應用層級式預測方法於季節性資料分析之研究─以家電製造商之出貨量為例

Abstract

需求規劃的結果高度影響著供應鏈整體網路規劃決策的品質，顯示出需求規劃在企業營運活動中扮演著舉足輕重的角色。然而，需求資訊往往在經過供應鏈層層傳遞後逐漸被扭曲，這種上下游間不對稱的需求資訊將嚴重傷害供應鏈規劃的品質。隨著產品多樣化及其生命週期愈益縮短，若對每一產品細項皆進行預測，其所需之實質成本與時間成本不僅龐大，也未必能得到較準確的預測結果。然而，聚合的預測雖然較為準確，但其所提供的資訊有限，無法因應企業中更為細部之規劃及作業層級之決策。因此企業為了提升需求預測的效益，應有效運用聚合與分解策略進行預測。本研究將以季節性資料—冷氣機製造商之冷氣月出庫量需求為實證分析對象，利用層級式預測方法，將各層級間資料聚合與分解，建立一套有效之預測模型，進而提高預測準確度。 本研究利用時間序列方法中的季節性ARIMA轉換函數模型，配合產品層級架構分類，先分別預測出產品不同層級之銷量，並利用Top-down、Middle-out及Bottom-up預測方式，由各產品階層向上聚合或依比例向下分解至其他階層，最終獲得產品層級架構中所有分類項目之預測值，建構出一套完整之季節性資料層級式預測模型，接著再進行不同預測方式間預測能力之比較，以找出最適預測階層以及預測方式。 經比較Top-down、Middle-out及Bottom-up三種預測方法之後，本研究發現Bottom-up為冷氣機出庫量之最適預測方法，並將預測結果依不同之層級類別作呈現，可提供不同作業層級之決策依據。如此所建構出之預測模型，不僅改善了廠商原先所利用的預測準則，亦能掌握出庫量時間序列之變動型態，使廠商未來於產品規劃及通路出貨量配置時能更有效掌握相關資訊，做出正確決策考量。

The results of demand planning highly influence the quality of supply chain network planning. It reveals that demand planning plays a very important role in business operations. However, demand information is often distorted through supply chain. The phenomenon of asymmetric demand information seriously harms the quality of supply chain planning. As a result of product diversity and shorter product life cycle, forecasting for every single product item will cost too much in physical and time aspect, and also unlikely obtain accurate forecasts. Although forecasting for aggregating demand is more accurate, it has difficulty to handle operations planning and decisions in more details.. In order to enhance the benefits of demand forecasting, managers must take advantage of aggregation and disaggregation strategies in demand forecasting. This research takes seasonal data from monthly sales data of an air conditioner manufacturer in last four years as an example. We apply hierarchical forecasting methodology with seasonal ARIMA transfer function to aggregate and disaggregate data between product levels, and come up a effective forecasting model. At first, we forecast for sales at different levels respectively, and then aggregate upwards or disaggregate downwards to other levels by using Top-down, Middle-out and Bottom-up approaches. Finally, we obtain forecasts of all items in the hierarchical product structure. After establishing this complete hierarchical forecasting model of seasonal data, we compare forecast predictability generated by different forecasting approaches in order to find the most suitable forecasting approach. After comparing three forecasting approaches, Top-down, Middle-out and Bottom- up, we find that Bottom-up is the most suitable forecasting approach for sales of air conditioner. The forecasting model we build up not only improves the forecasting predictability of the firm but also gets a grip of the pattern of sales series such that the firm can make accurate decision in product planning and shipment allocation.