Digitized Honors Theses (2002-2017)

Date of Award

5-2016

Document Type

Undergraduate Thesis

Degree Name

BS

Faculty Mentor

Ermanno Affuso, Ph.D

Advisor(s)

Rajarshi Dey, Ph.D., James Swofford, Ph.D, Matt Wiser, Ph.D

Abstract

Sports bettors and enthusiasts can apply econometric models to estimate results in international soccer. This paper first argues that a well-specified negative binomial regression model better fits the scoring distributions of international soccer than a Poisson regression model. Second, the paper argues that a model with ranking and non-ranking variables is preferred to a model with only rankings. The proposed DAG-ER model finds that non-ranking variables are statistically significant and add in-sample prediction power. A team with a weaker Elo ranking than its opponent is modeled to score fewer goals. Historical mean number of goals scored against the opponent is positively related to number of goals scored, while distance traveled to the match is negatively related to number of goals scored. In comparison to a model with only Elo rankings and a model with only FIFA rankings as predictors, the DAG-ER model is more correct in overall win-draw-loss predictions, total goals predictions, and exact score predictions. As event data such as shot and possession statistics are not available worldwide, the DAG-ER model was designed to be globally-applicable and utilize relatively accessible predictors. Sports bettors and enthusiasts benefit from incorporating variables beyond rankings alone in their prediction strategies.

Comments

© 2016 James Mozur ALL RIGHTS RESERVED

Download

Share

COinS