Now that the offseason is (more or less) over, it’s time for our semi-annual scorecard. How’d we do? Here’s what our trade log says, counting all trades that started with the early November 2022 swap that sent Sam Hilliard to Atlanta:

• Total trades: 64
• Total accepted by our model: 58
• Acceptance rate: 90.6%
• Average margin of error: 2.6

 

So our model got over 90% of all trades right. Its margin of error was a little higher than our historic average of 1.9. Now let’s dive into the details.

• Hits:
• Odorizzi/Allard
• Hunter Renfroe to the Angels
• Kyle Lewis to the D’backs
• Winker/Wong
• Varsho/Moreno
• Joe Jimenez to the Braves
• James McCann to the Orioles
• Miguel Rojas to the Dodgers
• Michael A. Taylor to the Twins
• Adalberto Mondesi to the Red Sox
• Barnes/Bleier
• Benson/Boyd
• Puk/Bleday
• And a whole bunch of minor ones

 

Gray area (Half a hit, half a miss): 

Sean Murphy to the Braves (3-way; A’s/Braves half was fair; A’s/Brewers half was not)

Misses:

• Teoscar Hernandez to the Mariners
• Arraez/Pablo Lopez
• Cole Irvin to the Orioles
• Gregory Soto to the Phillies

 

There weren’t a lot of high-profile trades this offseason, but we did hit on many of those medium-profile ones, including Renfroe, Winker/Wong, McCann, Rojas, Barnes/Bleier, Puk/Bleday, and others. Our biggest misses were Arraez/Lopez, the Murphy 3-way (where we were half right), Irvin and Soto. Not one of our best offseasons, but acceptable enough.

The biggest causes of misses:

$/WAR adjustment

Early on, we noticed there was more money being thrown around. Whether it was due to inflation, a robust free-agent class, a big cash distribution to all teams from BAM Media, a sense of certainty due to Covid having passed and counting on gate revenues to be fully restored, resolution of the new CBA, or all of the above (the most likely answer), it drove up player prices overall, such that we needed to adjust our dollars-per-WAR estimates.That in turn affected trade values. So trades that happened before that $/WAR adjustment were a little off.

QO adjustment

A knock-on effect of the $/WAR adjustment is that more players might be candidates for a QO post-2023 than first appeared. We base this on our AFV estimate. Before the above adjustment, some players, like Teoscar Hernandez, looked like they might not be considered candidates to be offered $19-$20M for one year on a Qualifying Offer. After the adjustment, he now looks closer. This, then, means that the team that acquired him (the Mariners) is likely to receive an additional benefit of a draft pick should Hernandez decline the QO next offseason. That has been added to his trade value.

Arbitration adjustment

For arbitration-eligible players, we start each offseason with an estimate. As the actual salary figures come in, either due to agreements made between the player and team in advance of arbitration, or the results of the arbitration hearings, we update them. This often affects surplus value, as it involves salary changes. For players with multiple years of control, this can be significant, as the current number will influence the number for subsequent years. So if the salary number comes in higher than expected, it will have a domino effect of making future years higher, which reduces surplus value.

Prospect updates 

Because we rely on public prospect evaluators, we are beholden to their publication timelines. So trades that are made prior to those updates could be a little off; we sometimes see upgrades or downgrades after the fact, which often has a resolving effect of making a trade that was off at the time look much closer.

Reliever model

We knew our reliever model wasn’t perfect. But after the Gregory Soto trade, we knew we had to take a more serious stab at fixing it. That was the last straw. Prior to this, we had baked in more volatility, to align it more with deadline deals. The idea was that, because relievers are notoriously volatile, and because supply/demand dynamics are more skewed towards recency, we should match those ups and downs.

But that resulted in a weird situation where we were off too much in the offseason. Over the winter, teams are more conservative – they tend to look at track record more than hot-hand recency. So we re-adjusted it for stability. We’ll continue to use stability as a guiding principle even at the deadline.

We also had a fundamental problem with leverage stats, and how much we were weighting them. 

So after these changes, the model tracked the market much more closely, both in trades and free-agent signings (which we use as a $/WAR benchmark). A.J. Puk went from zero to 4.1 in the updated model, which meant the Bleday trade came out pretty even. Richard Bleier rose a bit as well, which meant the Barnes trade came out more even. And cross-checking it against free agent deals, we found that most of them correlated more closely to the market – that is, with a surplus closer to zero (since free agent deals should, in theory, represent full market value).

Backtesting

One trade that had all five of these adjustments was the one that sent Teoscar Hernandez to the Mariners. 

Values of players at the time of the deal:

Team	Player	Value	Team	Player	Value	Difference
Mariners	Hernandez	8.4	Blue Jays	Swanson	10.4
				Macko	5.4
Total		8.4	Total		15.8	7.4

So the trade, as posted originally, was off by $7.4M. It looked like a good deal for the Blue Jays, which surprised many.

Values of players now, after adjustments:

Team	Player	Value	Team	Player	Value	Difference
Mariners	Hernandez	15.3	Blue Jays	Swanson	14.5
				Macko	4.6
Total		15.3	Total		19.1	3.8

Hernandez benefited from multiple adjustments: $/WAR, the additional value of a draft pick because the $/WAR adjustment up triggered a higher probability of a QO, and a slightly lower arbitration salary (because he lost his case). Swanson’s number was upgraded as a result of the reliever model update; and Macko was downgraded slightly in our sources’ prospect updates. The result is a deal that now looks much closer on paper than when originally reported – albeit still a positive one for the Blue Jays.

Meanwhile, the Luis Arraez/Pablo Lopez trade still looks like an overpay by Miami, but it’s also worth looking at again:

Values of players at the time of the deal:

Team	Player	Value	Team	Player	Value	Difference
Marlins	Arraez	26.6	Twins	Lopez	38.8
				Salas	20.5
				Chourio	0.1
Total		26.6	Total		59.4	32.8

Values of players now, after adjustments:

Team	Player	Value	Team	Player	Value	Difference
Marlins	Arraez	28.5	Twins	Lopez	38.8
				Salas	8.7
				Chourio	1.1
Total		28.5	Total		48.6	20.1

Arraez went up a tad due to the $/WAR adjustment; Salas was downgraded pretty heavily by prospect evaluators, which accounts for most of the difference (Lopez had already been adjusted for $/WAR). So it’s still a major overpay by Miami, but not quite as egregious as it first looked.

Now, you might think, why couldn’t we get these closer at the time of the trade?

The reality is, we have to read the room. We start each offseason with the data we have. But it’s incomplete. The model is dynamic – we are always calibrating it to the market. We bake in default assumptions for $/WAR, for example, based on history. But if we see it spiking, we’d be foolish not to adjust for that. 

And on the prospect side, we can’t force evaluators to adjust their timelines for us – and in fact, they often need the time to process information so they can get their ratings right. We’d rather be right and late than wrong and early.

So, in effect, each November, when the offseason starts, our numbers represent a first pass. As the hot stove gets going, once we get enough data points, we calibrate those in a second pass. As we get new prospect data, we update again. So it’s a process. If a trade happens early in that process, it might be a little off. The later the trade, the more likely it’s going to be close to our estimates, because at that point, the calibrations and prospect updates will have likely occurred.

Is there a better way of doing all that? Can’t we be perfect on the first pass?

No. We talk to front offices and agencies, and our impressions are that they’re engaging in a similar process. The free-agent market is fluid, which in turn makes the trade market fluid. Teams also use slightly different player evaluation models, so when they talk trades and compare packages, these will sometimes change in both the discovery and negotiation process. Overall, it’s an inefficient market, and we’re all adjusting to it as we go.