The NBA, it is often said, has no offseason anymore — and so, we are determined to keep pace. That’s why we have already reloaded the FiveThirtyEight NBA prediction model to include updated projections and probabilities for the 2019-20 season, three months before the regular season actually tips off. Because the NBA never stops.
What’s new in this year’s version? Just like last season, our predictions are driven by a multi-step process, but this time we have included a new defensive metric, a boost for certain players in the playoffs and more. Here’s an overview of how the system works:
- Players are projected using CARMELO, our forecast system that uses comparable players from NBA history to predict a player’s future career arc. This year, CARMELO incorporates DRAYMOND, our new defensive metric that better accounts for a player’s shot defense, in addition to the existing metrics that look at a player’s impact on team defense while on the court. All of the fancy projections boil a player’s contribution down to the expected number of points he’ll improve a team per 100 possessions on offense and defense, relative to an average player. It’s important to note that we plan to add more depth to these ratings before the season proper. Specifically, we want to overhaul the box score-based aspect of our blended stat, but we’ll have more on that in the coming weeks and months.
- Player projections are then compiled at the team level using a depth-chart algorithm that assigns minutes to each player at each position based on both past positional usage and a team’s rank-ordered preference for using its players at full strength. The composite of each team’s individual player ratings forms its offensive and defensive projections, which then feed into its CARMELO rating — an Elo-like representation of team strength in which an average team scores as roughly 1505. This offseason, the algorithm is authorized to assign minutes to a generic “replacement-level” player at any position if a roster is incomplete, a wrinkle we don’t use during the season itself.
- These depth charts and CARMELO ratings are then generated on a game-by-game basis for the entire season, factoring in who is available each game because of injuries, suspensions or rest. One new feature this season is a “load management” setting to account for reductions to regular-season minutes for players such as LeBron James and Kawhi Leonard as they conserve energy for the postseason. For playoff games, we generate different depth charts in which top players play more minutes; we also added a new feature to give players with a demonstrated history of playing better — or worse — in the playoffs a special boost or deduction to their base ratings.
- Finally, the game-by-game ratings are combined with adjustments for home court, travel distance, rest and altitude to form win probabilities for each game, which feed into our season simulations. Specifically, we get our projected records and team odds by running 50,000 simulations of the schedule.
In honor of our 2019-20 interactive officially launching, let’s take a tour through the predictions that have emerged from this way-too-early edition of our forecast:
Western Conference
The West looks absolutely insane next season. All but four of the conference’s 15 teams are projected to have a .500 record or better, and the average full-strength CARMELO rating for Western Conference teams is 1580, which equates to about 49 wins of talent in a vacuum. Our pick for the top seed — with a surprisingly comfortable margin over the rest of the conference — is the Rockets, who got better after snagging Russell Westbrook in a recent trade. A grand total of three projected wins then separates each of the next five teams in the West pecking order: the Nuggets, Lakers, Warriors, Jazz and Clippers. Out of that group, the Clippers, Warriors and Lakers have superior playoff projections thanks to better top-end talent (hi Kawhi, Steph, LeBron and AD) and postseason experience on their rosters. The final few playoff spots in the conference should be up for grabs, among the mainstay Trail Blazers (whom our model consistently underrated last season), the up-and-coming Mavericks and Pelicans, the ever–puzzling Timberwolves or a few fading postseason relics in the Thunder and Spurs.
At the bottom, CARMELO thinks the Suns have improved quite a bit from last season, anticipating Phoenix to leap from 19 wins to 36 on the strength of better play from Devin Booker and DeAndre Ayton, plus the acquisitions of better-rated talent such as Ricky Rubio, Dario Saric and even underrated big man Aron Baynes. It doesn’t have much faith in the Kings to replicate last season’s breakout; in fairness, it never had much faith in Sacramento to begin with. We’ll see if the Kings keep defying our computer. And finally, it is no surprise to see the Grizzlies in the basement after trading away franchise cornerstone Mike Conley.
Eastern Conference
The Eastern Conference will be less of a bloodbath than the West, though the battle at the top should still be fierce. Similar to the Rockets out West, the new-look 76ers are CARMELO’s pick for the No. 1 seed by a decent margin over the Giannis-led Bucks. Behind them, the Celtics, Heat and defending champion Raptors form the next bloc of teams, and all have at least an 86 percent chance of making the playoffs despite plenty of offseason roster shakeups. The Magic and Pacers round out the group of solid playoff picks in the conference. Orlando is building off of last season’s surprise playoff berth, while Indiana is hoping to bounce back after losing Victor Oladipo to injury for most of 2018-19. And the Nets, Bulls and Pistons each have about a coin flip’s shot at some of the last few playoff spots — which might be a shock given Brooklyn’s multiple high-profile offseason moves.
After that trio of teams, the dregs of the East are pretty clearly delineated. The Hawks showed promise last season but are still some distance from contending; the Wizards and Hornets are trending down (particularly with John Wall injured for Washington and with Kemba Walker saying goodbye to Charlotte in free agency); the Knicks are, well, the Knicks; and the Cavaliers are still deep in rebuilding mode.
Although there is always room for surprises, the East playoff field seems more certain (aside from the few teams fighting for the No. 8 seed) than the West. Eleven different Western Conference teams have at least a 36 percent chance of making the postseason, including five teams between 36 and 55 percent in CARMELO’s playoff odds. That’s just one part of an NBA landscape that should be fascinating to watch, even during the regular season. For now, we consider the Sixers and Rockets our NBA championship co-favorites, but that could very well change — and knowing this crazy offseason, it probably will.
As difficult as it is to remember, we were lamenting the Warriors’ runaway dominance at this time last year. But after a series of unforgettable signings, trades and injuries that reshaped the league, this should be one of the most wide-open seasons in recent memory. And our prediction interactive can help you follow the race for the 2020 championship at every step along the way.
Check out our NBA player ratings.
Also known as the Career-Arc Regression Model Estimator with Local Optimization.
Also known as the Career-Arc Regression Model Estimator with Local Optimization.
Also known as the Defensive Rating Accounting for Yielding Minimal Openness by Nearest Defender.
Also known as the Career-Arc Regression Model Estimator with Local Optimization.
Also known as the Defensive Rating Accounting for Yielding Minimal Openness by Nearest Defender.
For example, a full-strength Brooklyn Nets squad would feature Kevin Durant at No. 1, followed by Kyrie Irving, Spencer Dinwiddie, and so on.
Also known as the Career-Arc Regression Model Estimator with Local Optimization.
Also known as the Defensive Rating Accounting for Yielding Minimal Openness by Nearest Defender.
For example, a full-strength Brooklyn Nets squad would feature Kevin Durant at No. 1, followed by Kyrie Irving, Spencer Dinwiddie, and so on.
At the moment, we are still using the 2018-19 schedule because the 2019-20 version hasn’t been released yet. When it is, we’ll switch the schedule used for our simulations.