Comparison of Projected Wins of Three Projection Systems in Major League Baseball

Player Empirical Comparison and Optimization Test Algorithm (PECOTA), sZymborski Projection System (ZiPS), and FanGraphs are three well known projection systems in Major League Baseball (MLB). In this article, we will compare the effectiveness of these three projection systems using data from the seasons of 2013-2024. With different assumptions on correlation or covariance structure of the total numbers of games won by MLB teams in a season, three models are developed. Based on these models, we test the null hypothesis that the projected winning percentages are plausible values of the actual winning percentages for MLB teams. P-values of the Mahalanobis distance between the observed wins and projected wins are computed to evaluate the effectiveness of these three projection systems. Bonferroni confidence intervals, confidence ellipsoids, and Benjamini-Hochberg procedure for multiple hypothesis testing are also used to compare these three systems. Simulations are generated as well. The checking of the validity of normality assumption is also given.

Keywords: PECOTA, ZiPS, FanGraphs, Mahalanobis Distance, Matchup Games.