Predicting Outcomes of NBA Basketball Games

Abstract

Models are developed to help explain the point spread of an NBA basketball game based on significant in-game statistics. The models are based on a stratified random sample of 144 NBA basketball games played over a period of three years between 2008 and 2011. The in-game statistics found to be significant include the following: field goal shooting percentage; three-point shooting percentage, free throw shooting percentage, offensive rebounds, assists, turnovers, and free throws attempted. The models were validated using a random sample of 50 NBA games from the 2011-12 season. When these in-game statistics were known, the models were able to accurately tell which team had won the game between 88% and 94% of the time. The models were then used to make predictions using various methods to estimate the in-game statistics ahead of time. The models were used to predict winners of each game for the 2013, 2014, and the 2015 NBA Championships before any of the playoffs began. The models correctly predicted the overall champion each time. Predictions are also made for the 2016 NBA playoffs.