Can One Predict Next Year's Winning Percentage Using Ols Regression on Baseball Statistics?

Abstract

As part of a project for the University of Chicago GSB MBA program, we created a regression model to predict a baseball team's winning percentage using aggregate team statistics from the prior year. If we could judge our results on the same metrics as baseball players, our result would prove successful (in-sample R2=27%) - we batted .270. However, statistically speaking, the result of our regression indicates it is extremely difficult to predict - by more than 27% of the variance - next year's winning percentage for a baseball team. However, when assessing our model's validity by the standard deviation of predicted results vs. actual results, our model is proves to be at least marginally useful. In the following paper we will discuss how we came up with the metrics we used, what tests we performed to better account for predicting in-sample variance, and our conclusion.