台灣汽車產業需求預測及季節指數之研究-以ARIMA季節相乘模式分析

Abstract

自1997年起因受亞洲金融風暴影響，汽車市場漸趨衰退，各廠家為了生存，開始精簡人事、簡化車種、加強競爭力等各項措施。2000年進入WTO之後，關稅逐年減少，各車廠競爭由國內競爭轉移為國際性的競爭。因此，在獲利方面也由過去的暴利時代進入了微利時代。在每月實際銷售數字與預測數字差異很大之情況下，將導致車廠之生產及存貨管理相當困難，再加上裝配人員調度困難，造成車廠成本之增加。因此若能提高汽車產業短期需求預測之準確性，並計算出可靠度高之季節指數，將可大大減少國內車廠及進口車廠之成本。 國內許多產業的銷售量受到陰曆節日影響甚巨，本研究也針對此點進行初步研究探討，嘗試將一般的陽曆銷售台數轉換為陰曆銷售台數來進行需求預測及季節指數分析。 本研究應用時間數列法之ARIMA季節性相乘預測模式，以單變量探討台灣汽車產業未來之需求量，利用1997年至2003年各月份總市場銷售量之資料數據，分析台灣汽車產業未來之需求量所呈現之ARIMA預測模式。並將陽曆之每月銷售台數，依照不同年月陽曆與陰曆間的對應關係，轉換為陰曆的每月銷售台數。然後相同地應用ARIMA預測模式，預測陰曆未來每月的銷售台數，然後再依照陽曆與陰曆間的對應關係，轉換為陽曆的銷售台數，並比較此兩種方法所預測的結果與實績之差異性。研究結果顯示陽曆最適ARIMA預測模式為ARIMA(1,1,0)*(2,1,0)12；陰曆最適之ARIMA預測模式為ARIMA(1,1,1)*(2,1,0)12。所建立之預測模式也顯示出台灣汽車市場需求具季節性因子。且由陰曆之銷售台數所作之需求預測轉換為陽曆之後的結果較接近市場實績。 本研究預測與分析之結果希望能提供政府有關單位及國內相關產業經營管理上之參考。也期望本研究所採取之方法可作為後續有關汽車產業需求預測之評估與參考。

Influenced by the financial crisis in Southeast Asia since 1997, the automobile market has declined gradually. Most of the automobile companies started to take actions such as reducing the personnel size, production cost and strengthening competitiveness in order to survive the competition. After entering WTO in 2000, the tariff was reduced year by year. The competition has moved into an international one instead of a domestic one. The automobile industry actually has entered the era of meager profits. If the prediction of auto demand is not close to real sales, it will be quite difficult to implement the production and inventory management. It is also difficult to arrange the manpower and the cost of production would certainly increase. If we can improve the accuracy of short-term demand forecasts by computing reliable seasonal indexes, it will reduce the operating cost of domestic auto factories greatly. The sales volumes of autos are influenced extremely by the festival of the Chinese lunar calendar. This research supports the finding of previous research for this point as well. We will transfer the sales volume of solar calendar into the sales volume of lunar calendar and try to come up with more accurate forecasts. This research forecasts the auto sales with the monthly sales data from 1997 to 2002. We then transfer the monthly sales volume of solar calendar into the monthly sales volume of lunar calendar according to the relation between solar calendar and lunar calendar. Next, we forecast via the ARIMA model and compare the difference of forecasts of two different methods in 2003. Our results show that the fittest ARIMA model is ARIMA (1, 1, 0) *(2, 1, 0) 12 for solar calendar and the fittest ARIMA model is ARIMA (1, 1, 1) *(2, 1, 0) 12 for lunar calendar. Results also show that there are seasonal factors in the automobile market. The forecasted sales of the lunar calendar are much closer to the actual figures of market than those of the solar calendar. The proposed models in this research can be helpful to related government departments and the auto industry. The results are expected to be a starting point for follow-up research of the automobile sales forecasts.