Statistically Enhanced Performance in Salary Cap Fantasy Football

• Activity Number: 617
• Type: Contributed
• Date/Time: Thursday, August 7, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Sports
• Author(s): Elizabeth G. Hill*+ and Robert Hill
• Companies: Medical University of South Carolina and South Carolina College of Pharmacy
• Keywords: Fantasy football ; Bayesian hierarchical model ; Dirichlet process prior ; Data Scraping ; Prediction ; Integer linear programming

Abstract

In salary cap fantasy football (SCFF), players are assigned salaries based on their perceived preseason value. Each week SCFF owners build teams with a specific composition of positions within a salary cap constraint. For owners who rely on expert projections to pick weekly lineups, a season can hinge on the projections' accuracy just as much as the tipped pass, the bad call or the kick that sails wide left. We present a two-stage approach to SCFF play - weekly prediction followed by team selection. We predict players' fantasy points using a Bayesian hierarchical model that incorporates the cumulative history of players' actual fantasy points, matched projections from multiple experts, and the current week's expert predictions. We model expert bias as arising from a Dirichlet process mixture, facilitating identification of classes of players for whom a given expert predicts fantasy points with common bias. Team selection is then based on a 0-1 integer linear programming algorithm that maximizes the total predicted bias-corrected fantasy points subject to salary cap and team composition constraints. We demonstrate our approach using 2013 NFL regular season data.