Over the last couple of seasons, I’ve been tracking Game Projections from a number of different models. I want to continue the tracking, but instead of manually tracking the models I want to automate the process. This will make my life so much easier for me.
Let me know if you want me to include your model. You’ll just need to setup a CSV or Google Sheets file with: Date, Home team, Away team, Home probability and Away Probability.
Tracking Performance:
I will be tracking Log loss and comparing it to the other models as well as the implied odds. Implied odds is the calculated probability based on the betting lines. The betting line is found on Betstamp and it’s the best possible line from Bet365, Unibet, Bwin and 888Sport (bookmakers available here in Denmark).
Log Loss
Log loss is a way to evaluate model performance, and it’s calculated like this:
If a team is 60% favorite, then the log loss will be the following:
The model log loss is simply the average log loss for every game. The lower the log loss the better the model.
Bet Results
I will also be tracking betting results calculated based on 0.3 Kelly Criterion. Every time a model sees an edge the bet size is calculated based on the Kelly Criterion. So, the bet results aren’t based on actual betting suggestions, but based on a calculation. Sometimes, it may be smarter to fade the model.
Kelly Criterion:
The model evaluations will look something like this:
Showing Game Projections before Game time:
As a new feature, I want to include other models in the Game Projections Visualization. This way you can easily compare the projections of different models. And you would be able to select multiple models to calculate the average win probability.
Obviously, this requires game projections to be uploaded before game time.
Contact Me:
Let me know if you have a model that you want me to track during the season.
You can reach me on Twitter (@HockeySkytte) or on email (lars.sunesen.skytte@gmail.com).
Best of luck to all the modelers out there.
Hockey-Statistics 