NBA: A new player-based model for handicapping games

1073

A new player-based model for handicapping NBA games

Those who handicap the NBA on a daily basis know how insane the grind can be. Most days are spent pouring through injury reports and news stories, trying to find out the latest on which players will and won’t play. When you find the information it is hard to calculate the value of certain players outside of stars like LeBron James or Nikola Jokic.

 

So, during the offseason, I went on a quest to work on a solution. I was inspired by bettors I have interacted with on social media like Raheem Palmer (The Ringer) and Matt Meiselman (Value Betting) who have player-based models that can account for individual absences. It is an invaluable tool to use throughout the regular season.

Those familiar with my work on the website know that I spent the summer learning how to craft a rudimentary NFL model that I am currently using for the SuperContest. I decided to use some of those principles to work on my own player-based model for the NBA season.

Before diving into the model itself, I want to acknowledge Andrew Mack, who wrote the two books that helped me build both Jon Von Model 1.0 (NFL) and Jon Von Model 2.0 (NBA). Those who wish to get into building models should check out his books Statistical Sports Models In Excel Vols. 1 and 2 as they were resources I leaned on quite a bit in building these out.

I would like to note that much like the NFL model, this is an experiment. This model is the first attempt at a complex process, much like Jon Von Model 1.0 is for the NFL. This exercise aims to learn about model building and maintenance throughout the season, not to get rich by firing off locks. If anyone reading this piece today learned anything or was inspired to start a model themselves, then this experiment has done its job.

With that out of the way, let’s dive in.

The Model

Jon Von Model 2.0 is a player-based model that heavily weighs Box Plus/Minus (BPM) as a factor to give each player an individual value to a point spread. For those who do not know,  BPM is a metric on Basketball Reference, and it is defined as a basketball box score-based metric that estimates a basketball player’s contribution to the team when that player is on the court.

Right off the bat, I want to note that I believe BPM to be a somewhat flawed statistic to build a model on, as it uses information from the traditional box score and nothing from play-by-play or tracking data. I am using it for this model because it is a simple catch-all statistic that can at least give us numbers that are somewhat in line with the betting market. As the season progresses, I will experiment with different versions of Jon Von Model 2.0 and try to build out different models based on other stats.

A model in this format is a must for the NBA. Jon Von Model 1.0 is a team-based model that factors in the quarterback play. It is based on regression testing and finding a correlation to a team’s home margin of victory. That was the initial path I went down for this model, but the results were all over the place. Correlation kept coming back well under 0.5 and the randomness of the NBA season made it clear that basing a model on teams rather than individuals would be a waste of time.

Jon Von Model 2.0 projects the percentage of minutes a player will be on the floor and assigns those players a spread value based on the BPM per minute played and expected minutes. That process gives us a player’s projected value to the spread. Once that value is found for every player, they are added together to give a team an estimated power rating. The process from there is simple. Find the difference between the two teams’ rating, factor in homecourt and spit out a projected spread for a game.

The Projections

I will post the game-by-game projections once we get closer to the NBA regular season. For the sake of this column, I figured it would be worth posting how Jon Von Model 2.0 has the teams rated overall.

Jon Von Model 2.0 is somewhat in line with the betting market with the top two teams in the NBA. The model differs because it clearly gives Milwaukee an edge on a neutral of 1.045 points. Damian Lillard and Giannis Antetokounmpo are both very strong when it comes to BPM, as is Brook Lopez, so the model is slightly higher on the Bucks as we head into the season.

Two other ratings stick out when reviewing what Jon Von Model 2.0 spits out: Philadelphia and Denver.

Joel Embiid was second last season in BPM (9.2). He is one of the most individually valuable players when you account for minutes played, and he also has a very good team around him if it is actually whole. James Harden was factored into these projections, and he finished 18th in BPM last season. Those two alone give the 76ers an extremely high rating.

Denver is rated as the fifth-best team in the NBA by Jon Von Model 2.0 despite having Nikola Jokic who led the league in BPM, but it speaks to the turnover on the Nuggets’ bench. Reggie Jackson, Zeke Nnaji and Peyton Watson are all expected to play large roles for this team off the bench, and their BPM numbers have weighed down this team’s rating. Christian Braun is also being rated as last year’s version of himself — something I must learn to fix — which also waters down the projection. I expect their rating will improve as the BPM numbers grow for the players taking on larger roles.

One last note on these projections: There is clearly a line of demarcation between the worst teams in the NBA and the rest of the league. Indiana projects to be about 3.1 points better than Washington on a neutral court. That difference is wider than that of Washington and Charlotte, the worst team in the NBA according to Jon Von Model 2.0, which is only 2.11 points.

Conclusion

This is going to be an arduous journey this season. Building this model was much more challenging than building Jon Von Model 1.0 for the NFL. This is also going to be a model that I believe is going to have a much lower success rate.

The NBA is a brutal sport that has an insane amount of variance throughout the regular season. This model attempts to predict those outcomes by assigning numerical values to players through a single statistic. It will not account for scheduling situations or motivation, so there are sure to be stretches in which Jon Von Model 2.0 will take it on the chin. Still, this is a fun experiment and a learning experience that will make me, and hopefully you,  better as a bettor when all is said and done.