Tuesday, August 30, 2022

How Much Trade Value Did the Rays Give Up in the 2022 Draft?

 While the Rangers shocked everyone by taking Kumar Rocker 3rd overall, the Rays made their own puzzling decision by taking Xavier Isaac, a high school first baseman ranked 113 by Pipeline and unranked by FanGraphs, in the first round. At the time of the pick, I assumed the Rays were planning on cutting an under slot deal for Isaac in the first, and going after another expensive prep player later. Instead, the Rays paid him full slot value and drafted mostly college players the rest of the draft. It’s hard to give a complete scouting report on Isaac; he was injured for the entire showcase circuit, but what I would like to focus on is how much did the Rays sacrifice (or gain) in trade value by making this pick. There are two ways to get value out of a drafted player, either by the player making the major leagues, or whatever you get out of trading the prospect. In this analysis, I want to focus on the latter, because the Rays have consistently been in the playoff hunt and will at least consider trading prospects to improve the major league team, so drafting with trade value in mind is fairly practical.


To start, I want to lay out what the Rays got on the first day of the draft. I will be quoting the prospects’ value in terms of FV given by FanGraphs. 


Round (Pick)

Name

FV (FanGraphs)

1 (29)

Xavier Isaac

35+

2 (65)

Brock Jones

35+

CB-B (70)

Chandler Simpson

40

CB-B (71)

Ryan Cermak

35+


In comparison, the 30-45 range on the FanGraphs board was valued as 45 FV, so the Rays passed up on drafting a good player to go off the board and select Isaac. One thing to note is that Eric seems to be a bit colder on Jones and Cermak than industry consensus, and therefore the Rays draft looks fairly weak. 


Let’s look at the 2022 trade deadline and see what type of players the Rays could get with 2 40 FV’s and 2 35+ FV’s (rounding up on Jones) and compare that to what they could get with a 45 FV, 2 40 FV’s and a 35+ FV. Looking at how FanGraphs broke down the deadline, with the 45 FV, the Rays could get someone like Gallo, a fallen star with an expiring contract, and a couple relievers/platoon bats with the rest of the class, or packaged 45 and the 40’s for Tyler Mahle, a mid rotation starter with 1.5 years of control left. In comparison, the best the Rays could have done with the current class is get a player like Syndergaard, a 4th starter on an expiring contract or Jorge Lopez, a pop up reliever this year who has been dominant, but probably settles into a 7th-8th inning role. As you can see, while measuring trade value is inexact, based on these assumptions there is a large difference in trade value that the Rays gave up to select Isaac.


Now, if Isaac has a strong first half of the 2023 season, the pre draft rankings on Isaac do not matter, especially since teams did not have much to evaluate off of and will be updating their beliefs quicker than usual on him. However, I think the opposite is in play here as well, where if Isaac is just mediocre or below average, teams will write him off quicker than usual. 


To give a hypothetical example, imagine a draft prospect’s rankings from each team as a distribution, so that for a 45 FV prospect, most teams rate the prospect as a 45 FV. Now imagine you are running a team. You are picking 15th, and can choose between Player A, the consensus 15th best player, and Player B, the consensus 50th best player. If you choose A, you are right in the middle of the distribution, whereas if you choose B, you are on the far right of the tail. When you are on the far right of the tail, very few teams want to trade for B for the same amount of value that you think he is worth, whereas that is not the case with A. Furthermore, it’s hard to benefit from variance when you take player B over A. For instance, if player B has a breakout performance, it brings the mean consensus rating to where the right tail used to be, and unless there are a few teams that suddenly value him higher than you pre draft, you haven’t gained any trade value from the breakout performance. In contrast, if player A breaks out and consensus rating increases, you get that increased value since you drafted him at consensus value.


In summary, it’s very difficult for the Rays to do well by drafting Xavier Isaac. A lot of things have to work out well for them to do better than the median value of the slot. However, if there’s one team that has made many people look stupid, it’s the Rays, and so I look forward to see how this plays out in the future and what there is to learn if this works out.


Sources:

https://blogs.fangraphs.com/ranking-the-prospects-traded-during-the-2022-deadline/


Saturday, August 13, 2022

Statistical Inference vs Statcast Models

Since I am currently working on a Stuff model (done in Stan, and hopefully publishing a writeup in their case studies documentation), I have been thinking a lot about the philosophical differences in models that are done with Statcast data versus models built with game level data.

In general, I feel like there is an implicit assumption that models built from Statcast data are always going to be better than ones built with just statistics. I think that viewpoint typically makes sense. Statcast data is more granular, which makes it more feasible to give individuals credit. Fielding statistics as a whole have benefited from Statcast data. It's really hard to properly attribute skill to fielders without granular level data, and new statistics such as Catch Probability and OAA are really impressive pieces of work that show how important Statcast data is and how much we were missing it beforehand.

However, I want to push back on the idea that Statcast models are always better than statistical models. I think this can be especially true in pitching, which may be surprising. Two places where Statcast models can be beat by a well calibrated statistical model is in deception and pitch mix interaction. Deception is something inherently visual and means different things to different people, which makes it hard to quantify. Furthermore we run the risk of overfitting to small data based on this, which can make the overconfident in their perceived edge against others and lead to ruin.

Asking if you want a good stats only model versus a good Statcast model is like the beer or tacos question. You want both, and ideally want them to converge to the same prediction. However, acknowledging the strengths and weaknesses of both approaches and not defaulting to one approach is beneficial in the long run.