基于几种风险测度的比较研究

发布时间：2018-03-05 15:20

本文选题：风险测度　切入点：一致性公理　出处：《华中师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：在经济全球化、金融一体化的背景下,证券市场发挥着越来越重要的作用,但也不可避免地带来了极大的风险。而新兴的中国金融市场,受国外投资策略及国内经济环境的双重影响,具有其特有的性质。因此,研究现有的几种风险测度方法及其在中国证券市场上的适用性非常有必要。本文在现有文献的研究基础上,从理论和实证两个方面对均值-标准差、VaR、CVaR、熵凸风险测度、等熵风险测度、修正后的熵凸风险测度和等熵风险测度、HMCR和极小值测度的风险识别能力进行分析。在理论方面,主要从一致性公理和随机占优一致性以及风险测度的凸性、信息容量、风险测度值大小方面对几种风险测度的风险识别能力进行分析。在实证方面,一个角度是采用Spearman秩检验法检验几种风险测度的风险识别能力,另一个角度就是通过组合选择考察几种风险测度的实际择股能力。其中,对择股能力的研究比较重要,分单阶段、多阶段和不同市场组合进行分析。本文的研究样本为上证50指数及其成分股收盘价日对数收益率。可得以下结论:在理论层面,标准差测度不满足单调性,VaR不满足凸性,熵凸风险测度、等熵风险测度、HMCR和极小值测度为一致性风险测度,满足凸性的条件。除标准差之外,其它几种风险测度都可以表示为谱风险测度的形式,其不同点在于对权函数的选择,HMCR和等熵风险测度涵盖了整个收益率分布的信息量,相应的具有较大的风险测度值,VaR和CVaR分别仅涵盖了收益率分布的一个尾部点和一段尾部区间。标准差不满足一阶随机占优一致性,VaR和CVaR分别满足一阶和二阶随机占优一致性,熵凸风险测度、等熵风险测度和HMCR具有更高阶的随机占优一致性。从理论上推测,满足一致性、凸性、涵盖越多的信息容量、具有较大的风险值及满足较高阶随机占优一致性的风险测度与其具有较高的风险识别能力是一致的。在实证层面,根据几种风险测度在不同置信水平下的Spearman秩相关系数,可知具有较高阶随机占优一致性的风险测度同样具有较高的风险识别能力。组合选择的结果也证明了这一点,即HMCR(p=3)、极小值测度、熵凸风险测度、等熵风险测度的择股能力优于CVaR、VaR和标准差,当然这一结论受市场组合的影响,并在股市较为稳定时比较明显。在对几种风险测度的择股能力进行研究时,考虑了多阶段及不同市场组合下的影响。根据多阶段组合选择结果,几种风险测度的组合优化能力显著增高,且多阶段组合优化过程消除了样本内异常值的影响,使得组合优化结果更符合理论预期。根据本文研究的五种市场组合的优化结果,认为投资者在根据样本内组合权重进行投资时,基于几种风险测度的积极型投资策略更适宜预测股市为熊市的情况,指数型被动投资策略较适宜预测股市为牛市或股市波动较大的情况。
[Abstract]:In the context of economic globalization and financial integration, the securities market is playing a more and more important role, but inevitably brings great risks. The dual influence of foreign investment strategy and domestic economic environment has its unique nature. It is necessary to study several existing risk measurement methods and their applicability in China's securities market. Based on the existing literature, this paper makes a theoretical and empirical study on the mean-standard deviation VaRCvar, entropy convex risk measurement. Isentropic risk measure, modified entropy convex risk measure and isentropic risk measure HMCR and minimum measure are analyzed. In theory, the consistency axiom, random dominance consistency and convexity of risk measure are analyzed. In the aspect of information capacity and risk measure value, the paper analyzes the risk identification ability of several kinds of risk measures. In the empirical aspect, the Spearman rank test method is used to test the risk recognition ability of several risk measures. Another angle is to examine the actual stock selection ability of several kinds of risk measures through combination selection. Among them, the research on stock selection ability is more important, which is divided into single stage. The study sample of this paper is the daily logarithmic return rate of Shanghai Stock Exchange 50 Index and its constituent stock closing price. The following conclusions can be drawn: at the theoretical level, the standard deviation measure is not satisfied with monotonic VaR and does not satisfy convexity. Entropy convex risk measure, isentropic risk measure HMCR and minimum measure are consistent risk measures, which satisfy the conditions of convexity. Except for standard deviation, several other risk measures can be expressed as spectral risk measures. The difference is that the selection of weight function and isentropic risk measure cover the information of the whole return distribution. The corresponding risk measure values and CVaR only cover a tail point and a tail interval of the return distribution respectively. The standard deviation does not satisfy the first-order random dominant consistency and the CVaR satisfies the first-order and second-order random dominant consistency, respectively. Entropy convex risk measure, isentropic risk measure and HMCR have higher order random dominant consistency. The risk measure with larger risk value and satisfying higher order random dominant consistency is consistent with its higher risk identification ability. At the empirical level, according to the Spearman rank correlation coefficient of several risk measures at different confidence levels, It can be seen that the risk measure with higher order random dominance consistency also has higher risk identification ability, and the result of combination selection also proves this point, that is, HMCRP p3, minimum measure, entropy convex risk measure. The stock selection ability of Isentropic risk measure is better than that of Cvar VaR and standard deviation. Of course, this conclusion is influenced by the market combination and is obvious when the stock market is relatively stable. The effects of multi-stage and different market combinations are considered. According to the results of multi-stage portfolio selection, the combination optimization ability of several risk measures is improved significantly, and the influence of outliers in samples is eliminated by the multi-stage combinatorial optimization process. According to the optimization results of five market combinations studied in this paper, it is considered that when investors invest according to the weight of the portfolio in the sample, The positive investment strategy based on several risk measures is more suitable to predict the stock market as a bear market, and the exponential passive investment strategy is more suitable to predict the situation that the stock market is a bull market or that the stock market fluctuates greatly.
【学位授予单位】：华中师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F224;F832.51

【参考文献】