| Summary: | 碩士 === 國立清華大學 === 工業工程與工程管理學系 === 93 === For demand-oriented and fast changing industries nowadays, if enterprises are capable to forecast market demand or customers’ needs, then it can improve the coordination of supply and demand and increase the industry competition. However, with technology progress, products are trending to various types with small amounts. If enterprises forecast every individual product item, it will increase the complexity of forecasting activity. To manage a great deal of forecasting tasks efficiently, the hierarchical forecasting methodology comes into being. In the past, some studies indicated that top-down was a better forecasting approach since the variances of individual items could be canceled out by aggregating. On the contrary, other studies considered that aggregating individual items would lose some of valuable information and result in biases instead, so they preferred the bottom-up forecasting approach. As for which approach is better, there has not been consistent agreement so far.
TFT-LCD (thin film transistor-liquid crystal display) is one of the key industries with high clockspeed at present in Taiwan. In this study, LCD monitor is taken as empirical research subject. Through hierarchical forecasting methodology, we carry out demand forecasts at each product level, and then look for the best level and forecasting approach expectantly. The hierarchical forecasting methodology proposed in this paper consists of five stages. First, a few product attributes are considered to build product hierarchy by using cluster analysis. Second, exogenous driven factors that affect demand quantity significantly are identified at each level of product hierarchy. Third, a number of forecasting techniques are combined, including regression analysis, transfer function, and simultaneous equations model that considers causal relationship between quantity and price to forecast future demand at each level of product hierarchy. Forth, various forecasting approaches and disaggregation proportion methods are adopted to obtain consistent demand forecasts at each level of product hierarchy. Fifth, forecast errors with different forecasting approaches are assessed respectively to find out the best forecasting level and approach.
Different from previous studies regarding top-down or bottom-up as the best forecasting approach, the research results in this paper show that based on the built product hierarchy and forecasting models, the middle level of product hierarchy is the best level to be forecasted, and the best forecasting results of all levels can be obtained by using middle-out forecasting approach. These results not only break through the standpoints of related hierarchical forecasting studies in the past, but also could be the guide for LCD manufacturers and brands to forecast market demand in the future.
|
|---|
