The Real Moneyball: Modeling Baseball Salary Arbitration

Abstract

This project involves developing four modified logistic regression models to predict the true value of baseball players as determined by salary arbitration panels. Two models are for pitchers based off of their six-year-average performance statistics and platform season statistics, respectively. The other two models are for position players based off of their six-year-average performance statistics and platform season statistics, respectively. The variables included in these models were determined using principal components analysis, sparse principal components analysis, sparse linear regression, and sparse logistic regression. The four models are then compared in terms of their AIC values, log-likelihood values, and mean cross-validation errors to come up with final models for pitchers and position players.