PREDICTING UPSETS: THE 2017 NCAA MEN’S BASKETBALL TOURNAMENT

Abstract

In 2017, approximately 70 million March Madness brackets were completed worldwide, and more than $10 billion was wagered on the NCAA Division I Men’s Basketball Tournament both legally and otherwise (Goldberg, 2017). The difference between winning and losing in these contests is often one’s ability to correctly predict a few game outcomes where the lower-seeded underdog defeats the higher-seeded favorite in what is known as an upset. As the tournament has expanded and its popularity has increased, the term “upset” has become synonymous with March Madness. These games provide elation or heartbreak for players, coaches, students, fans, alumni, bettors and neutral viewers. On average, there are approximately six upsets annually in Round 1 alone of the NCAA Division I Men’s Basketball Tournament. Many theories have been proposed that attempt to identify which matchups will result in upsets. However, most theories rely on “gut feelings” or have a subconscious bias for or against certain teams based on the individual’s rooting interest.Our prediction method combines data sets published by the basketball analytics industry’s foremost experts with historical tournament seed data to accurately predict which March Madness tournament matchups are most likely to result in upsets. It removes personal bias by considering only independently-generated data from neutral sources and provides an objective evaluation of all teams participating in the 2017 NCAA Division I Men’s Basketball Tournament.