Bonus Boys
Plus an interview with a NBA truth seeker
This is The F5, a twice-weekly newsletter that helps readers be less wrong about the NBA. If you enjoy it, think its interesting, or want to learn how to make the visualizations you find in this newsletter (below), please consider a paid subscription!
One of my favorite stats to track over the course of the season is how often NBA offenses are in the bonus. On average, getting into the bonus is worth about two points of Offensive Rating. So the more time your team can spend in the bonus the more efficient your team will be overall.
In the chart below I’ve mapped out how often each team is spending in the bonus this season compared to last season
What Denver lost in spacing when they replaced Kentavious Caldwell-Pope with Christian Braun they gained in bonus equity. Braun, unlike Pope, gets to the rim and puts pressure on the opposing team to defend without fouling. This season, Braun is averaging double the amount of free throws per game compared to Pope last season. Additionally, Russell Westbrook’s flurry of activity contributes to Nuggets leading the league in bonus time, which neatly matches up with a decline in the Clippers bonus activity.
Meanwhile, I think one reason the Lakers have a top five offense despite their inherent roster flaws is by getting into the bonus early and staying there. The Lakers have a 117.2 Offensive Rating on the season, but that number balloons to 121.6 when they are in the bonus, according to pbpstats.com. Another way of thinking about this is that the Lakers play like an average offense when they’re not in the bonus but turn into one the league’s most lethal offenses when they are.
Lastly, the Bucks have an average offense this season and I think a big reason for that is that they don’t rack up easy points by being the bonus. As of this writing, they’re 27th in time spent in the bonus. That’s in stark contrast to last season where the Bucks spent more time in the bonus than all but one team (Orlando).
A refreshing interview with Kostya Medvedovsky, creator of DARKO
If you’ve spent any time on NBA Twitter (or its equivalents) you’ve likely come across a DARKO chart. DARKO, which stands for Daily Adjusted and Regressed Kalman Optimized projections, is the single most trusted all-in-one metric by NBA executives and is a favorite tool of every NBA nerd I know.
In basic terms, DARKO attempts to estimate how impactful a given player is just like any other alphabet soup all-in-one player impact metric. However, unlike its brethren, DARKO is a forward-looking projection system (like EPM is now). In practice, this means DARKO is less equipped to answer questions like, “who was the best player last season?” and is better at answering questions like, “are we sure this guy is good?”
Kostya Medvedovsky originally developed the idea for DARKO in 2015 while working as an antitrust lawyer. At first, DARKO lived online as a simple Google Sheet. Later, Medvedovsky was approached by Andrew Patton who volunteered to turn DARKO into an interactive web-app. Medvedovsky credits Patton for the wide-use of DARKO online and says DARKO wouldn’t have exploded in popularity the way it has had it not been for Patton. People love charts.
When he’s not developing DARKO, Medvedovsky works at a hedge fund doing merger arbitrage, which he says despite its name, “has less hedging, and less arbitrage than you'd expect.”
I talked to Medvedovsky about DARKO’s origin story, his idea to improve the NBA regular season, and whether Kevin Durant and Lina Khan are overrated or underrated.
This Q&A has been edited for length and clarity.
The F5: How do you explain what DARKO is in simple terms? In other words:
KM: At high level, it's a NBA box score and impact projection system. The big difference between it and most other competing systems is that it's built from the ground up to be a projection system, as opposed to a backwards looking indicator of how good someone has been. These are obviously closely related, but can lead to big differences in cases. It tries to remove as much noise and luck as it can from what's been observed, and return as close as it can to a projection going forward. I think it works pretty well at that, but your mileage may vary.
What's the backstory on DARKO? My understanding is that you were originally a baseball guy who realized that some of the lessons learned there could be applied to basketball.
Like a lot of people in sports stats, I was originally a baseball guy, with Rob Neyer being my real entry point into learning about it. Baseball is very "simple" in some ways, in that the bulk of the game can be reduced to a series of one-on-one matchups, and the sample size is huge, so it's very easy to run almost any analysis you want, provided you have the data. This is fun, but its also a bit boring after a while in my opinion. It becomes very easy to understand what makes players good. It's not that the problem is "solved", but it's somewhat less dynamic than other sports at least.
Anyway, so I was playing a high stakes fantasy baseball league with Dan Rosenheck (who runs the Economist's election model), and coincidentally, another league member was Jay Caspian Kang (then at Grantland). Kang was starting up a new salary cap fantasy basketball league and he invited me to join. I knew very little about basketball at the time, but I did know a bunch about fantasy sports, and one of the keys to fantasy sports is having a projection system you can lean on. Baseball has all these projection systems, like Nate Silver's PECOTA system, or Jared Cross's Steamer system, that essentially try take each player's performance from the previous seasons and try and extrapolate it to the next season using a combination of statistics and machine learning. So as I got into fantasy basketball, I was pretty surprised to see there was nothing similar out there for basketball. Rather, every set of fantasy basketball projections I could find was someone eyeballing each player's history and trends and taking their best guess. Some of these systems are actually pretty good, but they have a bunch of biases as well, so I wasn't fully satisfied with them.
So I initially built DARKO as a way to compete in Kang's fantasy basketball league, which is why DARKO has this heavy focus on box-score stats, and in particular is built around answering the question of how "real" are in-season box-score improvements. I was just trying to win a fantasy league.
With one or two exceptions, almost everything in DARKO is a concept, or a downstream of a concept I first saw employed in baseball, usually by Dan Rosenheck or Tangotiger. Without those two in particular, DARKO would not exist.
Has an NBA team or other entity every approached you to buy the rights to DARKO and make it private?
Yeah, I've had a few offers, both from teams, as well as other groups in the gambling space and elsewhere. I have a fairly intensive day job, so I've never really had the time to devote to betting DARKO fulltime, but my understanding is DARKO projections can move the lines on both spreads and props, so I've had a few people ask to buy DARKO data, but they wanted me to take down the site as part of the deal. I obviously understand that desire, but I'm also lucky enough that DARKO isn't my main source of income, so I've been reluctant to enter into agreements like that.
The closest I've come to actually making the site private has been offers from teams, who understandably would want me to take the site down if I took a job for them. A baseball team actually made me a very compelling offer a couple years back that would have let me keep the site up, but I wasn't able to make it work with the rest of my life (it would have required moving my family). I would very much like to work for a team someday, but the timing has never quite worked out there. Someday.
But generally, it's pretty important for me that the site be public, and not be behind a paywall. I think it helps with my credibility in talking about sports analytics to have it as sort of a "resume" piece, and I think a lot of people really enjoy the site and the charts and stuff, so I'd need to be pretty blown away by the economics to take a deal that requires me to take down the site entirely.
Have you heard of any NBA agents using DARKO as a point of reference for why their client deserves X amount of money on their next contract? If not, do you think agents should be doing that?
So I haven't heard of anyone using it as explicitly advocacy like this, but I have heard of NBA agents looking at DARKO internally to try understand which players are worth targeting as signings, or vice versa, which players maybe don't have very promising futures. And likewise, I am told DARKO is used by players/agents trying to understand their own leverage in contract negotiations, e.g., do they take an extension, or wait for free agency and test the market. DARKO has some tools that are pretty useful for this, like the longevity predictions, which tell you how long a player projects to remain in the league based on their performance profile and age, so I can see why that's important if I'm deciding whether to risk free agency or not.
The advocacy idea is an interesting one though. Maybe I should make some inquiries...
How did it feel when DARKO was named the most trusted all-in-one metric by NBA executives? Did anything change for you?
Yeah, this was big for me.
I built DARKO as a side project, essentially in between billable work as a lawyer, trying to scratch that sports analytics itch for me. And while it had some good traction with some groups before that article, things really blew up after. I started getting a lot more outreach from groups looking to leverage DARKO after that. There's a lot of analytics tools out there, with 50 different acronyms, and it was nice to have validation that the work I was doing was being noticed, not just falling into the void.
The other thing is that I think that poll validated the approach I've taken to making DARKO's architecture transparent (e.g., the landing page for www.darko.app is a long explainer of the model). I think the reason a lot of execs and others have responded to DARKO is not just the results, but the fact that they understand what DARKO is doing under the hood, so they know its resulting strengths and weaknesses. There are obviously some downsides to making the architecture open in terms of inviting competition, but I think on net, I've gotten far more acceptance by being transparent with ideas and research than if I'd taken a super secretive approach. That's been good for me, and I like to think it's good for the sports analytics ecosystem broadly.
Can you give me a preview of your Kevin Durant is the "most overrated player of the modern era" take?
Sure. This is an idea that's tough to truly prove out, since it depends who is doing the initial "rating" (that I'm saying is too high), but high level, this is a response to some takes out there calling Kevin Durant a top 10 guy of all time. The basic concept is simple - while Durant's box score numbers are wonderful, and he feels like a prototypical guy you build around, if you look at the various lineup-specific results, his units have actually been somewhat underwhelming. The starkest case here is looking at the Curry/Durant Warriors teams:
Curry and Durant together played at a +15.6 rating, which is great of course. But then you see, the units with Curry, but no Durant were still a +10.94 squad, and that's across an awful lot of minutes. That's still "best team in the NBA" sort of territory. The opposite however, of Durant but not Curry is a mere +3.14 (again, a lot of minutes).
So that's why when you do something like run long-term RAPM results, Durant scores well, but is in the realm of an ordinary superstar, and not some inner circle hall of fame guy. He also of course lacks a lot of the traditional achievements you'd want to see there. He's won one MVP, and despite winning two finals MVPs, I'll challenge anyone who says he was the best player on those Warriors teams.
I am also not really a fan of his career free agency choices. He's free to make those choices of course, but they obviously impact his legacy too.
You had one of the better ideas I've heard on how to make the regular season more meaningful. Can you tell me about that?
Why don't teams care very much about the regular season? It's because there's not much gap between a team's chances of going far in the playoffs if they win 52 vs. 48 games or something. Teams want home court, but player health is way more important to them, and there's often a pretty big cushion available, so teams will strategically rest their players. And that's the playoff teams. If you're out of the playoffs, you especially don't care about the difference between winning 28 and 32 games. You'd probably rather win fewer if you have your own pick. Another way of putting this is that teams' incentives in terms of regular season efforts are highly non-linear. There's lots of discontinuities and cliffs, where teams start trying to win, stop trying to win, or even actively start trying to lose.
So the idea is to try and add a linear component to counteract this. The NBA has 1230 games. What if we took $1.23B of league revenue from the national TV contracts, and instead of paying it out equally to each team, we paid each team $1 million for each game they won (taken out of the teams' share of BRI). So when a team is deciding how much they're going to try in some regular season game in late March, they have a pretty clear incentive (or rather, ownership has a pretty clear incentive) to try and win every game. There's no game that can be written off as "well, this doesn't matter for playoff seeding" anymore. There's always a force pushing a team to win every game.
Now, to be clear, this doesn't solve everything. Teams are still going to be weighing rest for the playoffs, and lottery seeding in making their decisions. You can't take that away. But when the difference between 48 and 52 wins is $4 million, then I would expect ownership is going to give management a pretty big push to try and win those games rather than having a bunch "DNP Rest" losses down the stretch. And likewise, it makes the cost of the sort of really egregious tanking we've seen pretty material in financial terms. Sabotaging your 32 win roster and turning it into a 20 win team is a no brainer under today's incentive structure. It's a very different picture if that sort of decision costs you $12 million. I think that sort of impact would be pretty helpful in terms of making the quality of the regular season game more intense.
Lina Khan -- overrated, underrated, or properly rated
Oh boy.
So I was an antitrust lawyer for about ten years, and had 15 minutes of fame for coming up with the term "Hipster Antitrust" back in 2017, which caused me to cross paths with Lina a few times. And I'm generally sympathetic with a bunch of where the current movement is coming from. I remember attending a talk by Jason Furman, Obama's CEA Chair in 2017, about "Has Antitrust Failed", which really opened my eyes to a bunch of evidence that consolidation was increasing across the economy, despite the efforts of antitrust enforcers. There's really good reason to have aggressive antitrust enforcement, and I think we've likely underenforced somewhat substantially. This is particularly true for non-reportable horizontal mergers (below around $100 million), which don't need to be disclosed to the FTC/DOJ, and can have pretty substantial anticompetitive impacts in local markets.
But antitrust enforcement isn't just a dial you turn up from "lax" to "aggressive" if you think there's underenforcement. The details matter, and I think that's where the current FTC has erred. They have ended up focusing hugely on vertical mergers and vertically integrated players, with disproportionate scrutiny given to tech, and large market cap companies. And that's not a surprise. Lina came to prominence by challenging something called the consumer welfare standard - the idea that the primary goal of antitrust enforcers should be consumers welfare or low prices. And at the time (in 2017), perhaps challenging the importance of consumer welfare made sense. It was a period of really low inflation, so there was perhaps some complacency around the importance of maintaining low prices. But there are tradeoffs to taking your eye off the ball of prices and focusing more on vague notions of corporate power and vertical consolidation. And it strikes me, especially in a period of high inflation, that those tradeoffs may not be worth it (if they ever were).
So when I see the FTC devoting huge amounts of resources to challenging deals like Meta's acquisition of Within, or Microsoft's acquisition of Activision, I have some skepticism that this is really the best use of FTC resources. Those sorts of court challenges are dubious on the merits, but also distract from other investigations that the FTC could be undertaking. We hear all the time about the FTC and DOJ's limited resources, and I'm broadly sympathetic, but when we actually see how they're using a lot of those resources, it calls into question the priorities. And I saw this when I was still practicing - I had smaller, less sexy deals with real potential competition questions go by without so much as a phone call from the Khan FTC, but had other deals garner major attention when it was a large market cap company doing a totally innocuous deal. So when you look at the net statistics, the total amount of enforcement is not actually wildly different than before, but a lot of the specifics are quite different, and I think mostly not in a good way.
I should add, I am a fan of a lot of what she's done on the consumer protection side (one click cancel, more pricing transparency on junk fees), and I'm fairly sympathetic with the logic behind her proposed ban on noncompetes, even if it seems like she may have exceeded the FTC's legal authority in imposing it.
What’s one thing you can’t live without during the NBA season?
This site is easily the best value in fantasy sports. When I spoke earlier of being surprised at the lack of NBA computer projection systems, and how they're mostly "hand curated" instead, one of the sites I had in mind was Basketball Monster. As near as I can tell, the way their projections work is that Josh Lloyd and a couple others there put together "per 36 minutes" projections for every player in the NBA in every relevant fantasy stat, and then come up with season-long minute projections. And then they update those, by hand, throughout the season. And like, it's pretty good!
I still actually use their minutes projections before the season as the basis for my win projections for teams, despite having a projection system at my fingertips. Especially for playing time stuff, it's hard to beat that much manual effort.
I'm just amazed at the effort that goes into the work product they produce, and that they mostly do a pretty good with such an approach. Before I built DARKO, I leaned on this site almost exclusively in my fantasy league, and was hugely successful as result. And even now, I still check the site every day, and in particular, their Daily Box score link which will provide little blurbs on each player's performance from a fantasy POV.
The box score format is so clean, the blurbs are easy to read. No ads. It's honestly one of the best places to keep up with the league even outside of a fantasy POV.
Loved the Lina khan section I learned a lot