Peter Tanner is the owner and operator of MoneyPuck, which has expanded into one of the essential stops for any hockey fan looking to understand more about the game. Yesterday, Tanner kindly answered some questions from Japers’ Rink writer Greg Young about their process, why their model is (relatively) optimistic about the Caps chances, and what’s to come from the site.
Greg Young: So you’ve been developing MoneyPuck for awhile now...what was the inspiration behind you starting the site?
Peter Tanner: The website started with in-game win probability in 2015, which I had previously seen for the Superbowl. I was really just interested in building it for my own interest, but it was cool to see it catch on. Before building the website I did research into NHL concussions, which the New York Times picked up, as well as writing an article for 538 about predicting which coaches would get fired. That helped motivate me to do more in hockey analytics.
GY: Without divulging any more than you’re comfortable with, what are the components that go into estimating the percent chance that a certain team wins on any given night?
PT: The main factor in the model is expected goal metrics for each team, with expected goals in recent games given more weight. Being the home team, the rest each team has, shooting talent, goaltending talent, and penalty discipline are also significant factors. There’s a full explanation of our model on our website.
GY: Your model has consistently been one of the more optimistic ones towards the Capitals. Do you think there’s anything that you’re catching about the Caps that anyone might be missing? What do you think explains the discrepancy?
PT: I’m a bit surprised most people are picking the Bruins so strongly. These are two pretty evenly matched teams, with the Caps having home ice advantage. The Capitals have been on a good streak with expected goals since mid April, though the Bruins have as well.
GY: About the Caps vs. Bruins series: what do you think are the key factors that will determine this series?
PT: Not sure if I have a good answer for this one beyond the goaltending and luck/randomness questions below.
GY: Obviously Capitals fans are deeply concerned about the goaltending situation. How do you see that shaking out, and how does that impact the Caps chances?
PT: Both Vitek Vanecek and Ilya Samsonov let in about one extra goal every four games compared to an average NHL goalie this season. Boston’s Tuukka Rask has been a slightly above average goalie this year (saves an extra goal once every six games). If they play at their season average performance, that’s 2.5 goal advantage to Boston over the course of the series because of goaltending. A general rule in hockey is that 6 goals is equivalent to one ‘win’, so it’s about a coin flip that goaltending talent will cost the Caps a game in the series. Goalie performance can fluctuate widely in small samples and they are unlikely to actually play ‘average’ over the course of a series.
GY: As has been detailed elsewhere, there is quite a bit of chance that goes into hockey. How does your model account for that and what do you think is a good way for a hockey fan to think about the randomness element in games?
PT: There is a ton of luck in hockey. Our models account for it by not being very confident in their predictions. Most individual game predictions are close to 50/50. Having said that, this has been a unique year in the NHL with very few average teams, which has made games easier to predict. The 3rd and 4th rounds of this year’s playoffs may have some surprises too since there’s been no play outside divisions yet this year.
GY: Anything else exciting that you have planned for MoneyPuck that’s coming up?
PT: Currently all our game predictions are made the night before the game and do not factor in starting goalies. We recently built a bot that identifies tweets from journalists to identify who is starting in net. These goalies are now listed on our homepage within minutes of the tweet. For next season we hope to factor starting goalies into our game predictions.