Modeling Games with Uncertainty and Third-Party Selection Case Study - NCAA I March Madness

The goal of this project was to model an economic game where there are there is a set of competing individuals with limited information about each other and who have some flexibility in their schedule of competition. Depending how each individual compares against the others, an outside party will select individuals. Therefore, the goal of everyone is to engineer their schedule so that their performance will be best maximized in light of the strength of their schedule so that they get selected by the outside party. This project was inspired by the firms who have a set local competition but then can choose to compete elsewhere against other firms. Each firm has access to the previous firm’s performance but cannot be certain about its future performance. Each firm would like to wisely choose its schedule so that it can maximize its performance and be selected by the outside party. In this project, this model is being studied by mapping it to the NCAA March Madness tournament. In NCAA Division 1 Basketball, there are 353 teams that compete in a regular season. At the end of the season, the top 68 teams are selected to compete in the NCAA tournament. 36 of the selected teams are selected by a committee based on their performance in the regular season. These are called “at-large” bids. In this project, we have trained a neural network to act as the NCAA Selection committee which can correctly predict whether a team gets an automatic bid or no bid with ~97% accuracy. Furthermore, we simulated 13,392 teams to see how different schedule strategies at different relative strengths would affect the likeliness of a team to achieve an at-large bid. In the Ivy league, there is a strong in correlation between selecting a schedule with a larger portion of teams who were stronger than you the year prior and achieving an at-large bid.