It’s been a while since I posted about any of my predictive models. While I’ve been busy posting season reviews for each team, I have also been working on the models in the background. It’s all come together and it’s time to share the results just in time for the offseason. With these models, we’ll be able to project next season’s standings based on current rosters and watch how each team progresses through the offseason relative to their competition.
The Models
First, a reminder of the models that I’ve shared previously. Here’s a list, along with links if you’d like to read more about them.
- Team Points Predictor – predicts team point percentage
- Team GF/60 Estimator – uses a weighted average of player 5v5 on-ice GF/60 to estimate team 5v5 GF/60
- Team GA/60 Predictor – uses team 5v5 save percentage and team 5v5 SA/60 to predict team 5v5 GA/60
- Team SA/60 Estimate – uses a weighted average of player 5v5 on-ice SA/60 to estimate team 5v5 SA/60
- Forwards On-Ice GF/60 Predictor – uses player statistics to predict 5v5 on-ice GF/60
- Defensemen On-Ice GF/60 Predictor – uses player statistics to predict 5v5 on-ice GF/60
In addition to the models I’ve previously shared, I’ve also created some new models to predict individual player statistics for next season. The approach is similar to the Team Points Predictor model using historical player statistics to train and test the models. These new models forecast the following player statistics:
- 5v5 Goals/60
- 5v5 First Assists/60
- 5v5 Second Assists/60
- 5v5 On-Ice SA/60
- 5v5 SV%
Input data for all models is from Natural Stat Trick.
Rosters & Ice Time
Based on contract data from CapFriendly, I’ve built rosters for each team based on the following major assumptions for the start of the offseason:
- All UFAs depart
- All RFAs are re-signed
- Players on LTIR that are not expected back for the start of the season have been left on LTIR
With the rosters in place, I utilized a set of categorical roles at each position to allocate 5v5 ice time. Players were allocated ice time similar to their 2020-2021 season actuals, then adjusted based on the current (UFAs removed) roster so that the total allocated ice time is equivalent to 48 minutes of 5v5 played per game. It’s important to note that I haven’t necessarily adjusted the ice time to be optimal, instead preferring to allocate it in the way the team seems most likely to. The player roles are not significant aside from the ice time allocation and, as such, have no bearing on the skill of the player. They are simply used for allocating ice time in the model.
2021-2022 Season Projections & Offseason Tracking
With the roster building completed as above, we now have a roster of contracted players plus RFAs for each team that represents their current projected opening day roster. With these rosters, we can run the models and get the projected standings for next season. While the Team Points Predictor does output a point percentage, it’s important to keep in mind that projections at this stage are likely all low since no pending UFAs are included in the models and teams have lost varying number and quality of roster players as UFAs. Some will re-sign while others will wear different colors next season and we can update the models as these signings take place. We also need to keep in mind that teams have varying amounts of available cap space to improve their roster with over the course of the offseason. I fully expect the rankings to change as the summer goes on.
With this initial projection giving us a snapshot as the offseason gets underway, we can track the progress of each team throughout the offseason. I will periodically update the models throughout the summer to see the effect of offseason free agent signings, trades, and the expansion draft. We’ll be able to see the 2021-2022 season standings projections update as the teams add and move players over the offseason.
Update (June 17): I resolved some issues to correct projections for a few players not mapping correctly through the models and updated projections for key players that missed the entire 2020-2021 regular season. The updated 2021-2022 projections are included below.
What If?
While I’m excited to watch the offseason unfold through the lens of these models, they also provide a fun tool as an armchair GM. We can make hypothetical roster moves and check the model results to see how it evaluates the changes. Have a move that you’d like your team’s GM to make? Leave a comment or find me on Twitter.
Initial 2021-2022 Season Projections
You can find the full set of season projections at the links below. Check back through the offseason to see how roster moves re-shape the landscape before the puck drops next fall.
