基于极值理论的卢布汇率与布伦特原油风险测度及时变相关性分析

发布时间：2018-06-25 00:46

本文选题：极值理论 + 时变Copula　；参考：《浙江工商大学》2015年硕士论文

【摘要】：自2014年以来,原油价格发生了巨幅下跌的状况,而且白金融危机以来油价从未像现在跌得如此猛烈。而原油价格暴跌受到影响的首当其冲是石油出口大国俄罗斯,作为俄罗斯货币的卢布也出现异乎寻常的暴跌,卢布贬值与近期的油价暴跌有着相应的联系。面对着原油暴跌以及卢布危机这样的极端金融事件的发生,风险状况的精确度量显得尤为必要和紧迫。已有研究表明基于极值理论的VaR模型能够较好地估计金融市场的极端风险值,然而现实的金融市场数据往往不能满足独立同分布的前提假设,因此本文首先采用GJR模型、EGARCH模型和GARCH模型结合t分布、GED分布和SKST分布处理布伦特原油及卢布汇率对数收益率序列,得到收益率序列的标准残差序列,接下来对服从独立同分布假设的标准残差序列运用阈值POT模型计算VaR值和CVaR值,最终计算得到单一资产的风险值。考虑到在模型回测中通常使用的Kupiec检验忽略了数据的时间变化特征,本文模型回测采用Christofferson有条件覆盖模型,其在Kupiec检验的基础上考虑了超出值序列的时间易变性。实证研究表明：在较低置信水平下,各模型对两资产序列极端风险状况的测度均失效,而在较高置信水平下,各模型均显著有效。各模型对两资产序列上尾部的检验值大小均比较接近,而且均在较高置信水平下表现出模型的有效性；而对于两资产序列下尾部极端风险状况的测度模型中均为GJR-SKST-POT模型最优,而且在此模型下的检验值均是明显小于其他模型的检验值,说明对于两序列下尾部风险测度来说,GJR-SKST-POT模型确实优于其他模型。为研究原油市场与卢布汇率市场之间相依结构,即卢布危机受到原油暴跌的影响大小,并且考虑到金融市场的相关性总是随时间变化的,本文采用三种时变Copula模型以及对应的三种常相关Copula模型研究两市场之间的相关性。由于Copula模型具有不受边缘分布的限制的优点,可以将边缘分布与Copula模型分开研究,本文利用前文得到的综合最优模型GJR-SKST-POT模型作为边缘分布,结合Copula模型测度资产相关性。实证研究表明：采用时变SJC Copula模型描述资产序列之间的相依结构最为准确,且时变SJC Copula模型测度的上尾部平均相关系数也大于下尾部平均相关系数,说明了两资产市场在牛市阶段比在熊市阶段更容易出现联合极值现象。通过得到的相关系数大小来看,两资产序列之间的相关性并不如想象中的大,但在其他的诸如西方国家对俄罗斯的制裁以及美元走强等因素的共同影响下,两资产序列之间的相关性已经相当可观,说明了原油价格的暴跌确实是卢布暴跌的主要原因之一。
[Abstract]:Crude oil prices have fallen sharply since 2014, and oil prices have not fallen as hard since the financial crisis. Russia, the major oil exporter, has been the first to be hit by the collapse in crude oil prices. The ruble, the Russian currency, has also suffered an unusual collapse, with a corresponding link between the devaluation of the ruble and the recent collapse in oil prices. In the face of the collapse of crude oil and extreme financial events such as the rouble crisis, the accuracy of the risk situation is particularly necessary and urgent. Previous studies have shown that VaR model based on extreme value theory can estimate the extreme risk value of financial market well. However, the actual financial market data often can not meet the premise of independent co-distribution. Therefore, in this paper, GJR model EGARCH model and GARCH model combined with t distribution GED distribution and SKST distribution are used to deal with the logarithmic yield series of Brent crude oil and rouble exchange rate, and the standard residuals of the return series are obtained. Then the VaR value and Cvar value are calculated by using threshold pot model for the standard residual sequence with independent co-distribution hypothesis, and the risk value of a single asset is finally calculated. Considering that the Kupiec test, which is usually used in model retrieval, neglects the time variation of data, the model retesting adopts Christofferson conditional covering model, which considers the time variability of the value series on the basis of Kupiec test. The empirical study shows that under the lower confidence level, each model fails to measure the extreme risk of the two asset series, but at the higher confidence level, each model is effective. Each model is similar to the test value on the tail of the two asset sequences, and shows the validity of the model at a higher confidence level, while the GJR-SKST-POT model is optimal for the extreme risk situation of the tail in the two asset sequences. Moreover, the test values under this model are obviously smaller than those of other models, which indicates that the GJR-SKST-POT model is indeed superior to other models for tail risk measurement under two sequences. In order to study the structure of dependence between the crude oil market and the rouble exchange rate market, that is, the magnitude of the rouble crisis affected by the collapse in crude oil, and taking into account that the correlation of financial markets always changes over time, In this paper, three kinds of time-varying Copula models and three corresponding frequent correlation copula models are used to study the correlation between the two markets. Because the Copula model has the advantage of not being restricted by the edge distribution, the edge distribution can be studied separately from the Copula model. In this paper, the GJR-SKST-POT model is used as the edge distribution and the Copula model is used to measure the asset correlation. The empirical study shows that the time-varying SJC Copula model is the most accurate method to describe the dependence structure between asset sequences, and the upper tail average correlation coefficient of time-varying SJC Copula model is larger than the lower tail average correlation coefficient. It shows that the two asset markets are more prone to joint extremum in bull market than in bear market. Based on the magnitude of the correlation coefficient obtained, the correlation between the two asset sequences is not as large as expected, but under the combined influence of other factors such as Western sanctions against Russia and the strengthening of the dollar, The correlation between the two asset series is already considerable, suggesting that the collapse in crude oil prices was indeed one of the main reasons for the ruble's collapse.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F224;F416.22;F835.12

【相似文献】