股票市场尾部风险与尾部相关性特征研究

发布时间：2018-06-06 23:29

本文选题：尾部风险 + 尾部相关性　；参考：《电子科技大学》2012年博士论文

【摘要】：频繁发生的金融危机一次又一次给投资者、金融市场甚至全球经济带来严重的不良后果。在这些危机中，市场上呈现出与正常情形下不同的特殊特征：单变量的情形下，投资者面临着发生极端损失的“尾部风险”；在多变量情形下，金融市场或金融资产间存在着相对更强的“尾部相关性”。如何把握并应对这两种特殊的市场特征，无论对于投资者和风险管理者还是政策制定者和监管者，都是一个至关重要的问题。因此，本文综合运用各种灵活的计量经济方法来捕捉危机时期中出现的“尾部风险”和“尾部相关性”特征，并进一步分析其对风险管理、资产配置和资产定价的影响。首先，针对单个资产所面临的尾部风险，本文引入检验效力更强的鞍点技术返回检验方法对各种风险模型的VaR和ES预测准确性进行了严格的再检验，重新探讨了何种模型能够最为准确地捕捉单变量情形下的尾部风险。基于中国股市的实证分析发现，简单的GARCH-Normal模型无法合理地捕捉中国股票市场的风险特征，而最好的模型为GARCH-EVT模型。进一步基于更多成熟股市和更多风险模型的研究同样证实，为了得到足够准确的风险预测，则有必要借助极值理论EVT来对金融资产收益率的分布尾部进行单独建模。仅使用一种分布形式很难同时捕捉到分布尾部和分布中间的特征，即使是以往文献中推荐使用的能同时捕捉分布偏斜特征和厚尾特征的有偏学生t分布。此外，通过对GARCH类模型中影响风险预测准确性的两个维度的相对重要程度首次进行正式统计检验发现，，残差分布的尾部设定对VaR和ES预测的影响要强于波动率方程形式。其次，针对尾部相关性对风险管理的影响，本文首先提出一种基于多元Copula函数模拟的方法来计算组合中个别资产的风险贡献，从而实现了对不同资产风险贡献区别的显著性检验。同时由于Copula函数在刻画资产间非线性相关结构以及尾部相关性特征方面的优势，使用本文方法计算所得的风险贡献结果还变得更为一致而不再受置信水平和风险度量指标的影响。此外，本文还引入一种更为灵活的多元相关结构建模工具，正则藤Copula函数，以克服现有研究中可选择的多元Copula函数类型的有限性以及存在的不同缺陷。基于上海、香港和台湾三个股市的实证分析证实了正则藤Copula在刻画多元相关结构方面的优越性。更具有实践意义的是，基于不同交易策略和不同模拟样本的风险预测结果进一步表明，使用正则藤Copula函数来对多元相关结构进行灵活建模，可以带来更为稳健和准确的组合VaR预测绩效。再次，针对尾部相关性对资产配置的影响，本文采用马尔可夫转换Copula模型来同时捕捉资产间相关关系的非线性和时变性特征，并基于该模型设计了一种选择组合调整时机的方法。基于中国股市中两类股票组合（高风险和低风险股票组合）的实证结果证实了金融资产间相关结构的依状态转换特征，从而指出以往文献中在较长投资期限内基于一个固定模型所构建的静态策略是不适宜的。本文提出可以借助马尔可夫转换Copula模型预测未来状态转换的时刻，采用状态变化后新的Copula函数类型来重新预测分布并计算的新的组合权重。样本外资产配置绩效分析表明，相比文献中已有策略，本文的择时策略确实能给投资者带来更高的平均已实现收益率和确定性等价收益率。最后，针对尾部相关性对资产定价的影响，本文关注了个别股票与整个市场之间的尾部相关性，并分析了其对股票收益率的影响作用。卖空限制的存在往往导致远比上涨风险更为严重的极端下跌市场风险的产生，然而线性的Beta却无法对其区分。本文使用尾部相关性系数来捕捉这种个股随整个市场同时暴跌的极端下跌市场风险。基于上证A股的实证分析证实了个股与市场间尾部相关性是普遍存在的，而且更为值得关注的是，这种尾部相关性对沪市中股票收益率具有显著的解释能力，其解释能力即使在控制了其他定价因素（尤其是线性Beta）的影响后依然存在。因此，尾部相关性系数提供了一种刻画市场风险的新角度，可能包含着已有定价因素之外的信息而有潜力成为新的定价因子。
[Abstract]:Frequent financial crises have brought serious adverse consequences to investors, financial markets and even the global economy. In these crises, the market presents a special feature different from the normal situation: in the case of a single variable, investors face the "tail risk" of extreme loss; in a multivariable case, gold is in the case of gold. There is a relatively stronger "tail relevance" among market or financial assets. How to grasp and cope with these two special market characteristics is a crucial issue for both investors and risk managers, policymakers and regulators. Therefore, this paper applies a variety of flexible econometric methods to capture. The "tail risk" and "tail dependence" characteristics in the crisis period are analyzed, and their impact on risk management, asset allocation and asset pricing is further analyzed.
First, in view of the tail risk faced by a single asset, this paper introduces a more effective inspection method of the saddle point technology return test for the VaR and ES prediction accuracy of various risk models, and reexamines what model can most accurately capture the tail risk under the single variable condition. Based on the Chinese stock market The empirical analysis shows that the simple GARCH-Normal model can not reasonably capture the risk characteristics of the Chinese stock market, and the best model is the GARCH-EVT model. Further research based on more mature stock markets and more risk models also confirms that in order to get enough accurate risk prediction, it is necessary to use the extreme value theory EVT to finance the finance. The distribution tail of the rate of return is modeled separately. It is difficult to capture the characteristics of the distribution tail and distribution at the same time using only one form of distribution, even if it is recommended in previous literature to capture the t distribution of skewed and thick tail characteristics at the same time. In addition, the risk prediction is influenced by the GARCH model. For the first time, the relative importance of the two dimensions of the certainty is carried out by formal statistical tests. It is found that the effect of the tail setting of the residual distribution on the prediction of VaR and ES is stronger than the wave rate equation.
Secondly, in view of the effect of tail correlation on risk management, this paper first proposes a method based on multivariate Copula function simulation to calculate the risk contribution of individual assets in the combination, thus realizing the significant test of the difference between different asset risk contributions. At the same time, the Copula function is used to describe the nonlinear correlation structure between assets and the relationship between assets. The advantages of the tail correlation characteristics, the results of the risk contribution calculated using this method have also become more consistent and no longer affected by the confidence level and risk metrics. In addition, this paper also introduces a more flexible multivariate correlation structure modeling tool, the canonical Copula function, to overcome the choice in the existing research. The finite nature of the type of meta Copula function and the existence of different defects. Based on the empirical analysis of three stock markets in Shanghai, Hongkong and Taiwan, the advantages of the canonical Copula in the characterization of multiple correlation structures are confirmed. Flexible modeling of multivariate correlation structures with regular rattan Copula functions can bring more robust and accurate performance of combined VaR prediction.
Thirdly, in view of the effect of tail correlation on asset allocation, this paper uses the Markov transformation Copula model to capture the nonlinear and time-varying characteristics of the correlation between assets, and designs a method for selecting the timing of combination adjustment based on the model. Based on the two types of stock portfolios in China's stock market (high risk and low risk stock groups) The empirical results confirm the state transition characteristics of the related structure between financial assets, and then point out that the static strategy based on a fixed model in the long term literature is not suitable in the previous literature. This paper proposes that the Markov transform Copula model can be used to predict the time for the transition of the future state and adopt the state change. The new Copula function type is used to re predict the new combined weights of distribution and calculation. The performance analysis of asset allocation shows that, compared with the existing strategies in the literature, the timing strategy of this paper can indeed bring higher average realized yield and certainty equivalent yield to investors.
Finally, in view of the effect of tail correlation on asset pricing, this paper focuses on the tail correlation between individual stock and the whole market, and analyzes its effect on the stock returns. The existence of short selling limit often leads to extreme market risk which is far more serious than the risk of rising, but the linear Beta is not possible. This paper uses the tail correlation coefficient to capture the extreme falling market risk of this stock with the whole market falling at the same time. Empirical analysis based on the Shanghai Stock A shares confirms that the tail correlation between the stock and the market is common, and it is more worthy of concern that the tail correlation has the stock returns in the Shanghai stock market. The explanatory power, even after controlling the influence of other pricing factors (especially linear Beta), still exists. Therefore, the tail correlation coefficient provides a new perspective of market risk, which may contain information other than the existing pricing factors and have the potential to become a new pricing factor.
【学位授予单位】：电子科技大学
【学位级别】：博士
【学位授予年份】：2012
【分类号】：F832.51;F224

【相似文献】