運用ARIMA與向量自我廻歸模式探討新竹科學園區半導體產值預測

Abstract

新竹科學工業園區自1980年設立以來，三十年來成功的發展經驗塑造了台灣高科技產業今日的風貌，所創造的產業群聚效應與專業分工技術創新，使得台灣已名列全球半導體及資訊工業大國。以園區六大產業中，半導體產業產值佔整體園區70%，而在台灣半導體產業中，所佔比例將近50%，可見其重要性。產業產值是反映企業生產成果，也是象徵著產業興衰與否的的重要指標，本研究以新竹科學園區半導體產值為研究目標，並以新竹科學園區管理局所提供1983年至2012年半導體產業產值、資本額及研發費用資料為樣本，然後以EView 8.0套裝軟體進行時間序列預測之模式適配，研究方法以自我廻歸整合移動平均模式(ARIMA)與向量自我廻歸(vector autoregressive, VAR)為建立半導體產值預測模式，研究結果顯示: 1.VAR模式對產值預測能力相對其他ARIMA模式較佳，因加入資本額及研發費用兩種變數後，多變量預測模式對未來的解釋能力比單變量預測模式佳。 2.從Granger因果關係檢定結果顯示，資本支出與研發費用的投入會影響半導體產業未來的產值，所以資本支出與研發費用為觀察半導體產值的重要領先指標，也驗證半導體產業是高度資本密集與技術密集的產業。 3.此模型在預測產值，年度愈近準確度愈高。 透過此預測模式之預測結果以提供政府及半導體業者決策之參考。

Since Hsinchu Science Park (HSP) had initiated from 1980, over the past thirty-ish years, its successful expansion had be playing a major role of shaping the entire technology industry of Taiwan. It created the industry cluster effect with technological innovation and specialization, which ranks Taiwan one of the biggest semiconductor nations in the world. Among the six major industries in the Science Park, the semiconductor industry accounted for 70% of the entire revenue. And in Taiwan's semiconductor industry, HSP almost account for nearly 50 percent of entire industry. The revenue is a reflection of achievements, but also as important indicator to the rise or fall to the industry. This research focuses on semiconductor’s revenue from firms in Hsinchu Science Park based on the data (1983 to 2012) of industry output and amount of R & D expenses providing by Hsinchu Science Park Administration. Then those data were processed with software, EView 8.0, to do time series prediction of matching the correct study model, and ultimately to create the forecast model for the entire semiconductor industry by using autoregression integrated moving average (ARIMA) and vector autoregression (VAR). The research shows: 1.VAR model does a better job in predicting the revenue than other ARIMA models. By taking the data of capital amount and R&D expenses into our study, the multivariable prediction model has better forecast capability than the single variable prediction model. 2.The test results of Granger shows that capital expenditures and amount invests in R&D affect the future revenue to the semiconductor industry. Capital expenditures and amount invests in R&D are leading indicators of revenue forecast. The semiconductor industry is highly capital-intensive and technology-intensive industries. 3.This research model produces more accurate data to if the year target is more recent which could be a good reference data to both the industry firms and the government in making decision