Doctor No
Registered User
The last thread was getting a bit tangled, so I'm starting a fresh thread for the 2016 Stanley Cup playoffs.
Some of the clarifications from the last thread, all here in one handy place:
What is a Simple Rating System? Essentially, it's the sum of a team's (1) goal differential, and (2) average opponent's strength. The second component is the weighted average of a team's opponents' SRS estimates, which leads to an iterative component (for the fellow nerds out there, although it's iterative, the matrix is non-singular so you can't just find the eigenvectors).
How do you use a SRS? For a single game, the home team will be favored over the road team by (Home team's SRS) minus (Road team's SRS) plus (Value of home ice).
What are some flaws of SRS? Because of its buildup, the base SRS algorithm considers all games equally - there is no "recency" factor. It also considers all goals the same (the seventh goal in a 7-1 win is the same as a game-tying goal with 0:03 remaining).
The SRS also can't see trades or injuries, or "momentum" if you're into that.
What are some improvements of SRS? Remember that the first "S" in "SRS" does stand for "Simple". The first two issues described above can be adjusted for. When I'm actually using this to make money, my algorithms adjust for these factors, and some other things which I've found increase predictive accuracy.
I publish the base SRS here (instead of my proprietary methods) because it's easier to explain, and because it's open-source. My goal from this is to get enough people interested in these types of things to develop their own ideas and thoughts ("hey, the SRS model keeps underpredicting Anaheim. I bet I can do better than that!")
Some of the clarifications from the last thread, all here in one handy place:
What is a Simple Rating System? Essentially, it's the sum of a team's (1) goal differential, and (2) average opponent's strength. The second component is the weighted average of a team's opponents' SRS estimates, which leads to an iterative component (for the fellow nerds out there, although it's iterative, the matrix is non-singular so you can't just find the eigenvectors).
How do you use a SRS? For a single game, the home team will be favored over the road team by (Home team's SRS) minus (Road team's SRS) plus (Value of home ice).
What are some flaws of SRS? Because of its buildup, the base SRS algorithm considers all games equally - there is no "recency" factor. It also considers all goals the same (the seventh goal in a 7-1 win is the same as a game-tying goal with 0:03 remaining).
The SRS also can't see trades or injuries, or "momentum" if you're into that.
What are some improvements of SRS? Remember that the first "S" in "SRS" does stand for "Simple". The first two issues described above can be adjusted for. When I'm actually using this to make money, my algorithms adjust for these factors, and some other things which I've found increase predictive accuracy.
I publish the base SRS here (instead of my proprietary methods) because it's easier to explain, and because it's open-source. My goal from this is to get enough people interested in these types of things to develop their own ideas and thoughts ("hey, the SRS model keeps underpredicting Anaheim. I bet I can do better than that!")
Last edited: