WITHDRAWN: Experts vs. Equations: A Case Study in the Prediction of MLB Games

• Author(s): Chamont Wang and Danielle Zanghi
• Keywords: Neural Networks ; Regression ; Stochastic Gradient Boosting ; Variable Importance ; Marginal Effect ; Ensemble Model

Abstract

In summer 2009, the lead author of this project spent months tracking the experts' predictions of the Major League of Baseball games from the ESPN website, http://espn.go.com/mlb/lines. The experts who posted their predictions are from ATS411.com, SPORTSBETTING.COM, Sportsbook.com, and SBGGLOBAL.com. Our lead authorfound that the experts are correct 58% of the times.

In our study, we used JAVA code to download the data from the MLB website, http://erikberg.com/mlb/. Our training data consists of box scores of 2008, 2007, and 2006 (n = 3*2,400 = 7,200 games). We then try to predict the outcomes of the matching teams in 2009. In the process, we tried to identify the variables that are most important in our prediction. In addition, we try to compute the marginal effects of neural network and other so-called "black box" models to help interpret the impact of the relevant factors.