原油價格近年來因為不斷上漲而受到各方矚目,由於原油商品價格之波動性較其他商品來得大,目前多數致力於此方面的研究中,皆以考慮原油價格波動性之模型進行分析。本研究擬比較多種波動性模型在西德州及布蘭特原油的配適度與預測績效,以提供原油價格波動性相關實證研究之基礎模型參考。 本研究採用了以報酬及變幅為基礎的波動性模型,其中以報酬率為基礎的模型有:Bollerslev(1986)提出的GARCH模型及Nelson(1991)提出的EGARCH模型;以變幅為基礎之模型則為Chou(2005)所提出的CARR模型。將原油價格配適於不同的波動性模型後,依據估計結果進行樣本外預測,比較各模型之預測誤差。此外,本研究亦採用DM檢定為統計上之顯著性檢定,找出較能擷取原油價格波動特性的模型。 利用樣本區間為1995年至2007年之西德州和布蘭特原油價格週資料,本研究得到之實證結果為:就波動性模型樣本內配適而言,CARR模型的表現較佳,且該模型對於短期的波動反應亦較敏感;就波動性模型樣本外預測誤差相較,CARR模型估計出的預測誤差值也明顯較小;根據DM檢定,在短期的預測上,CARR模型顯著優於其他模型,但在長期預測方面,則無法比較各模型之優劣。 就實證結果而言,無論所選用之樣本為西德州原油或布蘭特原油,經由模型估計、預測誤差比較、統計上之顯著性檢定,以變幅為基礎之模型和以報酬為基礎之模型相較之下,更能擷取原油價格的波動性。
This paper uses different heteroskedasticity models to oil price data and compares the performance of them. The goal of this thesis is to find the model which is better for modeling and forecasting oil prices. There are two branches of volatility models. One is the return-based model, and the other is the range-based model. For these two branches, this paper uses GARCH model, EGARCH model, and CARR model. The DM test is also used to provide the statistical result. Based on the empirical results, CARR model is more sensitive to the volatility than other models. The forecast error of CARR model is also smaller. Furthermore, out-of-sample forecasts of CARR model is dominant based on the result of the DM test, but this domination only exists when the volatility prediction is a short-horizon phenomenon. To conclude, the range-based model such as CARR model has better forecast performance than the return-based model such as GARCH and EGARCH models. By testing of the model predictability, CARR model performs better than others as well.