A Markov chain model for predicting Major League Baseball

In this report, we present a Markov chain model for predicting the scores and the winning team of Major League Baseball (MLB) games. We discuss how a baseball game can be viewed as an infinite horizon discrete-time Markov chain with finite state space. We demonstrate how standard Markov chain theory can be used to obtain analytical solutions for the expected runs and win probability in a given MLB matchup. We improve upon previous models by incorporating pitching and more complex baserunning, and then demonstrate the effect of these changes by comparing our model to historical data. We also discuss computational methods for solving the model. Finally, we test our model on games from the 2015 MLB season.