Weekly Playoff Probabilities: The Beginning

Last season I took score-adjusted fenwick and applied it to a Monte Carlo simulation of the NHL season to generate playoff probabilities. We start the season this year with an updated model. First, let’s take a look at where everyone stands today.

First and foremost, I wouldn’t read to much into these numbers at this point. Only a handful of games have been played by each team, subjecting their record to massive fluctuation as a result of unrepeatable measures.

The model I constructed this year has an added wrinkle. Last year I fixed PDO at essentially 1. My assumption was that PDO was virtually unrepeatable, and the best prediction would be 1 for any team, for any length of games. But I realized that considerable variability must be expected. The model still only applies even strength excluding empty net data. Perhaps in future iterations I’ll adjust it to include power-play data, which would likely increase the spread between teams.

This year I built the model to allow for normal variance in PDO. As a result, the mean points doesn’t really change for teams. However, and most importantly, playoff probabilities do. We see a much broader distribution of potentially points that each team can accrue, and hence, playoff probabilities closer to 50% than in the year prior. We’ll see how this all plays out this year, but I think we’ve got an improved model over last year.

As a side note, I can’t believe the NHL didn’t try to fix the point system with all the CBA talks. If they want to encourage scoring they need to move back to a zero sum game where total amount of points awarded during regulation equals the total amount of points awarded for OT/SO games (right now it’s 2 vs. 3). The NHL is encouraging boring hockey by enticing teams to play for draws. The idea of calling more bullshit penalties to try to increase scoring is so fucking stupid its driving me crazy.