2021 NHL Draft Predictions (Mostly QMJHL players, some WHL players) (Model version 2021.1.0)

06.17.2021 : release, 2021

New predictions are posted for prospects that are eligible for the 2021 NHL Draft:

As a reminder, cross-validation data is generated by essentially testing the model against the players used to train the model. Since we already know something about the outcome for these players, this allows us to judge the model's performance. For example, we can see what the model would have thought about drafting JG Pageau, Brayden Point, or Matt Barzal (oops).

For this season, with the WHL playing a shortened season, the OHL playing no season at all, and the USHL absorbing some players who probably would have gone elsewhere in a normal year, I've modified the model & output accordingly. In particular:

  • Only predictions for QMJHL and WHL players are provided. As a further disappointment, many WHL players didn't meet the minimum number of games played threshold, so only a handful of WHL players were evaluated.
  • In previous years, the model output for a given player was comprised of probabilities that they would become a superstar, a solid player, a "grinder", or a total miss. For this year, I simplified. For forwards, I only modeled probability that a given player would become a top-6 forward. For defensemen, I only modeled probability that a given player would end up on the top-pairing. This actually improved the accuracy of the model, but some of that is just a result of the decrease in resolution.
  • I did do something that contains some data for OHL and USHL prospects, more on this below.

Not all model/code updates since the last model release were directly-related to COVID, however. Looking at the data pages, the most obvious addition since last year is the raw stat-rank data for each player. Some of these stat names are somewhat self-explanatory, and I plan to provide some details about others in the future. These stat ranks can provide some interesting insight, but do not directly correlate to the model probabilities, and are probably best considered an unrelated way to look at a given player. I suppose there are now three ways to look at a player -- its stat ranks, its probabalistic model output, and its near neighbors. The output of the probabalistic model remains my focus for this project.

Time to move on to some player analysis.

But first...

As I mentioned earlier, I was compelled to at least do an experiment or two that produced output for OHL and USHL players. This is mostly for fun, but here's a model based on data from the 2019-2020 season:

It's interesting how some of these players' projections have changed since the end of the 2019 season.

Okay, now onto players of interest.

Players of interest

Looking at the predictions for defensemen, it appears that the model doesn't project any of the prospects graded as potential top-pair defensemen. A gloomy and slightly boring result. The cross-validation data provides proof that the model isn't simply calling all samples a miss.

In general, due partly to the COVID-impacted 2020 season, and also because (it seems) more top-50 prospects than ever are from non-CHL leagues, the model didn't get a chance to grade many top prospects, and the output lacks a little sizzle as a result.

At forward, there are still some potential value picks:

  • Riley Kidney -- Kidney has an aggregate scout rank around 70, and while it's a little dubious to compare him to players drafted around 70th overall, we'll do it anyway. Looking at the cross-validation data, you'll see it's not common to find a player with a 15% chance to become a top-6 forward around 70th overall (Though Jordan Weal is an exception). Kidney ranks high in most P60 stat ranks, and has Ryan Spooner among his comparables. The model doesn't consider playoff stats, but putting up 17 points in 9 playoff games seems notable.

  • Olivier Nadeau -- Although a 12% chance of becoming a top-6 forward isn't very high, like Kidney, it's still significantly more (on average) than a typical 90th overall selection. Probability-wise, a comparable for Nadeau is Linden Vey, who was drafted 96th overall in 2009. Additionally, Nadeau ranks 1st (among players graded) in almost every stat related to assists. Seems like he may be worth more than a 4th round pick.

Little light in this section this year, but moving on ...

What's Next?

Shot data is available for some or many of the leagues I typically produce predictions for. Example:

I don't think the amount of historical shot data required to be useful for this project exists yet, but, obviously that will not be the case a few years down the road. Probably not the worst idea to start tooling up now.

Some other ideas are in the mix also, but we'll see what happens for next year.

For the immediate future, I'm looking forward to digging into some of my own results a little more. Lots of data to look at with this release. Follow me on twitter as I ramble about what I find.

Stay healthy and enjoy the draft!