4/21/2020

Author: Chris Idso Project Name: NHL Expansion Draft Simulator Description: An attempt to simulate the NHL expansion draft, the process by which a new team selects players from the other teams, using the rules set forth by the NHL. The idea being to get a feel for how strong the resulting team would be compared to the rest of the league. And to learn as much as I could about hockey in the process.

Introduction:

Expansion in sports leagues usually (but not always) happens in pairs, for ease of scheduling, and for balance in divisions and conferences. And there is a sort of friendly competition between expansion teams to be the first to win a championship.

Seattle sports fans haven't seen an expansion draft since the 1970's when the Seahawks and Mariners joined their respective leagues. We will get another chance in 2021 when the Seattle Kraken joins the NHL and starts playing.

Our sister expansion team is already in the league. They are called the Vegas Golden Knights and they joined the leage NHL in 2017. They did so well in their expansion draft and in their first entry draft that they made a run at the championship (Stanley Cup) in their first year but ultimately fell just short.

The Expansion Draft:

So, how did they do so well? I was curious about this, because if they could do so well then perhaps the Kraken could do just as well. Fortunately, information about this was available on-line, where I could look up the NHL expansion draft rules and find a list of transactions made by the Knights related to their expansion draft and the entry draft that followed.

It seems there were two main reasons for their success. One was the rules for the expansion draft, which do not let a team protect as many players as they would like, which leaves quite a few good players on the table. The other was the trades they were able to make, which allowed them to acquire more draft picks (some in the first round).

Trade Example:
Golden Knights to one of the existing teams: We see that you left a couple of good players out there (A and B). We know you want to keep player A, so how much would you give us to take player B instead of player A?

I also found that the NHL had changed the expansion draft rules, and the existing teams may not have been properly prepared, or did not realize the impact the changes would have.

So with this information in the back of my mind I began designing an expansion draft simulator. I was determined to incorporate as many of the expansion draft rules as possible. And there would need to be a way to calculate the relative strength of the teams in the NHL, as well as a way to calculate how good a player is (something the NHL has trying to do for some time with varying degrees of success). There would also need to be a way to decide which players to protect, and a method for selecting and grading the expansion team.

Other considerations include players exempt from draft, teams exempt from the draft, players playing too few games to be protected or selected, and players too old (38, 39, 40+ years old). I did not consider goalies in this project.

Along the way I tried to put myself in the shoes of a general manager, deciding which players to protect, and in the mind of Seattle's general manager, deciding which players to take.

Expansion Draft rules:
1) Expansion team can take only one player from any given team.  
2) Expansion team must take 14 forwards and 9 defensemen and 3 goalies.
3) A team can protect a) any 8 skaters or b) 7 forwards and 3 defensemen.
4) A team has to leave 2 forwards and 1 defensemen unprotected who played
    at least 40 games in the current season or 70 games in the last two seasons. 
5) Players with NMC's must be protected and count against the abovementioned 
    limits.  Players with NMC's can waive that clause.
6) Players in their first or second year are exempt from the expansion draft. 
7) When Seattle drafts in 2021 the Vegas Golden Knights will be exempt.   

The Data:

The data for this project was available on line at NHL Reference. You just find the data you want (in the form of a big table) and convert it to text, and then copy it into a spreadsheet and save it to a csv file. Then it loads easily into a dataframe.

I was able to get data for skaters for the years 2011 to 2019, and data on player debuts from 2010 to 2019. I also found data on-line for players in 2019-2020 who had a No Movement Clause (NMC) in their contract. The expansion draft rules address NMC's so this has been added too (but only for 2019-2020).

Some interesting hockey info: 1) Useful player stats include TOI (time on ice in minutes), +/- (goal differential) and PTS (goals + assists) 2) The stats in (1) above do not make sense at first. Ex: A player can record many points, and play lots of minutes, and all or most of the 82 games in a season, but still have a negative +/-. 3) A team is allowed to suit up 20 players for a game. They usually go with 12 forwards, 6 defensemen, and 2 goalies. 4) Average shift time for a player is in the neighborhood of 40 seconds, but this varies. Different for players on the ice during a powerplay or when killing a penalty. And different for best players on the team than for other players. Also different for defensemen (longer shifts) than for forwards (slightly shorter shifts) - due to defensemen not having to skate as hard. 5) Hockey teams can expand their rosters late in the season, just as baseball teams can.
6) Some players are on the ice when the team is shorthanded, the team is on a powerplay, and when the teams are at even strength.
7) Some players are on the ice when the team is on powerplay, when the teams are at even strength, and not when the team is shorthanded. 8) Some players are on the ice only when teams are at even strength.

As you might expect the last four items in the list above make the problem of evaluating players difficult, since it seems unlikely that there is one formula that could be applied to all forwards and defensemen, and would involve all three of the abovementioned player stats.

This proved to be the biggest challenge in this project. Other challenges include how to handle trades (more than one line in the data for traded player), handling of a changing position list in the data, a strike-shortened season (2012), and a season shortened by coronavirus (2019).

My finished project consists of a series of functions:

1. load_setup(year) - Takes the year (2011 - 2019) and adds some variables, does some cleanup (trades) and removes players who played <=20 games during the season. These players are probably callups late in the season and can have weird stats.
2. calc_base_stats - Sums the +/- stats for the players on each team and calculates the mean and std over all the teams. Does this with the PS stat and the Ztot stat too. Uses this later to evaluate the results of the draft.
3. calc_player_value - evaluates the players using their +/- and PTS stats. Score for each player is recorded in the Ztot variable. Ztot = Z-score(+/-) + Z-score(PTS) for each player.
4. calc_team_strength - Calculates team strength and stores it in a dataframe for later reference when drafting a team.
5. remove_first_second_years - Removes first and second year players, since they are exempt from the draft. Thought they should remain in the data during calculations.
6. reduce_vars - Reduces the variable list to only those variables that are needed from this point on.
7. team_processor - For each team decides which players to protect and then creates a dataframe with the unprotected players.
8. team_selector_by_team_strength - Selects a team by reading from the team_strength list which is ordered by team strength. For each team selects best available player from the unprotected players list.
9. team_selector_best_top_down - Reads the unprotected players list from the top down and selects the best players, tracking the teams the players play for to keep only one player from any given team.
10. team_selector_fwd_first - Reads the unprotected players list from the top down and selects all the forwards first and then all the defensemen.
11. team_selector_def_first - Reads the unprotected players list from the top down and selects all the defensemen first and then all the forwards.
12. team_selector_alternate - Reads the unprotected player list and selects nine forwards and nine defensemen in alternating fashion, then adds five more forwards to round out the team.

NOTE: For each of the team-selecting functions the unprotected player list gets sorted by Ztot. When the team-selecting functions run they produce a list of the players selected. At the end of the list (you might need to scroll down a little) they write out three versions of the team score. One uses the sum of the players' +/- stats, one uses the PS stat, and one uses my own Ztot stat, all of which are evaluated using the team stats calculated by the calc_base_stats function, to determine how many standard deviations away from the mean the score is.

13. simulate_nhl_exp_draft - One function to control them all. So you only have to call one function.

Also included in this project, are some juypter notebooks. One (project_demo) has the functions above in it so you can run them. Another one (project_walkthrough) has some code I wrote as I was working on this project. It will run too, and will give you an idea of what this project does, and of the problems I faced, how I dealt with them, and of my thoughts as I worked on this.

There is work that still could be done on this project. I may need to find a way to correct the player evaluations for the strength of the team they play for. And I might be able to add age constraints for players protected or for players selected.

Ex: Only protect players who are 35 years old or younger, or maybe 25 years old or older. Or only select players between 25 and 30 years old.

Conclusion:

I have run this simulator many times now, and it looks like the results (+/-) are mostly in a range between 1.2 and about 1.8. So it looks like the Kraken will be pretty good. I expect the other teams in the league will be more prepared for this expansion draft than they were for the last one (in 2017) so the Kraken probably will not do quite as well as the Knights did.

When the Kraken joins the league we will be in the same division (Pacific) as the Knights, so there should be some really good games between Seattle and Vegas!!!

If you have any questions for me, or ideas on how this project might be improved, you can send me an email at [email protected].

Cheers!