software for "Extreme Value Estimators for Censored Inventory Demand"
This study explores extreme value theory (EVT) in the newsvendor problem for inventory control and demand forecasting under censored demand where inventory stockouts obscure true demand. EVT models, particularly the Generalized Pareto Distribution (GPD), focus on the extreme tails of demand distributions, enabling accurate estimation of high-demand events. The paper benchmarks competing methodologies applied to a multiperiod newsvendor model, examining their effectiveness in minimizing costs associated with stockouts and overstocking. The research uses Fisher-Tippett-Gnedenko (FTG) convergence to justify the application of EVT. The study benchmarks competing methods using simulated demand time series and dynamic reorder policies, analyzing the trade-offs between computational complexity and accuracy in cost minimization. By exploring parameter spaces, such as demand variance and tail behavior, the research highlights scenarios where EVT outperforms benchmark models. Ultimately, the findings support EVT's role in optimizing inventory management under extreme demand variability, contributing to the broader field of stochastic inventory theory.