本研究採用Lee and Strazicich (2004)具一個結構改變點的LM(Lagrange Multiplier)單根檢定與Lee and Strazicich (2003)具兩個結構改變點的LM單根檢定,重新檢定1995年1月至2010年12月間,布蘭特與西德州原油現貨、期貨價格共8筆數列月資料,透過該檢定模型的特性找出可能的結構改變點,與存在結構改變點的數列是否為趨勢定態,並透過不同的資料長度與頻率檢定與更新石油價格數列的行為,另外,利用具結構改變點的概念預測石油價格。 結果顯示在不同的資料長度情況下,布蘭特與西德州的油價現貨與期貨八筆數列均無法拒絕數列含有單根的虛無假設,表示即使加入結構改變點,仍無法有效解釋數列定態情形,與Maslyuk and Smyth (2008)檢測出石油價格為隨機漫步模型一致,但其中不論是具一個結構改變點或具二個結構改變點的模型中,幾乎所有的趨勢項均顯著,表示檢測出的結構改變點將致使數列擁有不同的時間趨勢,這些時間趨勢的改變與國際特定事件有很大關聯。另外,對於2007年後石油價格走勢使用不同頻率資料作檢定,得到西德州原油價格大致上均呈現趨勢定態的結果。預測方面,由於前述檢定數列為非定態,利用取差分後線性自我回歸模型AR(1)與Random Walk做預測,發現前者預測表現優於後者。
Minimum Lagrange multiplier (LM) unit root tests have two kinds, one is endogenous structural break with one and the other is two endogenous structural breaks proposed by Lee and Strazicich (2003) and Lee and Strazicich (2004), respectively. The research period concerned Brent and West Texas intermediate crude oil spot and futures prices from January 1995 to December 2010. The investigation adopted both approaches to estimate the possible trends, potential break points, as well as the stationarity of the spot and futures crude oil prices. In addition, the study considered different horizon and frequency characteristics of the data to examine whether there resulted in different situations. Moreover, we further utilize the concept of structural breaks to forecast the crude oil price. Under different kinds of data horizons, the empirical result indicated the eight crude oil prices series cannot be rejected the unit root hull hypothesis, and implying that even taking the structural breaks into consideration which still could not explain the stationary of the series. Consistent with the Maslyuk and Smyth (2008), the crude oil prices move as random walk process. No matter in one or two breaks models, the findings all pose significant time trends. The possible reason is the time trend has already altered, and the international specific events can explain this kind of change. Finally, we further employ weekly data to test WTI crude oil prices in ex-ante 2007 period, and the numerical result reveals that the oil prices are trend stationary. In the forecasting aspects, the study applied autoregressive models with one lag perform a satisfactory job than Random Walk in terms of forecasting accuracy.