At the 2020 NFL Draft, Bill Belichick did what he usually does: He traded down. Like clockwork, the New England Patriots head coach has a tendency to move down in the draft in exchange for a few extra picks, a savvy move considering the random nature of the NFL draft.
The NHL Draft isn’t nearly as random as the NFL’s, but there are inefficiencies that can be exposed and an opportunity to glean value. As it turns out, trading down to acquire extra picks was measured to be the optimal play at the NHL Draft, too. Every pick is a lottery ticket and the more a team has the better their chances of finding an NHLer.
Advertisement
That ideology was based on figuring out the value of a draft pick – a fundamental research topic in any sport. Belichick probably went by his own intuition to figure things out, but the idea is supported by math and applies similarly to hockey.
The most well-known and oft-used draft pick value model was created by Michael Schuckers in 2011. In his work, Schuckers based the value of a draft pick using data from 1988-1997, seeing the probability of a draft pick playing 200 games and how many games a draft pick ended up playing in the NHL. In 2016, he then improved that by looking at ice time over a player’s first seven years post-draft – the amount of time a player is under team control (something I believe Dawson Sprigings, now with the Colorado Avalanche, first suggested here). It was foundational work that shaped how analytically inclined fans saw the draft, work that I used every year at draft time.
In 2013, Eric Tulsky, now with the Carolina Hurricanes, made his own draft model, except he looked at how draft picks were valued in trades, creating a marketplace baseline. It allowed a direct comparison between how front offices viewed their picks and how those picks historically panned out according to Schuckers’ work. It showed there was a lot of room to exploit the market to gain positive value. Back in 2015 at Hockey Graphs, Garret Hohl showed just how much by comparing the two models.
That remains true in most cases, but it may not be to the degree that was originally shown by Schuckers and others who created models. Draft pick value has become collective knowledge in the analytics community, but it’s important to check and confirm that those hypotheses still hold true over time, especially as we create better tools to answer those questions.
In order to measure draft pick value, you need to be able to measure player value and Schuckers was hamstrung by the data he had access to. Games played, points and ice time can work to an extent to measure career quality, but points can be misleading and some minutes are obviously better than others. Our current measures of player value (GSVA, WAR) are not perfect, but they’re definitely an improvement over what’s been used before which can help create a stronger draft pick value model.
Advertisement
There’s another benefit to that which is putting draft pick value on a more understandable scale: Wins. That can allow for better understanding of trades by putting players and picks on a similar scale, and also provide a more tangible unit to what teams can expect from their future at the draft.
The other issue was time. NHL data is still not “great,” but it only became even “good” during the 2007-08 season with the introduction of real-time stats. When Schuckers built his models, it would’ve been impossible to even use that data if he wanted to due to the lack of sample where that data was available.
In 2020, we now have access to 13 seasons of better data and while that’s still not a very large sample, it’s probably enough to start gleaning insight from, especially if our focus isn’t on a player’s whole career, but only on his first seven seasons where the drafting team has control over him. In order to expand the sample as much as I could, I also included incomplete seven-year sets at both the beginning and end of the sample. Sidney Crosby’s first two seasons might be missed by the better data era, but we can still include his next five. Same goes for Connor McDavid, who has only completed five seasons so far. (For my draft pick value model, I looked at drafts from 2000-2019).
Both players were selected first overall and on average the top selection earns about 17.5 wins in his first seven seasons. The next pick is at 15.7 before a steep drop-off as the average first-round pick (outside the top two) is worth just five wins. It’s only within the top seven picks that a team can expect an average of seven wins or more (one win per season). Early in Round 3 is when a team can expect just a single win total from the player’s first seven seasons.
Here’s the full draft pick value curve (fit using an average of an exponential and power law fit where the goal was to find the lowest average error – similar to Tulsky’s marketplace model).
Like every other draft pick value model, it shows that the NHL is reasonably efficient on average at finding the best talent at the beginning of the draft. There are sleepers that surprise later but they are very rare and difficult to find. By this value model, late-round picks are nearly worthless with the average seventh-round pick being worth about 0.2 wins, 25 times less than an average first-round pick and 88 times less than the top pick.
Advertisement
Though the shapes are similar, it’s that detail where this model based on GSVA differs from Schuckers’ based on ice time which may be overvaluing lesser picks. There, the top pick is only worth 28 average seventh-rounders. Relative to the first-round pick, the value of an average seventh-rounder by Schuckers’ model would equate to the value of a late third-rounder by this model.
That all adds up when both models are compared to the marketplace model compiled by Tulsky. Both models still believe that the market underrates draft picks after the first round, but my model believes it’s significantly less than previously believed where picks in Rounds 5, 6 and 7 had a supposed value 15-30 times greater than their trade value. Not to appeal to authority, but those values would presume that the trade market for draft picks was fundamentally broken and I don’t think that’s the case. Not to that degree, anyway.
A draft pick value model based on GSVA is a lot closer to the NHL’s draft pick trade market. Even in the first round, Schuckers’ model may be overrating how close other top picks are to the first one. Nathan MacKinnon and Dylan Larkin may both play just over 21 minutes per game but it’s obvious one is providing much more value to his team.
Schuckers’ work on draft pick value was extremely important and the basic principle still likely holds true: Trading down is generally the play to acquire more draft value capital. The Belichick method is still optimal for the NHL. But with the older model being based on measuring player value in an incomplete way, it’s time to move on to a model that uses better data and offers a more complete picture.
When we do, we find that front office evaluators likely weren’t nearly as far off on draft pick value as previously thought, but that the lesson brought upon by analytics wasn’t wrong either.
Data via Evolving Hockey and Hockey Reference.
(Photo: Bruce Bennett / Getty Images)
ncG1vNJzZmismJqutbTLnquim16YvK57kHBwcWxmaHxzfJFpZmluX2Z%2FcK7YZquhnV2jwq6uxKuqZqqVq7a0tdOipaBlpJ2ybsDRrpxmrpGhwqZ5zp9kmmWUp66nwIypoJyjXw%3D%3D