Forecasting 2017–18 NBA Win Totals
Simulating the performance of each NBA team
With less than a week until the start of the 2017–18 NBA season much discussion has been made over how each newly constructed “super team” will perform. Can Houston and OKC challenge Golden State in the West? Will Philly make the playoffs?
When Vegas announces its win projections for each team prior to the season, degenerate gamblers take notice and hot takes spew from all angles. In the basketball analytics community, though, the conversation is civil, but competitive. Predicting how each team will perform is equal parts art and science.
There is almost an infinite amount of sophistication that can be built into any win projection model. With the season so close at hand I wanted to build a projection system that maintained simplicity, but modeled the intricacies of an NBA season. First, I forecasted each player’s “value” — a blend of plus-minus statistics — based on their performance over previous seasons and adjusted based on players of a similar age and skill. I made adjustments for draftees based on draft position and mean regressed sophomore players, too.
A player’s variance in performance was also modeled. This accounts for the fact that elite players not only bring a greater value to the team, but do so consistently. Each player’s minutes were forecasted and manually adjusted based on injuries, acquisitions, etc. On a team level, each game the players’s contribution are simulated and the team value is adjusted based on home court advantage, the number of possessions in the game, and the number of days rest each team has preceding the game, based on the effect this has on team’s historically. After all of this, the season is then simulated 1000 times.
Both player and team projections are not perfect, by any means, but illustrate a unique picture of the upcoming NBA landscape.
Western Conference
The migration of stars to the Western conference has become one of the summer’s biggest headlines. This is ever-present in the forecast of each Western conference team. For both conferences the model appears to overestimate the value of the best teams and underestimate that of the lower teams, rather than making conservative estimates. There is no real surprise in the top four teams order, but maybe in the magnitude of their win totals. Time will tell, but it appears the race for the final four playoff spots will be incredibly intriguing.
Eastern Conference
While lacking the fire power of the West, the East will be an exciting finish with middling teams able to possibly gain a home-court advantage in the playoffs. The model extremely undervalues both Boston and Washington. Washington was simply not a darling of the plus-minus system used, with Markieff Morris being one of the most poorly rated player. However, with a simple injury to a top starter or another wild senario it would be of no surprise to see this team at their projected 42.6 wins.
Boston, however, is projected at over 12 wins below their Vegas (Westgate) line. This is primarily due to the fact that they are relying on numerous first and second year players, which the model devalues highly. Secondly, while the Isaiah Thomas-Kyrie Irving trade appears to benefit Boston in the long run, the analytic model suggest that it might be a step back in the short term.
Parting Thoughts
In time I would love to make many adjustments to this model. Namely, reducing the variability of the minutes projection (including looking at time played at different positions and injury history), incorporating draft projections, how trades and acquisitions affect a player’s performance (does ‘Melo take shots away from Paul George, probably), as well as more robustly modeling a player’s growth from year to year.
These changes will hopefully come in time for another NBA season, but for now let us all sit down and watch some real basketball, finally.
Thanks to basketball-reference for data. For any questions on methodology or for further conversation reach out to me at aepowell[at]gmail[dot]com or on twitter @aptigers12.
