PMVP 2018 NBA Draft Model 2.0
Version 2.0 of forecasting collegiate talent for the summer’s NBA Draft
As all but 8 NBA teams have left for a summer vacation, their front offices have turned towards preparing from the busy summer schedule of roster building and player acquisition. The NBA Draft is the league’s bedrock for talent procurement. Therefore, forecasting how collegiate prospects will preform in the NBA is an art and science paramount to every team in the final two months of draft preparation.
Several weeks ago I introduced the first version of my Player Market Value Prediction NBA Draft Model. The model was a pre-NCAA Tournament forecast of college prospects, but had a number of limitations. In version two of PMVP, now with complete season data, I model the projected market value of each player — the value of their first contract after their rookie deal (~5th season) as a percentage of the salary cap. This gives a realistic value of the overall value a player will have on all aspects of the game.
Version 2.0 of PMVP also uses a similar random forest model as half of the forecast, but additionally includes a Bayesian Additive Regression Tree (BART) model. The BART model contains a prior weighted by a player’s Recruiting Services Consensus Index (RSCI) and a team’s offensive and defensive capabilities before estimating a posterior distribution. The mixture of these two models helps to reduce variation and take advantages of the strengths of each model.
The BART model provides a posterior distribution for each player, this enables us to more uniquely compare players, which I hope to do further in future posts. For instance, we can compare Arizona’s Deandre Ayton and Duke’s Marvin Bagley III. Ayton rates slightly lower according to BART, but higher in the PMVP mixture, than Bagley. However, we can see that though Bagley has the higher ceiling, he also has the lower floor. Thus, if a team is pulling teeth between the two big men, Ayton would provide a more stable guarantee of success, while Bagley might be a gamble worth taking. Team’s all go through discussions of what their risk tolerances are and this model design will help them make decisions accordingly.
It is easier, with any reasonable modeling exercise, to find quibbles and discrepancies with your own beliefs — and, trust me, there are several players in which the model and I thoroughly disagree. But, we must understand the model’s strengths to help in decision making and flaws in order to improve future iterations. Players like Brandon McCoy, De’Anthony Melton (only using last season’s statistics due to ineligibility in 2017–18), Tony Carr, and Trevon Duval are likely too highly favored by the model while Miles Bridges and Mikal Bridges are slightly undervalued.
The model, while naturally valuing younger players more, also finds that rebounding and play-making translate well into a successful NBA career. Therefore, a player like Trae Young, despite his lack of size, record-setting turnover numbers, and defensive liabilities, projects remarkably well for his ability to create opportunities for both himself and his teammates. This can also be a flaw in the model, as it heavily values players who have a remarkably large role within an offense, and damages a player like Mikal Bridges who has spent much of his career as an extraordinary complimentary player on largely successful teams.
In conclusion, there is a lot of talent hidden in this draft that will make the NBA decision makers sweat as the calendar approaches the draft. The NBA Draft is less about finding a diamond in the ruff and more about shifting through the noise to find the correct signal. Player Market Value Projections are a beacon of light at best, and a discussion starter at worst. But just maybe this will be the year your team will find its diamond.
Stay tuned for more NBA and draft analysis, but in the meantime you can find me on twitter or email me at aepowell95@gmail.com.
Data from basketball-reference and kenpom.com.
