Predicting the point spread in professional basketball in real time: a data snapshot approach

Predicting the point spread of a professional basketball game is difficult but important for many stakeholders. We propose a new approach to predict the point spread in real time using in-game data. The approach uses a snapshot from the current game to identify historical games that have the same snapshot. After identifying these games, we predict the point spread of the current game using information obtained from the historical games. Using data obtained from six seasons of professional basketball games, we compare the prediction error of this approach to that of a deep learning technique, a long short-term memory network, and a general linear model. The proposed approach performs nearly the same as both models without the need for resource-intensive training. We discuss the robustness of this approach for making real-time predictions as games are underway. The findings have real-world implications for game enthusiasts, coaching staffs, and, most importantly, bettors.