The season is not done, but we do have nine total position players that all six predictions have covered. It might be good to think about how each of us has performed in order to look forward as the final players come trickling in to agree to terms. Now, a few notes: (1) I am only considering position players here because my BORAS model is an entirely different model for pitchers as opposed to position players and I am unaware that anyone else has that issue; (2) None of our systems consider deferred money, so that is the elephant in the room that I will remain pretending that it does not exist; and (3) this comparison is based on yearly salary as defined by total divided by years. In reality AAV is very much impacted by years, but I think with how the different estimates come about that years are an aspect that is not exactly as well drilled down than AAV. I would also comment that contract years is a prediction that all systems get right roughly the same with differences of at most a year.
The table below shows the current performance of each model and reports on the difference between the predicted and actual spending, the regression coefficient from Least Squares Fitting, and standard deviations related to that coefficient.
Jim Bowden's estimates has been solid as shown by his estimates missing on total by only 0.2% spending and the Least Squares Fit (you want that to be 1 as it is the slope of the relationship between the estimate and the actual AAV). His estimates have also been very internally consistent as shown by the standard deviations. The BORAS model comes in behind Bowden with overall accuracy, but is more middle of the pack with internal consistency. The others have dealt with overestimating the market and being a little bit more dispersed.
A major issue for Cameron, Dierkes, and Heyman may be that they overestimated the price per win. For Cameron, I believe he begins with the assumption that the cost is 8 MM per win with that cost increasing 5% each season. My own calculations assumed 7 MM per win and a 5% increase with each year. If that was the only issue, then we should see a 14% difference between our estimates and that is not the case. I do think though that the cost per win is likely part of the reason for the difference. For individuals like Heyman who likely do not cost cost per win, it would be more akin with him thinking in general that more money was out there for the taking.
Anyway, the offseason chugs along and below are a smattering of available free agent position players and how the three currently best performing systems view them.
|Howie Kendrick||2 to 4||13.4||15||13|
|Yoenis Cespedes||5 to 7||18.8||21||22|
|Justin Upton||4 to 7||16.5||23||20|
|Dexter Fowler||2 to 4||12.1||15||14|
|Austin Jackson||2 to 4||12.4||10||10|
The apple amongst these oranges is Justin Upton. BORAS thinks Upton is good, but is not really impressed with him. The model largely sees him as being pretty good, but not valuing that he has been pretty good incredibly consistently. BORAS seems to appreciate exceptional years more than being an extremely good lunch pail slinger. Upton could very well be a big enough miss for BORAS to knock the model down a peg. To a lesser extent that might be true of any eventual Cespedes deal.
What the above table does make me think is that perhaps if the Orioles choose to grab another starting position player that perhaps Austin Jackson might be a good one to grab. Him and Fowler are similarly valuable players, but Jackson has a stigma lying over him of which I am unsure why. He wound up being a glorified bench player with the Cubs at the end of last season, which is a considerable red flag or, perhaps, a considerable opportunity to find a starting quality outfielder on the cheap.
As expected, the Upton deal changes the landscape here a bit. With him pulling in almost exactly what Bowden suggested and well beyond what BORAS pegged, we can expect to see quite a shift in model accuracy. For the sake of my model, it will likely need to perform well on the remaining deals to get back close to Bowden's predictions.