clock menu more-arrow no yes

Filed under:

How Good Are NHL GMs at Getting Value From the Contracts They Sign?

New, comments
2018 NHL Awards - Arrivals Photo by Ethan Miller/Getty Images

Last season, I wrote about the Detroit Red Wings’ performance in the 2016-17 season using the Goals Against Replacement (GAR) model created by Dawson Sprigings (@DTMAboutHeart). Dawson as since been hired by Kroenke Sports to work for the Colorado Avalanche, so he has stopped updating his model, at least publicly.

In the spirit of ensuring accessibility to those less-interested in the fine math details, the main body of the article will explain the basic ideas. Readers more interested in the methods will find a more in-depth appendix section at the end.

I wanted to do something similar this off-season, but an article from a few years back provided the idea to do something more. Emmanuel Perry, who runs Corisca.hockey started adding his own Wins Above Replacement (WAR) statistic. While not exactly the same as Sprigings’, it is similar.

This article will first take a quick look at how Detroit performed using this model. The reason for not spending as much time as last year is because it was pretty bad across the board as seen in their 5th worst finish in the league standings and because it would be pretty similar to last year’s article. Another reason is that the rest of the article should be more interesting in that it will look at areas I haven’t seen discussed as much.

Detroit’s Performance

Before taking at Detroit’s performance, here’s a quick summary of Manny’s WAR/GAR model. If you want a more in-depth look, here is the article in which he introduces it.

WAR/GAR via Corsica

Manny’s model uses nine components. Eight of these apply to skaters, while the ninth applies to goaltenders. He doesn’t explain in this article how the goaltending model works, in that article, but there is a followup article that goes into much more detail. Caution: it’s pretty much for serious math people.

The components are:

Offensive Shot Rates (RF)
Defensive Shot Rates (RA)
Offensive Shot Quality (QF)
Defensive Shot Quality (QA)
Shooting (Shooting)
Penalties Taken (PT)
Penalties Drawn (PD)
Zonal Transitions (OZF, NZF, DZF)
Goaltending

The individual components for skaters are compiled into Offensive WAR (OWAR) and defensive WAR (DWAR) as follows:

OWAR = RF + QF + Shooting + PD + OZF
DWAR = RA + QA + PT + DZF + NZF
OWAR + DWAR = WAR
1 WAR = 4.5 GAR

The stats for this model can be found by going to Corsica.hockey, clicking on Skaters on the top menu bar, then clicking on WAR. There is a box to check that will convert the whole sheet from WAR to GAR.

It’s interesting that the highest goaltenders on Manny’s model are not necessarily the players you might expect. Looking at the highest GAR players, you see a handful of goalies very near the top. My theory is that since they are out there for entire games at a time, they have more opportunities to provide value, either positively or negatively.

I ran a regression to see the correlation between a team’s total GAR, including goaltenders, and their points in the standings. There was a very strong correlation, with a p-value of 2.9 x e^-15, which indicates a high level of statistical significance. Here’s the plot of that regression:

Data: NHL.com, Corsica.hockey. Data adjusted and chart by @pflynnhockey

The last thing to point out before I get into the data is that for most of this article I used the GAR numbers from the regressions Manny ran on his model at the end of the regular season. For the individual Detroit numbers, the numbers from the regressions Manny ran after the playoffs were used. Appendix A will explain why.

So, here’s Detroit’s GAR broken down by component. The chart below it is their GAR/60.

Data: Corsica.hockey; Chart: @pflynnhockey
Data: Corsica.hockey; Chart: @pflynnhockey

(While Hicketts’ numbers are very good, it’s based on a very small sample size, so it’s too small to really make any judgments from.)

The most likely response to this is probably going to be “why is Dylan Larkin so low?” (Or maybe “how is Jared Coreau so bad if he only played 7 games”). If you remember the article on high-danger chances against, you’ll remember that Larkin was very low in that category. In this model, Larkin is very positive in both shot rate categories. He is, however, very low in the shot quality against component, with a -5.58.

To put that in context, it was second worst in the league for all players. Artemi Panarin was the worst in the league, and Connor McDavid was 7th worst, so it’s not like everyone in the bottom group is a bad player. The problem is that while McDavid’s shot quality for component was 16.2 to more than overcome his -3.92 QA, Larkin’s QF was only 0.4.

Larkin does a lot to contribute to the team, but this provides him an area of improvement. If his QA was the same as Tomas Tatar (2nd worst on Detroit at -0.63), Larkin’s total GAR would be over 6, easily the highest on the team.

It’s too much of a difference to say that it’s because of his linemates or his usage, although those two things could have something to do with it. This should be a topic to look at more closely in a future article.

Not surprisingly considering how they performed in the standings, Detroit doesn’t rate well from top to bottom, with more than 100 NHL players higher than Jimmy Howard’s team leading number.

How Much are NHL GMs Paying Per GAR?

At the beginning of this article, I mentioned that I saw an older article that gave me an idea of something different to do with this data.

In the article, Andrew Kerlson of Maple Leaf Hot Stove used GAR as part of a model to figure out what NHL GMs paid per GAR. He was then able to look at individual contracts and see if the team got their money’s worth for a season.

Since the cap is higher now, as well as the minimum salary, I redid Kerslon’s calculations using Manny’s GAR model. The other major difference is that Manny’s GAR conversion is 4.5 to 1 instead of the 6 to 1 conversion we’ve seen in previous models.

Here’s the short version of this was calculated (more detailed numbers in Appendix A at the bottom of the article, the numbers are rounded here:

  1. Total Salary paid for 2017-18 was $2.22B.
  2. Dividing by 31 teams gives us $71.67M spent per team on average.
  3. A team constructed of replacement players would cost $14.95M (23 players x $650k).
  4. The average NHL team paid $56.7M above a “replacement level” team.
  5. The average NHL team would have 19.5 wins more than a “replacement level” team (explained in Appendix B)
  6. The cost per win above replacement (step 4 divided by step 5) = $2.91M
  7. The cost per goal above replacement (step 6 divided by 4.5) = $646,349
  8. A player earning the minimum salary of $650k should be expected to produce 0 GAR (since they are replacement level).
  9. Every 1 GAR over that should be expected to cost an additional $646,349

This gives us an equation that’s easy to use: AAV = (GAR*$646,349) + 650k. Remember that this is based on what GMs actually paid, not necessarily on what they should have paid.

Let’s start with Detroit. The following chart plots AAV vs GAR for the 2017-18 season. The diagonal line is the line using the equation above. Anything to the right of the line is good value, anything to the left is poor value (for this season).

Data: Corsica.hockey, Capfriendly; Chart: @pflynnhockey

Player A is Jonathan Ericsson. Player B is David Booth, Player C is Dylan Larkin, Player D is Martin Frk. For players who were not with the team all year (trades or callups) I used the cap hit their salary cost Detroit. For example, having Hicketts up this season only contributed ~30k to Detroit’s cap hit. For players traded, I multiplied their GAR for the season by the percentage of their games they played for Detroit.

Here’s another way to look at this. The following chart shows how much value each player brought to Detroit, as measured through GAR:

Data: Corsica.hockey; Chart: @pflynnhockey

I made this into an interactive Tableau visualization, so if you go here, you can look at this for different teams and also look by position played. I’m trying to obtain a spreadsheet of all player cap hits for last season. With that, I can change the above visualization into one that will show contract value for every NHL player from last season

You can see here why the recent Mike Green signing will likely not be good for the team financially, to put it simply. Last year, he brought the team roughly negative $500,000 in value. Granted, he could do better this season, but it doesn’t look very promising if you hope he will live up to the $5.375M he re-signed for.

Now obviously, this is #NotGood. There was only one team last year that came out on the positive side of cap hit above what a replacement level team would cost and the value they got from their players. Can you guess which team?

Data: Corsica.hockey, CapFriendly; Chart: @pflynnhockey

It’s Winnipeg.

Detroit is...just a bit closer to the right side of the chart. As you’ll see with this next chart, Winnipeg had the highest Team GAR last season, and the benefited from performances from players on ELC contracts, most notably Patrick Laine. Connor Hellebuyck also did very well in GAR.

Data: Corsica.hockey, CapFriendly; Chart: @pflynnhockey

Obviously every team except one having negative value from their contracts looks really bad. But, based on the amount of GAR produced last season, it would have been impossible for every team to be on the positive side of contract value. This is because, as I mentioned in Appendix B, a replacement level team would be way below the salary cap floor.

For the 2018-19 season, the announced $79.5M salary cap and the average the league spent for the last three season in terms of percentage of the cap was used to estimate the value of GAR for next season.

That number is $711,332 per GAR. So based on this, a player’s value will be their GAR multiplied by $711,332, then with an additional $650,000 added.

I made a simple conversion tool that you can use by going here. It’s set it to view only, but go to File -> Make a Copy and you should be able to edit it by putting your own numbers in the fields with the boxes around them. Just to be clear, this is not measuring contract value for future seasons. In order to do that, you’d have to add a lot more to this, it’s a very simple tool.

The takeaway here is that even though it’s going to be very hard for a team to get positive value for all their players, some teams are doing much better than others. While we already knew how bad many of Detroit’s contracts are, we can see that Detroit is currently in a horrible position in terms of value for cap hit.

This can be fixed, and it doesn’t have to take years and years. Some of these bad contracts will be coming off the books in the next year or two. Combining that with getting value from ELCs from players like Zadina can quickly help to move Detroit closer to Winnipeg.

There’s a lot more to be looked at with this data. Just how overpaid on average are players being signed off of unrestricted free agency? Just how helpful is it to get value from ELC contracts? These questions and more will hopefully provided the basis of future off-season articles.

UPDATE: I finished putting in all the cap hits from last year. If a player didn’t play the entire season in the NHL, their accumulated cap hit was used. For each dashboard, you can filter by team, position, and time on ice.

Cost vs Value

Excess Value

Appendix A:

The numbers you see on Corsica for WAR/GAR come from the most recent time Manny ran the regressions on his model using the latest available data. At the end of the regular season, he ran the model, which created the numbers on the site from that time through the end of the playoffs.

At the end of the playoffs, he ran it again, which updated the numbers, not only for the players who played in the playoffs, but also for the players who hadn’t. The reason for this is that the more data the model has, the better it can identify another player’s performance compared to replacement level.

Manny explained it like this:

Say Zetterberg played 100 minutes against Crosby (obviously this is an exaggerated example), if the regressions ran after the playoffs say that Crosby is slightly worse than they did before the playoffs, then Zetterberg’s performance in his minutes against Crosby is retroactively adjusted due to changed expectations.

So for individual players, the number generated after the playoff regressions are the most accurate representations of how that player did because they are based on the most information available.

Because the second part of the article looks at contract value using league statistics, it doesn’t make sense to use the playoff numbers because it would skew things too much since players who played in the playoffs had more chances to provide value.

Appendix B:

Here are the specific numbers used to calculate $ / GAR spent by NHL GMs in 2017-18 and in 2018-19:

Salary Data: CapFriendly

As for where the 19.5 Wins Above Replacement number comes from, in his article How Much Do Wins Cost?, influential hockey analytics writer Hawerchuk says he uses Tom Awad’s GVT model, which determined that a replacement level team would put up around 52 points in the standings. The average NHL team typically puts up 91 points in the standings, for a difference of 39 points, or 19.5 wins.

Another thing to point out is that a replacement level team cannot actually exist because of the salary cap floor. However, since Manny uses a player making the league minimum salary as the definition for a replacement player, we have to use that as the comparison, even though it’s theoretical.