Abstract
This study develops a simulation-based framework to optimize Major League Baseball batting orders using a non-homogeneous Markov-chain model of inning progression. Using 2025 MLB regular-season batting event data from FanGraphs, plate-appearance outcome probabilities are estimated conditional on base-out context and opposing pitcher handedness. These probabilities define the transition mechanism for a nine-inning run-scoring simulator. For each team, a fixed set of nine hitters is selected and the optimal batting order is searched over the $9!$ permutation space using a Metropolis-Hastings procedure, a shallow exploration phase to identify strong candidate orders and a deep Monte Carlo re-evaluation phase to reduce simulation noise.
Across the league, the optimized batting orders vary with pitcher handedness, indicating that optimal lineup structure is conditional. At the team level, simulation-based run estimates are broadly consistent with observed offensive strength. Overall, the results suggest that optimal batting-order design is conditional on matchup context and that non-homogeneous Markov modeling coupled with stochastic search can generate interpretable lineup recommendations under realistic data and computational constraints.
Advisor
Lester, Cynthia
Department
Statistical and Data Sciences
Recommended Citation
Kang, Jun, "Optimizing MLB Batting Orders Using a Context-Dependent Monte Carlo Simulation and Metropolis-Hastings Algorithm" (2026). Senior Independent Study Theses. Paper 13404.
https://openworks.wooster.edu/independentstudy/13404
Keywords
baseball, MLB, markov chain
Publication Date
2026
Degree Granted
Bachelor of Arts
Document Type
Senior Independent Study Thesis
Code used in this research
Jun Kang Tables.pdf (81 kB)
Tables of simulation results
© Copyright 2026 Jun Kang