An NFL Gambling Manifesto Development of Binomial Probability Models and Neural Networks for Detection of Biases in the NFL Gambling Market

Abstract

Motivated by the rapid expansion in popularity surrounding sports gambling, especially with regards to the National Football League, this paper explores a statistical and economic approach to attempt to locate potential biases or profit-making opportunities in the NFL gambling marking. Using game and bet outcomes from the 2003 to 2013 NFL seasons, a variety of observable game and team attributes, associated with giving teams marginal advantages, are analyzed for their potential use in handicapping or predicting NFL straight bet outcomes. Then, a variety of binomial probability models that build on previous research are used to estimate changes in bet win probability. With the exception of double-digit home underdogs, the results do not show an overall home underdog effect in our sample, but instead show a mid-season bias against home underdogs. These previous sections results and variables are then incorporated into a committee of neural networks model, which is estimated using a back-propagation algorithm, to be used to predict bet outcomes. The model achieves higher than breakeven betting probability in the long run with several strategies, suggesting that the NFL gambling market is not fully efficient.