Which NBA Teams Are Best At Drafting?

Results

Among the 720 draft picks analyzed, Ben Simmons stood out above the rest. In his debut season he had a Win Share (WS) of 9.2 and won Rookie of the Year (ROTY). On the other end of the spectrum, Thomas Robinson (5th pick in 2012 by the Sacramento Kings) ranks worse than all other draft picks, followed by Luka Šamanić (19th pick in 2019 by the San Antonio Spurs). While Robinson went on to become a reliable role player a few years after, the Sacramento Kings passed on Damian Lillard and Andre Drummond. The enormous opportunity cost of acquiring Robinson is the reason for his low score.

Despite skewness, a draft score of zero is most common - a fairly typical drafy score that is neither bad nor good. Many of these draft picks are second round picks with low expectations and correspondingly low value performance.

The Brooklyn Nets are the best team in the NBA at drafting quick impact players while the Portland Trailblazers are the worst. The Philadelphia 76ers rank second and, to no surprise, the Boston Celtics are also among the top teams in third.

The ten top draft picks in the NBA:

Among the best draft picks, Ben Simmons and Karl-Anthony Towns stick out as first overall selections. Their high score is primarily due to the poor performance of the players drafted after them combined with their own impressive stats.

Among the ten worst draft selections, they seem to be concentrated around the late lottery or right outside it.

The best selection by each team:

And the worst selection by each team:

Methodology

The central idea behind this analysis is relatively simple: to evaluate draft selections, players should be compared against the alternatives available. If better players were available, then the draft pick can be considered bad. If better players were not available - or were drafted too far back to have been realistically considered at that draft position - then the draft selection was a good one.

It doesn’t matter how well players drafted in that position normally perform - in fact, you’ll find this methodology sometimes grades players considered disappointing as decent draft picks because the players taken next were even worse. For example, Andrew Wiggins actually receives a positive draft score because Cleveland avoided drafting Jabari Parker (2nd overall), who was far worse.

Therefore, the performance of each player drafted is compared against a weighted average of the players drafted after. Players who are drafted closer are given higher weights than players drafted further away since they were more likely to have been considered as alternatives at that position.

Some adjustments were made around which teams draft picks were attributed to. While a player is officially listed under the team that drafted them, if they were traded on draft night or any time before the beginning of the following season, I attributed the draft pick to the team that they were traded to. Any trades after that point did not affect attribution.

In order to measure player performance, the advanced statistic of Win Shares (WS) was employed. Win Shares is a player statistic which attempts to divvy up credit for team success to the individuals on the team. Win Shares are calculated using player, team, and league-wide statistics and the sum of player win shares on a given team will be roughly equal to that team’s win total for the season. Moreover, any WS values below zero are replaced by zero. A player that played poorly should not be considered worse than one that never played at all.

To form the weights, I use a modified version of Brown’s simple exponential smoothing model:

\[V^C_{yd} = V_{y(d+1)} +(1-\alpha_d)V_{y(d+2)} + (1-\alpha_d)^2V_{y(d+3)} + \dots \]

Where \(V\) represents the counterfactual (\(C\)) value (\(V\)) at draft position \(d\) in year \(y\). Alpha represents the smoothing factor at \(d\), which must be between zero and one. A high alpha places more weight on the players drafted closer while a low alpha spreads the weight more evenly among many players. The alpha is multiplied by the actual Value (\(V\)) of the player taken at \(d + 1\) (directly after) in year \(y\).

However, the alpha is not consistent across draft selections within any one year (though it is consistent in the same draft position across years). In top draft spots, there is usually only a very small number of players that could conceivably be drafted. In later draft positions, there is a large number of players could be considered. This can easily be seen by comparing actual draft positions against ESPN rankings of the best draft prospects. In early draft positions, ESPN rankings closely align with actual draft positions. By the second round, there are often huge variations between the two.

Therefore, each draft position has a unique smoothing factor. First, I determined the absolute difference between the actual draft position and the ESPN best player ranking of the same player for every draft between 2014 and 2020. In the rare instance a player not ranked by ESPN is drafted, their ranking was recorded as 101. I then found the average for each draft position and regressed this against actual draft position. The Beta coefficient was 0.317, which is then input into the following equation to the determine the alpha for each draft position:

\[\alpha_d = \frac{1}{0.317d+1}\]

Which yields alphas that decline at later draft positions. This approach gives a considerably higher alpha for the very top draft picks but considerably lower alpha for those in the second round. The end result is that the scores of top draft picks rely heavily on the performance of the players drafted directly after while the scores of later draft picks are compared more evenly to many players.

The alpha is used to calculate the counterfactual performance for each player. By subtracting the actual value from counterfactual value, the Net Rating for each player is determined.

\[NR_{yd} = V_{yd} - V^C_{yd} \]

The Net Rating can be thought of as the “excess value” a draft selection generated against the likely alternatives available.

To make draft scores comparable across draft years, I use z-scoring by dividing the Net Rating of each player in each year against the standard deviation of the net ratings within that year:

\[\text{Draft Score} = \frac{NR_{yd}}{\sigma_{y}^{NR}}\]

As always, if you have a question or a suggestion related to the topic covered in this article, please feel free to contact me!