Prospect dragons: Contextualizing NHL draft pick values
I’m tired of not understanding draft pick value so I wrote this.
In the NHL, like most professional sports, draft picks are highly valued assets. They represent a team’s future. Each is a wrapped present full of potential and promise of success to come. Smart teams stockpile them. “Draft and develop” is the chorus you’ll hear among smart hockey people with regard to team building.
In some ways, that’s true. Acquiring top tier talent in the NHL without high draft picks is exceedingly difficult. That’s at least partially due to the restricted free agency system that allows teams to retain a player’s rights in most cases through age 27 heavily tilting the scales in favor of teams’ ability to keep their best players. That combined with an accelerated aging curve for hockey players means that most great players will spend the majority of their prime seasons with the team that drafted them.
But another part of the reason you’ll hear the draft and develop refrain from most analyst types is that that the same restricted free agency system makes drafted players cheaper from a cap perspective. The more players a team has who can contribute while on an Entry Level Contract (ELC) or a first contract post ELC, the more cap space they have to either spend on other players or pocket for ownership.
Like most generally accepted sports management truisms, draft picks as assets to be hoarded by the smart with a vision for the future and wasted by the dumb who can only see the present has some rational grounding but is also an oversimplification.
Previous work on draft pick value
Many good analysts have done work on draft pick value. Michael Schuckers has probably the most cited work in this area and his paper is the basis for most draft pick valuation work. Eric Tulsky took a different approach using draft pick trades to assess their value as opposed to the on-ice results of the players drafted at a given slot. Our own GeoFitz did some work following Tulsky’s approach. And recently, Prashanth Iyer has improved on all of that by looking at the value available at each pick as opposed to basing the value on the player selected at a given spot.
All of this is to say that if you want to know the quantitative value of a draft pick, you have multiple approaches to choose from, each with their own strengths and weaknesses. I detailed those in a previous piece on the Lightning’s trade deadline acquisition of Barclay Goodrow so instead of rehashing them, I’ll just link to that piece for further background.
Even with so much good work in the area, I still have a hard time figuring out how to interpret draft pick value. Seeing the numbers is one thing but knowing what they mean is another. And so, as usual, I dug in to try to find a way to contextualize this idea further.
An attempt to contextualize draft pick value
If our goal is to understand draft pick value as opposed to putting a number on it, the way that makes the most sense to me is to figure out what type of player teams have traditionally been able to land at each draft slot. In this case, we don’t want to know a numeric value for a draft pick. We want to know the range of players we might expect to get with that pick. If you think of a team having the 10th overall pick, who is the player that comes to mind as the most accurate representation of that draft slot? That’s the question we’re setting out to answer. And it’s one that has many possible approaches.
The way we’re going to approach it today is to think in terms of what a prime season for a player drafted in a particular slot typically looks like. To do that, we need to include in our analysis all seasons for drafted players that could be reasonably defined as their prime. I chose to limit that to ages 23 to 28 based on what we know about aging curves in the NHL as well as only include seasons where a player played at least 60 games.
The other matter that we have to deal with here is the fact that for obvious reasons, players drafted earlier are more likely to make it to the NHL and play more prime seasons thus distorting our numbers. For example, if we want to calculate the average prime season among first overall picks and compare that to the 25th overall pick, we’re going to have a lot more seasons in the former bucket than the latter. And if we just take that as is and calculate an average, we won’t be accounting for all the 25th picks who never became regular NHL players.
To accommodate this, I set a minimum number of player seasons equal to the number of player seasons among first overall picks. Meaning that if later picks had less player seasons in the data set, I added dummy replacement level seasons thus dragging down the averages. A better analyst would use a more subtle approach but given the task here, this is sensible enough for me.
At this point, we can produce our own draft value curves. The following plot shows individual prime skater season standings points above replacement (SPAR) with a line fitted at the median, the high end of possible outcomes, and the low end of possible outcomes. The high and low ends are plus and minus two standard deviations from the median respectively. All data here is via Evolving Hockey. They recently added draft information to the stats on their site and I highly encourage you to explore it on your own.
The lines are fit using a generalized additive model (GAM). For our specific purpose of measuring the typical outcome based on historical data, this is a fine approach. It would not be fine if we were trying to predict future value as this would result in overfitting and so should not be interpreted as an “expected” SPAR or anything similar.
Our curve generally aligns with most of the accepted approaches. By the end of the first round, the difference in value between picks has decreased significantly. The median pick will be a replacement level player but the opportunity still exists to get a difference maker. By the third round, that opportunity is mostly gone and we see very little difference in pick value over the remainder of the draft. Pretty much everything from the third round on is a lottery ticket.
Once we have our measure of the median season SPAR at each draft slot as well as the upper and lower bounds, we can assign players who align with each of those numbers as examples. This part is a little touchy but I used a weighted blend of similarity in draft slot and in prime season average SPAR to identify the players to use as guides. The players displayed weren’t necessarily drafted where they’re shown. They’re displayed because they were drafted within a reasonable range of that pick and they are close in SPAR value to the values in our curves above.
For clarity, the way to read this is among first overall picks, the median prime performance has been similar to Ilya Kovalchuk, the high end outcome has been similar to Nathan MacKinnon, and the low end outcome has been similar to Alex Galchenyuk.
A few things become immediately obvious. One is that unfortunately, Galchenyuk is the prototype of an early pick who doesn’t reach his projected ceiling. On the flip side, if you’re drafting in the top five, the dream scenario is to come away with a player like MacKinnon.
But what’s more interesting than the extremes are the players at the median. Remember the question from in this article about the 10th overall pick? Would you have guessed Brandon Sutter as the prototype? Well, that’s whose career is mostly closely aligned with the median for 10th overall picks.
And if the 10th pick means the most likely outcome is a player like Brandon Sutter, does that affect your opinion of the value of the pick? Yes, we still have to allow for the full range of outcomes. We can’t ignore that landing someone like Dustin Brown is possible there.
But when I hear discussions of trades involving the 10th pick, it sounds like they’re thinking a lot more about a Dustin Brown outcome than a Brandon Sutter outcome. Never mind the Cody Hodgson outcome, which is ignored despite being just as likely as the Brown one.
Summary and takeaways
When we discuss the value of draft picks, we can use any of the quantitative approaches listed earlier in this article. But we also need to think of those values in terms of what they’re likely to mean on the ice. The 10th pick in the draft is valuable. But a team isn’t likely to get a franchise changing player there. And too often, picks in that range are discussed as untouchable.
Teams that are rebuilding should absolutely be exploring all possibilities to try to land star players and that includes getting as many early swings in the draft as possible. But for teams trying to win now, should they be so hesitant to move a late first round pick? Especially if they can get a later pick back as part of a trade?
If a team isn’t going to be picking until the latter half of the first round, the draft becomes more of a quantity game than a quality game. If those first round picks can get you good players now in a trade, teams should be less hesitant to move them as long as they can find ways to recoup picks later in the draft. If a team isn’t picking in the first half of the first round, they should be more focused on how many picks they have and less on where those picks fall. For a contending team, it hardly matters if they have one pick in every round or no firsts, one second, two thirds, no fourths, three fifths, no sixths, and two sevenths. They’re probably going to pull the same value in either of those scenarios.
So bad teams, yes, hoard those early picks. Sit upon them like a dragon and let no one near them. But good teams? Be creative. Flip those mid to late firsts for players who help you now and try to compensate with multiple picks in later rounds. You’re probably not getting a difference maker at pick 22 anyway. So stop looking at that wrapped present like it has your franchise savior in it and instead, be willing to move it if you have a chance to get better now.
Note: You can access the code used in this analysis on Github. The data sources require a subscription to Evolving Hockey so they are not provided but instructions for subscribers on which data to pull are.