With the NBA regular season wrapped up, I thought it would be good to do a recap post on the model. Because of the completely different nature of playoff match-ups and more variables that the model cannot account for, I’ve decided to not run the model for the playoffs. That being said, here are the model’s regular season results:
December 22 – April 10
Record: 180-146-8 (.539)
CLV (tracked since 1/18): -9.5
Units by Month (full months only):
From the trade deadline to one week after the all star break, the model went down 11.16 units, presumably due to trades and acquisitions not being able to be accounted for.
Win % by CLV:
Win % by Team:
Stats with New Combinations of Exclusions:
As the model works right now, it excludes a game from being picked if they have one of these characteristics:
- Team is on day 2 of back-to-back
- Spread for the game is less than or equal to 3
- The model/spread difference is not within one standard deviation of the average difference
|Back to Back Included||52.8%||+12.64|
|Spread < 3 Included||53.2%||+14.91|
|Difference Outliers Included||53.1%||+14.28|
|Only Back to Back Excluded||53.0%||+15.67|
|Only Spread < 3 Excluded||51.4%||+0.49|
|Only Difference Outliers Excluded||51.9%||+8.03|
If the model was ran the entire season using the exclusions as I have them now, the model would be up 16.80 units.
Based on these calculations, there are a few things I can change for next season. For one, I will stop running the model from the trade deadline until after the All-Star break so the statistics I use can catch up with roster changes. I also may place higher weight on picks that feature an away favorite, as those picks hit almost 9% more than any other combination. I will not make any change to the model exclusions, as no other combination produced better results.