Report on the Evaluation of Demand Forecasting Techniques for the Subsistence Commodity. Volume 1

摘要

This study involves the evaluation of quantitative techniques to improve subsistence demand forecasting in the Defense Logistics Agency. A wide variety of techniques were tested for approximately 4,000 item/warehouse demand series over the 1977-1983 time frame. The range of forecasting techniques analyzed included the following: autoregressions, autoregressions with seasonal terms: simple moving averages; single, double, adaptive and combined exponential smoothing; naive and the current DLA methods. The methodology, in effect, substituted individual and/or groups of methods for the DLA method to compute forecasts over the 1980-1983 period. Findings indicated that about 20% more variability in lead time demand was incurred during the 1980-1983 time period than achieved by a group of five methods identified in this study, during that period. Similarly, 14% more variability in procurement cycle demand forecast error was incurred during the 1980-1983 time period. Procedures are identified that can achieve further substantial reductions in variability of lead time demand through the use of forecast adjustments. A prototype subsistence demand forecasting system is described based on the recommended group of models in this study. This study also serves as the analytical basis for the development of the functional description for the Forecasting Module of Defense Integrated Subsistence Management System (DISMS). Volume I contains the body of the study report.