Estimating NHL Expected Free Agent Value

The following essay on Estimating Expected Free Agent value can be found in my book series "How to Win or Lose at NHL Free Agency" available on Amazon. It describes how I built my algorithm to estimate the market value of any given stat-line, and thus, determine the amount that any given player was over or under-paid.

If you are planning to grade or rank NHL contracts, you will need to create an algorithm to estimate the Expected Free Agent Value of any given single season stat line. This will give you an approximation of the market price of a player's annual production that can be used to measure the quality of the investment. Alternatively, if you are attempting to predict what deal a player might sign in the future, then age and past performance become more important factors. My own E[FA$] algorithm does not care what anyone has done in the past to originally procure their salary, it is only concern is what they’ve done that year to earn the money they are currently being paid based on the market value of the production. 

The next thing you’ll need to write your own algorithm, is a large database of NHL contracts and the statistics produced to earn those deals. My own database has over 6,500 entries. Salary is almost entirely a function of Points (with a slight premium for goals) and Average Time On Ice. When estimating forward value, scoring production is the most important factor; whereas for defensemen, logging big minutes is just as important as helping to put the puck in the net. The extent to which any given venture can be considered a success or failure is measured by how much more or less you are paying over or under the worth of their output.

My algorithm to determine expected free agent value is not necessarily the amount of money that I personally believe they should have been paid, but rather a reflection of what the market tends to pay. Some analysts prefer to construct their own models where value is estimated by wins above replacement or some other metric, but I prefer mine to be “Market Price” determined by what General Managers are collectively willing to pay on average, which is a function of supply and demand for talent. But obviously the “market” isn’t always correct and several dots on the scatter plots were bad decisions that turned out poorly.

The number of years that a contract lasts does affect what the salary will be, especially for older players. Teams will deliberately sign star veterans to extensions long past anticipated peak performance in order to get a few more seasons of quality productivity at a lower cap hit. This represents a flaw of crafting an algorithm to approximate free agent salary alone based on single season statistics. When I’m doing my annual free agency predictions, term absolutely factors into the equation. With younger players, often times teams will have to pay extra salary to buy years of service where the player could possibly be an unrestricted free agent. A restricted free agent might be willing to accept a lower salary if it means getting to the UFA market faster. Once a player attains that freedom, he cares more about salary multiplied by term. 4 years for $5M is more total money than $6M for 3 years.

Using the statistics currently available and attempting to determine what the market SHOULD pay can be a slippery slope. It is deciding an arbitrary number for value, versus taking a data set of player stats in their free agent seasons and correlating it to the salaries they obtained. That’s the real market value of a stat line, not setting my own standard for what I believe the market should pay for any given asset. The price the market pays is based on so much more than just a stat line. Teams pay scouting staffs to watch players and determine if they are a good fit for the organization. It may not be perfect and can lead to mistakes, but it’s also not crazy to say that NHL teams are better at identifying and valuing talent than arm-chair mathematicians.

FORWARDS 

If you’re embarking on a journey to predict what free agents might get paid, the easiest place to start is the forward position. The relationship between statistics and salary is strongest for forwards, also producing the greatest number of giant pay days. There are 3 times as many contracts with an average adjusted cap hit over $10M for forwards than defensemen (60% of the forward jackpots are centers, despite teams carrying twice as many wingers as centers). Term does factor significantly into my calculations when making actual free agent predictions every summer, but my E[FA$] algorithm is more about assigning a dollar value to a single season stat line for the purpose of determining how much a player was over/under paid any given year.

Centers tend to get paid more than wingers and that gap only increases with points scored. If you don’t score many points, you’re getting close to the league minimum regardless of whether you’re taking faceoffs or not. Whereas a 90-point center would earn roughly $1M more on average than a winger with equal production. In my own algorithm, all forwards are treated equally, with a slight premium for centers depending on point production. For statistical simplicity, my definition of center is a player who averages at least 6 faceoffs per game. There is a substantial “mushy middle” of those who mostly play on the wing, but still take draws, like Gabriel Landeskog and many others who play both positions.

DEFENSE 

Predicting salaries for defensemen is much more difficult than for forwards. The occurrence of low scoring players earning multi-million dollar salaries is far higher on the blueline. That may be in the process of changing while preferences for offensive puck moving D-men increases, but if you are looking at the historical data, ice time per game has a similarly high correlation to salary as does points. If an offensive specialist is also a defensive liability, it makes it harder to get him into games because the consequences of mistakes is magnified with defenders. Being able to “eat minutes” is a skill even if the point output is low.

In the past we saw teams spend more money to acquire big and physical specimens like Derian Hatcher, who became less effective after the obstruction crackdown and elimination of the red line. Speed and the ability to move the puck have become a more valuable commodity. Guys like Erik Gudbranson can still get paid, just not as much as past seasons. Mike Komisarek was one of the last dinosaurs to land a monster pay day and it blew up in Brian Burke’s face. This type of player has been slowly phased out of existence in a league that has been evolving towards speed and puck moving. Defenders who lack footspeed have become a greater liability in the new NHL.

The term “puck moving defenseman” is on now on every General Manager’s shopping list, as the game trends to greater emphasis on high octane offense. Scouts are scouring the amateur leagues looking for the next Cale Makar. Those who show proficiency in both ends of the ice have added value. Specialists who only play in offensive situations will have lower ice time and earn slightly less money.

GOALIES 

Goaltending can be such a fickle thing, whether you are trying to predict future performance or expected salary. As hockey statistics continue to evolve, so to is our understanding of what makes goalies good or bad. Some can be awesome or terrible on one team, get traded, only to see their performance boomerang in the opposite direction. Some analysts believe that goalies (like running backs in the NFL), have become interchangeable and are nothing more than a product of the system playing in front of them. For example; playing in a Barry Trotz system is clearly advantageous for goaltenders, looking at Islanders netminders before and after his arrival. Teams that play tight defense can create good goaltenders. Some would even argue that Martin Brodeur doesn’t belong on the Mount Rushmore of great goalies despite owning the record books because his career stats were boosted by strict adherence to the neutral zone trap by the New Jersey Devils.

If we look at a sample of 233 pending free agent goaltenders who played in at least 30 games and the salaries they obtained, the difficulty of predicting goalie salaries becomes more apparent.

Looking at goalie statistics the season before they become free agents and compare their numbers to what they eventually received on the free agent market, the variable with the highest correlation to salary is minutes played. The number of games a goalie plays has a substantially higher correlation to salary than save percentage. The biggest reason for this is that plenty of goalies can post an impressive SV% over a small sample of games. The best performance statistic available is Goals Saved Above Average which calculates how many more goals you saved or allowed compared to league average SV% facing the same number of shots. Minutes Played is effective because goaltenders are always in competition with each other for starts, and the quantity of starts is tied to the quality of play (or at least how much better you are than the alternative)

Many consider Save Percentage to be one of the best measures of a goaltender’s quality, but as you can see in the charts above, it is a terrible predictor for what you can actually expect a goaltender to get paid in free agency. Many in the analytics community love the Goals Saved Above Average metric (how many more/ less goals you prevented/allowed compared to league average save percentage based on your shot volume), which is a far better predictor than Save Percentage, but still lacks a strong correlation.

Games won certainly appears to be one of the best predictors of the conventional goalie stats, but that’s mostly because it’s highly correlated to the most important variable, minutes played. Including wins in my algorithm did not add to the accuracy. Mine is based mostly on minutes played and GSAA and has a correlation of 80% to actual free agent salary attained. Some of the outliers on the scatterplots above include goalies who signed extensions a year in advance, so their outlier season did not have an effect on the contract they eventually signed, like Carey Price. Some of the outliers under the regression line were goalies who were thrust into starting roles because of injury or the lack of a better alternative. Playing well in 30 extra games because a starter was injured doesn’t necessarily mean those minutes were earned, or that the success is replicable over a growing sample.