Developing an Economic Model Based on Performance

[Hyped]

In my Advanced Econometrics class in school, part of our mark (a large proportion)goes towards developing an economic model of something of our choosing. I've decided to try and develop a model for NHL players that would be able to negate negotiating contracts as much as possible. In the end I want my model to replicate what a player should be paid for the following season based on the previous season's statistics (ie: performance-based contracts). Off the top of my head, factors that I will be taking into consideration include:

goals
assists
ice-time
experience
+/-
shots
PIM
games played

Does anyone have any other ideas for variables that I could use in my equation? Right now, I just want variables that can easily be translated into numerical form. I will dig into those other variables later on. I just want a place to start right now. So does anyone have any ideas?

Click to read more...

 

[Snap Wilson]

team record

playoff totals in all of the above.

Ice time is probably the most important statistic of all of the above.

 

[boredmale]

Ice time is probably the most important statistic of all of the above.

some combination of ice time and games played

Salary = ($$MONEY$$) X ICE TIME X GAMES PLAYED

 

[Hyped]

team record

playoff totals in all of the above.

Ice time is probably the most important statistic of all of the above.

Would "team goals-for" be a better indicator than team record? Afterall, a team like the Devils usually does not have a whole pile of scorers (only Elias in the past), yet they usually have a very good record.

Playoff totals is a good one...

 

[Hyped]

some combination of ice time and games played

Salary = ($$MONEY$$) X ICE TIME X GAMES PLAYED

Salary is going to be my dependent variable in the equation because I want to see how each variable affects salary. I will be using average time-on-ice in games played for that variable in specific...

 

[USF Shark]

shooting percentage.......if you can't hit the net no matter how much you shoot, you're worthless. On the other hand, your goals represent a good portion of your shot total your value goes up...especially for defensmen, good poweplay Dmen have high shot %.

 

[boredmale]

Salary is going to be my dependent variable in the equation because I want to see how each variable affects salary. I will be using average time-on-ice in games played for that variable in specific...

How about this

Forwards

Salary =
500,000 * Games Played/82 (Base Salary)
+ $150,000*(Player time - 10:00)*(games played)/82 (don't count if negative)
+ ((Plus Minus) - (Team Average Plus Minus))*50,000 (don't count if negative)
+ Goals*30,000
+ Assists*25,000

 

[Hyped]

How about this


Salary =
500,000 * Games Played/82 (Base Salary)
+ $150,000*(Player time - 10:00)*(games played)/82 (don't count if negative)
+ ((Plus Minus) - (Team Average Plus Minus))*50,000 (don't count if negative)
+ Goals*30,000
+ Assists*25,000

That's a relatively simplistic system. What I'm trying to determine is how much effect each variable has on salary. I'm not trying to put exact numbers on what a goal or assist should be worth. If goals go up by 1, salaries should go up by X amount. That's what my model will be figuring out.

Here's an example for what I'll be trying to accomplish:

I want to find out how much salary should increase or decrease the next season if a player scores one more or less goal. So if they score 30 goals, I want to know how much their salary would be affected if they had scored 31 goals or 29 goals. If they had scored one more goal, how much should salary go up by? Or if they scored one less goal, how much should their salary decrease? That is, if goals are even stastically significant, which is what my model will be trying to figure out.

I doubt I've explained this very well; it's not an easy subject...

 

[Sens Rule]

This would be an interesting project. However it seems better suited to baseball then hockey as baseball is far more statistically quantified to individual players than hockey. For example in the early 80's when Rod Langway won 2 Norris trophy's over Paul Coffey, I think it would be very hard to find a statistical system that would place Langway higher or well even close to Coffey in terms of salary although according to the experts of the time he was better.

That said with a project like this I would focus on teams, as well as players. You could make the NHL be a league where there is far more revenue sharing than is currently practised. This would enable to better pay the players on those teams. Like assume that the 30 teams make 2.1 billion in revenue. (close to reality) that is 70 million in revenue per team. For simplification say that there is revenue sharing and players recieve 50% of revenue. So there is 35 million per team. Then make a formula that determines what the salary is per team based on their peformance in the regular season and playoffs. So let's say that the 30 million of the 35 million is allocated for regular season and the other 5 milion per team (150 million) is allocated to only the 16 playoff teams.

So each team has an average of 30 million for the regular season. let's say 25 million is allocated for the 30th place team and 35 million for the 1st place team and the 15th place team get's 30 million, with the other teams getting paid as to where they are placed on that scale. The 16 playoff teams get the 150 million portioned to them as to who far they go.

The method of determining how much of the salary an individual player gets is the same for all teams and also the same in the playoffs (for the playoffs throw out regular season and only count the playoffs stats)

I think the formula should use the base of ice time. After that players get paid for what they do. I would make the base stat to adjust from ice time. So you take all the ice time for the season (for simplicity I won't count regular season overtime or shootouts) 60 min x 5 players x 82 games. (I just remembered about Goalies - there would need to be a different system for them. For simplicity let's allocate 4 million for goalies and the other 26 million for reg players) So you have 26 million for all other players. Again for simplicity I will say there are 26 players that played in the past year as forward or D-Men.

So you take the total minutes played 24600 and divide up that by the minutes played of the 26 players (or however many players who played). That is your base salary for these players. Then you adjust that base using a formula adjusting it up or down based on goals, assists, plus minus etc. Because Defenceman get more ice time but score less points you could probably have the same formula for Defenceman and forwards. (and it is simpler)

The same system would go for the playoffs except that you need to adjust for who far a team makes it in the playoffs. For teams you can adjust by games or minutes played and that works on a pay per performance basis until the final round. Then you could give the team that wins the final 2/3 of the revenue and the loser 1/3.

Hope these ideas help you some.

 

[Sens Rule]

So with my plan above that is based on ice time, I can see what you really want to do is measure goals and stats. My system has not done that thus far but has allowed the performance of teams be measured.

I think ice time is a great measure of a players value if you take team performance into account. In a sport like hockey it is hard to quantify the value of a player by statistics, the coach does that by allocating ice time. I don't know how you could adjust the salaries except to rate the players on a team relative to other players on the team on how many goals, assists and their plus minus. This would be complicated.

But let me say this I would not care one bit about shots and shooting percentage and stuff like that as it doesn't matter really, only goals and assists and perhaps plus minus matter at least in my mind.

 

[SJeasy]

Not sure how you would do this. But, +/- is deceptive. If a line or d-pair is consistently playing the other teams' top lines, their +/- might not be as good relative to their own team although their value to their own team might be higher.

 

[hfboardsuser]

To address the problem of +/-, look into rewarding players who are good outscorers. If you've not heard of this concept before, check out the Library sticky in the Oilers board and read through the fine posts written by Lowetide, igor, mudcrutch and many others on the subject.

 

[DocHolliday]

Just to clarify, are you taking existing salaries and statistics to create an explanatory model? That's the route I would go. You could then see what variables explain salary to which degree, although I doubt you're going to find a linear relationship I would also suggest maybe a dummy variable classification system based on players roles, since stay-at-home defeseman command a salary that will cause some problems comparing the stats with those of a fourth-line plugger. IMO, that is your biggest challenge is seperating between different roles and perhaps the intangibles such as leadership that may inflate a salary.

 

[Hyped]

Just to clarify, are you taking existing salaries and statistics to create an explanatory model? That's the route I would go. You could then see what variables explain salary to which degree, although I doubt you're going to find a linear relationship I would also suggest maybe a dummy variable classification system based on players roles, since stay-at-home defeseman command a salary that will cause some problems comparing the stats with those of a fourth-line plugger. IMO, that is your biggest challenge is seperating between different roles and perhaps the intangibles such as leadership that may inflate a salary.

I cannot take existing salaries for the upcoming year into consideration because there are no stats to go with those salaries yet. But that doesn't really matter that much because the model I am creating is relative to the time at which it occured. All that would have to happen is a simple adjustment for the next year.

I may as well answer other replies in this one as well:

As far as my model goes (so far), I am only including forwards as I think defencemen and obviously goaltenders would both need different economic models.

Like I said earlier, I don't want to include any intangibles just yet that aren't easy to put into numerical form. Right now, I just want the basics and I will jump into that other stuff at a later date when I feel comfortable with what I've got. Even variables such as shots, +/-, and penalty minutes are okay because my model will show me whether they are valid statistics or not so it can't hurt to include them. I will be testing their significance as I go.

I've been thinking about how I would include ice-time in my model and came to the conclusion that I would take the player's average ice-time per game and multiply it by games played to get a number for their entire season.

I know a value could never be placed on a hockey player statistically as there are too many variables that can't be accounted for (leadership, intelligence, etc.), but this topic will at least keep me interested until the class is over.

Has anyone ever attempted something along these lines before?

 

[Weary]

Is the purpose of your model to derive the most accurate prediction, or a fair one? There should be some correlation between draft position and salary. But it's probably not fair to use it to decide what a salary should be. Also along those lines you could use the number of rounds the team won in the playoffs the previous year. As teams do better, their players tend to draw higher salaries.

You also may want to divide time-on-ice into shorthanded, even-strengh, and power play time. I have a feeling that the better paid players get a lot of the power play ice time.

 

[ChemiseBleuHonnete]

model like this do not really work imo... It's total ********.

 

[Hyped]

model like this do not really work imo... It's total ********.

Buh-bye...

 

[RingWraith]

A points per minute statistic might give a good indication of how effectively a given player is making use of the ice time he is supplied with.

 

[DocHolliday]

I cannot take existing salaries for the upcoming year into consideration because there are no stats to go with those salaries yet. But that doesn't really matter that much because the model I am creating is relative to the time at which it occured. All that would have to happen is a simple adjustment for the next year.

Well you could start with known salaries and stats from a previous seasons and use them to derive your model. That way you can see if the model holds over a period of time, or perhaps what factors affect that.