Introducing The Model Aggregator
We built an open source sports betting model that we’d like to share with you.
Basically, what our model does is it aggregates and standardizes predictions from a bunch of other models in order to arrive at a single prediction for each upcoming game.
Here’s an overview of how it accomplishes this:
- First it pulls data from over 50 unique models listed on ThePredictionTracker.
- For each game of the day, each model has a predicted spread which we can compare to the actual spread. We take the difference between these two values which we call the “spread-diff”.
- Since some models tend to differ with the spread more than others (higher standard deviation of spread-diffs), we standardize all of the spread-diffs by model. What we get is called a z-score in statistics.
- Now that the spread-diffs values are standardized, we take an average of them for each game and sort these averages from highest absolute (best pick) value to lowest (worst pick). Whether the average spread-diff for a given pick is positive or negative tells us whether the pick for that game is on the home team or the away team
The model is completely open source meaning you can copy, modify, and run it yourself. The code is available on GitHub along with a more detailed explanation and instructions on how to set it up:
View/Download the model aggregator here: wagerlab/model-aggregator
It will definitely need some fine tuning, but for now, it’s a great start. We’ll start posting the WagerLab Model Aggregator’s picks along with periodic evaluations of its performance, our theories behind how it could be improved, and more.