Despite the large number of systems that have been developed for rating and ranking major college football teams in the United States, only two prior studies have attempted to identify an ensemble of those models - i.e., a metamodel - that is most predictive of future game results. Using 29 college football rating systems, data for 5925 games during 2016-24, and a k-fold cross-validation process focused on model predictiveness, a five-system metamodel is developed to predict the victory margin for games in the ensuing week. The model achieves strong results vis-à-vis the opening, midweek, and closing betting lines, and is statistically significant in the presence of the opening line in validation and test samples. The model is potentially useful for the College Football Playoff committee, media, fans, the betting market, and oddsmakers, and provides a benchmark against which future rating systems can be assessed.
© Copyright 2025 Journal of Sports Analytics. IOS Press. Alle Rechte vorbehalten.