Notes
2020 Draft Recap - Best and Worst Values
October 10, 2020 This was a particularly deep draft compared to previous years, so most teams were able to stock up vs. the historical averages. At the top of the first round, Alexis Lafreniere predictably went No. 1. Behind him, Quinton Byfield was an exceptional No. 2 pick - at this slot, players are a decent distance from the pick before them, but the model actually had Byfield ranked first. Byfield didn't present the only exceptional value - Seth Jarvis also ranked highly by the model, but lasted until the middle of the first round. Sam Reinhart and Leon Draisaitl were players that put up similar stats to Jarvis in the draft years. And then, the conundrum. The Chicago Steel destroyed the USHL this past season (41 wins in 49 games). 4 players were drafted from the team. So how much of the players' production was a result of being on a team that was stacked? Considering the circumstances, NHLdx may be overvaluing them here. Time will tell.
On the other side of things, there is the question of skill vs. production. Amirov and Raymond, both highly touted, taken in the front half of the first round. They just don't seem to put up great numbers. Raymond had 10 points in 33 games in the SHL - that's fine, but not worth No. 4 overall. Amirov had 2 assists in 21 KHL games.
On the team side of things, below is how the rankings look sorted by total value added. On a per player (first-time-eligible forwards only) basis, Vegas, Arizona, and Colorado also had strong showings. Apart from Byfield, L.A. looks to have made excellent picks with Alex Laferriere (No. 83, +11) and Martin Chromiak (No. 128, +17). They had only one pick in the negative, but barely - Kasper Simontaival (No. 66, -1). Similar to L.A., Carolina was on the negative side only once, picking up good value with Jarvis and Noel Gunler (No. 41, +12)
Detroit did not have a single pick with an NHLdx higher than their slot value (which again, was difficult due to the strength of this draft versus past years).
We'll check back in 5 years from now. |
Where are they now?
October 6, 2020 Looking at simple linear correlation isn't the most exciting thing to do, so I want to visualize a couple of different ways we can assess the accuracy of a model, building on the beeswarm chart I posted earlier in this summer. Year-over-Year for Individual Players Firstly, how does the model perform via year-over-year look? Here are the Top 10 forwards selected in 2016. With the top forwards, there's a big question as to how elite they really are. Those who make the immediate jump to the NHL (Auston Matthews, Patrik Laine, the Matthew Tkachuk example below) typically see a corresponding jump in their NHLdx, as the model can obviously be a lot more certain of their NHL potential a few years down the road. (I'll take a look at transitions to different leagues and how that affects NHLdx at a later date.)
Depth Chart Expectations from Draft to Age 23 And then another way to see things is through a lens of where they end up in the depth chart. I'll let the alluvial graphs do most of the talking. Again, generally, players end up where predicted, but there are a few exceptions to the rule, as a few anticipated first/second-liners will fall short of their lofty expectations, and a few late-round picks will develop over the years into NHLers. A quick summary before the charts:
- Top 6 Players All expected first-liners end up in the Top 9, with 65% hitting their expected line, with most of the rest falling to the second line. The model performs pretty well with the high-end players. There's a little bit more scatter with second-liners - though again, they are most likely to hold that role once they make it to the NHL. 71% end up in the Top 9. - Bottom 6 Players The same thing that we see with the second line is applicable here as well - not 100% of players maintain their line expectation, but they don't stray too far from it. 63% of project third-liners end up between the second and fourth lines. - Fourth Line/Call-up/Not NHL As with the projections for high-end players, the model has a pretty good idea of where the low-end players will fall. 93% of the players with "Call-up/Not NHL" at Age 18 find themselves either outside of the league altogether, or playing on the fourth line. Overall, I'm pretty happy with how the model works - I'll try to tackle things from other angles once things settle down after tonight's draft. |
NHLdx Over Time
August 24, 2020 NHLdx evolves over time as certainty grows as the player ages. Here's a quick snapshot of the 2010-2015 drafts. Initially, there are quite a few players in the 15-35 range (bottom 6 players) due to very similar draft-year statistics, and we're not sure which ones will actually turn out yet. Over time, those players disperse over a wider range of possibilities - a select few will move up, some stay right where they are, and some drop down all the way into the zero range. Thus, we end up with a few more 35+ point players, and a few more 0 players by Age 23, and so we can see that middle range is a little less busy than at Age 18. Yes, that top data point is Connor McDavid. |
The 2020 Draft Class
October 5, 2020 (Updated) Alexis Lafreniere is NOT the No. 1 ranked player by NHLdx for the 2020 draft. That rank falls to Quinton Byfield, on account of his primary points numbers and adjusting for age (he's nearly a year younger than Lafreniere). It's a very close race, and you can't go wrong with either pick. Overall, this is somewhat of a top-heavy draft, and if you're picking in the Top 20, you should be okay. Comparables here are based mainly on NHLdx at age 18 and source league (e.g., for OHL players, their comps lean toward other OHL players), and not necessarily on style of play. The Top Two
Both players have excellent comparables - Hart Trophy finalists in MacKinnon and Draisaitl - for their NHLdx range. Mid-80s is roughly 10 points better than 2019's top pick, Jack Hughes, and much stronger than Nico Hischier in 2017 (his NHLdx was 52, the lowest of any No. 1 overall pick since 2010). Neither is anywhere near Connor McDavid (by far the highest rated player in the dataset at 120 NHLdx), but still pretty good. Not included here is a mention of Tim Stutzle, who has been in the conversation for No. 2. As you might have seen mentioned elsewhere, 3 German players could be taken in the first round (John-Jason Peterka and Lukas Reichel being the others). 18-year-olds as regulars in the DEL is a recent phenomenon. In the league's history, only 6 forwards - including 2020's trio - have put up double digits in scoring. 2001 was the last time it happened (Marcel Goc, drafted by San Jose and who had his best season - 30 points - with the Nashville Predators; Goc even played in the DEL as a 17-year-old) - Stutzle wasn't born until 2002. So, with regard to Stutzle and the other DEL players, we don't have any recent comparables in the dataset. As such, we are deferring to a "slot" estimate for their NHLdx, using the historical point production of players taken at their selection number in the draft/consensus pre-draft ranking. For Stutzle, this means looking at the players at No. 4. Is he Mitch Marner, or Sam Bennett? Historically, forwards taken No. 4 overall have reached 53 points. Based on scouting reports, this may be low. (Note: while many mock drafts have him at No. 2 or 3, the Elite Prospects June ranking had him at No. 8, hence the drop out of the Top 3)
Second Tier This is a good crop of players, projecting to be Top 6 forwards, and all should definitely crack the Top 10 or 20.
The biggest omission here is Lucas Raymond, who has been ranked as high as No. 3 in some lists. He got plenty of experience in the SHL this season, but his numbers were quite a distance away from some of the top Swedish players in recent drafts (Alexander Holtz had 9 goals/16 points compared to Raymond's 4 goals/10 points). Seth Jarvis gets the biggest jump via the NHLdx ranking, 12 spots above his consensus spot. His season (1.28 primary points/game) with Portland was in the range of the 2013-14 seasons of Leon Draisaitl (1.34) and Sam Reinhart (1.17) - they went No. 3 and No. 2 respectively in 2014. The full Top 30 is available here. |
The Model
November 28, 2019 How it works The #prospectpred model attempts a rough guess at a prospect's future point production. It is based on a fairly simple process called CART (classification and regression trees), looking at how certain factors earlier on a player's career influence the end result of peak NHL performance. CART classifies each player in the master database by their source league, establishing a discrete performance environment (i.e., so we aren't directly comparing the KHL to the USHL). From there, the model takes a number of inputs (that include age-adjusted points - using the method developed by Rhys Jessop and Josh Weissbock - and draft position) and compares them to the player's age-23 season to generate a predictive regression. The output for the target players is NHLdx (NHL draft/development expectations; expected points/82 games at age 23). Limitations
|
Who the Hell is Eeli Tolvanen?
September 30, 2019
The player that inspired all of this was Eeli Tolvanen, selected 30th in 2017. Not a single Predators fan has a clue on what he'll become at the NHL level. In 2017-18, for Jokerit of the KHL, he put up 19 goals and 36 points in 49 games, tied for the team lead in goals. He scored a hat trick in the season opener (again, his first game in the league). 19-year-olds aren't supposed to do that in the KHL. In recent history, the only player close to that kind of performance at age 19 is Kirill Kaprizov, who managed 27 points in 53 games. Going back a decade, some other names in the rough 20+ point range as 19-year-olds: Kuznetsov (32 points in 44 games), Panarin (21 in 40), Tarasenko (19 in 42). Good company - reason for excitement. And then 2018-19 happened - his "can't-miss prospect" was revoked - he needed 12 points in his last 10 games to claw his way to 35 on the season for the AHL's Milwaukee Admirals. So who is the real Eeli Tolvanen? Also, "Predators" and "Prediction" both = Pred. #prospectpred The Numbers Game NHLe, as developed originally by Gabriel Desjardins, and then Rob Vollman, and continued by analysts such as Emmanuel Perry and Byron Bader served as an inspiration for this project, as did the work on pGPS and SEAL (before those involved were picked up by NHL teams). Their work is very much appreciated. NHLdx differs from the NHLe method in attempting to project what a player will become in multiple years (at 22-24), not what they are currently/how they will perform in the next season (though similar approach to NHLe to include smaller leagues may be taken in future iterations), and also incorporates draft position. |