Predicting NHL Playoff Outcomes Based on Regular Season Data

• Activity Number: 316
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Sports
• Author(s): Nilesh Shah*
• Companies: University of Pittsburgh
• Keywords: hockey ; predictive modeling ; corsi ; fenwick ; simulation

Abstract

NHL playoff winners are difficult to predict based on regular season standings. Point standings can be influenced by factors that cannot predict future success, such as luck, shoot-out victories, and inherent variability. I propose a method to predict playoff success based on teams' regular season underlying statistics. Using data from the 2007-2008 season to the 2012-2013 season, I developed a cross-validated logistic regression model using possession based statistics, save percentage, and special teams play to estimate the probability of a home team playoff series victory for any match-up. I then used these probabilities to simulate the 2013-2014 NHL playoffs 10,000 times. The model was updated after each round of the playoffs. The initial model showed Stanley Cup Final participants Los Angeles Kings and New York Rangers as the 2nd and 3rd most likely teams, respectively, to win the Stanley Cup. The model will be updated using more recent research and data for the 2014-2015 playoffs.