A Computational Approach to mitigating tanking in the NBA

Inspiration

The NBA Draft serves as a lifeline for bottoming teams, replete with fresh young talent that they hope can bring them back to relevance. Unfortunately, while attempting to ameliorate the prevalence of tanking, the lottery system as well as its recent revision has somewhat contributed to perpetual tanking: the 2019 lottery is a prime example. Developing, young, but poorly constructed teams such as the Atlanta Hawks, Chicago Bulls, and the Phoenix Suns, have looked to the draft to build their teams, but were only able to land the 8th, 7th, and 6th picks, respectively. Meanwhile, teams like the Memphis Grizzlies and the New Orleans Pelicans landed the 2nd and 1st overall picks in the draft, which is a testament to the infallibility of the system for the simple fact that these two teams were the ones tanking: the Pelicans rested their star player for 4th quarters for over half the season, and as soon as the Grizzlies entered a mid-season slump, sent away their franchise cornerstone, Marc Gasol, and were simultaneously actively shopping Mike Conley Jr. We seek to determine an effective win-loss record that accounts for tanking throughout the season. This adjusted win-loss ratio will be utilized to determine drafting order. Hypothesis: We posit that by determining a quantitative measure of tanking and non-tanking games the NBA can better calculate the true win-loss ratios for teams near the end of a season. Furthermore, we predict that games where a given team has lower average lineup age, minutes played of the best lineup, offensive/defensive rating, and Pythagorean win shares, can generally be classified as a tanking game.

A shortcoming is that teams can be penalized for poor performance in the beginning of the year. Some of the applications, as previously states is that we could institute a new draft system without lottery or reformed odds that take into account tanking. The league can take action by punishing significant tanking with the loss of draft picks. For future work, we could add more features and do more case studies based on expert testimonials to refine model.

Built With

• jupyter
• python
• tensorflow