| Summary: | 碩士 === 國立臺灣大學 === 商學研究所 === 94 === For those companies who sell multiple products to end users, demand forecasting is a tough but important issue. In consumer product industry, retailers have to coordinate the supply chain to become more competitive. In generally, aggregated demand forecasting is more accurate than individual item forecasting. However, if we have only the aggregated demand, we will not be able to get useful information for decision-making in the operation level. If we forecast each end item’s demand, it will be too time consuming and the difficulty of forecasting will increase. As a result, this research makes use of the hierarchical forecasting methodology to make the demand forecasting more efficient.
In this research, the study of the hierarchical forecasting model is conducted in several steps:
1.Use regression and regression tree model to build the forecasting model.
2.Aggregate the lower level of product hierarchy or disaggregate the higher level to other levels to generate the forecasting values for other levels.
3.Compare the error rates and identify the best forecasting levels and forecasting methods.
To validate the proposed model, sales data of a local TV-shopping company are collected. Three different forecasting methods are compared, Top-down, Middle-out and Bottom-up, and the results show that ”Bottom-up” is the best forecasting method. In forecasting the sales at the bottom level, regression tree model has a better prediction accuracy than that by regression model due to the “re-classification” feature. Furthermore, the sales forecasting model built in the research performs better than the current forecasting methods of the TV-shopping company.
|
|---|
