觀察視窗的選擇對銷售預測績效之影響－以彈性Bass模型為例

The Impact of Observation Window Selection on Sales Forecasting Performance— Exemplified by a Flexible Bass Model

王金鍊(Chin-Lien Wang)

《嶺東學報》 39期 (2016/06) Pp. 245-271

摘要

四十幾年來, 針對Bass 模型的缺點, 所推出延伸模型不可勝數。主張釋放固定參數的假定並提出彈性的模型以反映多變的市場實況大有人在。在參數估計方面, 基因演算法被認為較傳統方法為佳。較早期的觀察值離現在較遠, 其資訊往往與目前沒有太大關聯而易使參數估計失真; 最近的觀察資料反而較能反映近期的變動。本研究提出一套彈性的Bass 參數估計方法, 透過捲動的五種觀察視窗, 搭配實數基因演算法, 根據color tv, IBM 主機與台灣手機銷售資料估計Bass 參數, 對每一預測期各進行十次參數估計與樣本外預測。 實證結果顯示, 在短期預測方面,最近三、四期視窗以MAPE 表示的績效都居前二名, 顯示在單期預測上,最近幾期資料就已足夠。至於中期預測方面, 仍以最近三期視窗績效最佳, 涵蓋參考期所有資料的視窗次之, 顯然最近的資料仍然是中期預測最重要的根據, 但其它較早期資料還是有其價值。

關鍵字

觀察視窗 ； Bass模型 ； 實數基因演算法

並列摘要

We have witnessed numerous extension models introduced over the past forty plus years to tackle Bass model’s limitations. Many researchers relaxed its constant parameters assumption with flexible models to reflect rapid changing nature of the real world. As a parameter estimation tool, genetic algorithm is much appreciated than traditional methods. Earlier observation data are far away from the present, offering information irrelevant of current forecast context, leading to unfaithful parameter estimation. However, most recent data could much better reflect recent changes. This study proposes a flexible parameter estimation model–a rolling window of five choices, coupled with a real coded genetic algorithm, to estimate Bass parameters based on historical data from color tv, IBM mainframe, and M-phone ROC; and conducted out-of-sample forecasting 10 times respectively for each forecast period. The empirical results show that the performance in terms of MAPE, the most recent 3-4 periods data windows occupied the first two positions in one-step-ahead forecast, implying that it’s sufficient to merely use the most recent data to conduct forecast. As the 3-step-ahead forecast is concerned, the window of most recent 3 periods still ranked first, and the window of the whole data ranked second, signifying the most recent data is most important in this regard also, nevertheless, earlier data remain valuable.

並列關鍵字

Observation window ； Bass model ； Real coded genetic algorithm

參考文獻

Abdul-Rahman, O.,Munetomo, M.,Akama, K.(2012).An adaptive Parameters Binary-Real Coded Genetic Algorithm for Real Parameter Optimization: Performance Analysis and Estimation of Optimal Control Parameters.International Journal of Computer Science Issues.233

Bass, F. M.(1969).A New-Product Growth Model for Consumer Durables.Management Sci..15,215-227.

Bemmaor, Albert C.,Lee, Y.(2002).The impact of heterogeneity and illconditioning on diffusion model parameter estimates.Marketing Sci.21,209-220.

Brindle, T.,Thiele, L.(1995).A Comparison of selection schemes used in genetic algorithms.,Zurich:Swiss Federal Institute of Technology(ETH) Zurich.

Cardinali, C.(2009).Monitoring the observation impact on the short-range forecast.Quart. J. Roy. Meteor. Soc..135,239-250.