the market, Bowden fared slightly better than a coin toss. Importantly, any GM making a trade of those players by acting on Bowden’s published guidance would, on average, lose $17.5M in value with each one.

How does that compare to our model? We went back through our trade log and matched the same subset of big-name players cited in Bowden’s proposals, and noted how the model did in each case against the market. Note that there was only one instance of each trade (unlike Bowden’s multiple possibilities), so our total number is smaller. 

Here are those results:

Number of big-name trades in Bowden’s proposals	Number accepted	Acceptance rate	Variance/Degree of loss
29	24	82.8%	5.5

Of the 29 big-name-led real trades, our model got 24 of them right, for an 82.8% success rate. That’s not as high as our aggregate number, because the latter includes a lot of smaller deals that are easily accepted. Big-name trades are harder to hit. And if you’re dealing stars like Mookie Betts or Francisco Lindor, who tend to have higher values, there’s going to be some variance, so a give-or-take of $5.5M seems reasonable to us.

In any case, the BTV model fared very well on such big-name trades as Arenado (we had him at -$43.7M at the time, and the Rockies had to include $51M in cash just to get a few low-level prospects), Lindor, Scherzer/Turner, Bryant, Baez, Lynn, Snell, Taillon, and many others.

So we’ve correlated both Bowden and the BTV model to the market. Now let’s compare them:

Source	Acceptance rate by the market	Variance/Degree of loss
Bowden	51.1%	17.5
BTV	82.8%	5.5
Gap	-31.7%	12.0

Using this method, we see that Bowden’s performance on big-name trades is 31.7% less accurate than the BTV model, and each Bowden trade is riskier than BTV’s by $12M.

Why might this be? There may be a few reasons.

Different approaches

As Nobel-Prize-winning economist Daniel Kahneman describes in his bestseller, Thinking, Fast and Slow (a book popular in MLB front offices), there are two systems in the brain – System 1 is based on gut reaction, and “operates automatically and quickly, with little or no effort”; while System 2 “allocates attention to the effortful mental activities that demand it, including complex computations.” 

Most front offices are led by Ivy Leaguers (David Stearns and David Forst, among others, graduated from Harvard; Farhan Zaidi graduated from MIT and has a Ph.D in Economics from UC Berkeley), and are supported by analytics teams who crunch numbers to arrive at precise valuations. They’re using System 2. It’s possible that Bowden neither takes that approach, nor is supported by an analytics team.

Here at BTV, we try to mimic what those analytics teams are doing, on a much smaller scale. We’re using System 2 as well.

So it’s possible we have a System 1 vs. System 2 discrepancy here.

Inconsistency in approaches

While it’s not clear what methodology Bowden is using in his trade proposals, there do appear to be inconsistencies in his approach. His proposals seem to vary by concept.

For example, on January 27, 2020, Bowden wrote an article outlining possible trade destinations for Nolan Arenado. At the time, we estimated Arenado’s surplus value at $27.7M. He had seven years left on his contract, and was still owed $234M. Our model showed that he was unlikely to opt out in 2021, because he was already being paid top dollar. Therefore, any team that acquired him would be on the hook for the entire amount. It seemed unlikely to us that any team would give up a big haul of talent on top of that burden.

Many of Bowden’s proposals, however, seemed to assume a team would do just that. For example:

"The Dodgers … are certainly one of the best trade fits for Colorado, especially if they were willing to center a deal around their top infield prospect, Gavin Lux. A deal that included Lux, top catching prospect Keibert Ruiz and a pitching prospect like Josiah Gray or Gerardo Carrillo is probably as good a value as the Rockies could expect in an Arenado trade." 

This deal was one of the most lopsided proposals we found in Bowden’s articles (mostly because Lux was a consensus Top 10 national prospect at the time). Here’s how it looked when plugged into our model:

Dodgers get	Arenado	27.7	Rockies get	Lux	80.8	Gap
				Ruiz	34.3
				Gray	17.7
Total		27.7		Total	132.8	105.1

It seems like a stretch to believe that the Dodgers would agree to assume Arenado’s $234M contract AND give up another $132.8M in value.

But then, later in the same piece, Bowden proposes a Cubs deal that focuses not so much on field value, but on money:

"For [the Cubs] to make a run at Arenado, they would have to do it with major-league players making money, like catcher Willson Contreras and outfielder Jason Heyward. Contreras will make $4.5 million this year… Heyward makes on average $21.5 million over each of the next four seasons… One of the problems with that idea is the Cubs can’t take on any money and they’d have to ask the Rockies to absorb the approximately $10 million dollar differential."

At the time, our model showed Heyward with a severely underwater contract, as he was still owed $96M, but only figured to produce $22.8M in field value. Here’s what that trade looked like:

Cubs get	Arenado	27.7	Rockies get	Contreras	20.4	Gap
	Cash	10		Heyward	-73.2
	Total	37.7		Total	-52.8	-90.5

 

So in this one, the Rockies take on significant negative value (all the more so because they’re kicking in another $10M), such that they lose over $90M in value in the deal. In the first example above, they made a profit of $105M. That’s roughly a $200M swing.

That seems to be due to a lack of consistency in the approach here. The first trade is based on field-value matching; the second is based on salary matching. 

Our observations are that most trades are based on surplus value – the amount of field value that exceeds the cost. They are not made by focusing on one of those factors or the other in isolation, but both together.

That is not to suggest it’s the only approach. At Fangraphs, the authors of the site’s most recent trade value series, Ben Clemens and Kevin Goldstein, point out that, as collaborators, they have subtle differences:

"We share similar overall viewpoints — we both think in terms of surplus value…  but the particulars vary in useful ways. Our perspective has been colored by our respective experiences. Kevin’s experience on the scouting and team sides (and mostly both at once) tilt his evaluations, while Ben’s outsider perspective and quantitative bent tilt his."

Notice, though, that they both start with surplus value as a framework; the differences from there are nuanced, and presumably focused more on the field-value part of the equation.

Absolute-ism vs. probabilistic thinking

We noticed doing this exercise that Bowden is prone to absolutist statements. Here are a few examples:

July 27, 2021:

Do you view a Trea Turner trade as at all likely, and what would be a realistic return for the Nationals? — Tom S.

Bowden: "I don’t see Turner getting traded at the deadline."

Turner was traded to the Dodgers with Max Scherzer two days later.

 

July 6, 2021 (with Maria Torres):

What are the Nationals going to do with Max Scherzer?

Bowden says: "Scherzer will finish the season as a Washington National regardless of how they play between now and the trade deadline."

See above.

May 21, 2021, in response to a question about the Giants:

"Bowden: The Dodgers and Padres are going to pass the Giants soon in the standings. The Giants will be a third-place team in the NL West no matter how many trades they make. They are really fun to watch, but enjoy it while you can because third place will be their final destination."

The Giants won 107 games and the division title.

The problem with these types of statements is not so much the content, it’s the delivery – there’s no equivocation. There’s no allowance for the possibility of being wrong. 

But in the trade market (and the world writ large), things are not this black and white – it’s all shades of gray. Every deal is a gamble based on probabilities, which means modern-day GMs think probabilistically, not in absolutes. “You’re placing bets,” Yankees GM Brian Cashman told The Athletic. “And there are risks to every bet. It’s like investments. There is no such thing as a 100 percent guaranteed return on your investment.”

Looking at the world with a more nuanced mindset tends to correlate better in more complex scenarios, which tend to have a mixture of signal and noise.

Keeping up with the prospects

We also noticed that, in many trades involving prospects, Bowden’s view of them seemed to be outdated. Recently, he proposed a trade of Ketel Marte to the Marlins for prospects J.J. Bleday and Braxton Garrett. Here’s how that one looks in our model:

Marlins get	Marte	37.3	Diamondbacks get	Bleday	11.9	Gap
				Garrett	2.6
	Total	37.3		Total	14.5	22.8

 

It appears lopsided because both prospects have fallen out of favor with most evaluators due to poor performance. Athletic reader Wylie B., in a comment in response to this proposal, offered this explanation:

"I think a lot of the Jim hate stems from what seems like a stubborn refusal to improve his prospect knowledge combined with a continued appetite for suggesting star-for-prospect moves. JJ Bleday and Braxton Garrett were former first round picks, which I'm sure is what attracted Jim to their names, but Bleday hit .212 as a corner outfielder in AA and dropped off of every Top 100 list, and Garrett is now 23 with walk issues and is likely just a 5th starter or reliever. Somehow this package lands Ketel Marte."

To be fair, Bowden later responded: 

"My prospect knowledge is based on my personal scouting and discussions with the 30 GM's. Not based on media or fan opinions. I saw Bleday in the Arizona Fall League and as an Amateur did you? I saw Bleday and liked him just like Derek Jeter and Mike Hill did and was just as impressed this fall with him. Prospects have ups and downs...I don't jump off the bandwagon when they're down. In addition, it's ok if we disagree on player evaluations....the game is about opinions.........."

Bowden is right – prospect evaluation is subjective, and opinions do vary. But it’s a bit difficult to imagine that Bowden, in addition to his work at The Athletic, MLB Network Radio, and CBS, has time to “personally scout” every prospect who could possibly be traded. There are teams of smart, dedicated people at Baseball America, Fangraphs, Baseball Prospectus, Prospects Live, and other outlets whose entire jobs are dedicated to that endeavor. 

Regarding this trade in particular: most of the prospect evaluators we follow, who also speak regularly with front office executives, saw Bleday in the AFL, and acknowledged his recent swing change. But they did not upgrade him in their ratings, because the sample size was too small, and the quality of the pitching this past fall in Arizona was considered well below par. We should note also that even at The Athletic, Bowden’s colleague/lead prospect writer Keith Law downgraded Bleday to #7, and has Garrett at #14 in the Marlins’ system.

But more broadly, we believe the best way to reduce the variance of subjectivity is to aggregate and properly weight a large number of inputs over time. This tends to smooth out the wrinkles, and more accurately correspond to market results.

What Bowden gets right

Of the 24 Bowden trades that correlated well with the market, the most consistent pattern we saw was that they were smaller. These included the 2021 deadline trades of relievers Brad Hand, Daniel Hudson, Ian Kennedy, and Yimi Garcia; depth starters Danny Duffy and Tyler Anderson; and the 2020 deadline trades of relievers Trevor Rosenthal, Brandon Workman, and Miguel Castro. 

Of the bigger names, the only Bowden proposals that correlated well to actual market returns were one Blake Snell proposal (to the Angels, for Detmers and Marsh); one Lindor proposal (to the Cardinals, for Liberatore, Torres and Williams); a 2020 deadline deal of Mike Clevinger (to the Mets for Conforto and Matz); and two Mookie Betts proposals (one to the Dodgers, for Ruiz, Downs and Gray; one to the Braves, for Waters, Muller and Contreras). 

That’s all we could find.

Conclusion

The BTV model is the only tool we know of that can measure trade valuation relatively closely. Bowden’s trades did not perform well relative to the model, nor did they perform well when measured more directly against the market. 

Further, we found no clear methodology in his evaluations. 

—-------------

If you’re interested, here’s the log of all 47 of these Bowden proposals, and how they performed against the actual market returns, as measured by the BTV model:

Notes:

*PTBNLs were assigned a value of 1, based on previous instances we’ve noticed in our trade log.

*To estimate the values of unnamed prospects Bowden suggested based on their team ranking (e.g., “Top 15 prospect”), we averaged out the values of that type of prospect on six teams that were in the midrange of team farm ratings at the time. In other words, for a Top 15 prospect, we looked at what a typical #15 prospect value would be on teams like the Blue Jays, Giants, Guardians, Marlins, Royals, and Yankees. For unnamed prospects that fell into a range (“Top 20-25 prospect” we did the same, but took the value of the number slightly higher than the midpoint – e.g., the average value of the #22 prospect on those types of teams). 

*In some cases, the value of one player in the package varied by date. For example, Bowden suggested including Dodgers prospect Diego Cartaya in a proposal on 7/7/21. Cartaya was valued at 16.9 on 7/2/21, but was upgraded to 19.8 on 7/19/21 For this, we used the value that existed at the time of the proposal – 16.9 in this case. As it happens, using this method tended to favor Bowden in these cases.