2010
MAS 2010 Proceeding
Inventory control performance of various forecasting methods when demand is lumpy
Authors: Adriano Solis, Somnath Mukhopadhyay, Rafael Gutierrez
Abstract
This study evaluates a number of methods in forecasting lumpy demand ? single exponential smoothing, Croston?s method, the Syntetos-Boylan approximation, an optimally-weighted moving average, and neural networks (NN). The first three techniques are well-referenced in the intermittent demand forecasting literature, while the last two are not traditionally used. We applied the methods on a time series dataset of lumpy demand. We found a simple NN model to be superior overall based on several scale- free forecast accuracy measures. Various studies have observed that demand forecasting performance with respect to standard accuracy measures may not translate into inventory systems efficiency. We simulate on the same dataset a periodic review inventory control system with forecast-based order-up-to levels. We analyze resulting levels of on-hand inventory, shortages, and fill rates, and discuss our findings and insights.
I3M Scientific Sponsors
I3M Industrial Sponsors
I3M Media Sponsors
