Improving Foresight Predictions in the 2002-2018 NFL Regular-Seasons: A Classic Tale of Quantity vs. Quality

Abstract

Utilizing a modied Bradley-Terry model, we develop a method of making foresight predictions of 2002-2018 NFL games by incorporating a home-eld parameter into previously established ranking models. Knowing only the home team and score of each contest, and taking into account previous predictions, we optimize this parameter considering one of two things: the quantity of correct picks to date or the quality of predictions to date as measured by a quadratic scoringfunction. Our main results establish that optimization of quality-rather than quantity-when making a prediction has higher overall accuracy.