台灣資訊產業產值預測模型之研究

Abstract

近幾年來，台灣電子資訊等相關高科技產業呈現高度成長，取代八十年代的金融產業成為台灣經濟成長的骨幹。然而高科技產業往往有週期性景氣循環，產品生命週期短，因此產值預測困難。它會影響政府產業政策以及整體環境之投資計畫，以及廠商投資擴廠之決策，對於企業經營的獲利或虧損影響甚巨。 有鑑於此，本文之研究目的則利用時間序列模型中自我迴歸模型(ARIMA) 、向量自我迴歸模型(VAR) 、貝氏向量自我迴歸模型(BVAR)結合產業群聚效應，探討台灣IT產業（光電產業、半導體產業以及電腦製造業）與其他週邊產業彼此之間是否具有長期之均衡關係，將收集10年間的季資料區分為兩部分，第一部分用來建立模型，第二部分則作為預測之比較值。 結果發現，在三個產業的模型驗證中均呈現VAR的表現較差，其次為AR，而BVAR則是驗證模型中最好的，證明BVAR模型可以用於少樣本的時間序列與動態預測中。而跟其他研究機構比較結果如下 (1)與IT RI的產業報告比較半導體產業：BVAR模型不論在1998年以及2001年產業大幅成長衰退時預測均較ITRI精準。(2)與III的產業報告比較電腦製造產業：BVAR模型預測較III精準。

The production forecasting of high technology industries is an important issue for entrepreneurs and governments, but it suffers from the situation of fast growth and frequent fluctuation. In this article, we propose a forecasting method that combines the clustering effect, different transformation of data, and non-informative diffuse-prior Bayesian vector autoregression (DBVAR) model to forecast the productions of technology industries. The BVAR model possesses the superiority of Bayesian statistics in small sample forecasting and holds the dynamic property of VAR (Vector autoregression) model. Three industries are examined to verify the proposed method. The subjects are: (1) Using Four Forecasting Models to Forecast of Total Production Output of Taiwan’s Photonic Industry, (2) Using Four Forecasting Models to Forecast of Total Production Output of Taiwan’s Semiconductor Industry, (3) Using Four Forecasting Models to Forecast of Total Production Output of Taiwan’s Computer Manufacturing Industry. It is found that the DBVAR models outperform the other three conventional time series models including the autoregression (AR), vector autoregression (VAR), and Litterman Bayesian VAR (LBVAR) models. Moreover, the DBVAR models also could exactly find the inflection point of the trend and give a promising forecasting. Our forecasting method is therefore concluded as a feasible approach for production prediction, especially for technology industries in volatile environment.