How well do Elo-based ratings predict professional tennis matches?

This paper examines the performance of five different metrics for forecasting men’s and women’s professional tennis matches. We use data derived from every match played at the 2018 Wimbledon tennis championships, the only grass court Grand Slam tournament. The metrics we use are the betting odds, the official tennis rankings, the overall Elo ratings, the surface-specific Elo ratings (Elo based in this case only on matches played on grass), and a composite of some of the above. The Elo rating system is a method of ranking players based on their past matches, weighted by the ratings of the players they competed against. The performance indicators we use are prediction accuracy, calibration and model discrimination. For men’s tennis we find that the betting odds outperform the other measures in terms of prediction accuracy and calibration. A weighted composite of overall and surface-specific Elo performs best in terms of model discrimination. For women’s tennis, we find that a weighted composite of overall and surface-specific Elo performs best in terms of prediction accuracy, while a weighted composite of the betting odds, overall Elo and surface-specific Elo performs best in terms of calibration and model discrimination.