[NYEnvy109]

Hello everyone. I wrote a post trying to backtest a system I developed in Excel recently. The system went 13-3 (+9.15 units).

I would like to post the system so we can work together and possibly "fine tune" it to be a reliable handicapping tool. I figured everyone in here knows something that I don't and vice versa so let's team up and develop a kickass model!

So here is the system:

NHL System	Away Team ("AT")	Home Team ("HT")
(1) Games Played
(2) OFFENSE: Total SOG
(3) DEFENSE: Total SOG
(4) Starting Goalie Save Pct.
(5) OFFENSE Total SOG + Opponent Defense Total SOG	AT (2) + HT (3)	HT (2) + AT (3)
(6) Predicted SOG	AT (5) / [AT (1) + HT (1)]	HT (5) / [AT (1) + HT (1)]
Predicted SCORE	[1- HT (4)] * AT (6)	[1-AT (4)] * HT (6)

As you can see, I try to predict the amount of Shots on Goal ("SOG") each team is going to have and then predict how many goals each team will have based on the opposing goalie will make depending on their save percentage.

I know obvious flaws in this system (e.g., key injuries; system relies a lot on goalie play; PP/PK) but I think it's a good foundation for a final product.

I would want to add in points for and points against but I dont know where. Also, maybe it'll be more accurate if we use SOG percentages instead of averages.

I'm just throwing somethings out there.

Hopefully we can get a nice discussion going.

Thanks!

[bowski44]

I would throw out the element of goaltending and focus more on predicting shots on goal differential. Everything I've read on shots on goal percentages are that it's very "luck based" and everything regresses to the mean (don't remember what the mean is though) much like BABIP for baseball. Shot differential has a very high correlation to puck possession and offensive zone time and this correlates to future success. I would google Fenwick if you are not familiar with the term and read as many articles as you can on this stat. Fenwick close might be one of the best stats out there; however, it has a small sample size of the game and you may want to use score adjusted Fenwick instead because it uses a larger sample size of data.

Fenwick = (SOG FOR + Missed SH FOR) - (SOG AGAINST + Missed SH AGAINST)

Fenwick close only counts the 1 goal differential in 1st and 2nd period and tie games in the 3rd

Score adjusted Fenwick counts all game scenarios but uses adjustments based on historical shots on goal data

What I would do is take these Fenwick numbers and then adjust them to their opponents stats. When I say opponent stats I mean you take every team they've played against and average out their Fenwick scores. For example say you have the best team in the league plays the worst defensive team in the league. Their stats will be misleading because they have played such weak competition


Eventually you would have a Fenwick score for a team: let's say +7 per game against opponents who give up 6 shots per game giving you a differential of +1 which shows this team is better than average etc... I would then take those differentials and try to develop a model or system that uses that data to predict scores or winning percentage or whatever you want to do... Be sure to let me know if you come up with something.

[orhane]

Do you create a minimum margin of victory?
Maybe it is useful to include standard variation into the equation. So the effect of differences in the average numbers are not so high.