Building the Winning Percentage Model to Predict Regular Season Results of NBA Teams Based on Regression and Time Series Analysis of Common Basketball Statistics
With the trend to apply statistics to predict sport games, the purpose of this paper is to find a model that can help to predict the percentage of games won for NBA teams during a season based on data for team and individual player performance. Multiple linear regression is used to build a predictive model, while time series analysis is used to assist with model selection. Great care is taken here, because statistical software will build a model regardless of collinearity, which means the model contains highly correlated variables, and despite whether regression assumptions are met. A general model that can predict all teams’ performance is found. The model basically fits every team, and even the worst predictions look decent. However, each team has its own philosophy, so each has different significant factors. Thus models built for individual teams perform better.