Forecasting Marine Corps enlisted manpower inventory levels with univariate time series models

Abstract

Accurately forecasting future personnel inventory levels by rank and occupational specialty is a fundamental prerequisite for development of an effective and functional staffing plan. This thesis develops and evaluates univariate time series models to create six- and twelve-month forecasts of Marine Corps enlisted manpower levels. Models are developed for 44 representative population groups using Holt-Winters exponential smoothing, multiplicative decomposition, and Box-Jenkins autoregressive integrated moving average (ARIMA) forecasting methods. The forecasts are evaluated against actual, out-of-sample inventory levels using several goodness-of-fit indicators including Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and Sum of Squared Errors (SSE). Among the modeling techniques evaluated, the multiplicative decomposition performed the best overall and represents an improvement over the Marine Corps' current forecasting method. This thesis recommends Marine Corps Systems Command, Total Force Information Technology Systems develop and introduce a multiplicative decomposition forecasting model into the Enlisted Staffing Goal Model. This forecasting technique should be implemented in phases, starting with the E-1 through E-4 population groups.