Exploring the Development and Use of a Multinomial Logit Model to Improve Betting Returns at Saratoga Race Course

Abstract

Inspired by the abundance of first time players, sentimental bettors, and "drunk" money prevalent at Saratoga Race Course during the summer racing season, this paper explores a statistical approach to systematically handicap horse races. A number of observable attributes, believed to affect winning potential, are analyzed for predictive capability. Using a multinomial logit model, fundamental win probabilities are calculated. The public oddsline is then incorporated into the fundamental model. Holdout validation methods are employed and three wagering strategies are tested. Following the procedures outlined in Bolton and Chapman (1986), Chapman (1994), and Benter (1994), positive expected returns are realized.