An Optimization Approach to Fantasy Football Draft and Roster Management
Open Access
- Author:
- Stauffer, Daniel
- Area of Honors:
- Industrial Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Guodong Pang, Thesis Supervisor
Catherine Mary Harmonosky, Thesis Honors Advisor - Keywords:
- Fantasy sports
Fantasy football
Multi-criteria optimization
Linear programming
Optimization
Mean-Variance - Abstract:
- Fantasy football is already one of the most popular games in the United States, and its player base only grows each year. The fantasy sports industry is worth billions, and this is in no small part to the game’s parallels to gambling—each fantasy player contributes money to be re-distributed based on a fantasy league’s final standings; in doing so, the players enter into a game of risk and expected return. The goal of this applied research is to look at techniques in two primary phases of the fantasy football season that can assist a fantasy player in making sound decisions. First, the research considers decisions made during the initial draft selection process of a fantasy football league. This piece of the research considers the tradeoff between the projected score of an NFL player and their demonstrated historical weekly variance in performance. This comparison of projected scoring and demonstrated variance is carried out using a bi-criteria optimization model. Considering the draft behaviors of opponent owners, simulations were carried out to assess the effectiveness of the mean-variance comparison in drafting fantasy teams. An efficient frontier of teams was created to demonstrate the tradeoff between expected score and demonstrated variance, and a set of teams from that efficient frontier were compared. As a result of the analysis, it was found that modeling player selection at the draft stage as an effort to maximize projected scoring and minimize total team variance could result in a team of players which scored near the maximum projected number of points with a substantially lower variance. The second part of this research considers season-long ramifications of roster changes and starting lineup selections made at single decision points throughout the course of the season. First, a model was created to maximize the total number of points scored by players in the starting lineups over the course of the season. Next, an algorithm was created using this maximization model to test the “value added” from making an “add/drop” move in free agency. In other words, the effects of a single-week roster change were assessed based on the move’s effects on the team’s scoring projection for the entire season. This novel consideration of the season-long impact of weekly moves provides a formal in-season decision-making method that projects to only improve the outlook of a fantasy team. Finally, the draft optimization technique and free agency assessment algorithm were considered in concert. A data-drive optimization model for the draft provides a fantasy team owner with a consistently high scoring team and model which is robust to the behavior of other fantasy owners at the draft. Additionally, a simple maximization technique allows the team owner to project their fantasy team’s final point output based on optimal starting lineups. Finally, at each week throughout the regular season, the team owner is given the ability through the free agency optimization algorithm to assess the long-term impacts of any immediate moves that may be under consideration. In aggregate, these techniques combine to produce fantasy football teams with high season scores and low variances which prove to be adaptable and robust throughout the season.