The results generated by this model are created using a (modified) Dixon-Coles' based technique. In short, team goals are considered independent Poisson functions, with each team having attack and defense strength impacting the Poisson lambda value. In addition, adjusters to increase low scoring game probabilities and tie game probabilities are used, as well as a home-ice advantage term.

Seasons are simulated using team characteristics that slowly revert to the mean. Reversion would occur at 33% to the mean over the course of a full season, prorated for actual regular season time remaining.

In contrast, playoffs are simulated using stagnant team characteristics from the time of simulation through the conclusion of the playoffs. If playoff predictions are generated prior to the season end, team capabilities are not reverted to the mean prior to simulation.

The model is updated nightly with the day’s scores and predictions are re-run. The biggest shifts appear after upsets but team capabilities are slower to respond as a total of ~10 years history is included in the model (with more recent games provided higher weight).

None of the content of this site should be considered ‘betting advice’ nor should the model creator be considered liable for any losses suffered if the model is used as input to betting techniques.

Follow updates to this model on Twitter, Github, or here. The model is written in the R language, and most functions are available in a R package.