Who’ll Win the NBA Finals?
We’re 30 games into the 2019–2020 NBA season. It’s time to make some predictions. I ran the numbers using the Elastic Model.
TL; DR
- The Rockets are the most likely team to win the NBA Finals. The Lakers are the second-most likely team to take the trophy.
- The Warriors will finish the regular season with the second-worst record in the NBA. This could lead to a strong draft pick.
- Toronto and Milwaukee are the most likely contenders from the East. But neither team will stand a chance in the Finals.
Let’s dig into the numbers. All predictions are based on data as of December 21, 2019
A brief overview of the Elastic Model
The ratings are based on a statistical model that predicts the winner of match-ups based on 3 families of variables: roster, performance, and game circumstances.
No human opinion has been injected into the model or rankings other than feature engineering.
Let’s look at the predicted season win-rates
I ranked all teams based on their projected win-rates for the season using the Elastic model and the FiveThirtyEight predictions. Then I grouped teams into three buckets denoted by green, blue, and red labels.
- Green: cities where the two models are in agreement on seeds.
- Blue: Elastic model is more bullish on the team than 538.
- Red: Elastic model is less bullish on the team than 538.
(Note that lower numbers mean higher ranks.)
In the Western conference, both models agree that the Lakers will win the most games this season. But they differ on the #2 seed: 538 has the Clippers whereas the Elastic Model has Houston.
In the East, the Bucks are projected to win the most games by both models. But, again, the models don’t agree on the #2 seed: the Elastic Model has Toronto and the 538 model has Boston.
Here’s the raw predicted win-rates for all teams using the Elastic Model:
Most regular season wins ≠ championship
Let’s see what happens if we take the top 8 teams from each conference and simulate the playoffs. The graph below is a bit busy. Here’s how to interpret it:
- Each dot represents a 7 game series.
- Going from left to right takes us through the playoff tree, from the first round to the NBA Finals.
- The y-axis represents the confidence in the predicted outcome.
- For example: According to the model, Houston will beat Toronto in the finals with a 75% probability. The Lakers will play the Thunder in the first round with a 90% chance of winning.
Why the Rockets over the Lakers?
As mentioned, the model is based on 3 families of variables: roster, performance, and game circumstances.
Let’s double-click on the roster component:
- Players are grouped into 25 archetypes using k-means clustering based on their normalized boxscore statistics.
- The algorithm quantifies the benefits of allocating more game minutes to each archetype.
In order to summarize the strength of a given roster, I grouped the archetypes into three categories based on the magnitudes of their model coefficients: max players > super players > everyone else.
This shows that, even though the Lakers are going to win more games in the regular season, they’re top heavy. The Rockets have a deeper roster and just as many max players.
And, in case you’re interested, here’s the top 10% players according to the model. The x-axis reflects the archetype strength (there are 25 unique strength values, one for each archetype, but only 5 are shown here).
Final words
Given the latest available data, the model is picking Houston. I’m going with the model.
But it’s still early in the season and lots of things can change. According to the model, the conference finals will be near coin tosses, which makes the playoff simulations even more volatile.
To illustrate this point, here’s a simulation based on an earlier date where the Lakers end up playing Milwaukee in the Finals:
I’ll follow up with a new post after the allstar break.
