Location

Caesars Palace, Las Vegas Florentine III

Start Date

29-5-2013 10:30 AM

End Date

29-5-2013 12:00 PM

Abstract

A variety of methods are often employed to forecast the outcome of sports events and to shed light upon the factors that influence contestants’ winning prospects. We explore the potential of combining forecasts to improve the accuracy of sports event prediction. Successful combination usually involves accurate and diverse individual predictions. Further, to accurately predict the outcome of sports events it is important to effectively account for the intensity of rivalry among contestants. Consequently, we develop modeling procedures and practices for addressing these challenges: First, in order to construct a large and diverse library of base models, we employ a range of surrogate measures of target output that facilitate the use of regression and classification. Second, a forecast calibration procedure is developed to enable average-based pooling mechanisms. Third, to effectively account for competition, a stacking paradigm towards forecast pooling is implemented by integrating conditional logit regression and log-likelihood-ratio-based forecast selection. The appropriateness of these procedures is confirmed by empirical experimentation using data related to horseracing events run in Hong Kong. The results may have important implications for regulators of associated betting markets (concerned with issues such as market efficiency).

Keywords

Combination forecasts; Horseraces; Market efficiency; Competitive event prediction

Disciplines

Gaming and Casino Operations Management | Mental and Social Health | Psychology | Statistics and Probability

Comments

Moderator: Ashok K. Singh

Session 2-2-F Forecasting

File: 18 PowerPoint slides

Attached file: Abstract

 
May 29th, 10:30 AM May 29th, 12:00 PM

Session 2-2-F: Combining forecasts to predict the outcome of horseraces

Caesars Palace, Las Vegas Florentine III

A variety of methods are often employed to forecast the outcome of sports events and to shed light upon the factors that influence contestants’ winning prospects. We explore the potential of combining forecasts to improve the accuracy of sports event prediction. Successful combination usually involves accurate and diverse individual predictions. Further, to accurately predict the outcome of sports events it is important to effectively account for the intensity of rivalry among contestants. Consequently, we develop modeling procedures and practices for addressing these challenges: First, in order to construct a large and diverse library of base models, we employ a range of surrogate measures of target output that facilitate the use of regression and classification. Second, a forecast calibration procedure is developed to enable average-based pooling mechanisms. Third, to effectively account for competition, a stacking paradigm towards forecast pooling is implemented by integrating conditional logit regression and log-likelihood-ratio-based forecast selection. The appropriateness of these procedures is confirmed by empirical experimentation using data related to horseracing events run in Hong Kong. The results may have important implications for regulators of associated betting markets (concerned with issues such as market efficiency).