École Nationale Supérieure de
Techniques Avancées

Abstract

Daily Fantasy Sports (DFS) has swept the nation, and with it has come a wave of websites which provide projections and tools for helping casual fans to pick their lineups. In DFS, fans pay to enter contests in which they select players to build a lineup. Each player has a play money cost and a position. Players must be selected to fill a specified combination of positions while staying under a salary cap. After the players finish their real life games, they are given a fantasy score based on statistics accrued during the game. Lineups in a contest are sorted by aggregate fantasy score and the top scoring lineups are paid out winnings based on where they ranked. We provide a Bayesian model for projecting NBA player performances and two novel optimization tools: contemporaneous lineup selection and a robust optimization counterpart. We try several optimization strategies and determine the best ones based on the Sharpe ratio of returns.