The Science of Trading Down
Using data science and statistics to create my own NHL Draft Philosophy
Since I was able to purchase video games for myself online, I’ve been obsessed with the sports management genre of video games such as Eastside Hockey Manager (EHM). So much so that I’m only a few hours away from 4000 hours on EHM! I’ve had lots of fun managing teams in the game as if I were an actual NHL general manager. Part of managing your team in EHM involves scouting players and drafting them. If any of my EHM buddies are reading this, then you know what it’s like to want to move up to get “your guy” in a draft, or trade down to, in theory, get the most you can out of your draft pick. The concepts of “moving up” and “trading down” in NHL drafts are important to every team’s draft strategy. In 2016, Pierre Dorion gave up the 80th overall pick to move from 12th to 11th to guarantee selecting Logan Brown. This kind of deal will happen at any point in an NHL draft, whether it be the first round or seventh round, teams will look to move up or down, depending on their draft strategy and philosophy. My experience with EHM isn’t exactly one-to-one to real-life hockey, however my experience with the game and following NHL drafts raised a question I wanted to answer: is there more benefit to trading down in an NHL draft, or is it better to move up?
In addressing this question, I had to determine what the actual “value” of each overall pick number is in the draft. To do this, I decided to use data on previous draft pick selections between 2005 and 2018, and assign each player based on the following metrics:
Max Cap Hit % - The player’s highest cap hit percentage of the cap ceiling when the contract was signed based on capfriendly.com, as a measure of the financial impact of the player. For example, Mark Stone’s max cap hit % is 11.95%,
TOI % - The player’s career average NHL time on ice, as a percentile based on position (forward, defenseman), as a measure of the players’ deployment in games. Mark Borowiecki is in the 31st percentile of time on ice.
GP % - The players’ total games played, divided by the possible number of NHL games played, as a percentile based on position. This is meant to be a measure of career longevity. For example, Ryan Dzingel’s GP% value is 44%.
This gives us the equation:
To no one’s surprise, Sidney Crosby and Connor McDavid lead the way in value score, with Matthews new career-high contract extension putting him right there in third! Alex Ovechkin is not included in this, as he was drafted in 2004 - which I’ve excluded from the dataset we are using. My childhood favorite Erik Karlsson led the way in value scores from defensemen (alongside Drew Doughty). I excluded goalies from this as well, as the sample size for goalies is much smaller than skaters, and measuring a goaltender’s value to match a skater’s values would be difficult.
I decided on only using draft data from 2005 to 2018, as many players drafted in 2019 or beyond have yet to reach their “prime”. Many players like Alex Newhook and Nils Hoglander, even Parker Kelly, have just started to play up to their potential, and others are still working to reach the NHL - thus many players’ value scores would not be truly representative of their career value.
We also choose to begin the dataset in 2005, as the salary cap era only began in the 05-06 season - thus player contracts before that would not be accurate to current players in the league.
Using each player’s value scores, I was able to create a non-linear model that estimates a draft picks value score based on the pick number.
Based on the visualization, we can see that the value scores decrease exponentially, which I believe is unsurprising. Earlier picks in the draft tend to be the very best, rarely is the player selected fail to make the NHL, let alone become an impact player. As we continue further along in the draft, the likelihood of a draft pick even making the NHL decreases rapidly, correlating with the exponential decrease in value scores.
In answering whether there was more value in trading up or trading down for NHL teams, I would have to understand what the “perceived value” of a draft pick is and compare it to the estimated value. If the perceived value of draft picks were to be higher, it would be safe to say that there is more value in trading down. If perceived value were lower than estimated values, there would be more value in trading up.
To determine the perceived values of pick numbers, I gathered all trades where teams moved up/down in a draft, where assets that were only picks in that specific draft were involved. For example, a trade where the 2020 3rd was traded for a 2021 4th and 5th would not be considered, nor would a trade involving non-draft pick assets, such as Alex Debrincat for the 7th and 29th overall pick in 2022. Trades such as the 20th overall pick for 27th overall and 77th overall in the same draft would be used. This dataset of draft trades consisted of almost 130 trades from 2005 to 2023, where a team moved down from a single draft position, acquiring only picks in that draft in return.
Using this dataset, I would calculate the perceived value of a draft pick as the sum of the estimated values of all draft picks received for the singular draft pick.
For example, the perceived value for the 11th overall pick based on the trade between the Arizona Coyotes and San Jose Sharks made during the 2022 NHL Draft is:
16801.328 (27 OA) + 13709.819 (34 OA) + 10640.253 (45 OA) = 41151.4
While the estimated value for the 11th overall pick based on our earlier model is 34838.214.
By calculating each perceived value based on these draft trades, I was able to make another non-linear model that estimates the perceived value of a draft pick.
Given that there is no trade data on picks before the 7th overall pick, the perceived value model starts from the 7th overall pick and ends at the 185th overall pick. We can see once again there is an exponential decay in the perceived value of draft picks as the pick number decreases.
Plotting the estimated value alongside the perceived value visualizes that the perceived value of a draft pick tends to be higher than its estimated value, suggesting that trading down will result in an increase in value for the team moving down.
An increase in value in this context means that the two draft picks acquired will combined have more value than the single pick acquired. Thomas Chabot (86548.11) would have a lower value score than Noah Hanifin (52166.87) + Colin Miller (26907.86), but Chabot is obviously (I’m not biased!) the better player among the three. This conclusion supports the quantity over quality argument - the team will acquire more assets with a greater total value, but it doesn't necessarily mean those assets will be of higher quality or contribute as significantly individually as the draft pick dealt away. To understand how quality can be impacted, we can look at the draft data through the lens of probabilities.
Using the value scores as a percentile value for the player set we had achieved earlier, I categorized players based on the role they played throughout their careers.
Generational - top 10 players in value score percentile (i.e McDavid, Karlsson)
Franchise - top 25 players in value score percentile (i.e Letang, Pastrnak)
Elite - players above the 95th percentile (i.e Stone, Chabot)
Primary (First Line/Top Pair) - players above the 90th percentile (i.e Duchene, Klingberg)
Core (Top 6 F/ Top 4 D) - players above the 85th percentile (i.e Turris, Hanifin)
Secondary (Middle 6 F/ 4-5 D) - players above the 80th percentile (i.e Pageau, Demers)
Support (Third Line/ 5-6 D) - players above the 70th percentile (i.e Byron, Gudbranson)
Role (Fourth Line / 6th D)- players above the 60th percentile (i.e Wingels, DeMelo)
Depth (13th F/ 7th D) - players above the 50th percentile (i.e Balcers, Borowiecki)
Cut - players below the 50th percentile (i.e Justin Falk, Francis Perron)
By categorizing it as such, we can create a model that can give us the probability of a draft pick being at least an X-type player throughout his career. For example, players picked first overall have a 76% chance of being at least an Elite player, while a player picked 30th overall has a 2.2% chance of being Elite.
Using these probability values, we can calculate the probability change of a team selecting at least one X talent level or better player when trading down as:
Back to the Coyotes/Sharks example, the probability change for the San Jose Sharks selecting at least one core+ player is:
(0.118 (27 OA) + 0.090 (34 OA) + 0.065 (45 OA)) - 0.294 (11 OA) = -0.021
This in essence means that the probability the Sharks selected at least one player who will be a core player or better decreases by 2.1%, however the probability the Sharks selected at least one secondary player or better increases by 1.5%!
By plotting each trades change in probability for the pick being at least X talent level after trading down, we can make the following conclusions:
Picking Generational talent - Will never increase after trading down, will decrease if pick traded is high.
Picking Franchise talent or better - Probability change is close to zero unless decreases if the pick traded is mid-first or better.
Picking Elite talent or better - Probability change decreases significantly if the pick traded is mid-first round or better, and slightly increases otherwise.
Picking Primary talent or better - Probability change decreases if the pick traded is mid-first round or better, usually ~1% increase otherwise.
Picking Core talent or better - Probability change decreases if the pick traded is mid-first round or better, increases by 1-4% otherwise.
Picking Secondary talent or better - Probability change almost always increases by 1-4%, with the increase being higher the earlier the pick traded is.
Picking Support talent or better - Probability change almost always increases by 1-15%, with the increase being higher the earlier the pick traded is.
Picking Support talent or better - Probability changes almost always increase by 1-15% - change exponentially decreases as the pick traded is later.
Picking Role talent or better - Probability change almost always increases by 1-25% - change exponentially decreases as the pick traded is later.
Picking Depth talent or better - Probability change almost always increases by 1-50% - change exponentially decreases as the pick traded is later.
Based on these findings, the argument for whether or not trading down is beneficial depends on context. If teams are looking for quantity > quality, trading down will always result in a qualitative value higher than the initial draft pick, however, if teams are aiming for quality, they should look at their position in the draft.
The probability change visualizations suggest two things:
Trading up into the top half of the first round will increase the probability of finding a player who is at least Core or better
Trading down from a pick that is NOT a top-half first-round pick will also increase the probability that at least one player selected with the picks acquired will be at least Core or better.
Every team’s ultimate goal with their draft picks is to maximize their capabilities. Some teams like the Golden Knights or the Pittsburgh Penguins tend to move their draft picks for win-now assets, but for rebuilding/retooling teams such as the San Jose Sharks, the Chicago Blackhawks, and the Calgary Flames, the ultimate goal is to build for the future and maximize their ability to select future impact players. The analysis suggests that in most cases, trading down presents teams with an increased probability of selecting quality future players, while also presenting the possibility of selecting multiple impact players. A prime example of this strategy in execution is when the Red Wings traded down from 18th to 20th and acquired 58th overall, and selected Anthony Mantha and Tyler Bertuzzi with those picks. It’s important to note that this draft philosophy would require trust in the team’s scouting abilities, to make good and well-informed selections at any point of the draft, unlike the 2021 Ottawa Senators. Hopefully, it works in Eastside Hockey Manager though, as I will be trying it on my next save!
Samee
I realized that I had written a lot more than I thought I was capable of when working on this. As they say, a few graphs and numbers can mean a thousand words! I hope you enjoyed reading this, NHL drafts are fascinating.