23 October 2017

Cup of jO's: BORAS Contract Pitching Model Accuracy

I have seen some off-site comments speaking negatively about the perceived accuracy of the BORAS pitching model contracts.  Now, these comments do not find their way here, so I have yet to really address them, but briefly.

2015/16 Model: R2 = 0.81
2016/17 Model: R2 = 0.87
Overall: R2 = 0.84

What does all of this mean?
Half of the contracts projected by BORAS to be above 8 MM per year are accurate within 1.7 MM.  The entire range of contract project misses are right on the mark to a miss of 7.1 MM (Jeff Samardzija).

Each year, the model is new.  Each year, the model assumes a certain growth rate of contracts.  Recently, there has been no growth in contract money, so the model projects cost per win to not go up this year.  The model also does not do anything to make up for lost time due to injury.  In other words, it would value a pitcher who lost a year to an injury to be equal in value that particular season as a pitcher who logged in 200 IP of replacement level work.

The intent of the model was to eliminate potential biases and see how close a model that knows very little can hit the mark.  So far, each iteration has done well.  We shall see what this season brings.


 ----------
EDIT

I have seen in some insulated message boards and argument about my use of AAV as the way to look at this model.  The reason why I use AAV and not total contract is because it is less applicable.  Putting the two together, from my perspective, is more or less an academic exercise because we get things like a 1.2 year contract.

However, if you do choose to go that route and want to see the accuracy of a contract with terms like 1.2 year at 12.6 MM per year, then for the pitching models they have a cumulative r2 of 0.87.  The is a better fit than the AAV as you would expect by adding a second meaningful variable.  However, I do not see 0.87 being meaningfully different than 0.84 in this context.

2 comments:

Pip said...

Jon as I understand it, you're a MD and a PHD in statistics. Where did you get your PHD?
( just wondering, it's a very impressive achievement)

Jon Shepherd said...

Nope.

PhD in Toxicology from UMD at the Medical Center. Masters in Environmental Toxicology from Clemson.

I do not really emphasize my training. Some people insist on being called Dr. such and such, but I find it pretentious. I would reiterate that there are many ways to acquire knowledge and mine was just one path. And really the only major way it contributed to my life was that it taught me to search for and accept reality as opposed to what I wish was true. I think that lesson is the most important one and it does not require 2 years in a masters program and a shade under 3 years in a doctoral program to learn that.