汽車市場存在多種車款,倘若以整體汽車市場之銷售量進行分析與預測則難以探討個別車款在整體市場亦或各車款間所提供之資訊,因此本研究著重於以個別車款銷售量來探討車款間之互動影響,而由於車款數眾多,參數的估計量上亦相對較多,因此使用兩個不同之因子模型來降低維度,減少所需的參數估計量,其中傳統因子模型(Classical Factor Model)係使用主成份分析法(Principal Compo-nent Analysis, PCA)來萃取共同因子,此一因子模型主要用以探討同期車款銷售量之共同因子,另一個Peña-Box因子模型則主要探討跨期間車款銷售量之共同因子,除此兩個因子模型外,另考慮自我迴歸模型(Autoregressive Model, AR)與向量自我迴歸模型(Vector Autoregressive Model, VAR)與因子模型進行比較,實證分析結果顯示兩個因子模型在整體車款上所得之結果相似,而在各級距的分析上則有所不同,顯示考慮時間變動與否將直接影響共同因子分析結果,Peña-Box因子模型提供了在考慮時間效果下車款間之長期互動關係,此亦提供管理者另一種切入角度,在各級距預測上則發現整體而言向量自我迴歸模型(Vector Autoregressive Model, VAR)提供了最佳的預測精準度,而因子模型在整體車款的預測精準度方面則以傳統因子模型(Classical Factor Model)勝過Peña-Box因子模型,因此因子模型可提供較豐富的車款互動關係,特別是Peña-Box因子模型提供長期互動關係之資訊,而傳統時間序列模型,向量自我迴歸模型(Vector Autoregressive Model, VAR)則提供較佳之預測精準度。
Since there are many car models in the automobile market, it is hard to obtain the information between different car models if the analysis is based on the entire auto market. Therefore, our study aims at discovering the interrelation between each car models. We conduct the classical factor model (Principal Component Analysis is adopted) and the Peña-Box model to reducing the dimensionality, due to the large amount of car models, whereas the classical factor model extracts the contemporaneous common factors and the Peña-Box model extracts common factors that timevarying effect is under consideration. Moreover, we compare the results of analysis and forecasting that obtained from factor models with the results of analysis and forecasting that provided form Autoregressive Model ( AR) and the Vector Autoregressive Model (VAR). The classical factor model and the Peña-Box model give similar results while the analyses are based on all brand-models, but if it comes to categories, these two factor models provide dissimilar information. The Peña-Box model provides the management a distinct view point while facing the changing market environment. Further, in general, the VAR model has the best forecasting accuracy at category-wide level, and the classical factor model outperforms the Peña-Box model at macro level forecasting. Generally, the factor models provide rich information in discovering the interrelation between each brand-model, while the Vector Autoregressive Model provides better forecasting accuracy. It all depends on the purposes of the user to define which model fits better.