Please use this identifier to cite or link to this item:
標題: 財務時間序列尾端行為結構變動之檢定方法
Alternative Tests for Structural Changes in Tail Behavior of a Financial Time Series
作者: 陳美源
關鍵字: 基礎研究;VaR;經濟學;尾端行為;結構變動;極值理論;風險值;靴跋法;分量迴歸;EVT;structural changes;tail behavior;bootstrap method;quantile regression
本研究計畫將探討如何檢定資產報酬分配的尾端行為(tail behavior)是否發生結構性變動(structural changes),資產報酬分配的尾端行為如果發生變動,將對於應用極值理論(valueat risk)估計資產報酬的風險值(value at risk)有所影響;Quintos, et al (2002)所提出檢定資料分配尾端行為是否發生結構性變動的方法,雖然具有不需假設結構變動時點為已知或外生的優點,但是文章中並未討論Hill 厚尾度(tail thickness)估計式中門檻值(thresholdvalue)對檢定方法表現的影響;本研究計畫首先將藉由模擬分析門檻值對Quintos, et al(2002)檢定方法檢定力的影響效果;進而本研究計畫將結合Danielson, et al (1996)所提出利用跋靴法(bootstrap method)決定最適門檻值的方法,進而透過模擬分析探討Quintos, etal (2002)檢定方法的表現。再者,Taylor (1999, 2000)利用份量迴歸(quantile regression)估計資產報酬的風險值,而資產報酬分配的尾端行為若發生結構性變動,則相對應的份量迴歸模型亦將發生結構變動,因此,對於份量迴歸模型參數是否發生變動的檢定,與檢定資產報酬分配的尾端行為是否發生結構性變動,將有異曲同工之處;是故,本研究計畫將建構在份量迴歸模型中參數一致性的檢定方法,並進行其檢定力的模擬實驗,進而比較其與Quintos, et al (2002)所提出檢定方法在檢定尾端行為是否發生結構性變動的能力。最後,本研究計畫將應用所討論的檢定方法,對於台灣的財務金融資料進行實證分析,期能獲致較可信賴的風險值估計結果。

In this research, estimation of the value at risk (VaR) via extreme value theory (EVT) isstudied under the case that structural changes exist in the tail behavior of asset returns. Thetest for structural changes in the tail behavior suggested by Quintos, et al (2002) isconsidered. As the test of Quintos, et al (2002) is constructed by looking up the index of tailthickness estimated by Hill』s estimator, its test performance will be affected by the thresholdvalue specified in the Hill』s estimator. In this research, the influences of the threshold valueson the performance of Quintos, et al (2002)』s test will be investigated first. A determinationof an optimal threshold value has been suggested by Danielson, et al (1996) via bootstrapmethod. Test performance of Quintos, et al (2002)』s test combined with an optimal thresholdvalue determined by Danielson, et al (1996)』s method will also be investigated by MonteCarlo simulations. Besides, Taylor (1999, 2000) suggest that estimate VaRs via quantileregression approach. It is clear that structural changes should be considered in the quantileregression models if the tail behavior of asset returns changes. Therefore, testing forcoefficient stability for a quantile regression at extremal quantile is coincident with checkingthe tail behavior change. In this research, tests for coefficient stability in a quantileregression model are going to be constructed. Their testing performance will also be studiedvia simulations and compared with ones of the test suggested by Quintos, et al (2002).Finally, all tests discussed in this research will be applied to check the constancy of the tailbehaviors of Taiwan financial data. And then, empirical estimations of VaRs for Taiwanfinancial data will be studied furthermore.
其他識別: NSC93-2415-H005-002
Appears in Collections:財務金融學系所

Show full item record
TAIR Related Article

Google ScholarTM


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.