My two previous articles walked readers through the process of building an international draft model. Part 1 introduced the project, discussed some of the challenges posed by international data, and looked at the relationships between international competitions. Part 2 discussed the model itself. I laid out the process of choosing which variables to include, and listed some of the interesting relationships I have found between production in international competitions and success in the NBA. Today, I want to highlight some of the model’s historical outputs and see what it says about next year’s prospects.
Best “prospect” performances of all time:
One good sign looking over the data is that the model nails the most obvious cases. The top-5 single performances of at least 100 minutes belong to Malone, Jordan, Barkley, Bird, and Robinson all during the ’92 Olympics. In fact, the first twenty-two performances all belong to American NBA superstars playing in some variety of international competition. Twenty-year-old Vlade Divac’s impressive outing at the ’88 Olympics finally puts the run to a stop. If I raise the threshold to 200 minutes, the list is less comically USA-centric. The top ten performances now belong to; Durant, Paul, Pau, Pau, LeBron, Ming, Dirk, Sabonis, Sabonis, Dirk.
Looking at actual prospects:
It is nice to see that the model accurately labels a list of superstars, many near the peak of their careers, as guys with great NBA potential, but that is not the desired goal of this model. The goal is to identify superstars before their destiny is clear. To get a sense of how the model does here, I projected each player drafted since the late 80s using only pre-draft performances. This means that Dirk’s performance at the 2002 World Championships does not help inform his ‘Expected Win Peak” (EWP), but his performance in the ’96 U18 European championships does.
Here are lists of the top scorers to log at least 100 minutes, 500 minutes, and 1,000 minutes before draft day:
Delfino was never as good as his projection, and Ricky, Kanter, and Valanciunas are all likely to underperform theirs as well. Marcelo Nicola did not cross the ocean to put the projection to the test, but his overseas production never matched his impressive early statistical potential. That said, it looks like the model does a reasonably good job identifying top international talent before it ripens, in some cases even when many drafting teams did not. Expanding on these examples, here are the projections for all lottery picks who logged at least 100 minutes in international competition before their name was called:
There is a nice collection of hits and misses on that list. Both those shared with NBA teams (as reflected in draft order) and those outside the general consensus. The model would have advised against some famous busts like Tskitishvili, Vesely, and Yi; and supported some of the wiser top international selections like Yao, Pau, and Dirk. There are a lot more stories to tell in the retrodictions, but for now I will leave you to tell your own. I also make the rest of my retrodicted outputs available in my Google spreadsheet, so you are not limited to the above lists. The sheet is sorted by ascending draft slot and descending draft class. International projections are given in the “intl” column.
International men of mystery:
One major theme looking at international prospects is sample size. Many prospects are drafted into the NBA before they have played in any of the major European professional leagues or international tournaments. Typically, the hype-train gets up to full speed while these prospects are still kids battling in 2nd or 3rd tier leagues, or riding the pine for a top-tier club. Often the only data I have for young prospects comes from junior competitions where they are likely to play about 200-300 minutes.
The result is a ton of players drafted without much information to work with. This is clear looking at the distribution of minutes played before being drafted:
Small samples are a huge problem, and they likely help explain a number of the model’s misses. In addition to some of the above lottery cases, the model was also overly excited by small samples of players like Kosta Koufos, Marko Milic, Arsalan Kazemi, and Martynas Andriuskevicius. My model is far from perfect, but even if it was, I would advise caution when interpreting results based on a couple of games. That said, basketball is a surprisingly low-variance sport. With the exception of three-point percentage, production is stable enough to trust a small sample further than most would think is acceptable (a topic for another day). This is why you should not completely ignore model outputs for low-minute prospects. In fact, scores for players evaluated on less than 100 minute samples correlate with observed performance around 0.6, which is pretty shocking considering the correlation is only slightly higher at 0.65 for players with more than 1,000 minutes.
It is also important to note that sample size is not just an issue for statistical projections. Whatever challenges prospect obscurity poses for statistical models, it seems to hit other approaches to evaluation equally hard. For example, both the model and at least one NBA team were overly impressed with Darko Milicic after nothing more than a few well-timed performances (though to the model’s credit it would have projected Darko below Melo, Wade, and Bosh). Teams are routinely left making decisions on little more than physical profile and workouts. Front offices can be quick to discount 1,000+ minutes of mediocre collegiate production if a guy measures out with a 7’3.5” wingspan and a 38” vertical, so it should not be any surprise that some got excited when Yi Jianlian did exactly that while giving scouts little else to work with (other than impressive dominance of a folding chair). Making decisions about players based on a single tournament against inferior competition is difficult from any angle, but it looks like the model handles this situations better than I thought it would.
Dennis Schroeder, Giannis Antetokounmpo, and Dante Exum are the most recent examples of international prospects drafted on limited information. Unfortunately, the model is not particularly excited about any of them. Schroeder is pegged at a win peak of only 1.2, and so far this looks to be accurately pessimistic. Assessment of Giannis’ rookie season was inflated by his likability, but he is definitely on pace to outperform his 3.7 expected wins. Hopefully Exum follows Giannis’ lead and performs closer to his draft slot than his EWP of 4.6 predicts…
Speaking of recent prospects, I suppose that is what most of you came here for.
Recent drafts have been full of international players, and collegiate players with international experience. Until now I had been working with only an NCAA draft model and unable to generate projections for these players. I am excited to finally have my international model up and running. Here is how players in the ‘12, ’13, and ’14 classes shake out:
Note: Anthony Davis’ international experience actually came a month after he was drafted, but I included him anyway just to hammer home how good he looks.
Focusing on the most recent draft, Nurkic and Jokic look like huge steals, Saric and Capela look like they were drafted in the appropriate slots, and Tavares, Inglis, Micic, and Gentile all look like great 2nd round values. 2014 will likely be remembered as a great draft for international talent.
I also included NCAA projections for comparison where appropriate. I am happy to see some relative stability between the international and NCAA projections. Tyler Ennis, who looked much better at Syracuse, is the biggest exception to this.
Another nice feature of international data is that it gives us a useful peak into upcoming drafts much earlier than college data. Not only that, but especially in years like this where many top collegiate prospects played in a junior international tournament, we even get a summer preview of some top NCAA freshmen’s NBA potential.
Here is how the 2015 draft looks based on international production:
Mario Hezonja and Kristaps Porzingis have been touted as top-tier international prospects, and the model agrees with this assessment. Hezonja’s 8.5 EWP makes him the best international prospect since 2011, when the model got a bit overexcited by Kanter and Biyombo. Porzingis looks like a solid lottery pick. The model also appears to like the other top NCAA prospects who played in the Junior World Cup. Okafor, Towns, and Winslow all score well It will be interesting to see if these scores hold up as they start testing themselves against collegiate competition.
Reaching a bit further, there are also some players too young for the upcoming draft who have looked impressive on the international scene. Sixteen-year-old Dragan Bender’s performance at the Junior European Championships gives him the 4th highest rating ever for a player under 18 years old. He joins Rubio, Splitter, Deng, and Kirilenko in the top 5.
I would say don’t forget the name Dragan Bender, but I doubt you will.
Fellow junior Croatians Karlo Zganec and Lovro Mazalin also look like strong prospects. The young Turks, who have rattled off a series of championships at the cadet and junior levels, look to have a deep and underrated collection of talent led by point guard Berk Ugurlu. I also recommend you keep your eye on Arvydas Sabonis’ son Domantas, whose ’14 Junior Championships performance trailed only Dragan’s.