Nylon Calculus: Projecting the future value of Dewayne Dedmon

Dewayne Dedmon recently exercised his player option with the Atlanta Hawks for the 2018-19 season. He’s not exactly a household name for most people — he’s been playing about 57 games per year and has jumped from team to team. Dedmon went undrafted in 2013 and has since played for the Golden State Warriors, Philadelphia 76ers, Orlando Magic, San Antonio Spurs, and the Atlanta Hawks, his current team.

This past season was his best yet, averaging career highs in points and rebounds, making the most of the career high in minutes that he was given and looking like a useful big at both ends, protecting the rim and stretching the floor. It was the perfect year to play well and if he could build on it, Dedmon could be in line for a big new deal next summer.

I was curious to know if this was a good decision on his end and wanted to use an ARIMA model to see if I could predict how well he can perform in the immediate future.

ARIMA Model for Dewayne Dedmon

ARIMA stands for autoregressive integrated moving average, and for the sake of simplicity, I won’t go into the deep end on the concept. Instead, I’ll use Dewayne Dedmon as a case study in order to see how an ARIMA model predicts his future NBA performance.

I’m using Dewayne Dedmon because his career stats aren’t totally consistent, meaning that we don’t really know much about him considering the relative inconsistency in his sample. However, it still might be possible to figure out using the ARIMA model because of the functionalities available that can help stabilize the data, like differencing.

ARIMA is a forecasting tool typically used to determine future stock performance, and it provides reasonable results based on a ton of data.

To understand ARIMA, imagine a flower shop owner is trying to get a sense of how his rose inventory will sell in the coming year. He’d bring out the books and review rose sales over the past twelve months to set generalized expectations. Sales records can also show timely trends like rose sales during the week up until Valentine’s Day, or how many arrangements of roses were sold during the summer wedding season.

He could also find wacky instances of bad weather messing up rose drop-offs to his shop, or the instance of the over-zealous and desperate high schooler cleaning his roses out during the week before prom. It’d be smart to prepare for another bad rainstorm sidelining transportation, but he could reasonably expect that the senior high schooler won’t be back.

It isn’t any different than trying to predict the future performance of a basketball player. We can analyze past data and set reasonable expectations for the future while weeding out craziness outliers. Today, this process can be statistically modeled. While the acronyms of different statistical models seem daunting and pretentious, these models are merely trying to use past data for analysis to improve the model accordingly.

The Method

I used Dewayne Dedmon’s career game log data from Basketball-Reference.com and used the R programming language to help forecast future performance (ARIMA). The first step is to reign in the inconsistencies during Dedmon’s career, which is done by testing for stationarity and differencing the time-series if it is non-stationary. Further analysis is done by comparing and contrasting Autocorrelation and Partial autocorrelation functions to confirm stationarity, and to finally determine the parameters of the final ARIMA function.

For the sake of simplicity, I’ll go ahead and post the before and after on Dedmon’s points results. The table at the end can provide a rough picture of what to expect from Dedmon moving forward.

Before:

The data above jumps a bit all over the place, and this is largely due to some instability Dedmon experienced over the course of his career. He didn’t get much playing time until he came to San Antonio and became the de facto starter in Atlanta. At this point, I had to determine that this time-series was non-stationary, and had to difference the process. It was arduous, and I did have to adjust for seasonality, but the upper and lower bounds of predictive output look reasonable.

After:

I was able to get a model where the bounds didn’t stretch to high noon and zero. The green line is just a reference to show where the blue line lies in regards to 10 points per game, and the prediction looks favorable. Based on the past data available, the model predicts that Dedmon should continue to average a little over 10 points per game.

The blue line extends about a month into the next season, as you can see the further the blue line gets away from its own history, the more flatlining starts to occur.

Look at the rest of his projections from similarly fitted ARIMA models:

The table above shows the final instances of the lower, middle, and upper bounds of the predictive model. In short, the model says with confidence that Dewayne Dedmon can be expected to average about 10.7 points and 8 rebounds per game.

Based on the model’s favorable view of Dedmon’s future production, I’d lean towards feeling he left some money on the table. Considering the politics in free agency, one really doesn’t know what’s going on behind the scenes, but I don’t think it’s out of line to think that Dedmon could reproduce 10 points per game. The stats above are also not the only ones that matter, but having a center on the roster with some touch around the rim is extremely valuable.

He’s opted in where there’s an up and coming center in John Collins, and with an organization as stable as the Atlanta Hawks, it’s difficult to assess if he’ll be assured of playing time. Thinking that Dedmon could’ve waited until after Clint Capela signs his own first offer sheet isn’t unreasonable, but hey $7.2 million for one year is still a bunch of money.

Next: Nylon Calculus: How much do the Heat need Wayne Ellington?

The model could use more data — think of stocks — but by the time there’s enough Dewayne Dedmon will be close to retirement. Forecasting a player like Lebron James or Chris Paul would be a lot easier, in that there’d be less differencing needed in order to get a stationary process for the ARIMA model.

Also, take note on the upper and lower bounds on plus-minus or steals. The plus-minus largely reflects his career at vastly different organizations. He’s had pretty good games in San Antonio but struggled to get off the ground in Philadelphia. His future projections in the steals category are unrealistic because he doesn’t get many steals anyways.