As someone who follows the Toronto Maple Leafs, it’s obvious that sometimes GMs lose NHL contract negotiations. On the contrary, some managers like Steve Yzerman consistently seem to sign players for below market value. This begs the question if NHL contract negotiations are a repeatable skill among managers. If they are, we will see some year to year correlation in their ability to sign players for above or below market value. Today, let’s dive into my new contract projection model to find out if some managers are consistently good at negotiations, or if your GM getting crushed is likely luck and therefore means nothing going forward.
To define negation skill I’m using my contract projection model to proxy “market value”. Since the model estimates each player’s signing AAV, signing players for less than expected will be considered good negotiation. On the contrary, if some team’s contracts always seem to sign their players for above what was expected, that’s how we will label them as poor negotiators. There is obviously more to it than this, but this looks like a good proxy.
A good negotiation does not necessarily equal a good contract. Take the Jack Johnson debacle in 2018 for example. Johnson was not a defenceman that helped his team win.
Worse yet he was 30 years old, so there was no reason to expect him to get any better. While Johnson was bad, his coach did not seem to think so. He was playing first pairing minutes the season before. Since time on ice has a massive influence on contracts, the model expected Johnson to sign for about $4,000,000 on a 5-year contract. Not because Johnson was good, but because players like him don’t tend to come cheap on 5-year contracts.
So when Jim Rutherford signed Johnson for a measly $3,250,000 per season, it will go down as a successful negotiation. Even though the Penguins are in the process of paying Johnson $7 million not to play for them. Just so they can get out from under his disastrous contract. So remember, this analysis is specifically about signing players for cheaper than their expected market value, not if that contract is actually good.
For some examples, let’s analyze Kyle Dubas’s 2019 offseason. Since Kyle Dubas was the GM at the time, he will take responsibility for the contracts being signed. The Leafs signed 11 contracts this offseason of various sizes. Here is what the model projected each player to make compared to what Dubas signed them for.
The model shows a mixed bag of results for Kyle Dubas in 2019. The most meaningful contract he signed was the Mitch Marner disaster. The model estimated Marner’s market value was around $9,500,000 per season and Dubas signed him for about $11,000,000. After Marner, Dubas’s offseason looks much stronger. He signed Kerfoot for about $1,000,000 per year less than expected, and Jason Spezza for almost $2 million less than expected. The Ceci contract was more than expected by about $1 million, but Dubas had no control there.
Overall, Dubas signed 11 players for about what was expected. The average expected AAV was about $2,900,000, while the average actual AAV was about $2,700,000. This means Dubas’s contracts were about $200k less than expected on average. This gives him an AAV above expected of -0.22, since he saved his team about 1/5th a percent of the cap, per contract in 2019. Note negative is good because you want to sign players for less than expected.
If negotiation is a skill, we will expect this number to be somewhat correlated with future results. Why? Well if something is a skill, it will tend to repeat itself over time. If something is pure luck, past results will not be correlated with future results. Is negotiation as I have defined it a repeatable skill? Let’s look at the data.
The year to year correlation between GMs average AAV above expected is small, but there is some signal in the noise.
We are dealing with super small samples here. GMs only sign about 3-10 contracts per offseason, so the year to year correlation of GM average value above expected is small, but it is statistically significant.
When regressing the previous season’s data on the following season, we see a meaningful relationship in the year over year GM value above expected. Note the very small P-value suggests the year to year correlation is almost certainly not noise. This is a strong sign that negotiation is a repeatable skill among general managers. So beyond contract projection models, it appears although some GMs are able to consistently sign contracts for below market value, and some GMs have consistently signed contracts for above market value.
Looking at the regression results above, the year to year correlation of GM average value above expected is even stronger in the second regression. This was done by weighing the data points based on the total size of the contracts signed. This way if a GM signed $100 million in contracts one offseason, that data point would be weighed more heavily than a rival GM who signed 3 league minimum contracts the same offseason.
So, weighing the results based on the total size of contracts being signed makes the data more repeatable, and it also makes more sense descriptively. Think about the Kyle Dubas example from above. On average, Dubas signed his 2019 contracts for below market value. The problem is that he signed about $115 million worth of contracts, but half of that money was given to Mitch Marner. So Dubas got crushed when it mattered most but did incredibly well on many smaller contracts. So was Dubas actually a good negotiator in 2019? Almost certainly not, and weighted numbers would reflect that.
Altogether, it looks like negotiation is a repeatable skill among GMs, and therefore worth investigating further. The next step is weighing the results based on the size of the contract so our estimates more accurately reflect whether or not they were successful in any given offseason.