Simplifying the Madness of March Madness

This paper investigates the effectiveness of maximizing the expected value of the best-performing entry in multi-entry betting pools for single-elimination tournaments, with a particular focus on March Madness. In such betting pools, participants select winners for each game, and their score is a weighted sum of the correct selections. Due to the top-heavy payout structures in these pools, we study whether maximizing the expected score of the highest-scoring entry, which has been successfully adopted in the sports betting literature, is a suitable modeling approach for March Madness betting pools. Unlike traditional methods that require modeling how other participants make their selections - an especially challenging task given that March Madness only occurs annually - this approach only requires win probability estimates for teams as input, and inherently provides a diversified position across multiple entries. We present an exact dynamic programming approach for calculating the expected maximum score of any fixed set of entries. Additionally, we explore the structural properties of this approach to develop several solution techniques. Using insights from one of our algorithms, we design a simple yet effective heuristic which delivers high-quality results when tested against high-roller betters in a real-world March Madness betting pool. In particular, our results demonstrate that the best 100-entry solution identified by our approach had a 2% likelihood of winning a $1 million prize in a real-world betting pool, showing that the proposed heuristic is competitive even against the best professional bettors.

Recommended citation: Jeff Decary, David Bergman, Carlos Cardonha, Jason Imbrogno, and Andrea Lodi (2025). "Simplifying the Madness of March Madness"; URL https://arxiv.org/abs/2407.13438.
Download Paper | Download Slides