NBA win predictions 2017-2018

Adapting my NBA season model for 2017-18

Posted by RDJ on October 16, 2017

Header image is Baller by Gul Selz used under license CC BY-SA

I realize that I’m cutting it a bit close before the season starts. The league took a week away from me relative to last year! I’ve been busy, but I wanted to generated and publish playoff probabilities before the season actually starts.

Get straight to it

Once again, these come with decimals, because math is forever.

Eastern Conference

Team Projected Wins 2016-17 Wins Improvement Playoff Probability
Boston Celtics 51.2 53 -1.8 89%
Toronto Raptors 49.9 51 -1.1 87%
Cleveland Cavaliers 47.9 51 -3.1 87%
Charlotte Hornets 43.7 36 7.7 72%
Miami Heat 42.8 41 1.8 68%
Washington Wizards 42.8 49 -6.2 69%
Milwaukee Bucks 41.3 42 -0.7 62%
Detroit Pistons 36.7 37 -0.3 44%
Indiana Pacers 36.0 42 -6.0 42%
Chicago Bulls 35.2 41 -5.8 40%
Atlanta Hawks 34.6 43 -8.4 38%
Orlando Magic 31.4 29 2.4 23%
Philadelphia 76ers 30.7 28 2.7 23%
New York Knicks 30.4 31 -0.6 22%
Brooklyn Nets 30.1 20 10.1 20%

Western Conference

Team Projected Wins 2016-17 Wins Improvement Playoff Probability
Golden State Warriors 61.2 67 -5.8 98%
Houston Rockets 53.8 55 -1.2 89%
San Antonio Spurs 52.0 61 -9.0 85%
Oklahoma City Thunder 51.5 47 4.5 83%
Los Angeles Clippers 47.7 51 -3.3 72%
Denver Nuggets 45.0 40 5.0 62%
Utah Jazz 43.8 51 -7.2 56%
Minnesota Timberwolves 41.8 31 10.8 49%
Memphis Grizzlies 41.1 43 -1.9 46%
New Orleans Pelicans 41.1 34 7.1 45%
Portland Trail Blazers 38.0 41 -3.0 33%
Dallas Mavericks 35.2 33 2.2 25%
Sacramento Kings 32.6 32 0.6 20%
Los Angeles Lakers 30.6 26 4.6 12%
Phoenix Suns 29.9 24 5.9 11%

The Method

These models are slightly different than what I used last year. I designed these models for a different project, focused around estimating playoff probabilities given the possibility of injuries or other correlated shocks. The simulation involves stochastically simulating player quality and minutes, so every team has a high variance around their actual performance.

Only 61 wins for the Warriors? Say what?!?

Yeah. That’s a low number. It’s much lower than I expected. I personally think there’s a value to continuity. The Warriors kept all their key players and added a couple of other handy pieces.

The model thinks the Warriors are by far the most talented team. But it considers them as a worse team than last year, probably in large part because of how bad Nick Young’s BPM has been over his career, and how many minutes he plays. If he plays a lot, and takes any minutes away from any of Golden State’s far superior players, this team might not be quite as good. I think there is a great chance his BPM will look a lot better on the Warriors, and his minutes end up vastly reduced from last year. The model doesn’t know any of that. It also doesn’t know that Steve Kerr and the Warriors can always backtrack away from using him, and revert to their existing fantastic line-ups, if need be.

Similarly, adding Derrick Rose, combined with aging, makes the Cavaliers look a lot worse. For them I see less options.

However, Nick Young is not the only reason the model predicts the Warriors will come in under expectations.

The West got better

After the Warriors, the Thunder and the Rockets are the next best teams in the league, by minute-weighted BPM. The Spurs, Nuggets, and Clippers also land in the top 10. This summer Paul George, Carmelo Anthony, Jimmy Butler, and Paul Millsap all went West, with only Gordon Hayward going the other direction. Not to mention good supporting pieces like Patrick Patterson and P.J. Tucker, who were pried away from my Raptors.

The model likes averages

Minimizing loss and adding in high variance to both player quality and playing time results in few outliers. Teams aren’t projected to win 70 games or 10 games, because for any given team that’s (usually) unlikely. The distribution of actual wins is always wider than the distribution of predictions should be. For example, the model sees Boston as a much improved team, yet still conservatively predicts a drop in wins. Likewise it sees the Knicks as a highly degraded team due to Carmelo’s departure, yet still predicts a similar quantity of wins. It’s not a very exciting model.

Some of these teams will be much better than expected, and some will be much worse, but right now we don’t know which.