本文針對台灣總體時間數列的長期趨勢進行分析。我們修正Perron(1989) 只允許一次結構變動且改變點已知的單根檢定,先利用Bai(1999)的方法,找出到底台灣總體資料發生幾次結構變動及其改變點估計值。再利用蒙地卡羅模擬方法,得出在多次結構變動下,對應不同長期趨勢模型單根檢定的小樣本分配,據此判別台灣總體資料為趨勢穩定式差分穩定。我們發現在六組總體數列中,實質 GDP,實質每人GDP與貨幣存量均包含四次結構變動,而實質每人消費,製造業平均薪資與工時則有三次結構變動,時點大都發生於1969年附近,第一次與第二次石油危機,及80年代後期的台灣財富重分配時期。在5%的顯著水準下,除了實質每人消費無法判定外,其餘總體變數均接受斷裂趨勢(trend break)假說。
In this paper we analyze the long-run properties of macroeconomic data from Taiwan. In contrast with Perron (1989), we do not assume the number and locations of breaks are known. We first determine the number of breaks and their locations using the method of Bai (1999), and for a particular model, we then simulate the finite-sample distribution of the associated unit-root test under multiple breaks. By comparing p-values of the unit-root statistics, we are able to distinguish between trend- and difference-stationarity of the data. For six macroeconomic series, the empirical results show that real GDP, real per capita GDP, and money stock have four breaks, while real per capita consumption, average manufacturing wage, and average working hour have three breaks. The estimated break dates are around 1969, the two global oil crises, and the years of wealth reallocation in late 80's. At the 5% level, all but real per capita consumption can be characterized as trend break processes.