Draft Signing Bonuses Pt. 2

            Today we’ll be looking at some specific example from our signing bonus model and discussing some more potential long term improvements that we may consider in the future. As stated before, one of the benefits of our Bayesian approach is that we can make a distribution of predicted bonuses that better allow us to encompass the inherent uncertainty that comes with the MLB draft. To demonstrate some examples of this, we’ve included some graphs below with predictions on both hitters and pitchers. The histogram and density plot show the distribution of predictions for one specific player, with the green line marking the actual bonus the player signed for.

Here we have two examples. One is JJ Bleday, the 4th pick in the 2019 draft who signed for $6.67 million. We see that he certainly had higher prediction than Bader (notice the x axes have different scales), but overall we still way under predicted what he ended up signing for. Harrison Bader on the other hand, the 100th pick in 2015, had predictions most dense around his actual bonus of $400,000.

Here we again compare AJ Puk, the 6th overall pick in the 2016 draft, to Daniel Mengden, the 106th pick in 2014. Again, we see that we’re usually underpredicting the very top of the draft, but our predictions seem to be pretty accurate after that with just these couple of players.

            Going forward, we have a few different revisions that we could consider. As stated before, conferences could be better adjusted for by seeing how certain stats are impacted by conference or better weighting certain conferences and/or strength of schedules. We also will probably want to include park factors down the road.

            Another possibility is to expand our model to more rounds of the draft and see how it performs when high end players are less likely. On top of this, we also may want to see if scout ratings are worth including (they likely are). For now, we’re interested in how stats impact bonus, but at the end of the day we’re interested in a model that does the best job predicting bonus, however that is. That may mean we could try a second version including scout rankings and/or grades on certain characteristics. This would almost certainly come from a public website.

            One more major improvement could simply come from including data from previous college seasons and summer ball. MLB teams will always be ahead of me since they’ll have a plethora of Trackman data and better development of their models, but even just getting more years of college data could help explain certain players. Overall though, this method of breaking down draft eligible players certainly seems to be superior to traditional approaches and allows us to better understand all the many possibilities that could occur in the draft.