【作者】 耿国靖；

【导师】 国世平；

【作者基本信息】 辽宁大学 ， 政治经济学， 2011， 博士

【摘要】 近年来,随着我国加入WTO,利率市场化、资本项目开放以及衍生金融市场的建立,金融资产所面临的市场风险也将日益突出和复杂。而且伴随着经济全球化和金融自由化,以及放松管制和金融创新、技术进步等因素,金融市场效率明显提高,而且市场规模也迅速扩大,与此同时金融市场的波动性和风险逐渐加大并日益明显。至此,金融风险的定量测度也日益重要。VaR方法是目前世界上应用比较普遍的风险定量测度方法之一。并逐渐成为金融风险管理的国际标准。VaR方法测量风险的特点有定量性、综合性、通俗性等,正因为如此而被许多银行、金融机构和监管机构广泛应用。将这种方法引入到中国的证券市场风险管理中有重大的现实意义。对于中国的证券市场,目前没有一种方法能在不同置信水平上有效而准确地估计市场风险值——VaR,本文在对各种VaR估计方法进行比较分析的基础上,对其中的一些方法进行改进,并以创业板综合指数的日收益率数据为样本,广泛借鉴国内外的研究成果,对创业板市场风险进行定量测度。文章所采用的研究方法主要是实证研究方法和比较研究,定量研究和定性研究等研究方法方法。采用这些方法对中国股票市场风险进行分析,从而获得不同置信水平下的VaR值,进而定量的度量中国创业板股票市场风险。如果想比较准确的估计VaR,就需要找到合适的模型。这样的模型能够较好地拟合收益率序列的分布。因此,本文我们首先对创业板综合指数日对数收益率序列的统计特征和分布进行分析,然后根据分析的结论选择更为合理的VaR估计模型。实证研究的结论是——通过正态性、自相关性以及ARCH效应的实证分析我们发现我国股市收益率序列具有尖峰厚尾性,弱自相关性,波动聚集性等特性。估计VaR值的方法有很多种,本文运用几个常用估计方法估计VaR,比如正态方法、GARCH模型法、历史模拟法和极值理论。然后应用Kupiec失败返回检验法对各VaR值有效性进行有效性检验。实证分析的结论显示——应用简单的平均正态方法估计的VaR在各个置信水平上都无效;而极值理论方法估计的VaR,在高的置信水平上无效,在较低的置信水平上有效;而运用历史模拟法和GARCH模型估计的VaR在较低的置信水平上有效,在高的置信水平上无效。因此直接运用这些模型估计VaR的值并不能得到在各个置信水平上都有效的VaR值,所以文中我们从两个方面对估计VaR的模型进行改进。我们在应用极值理论的POT模型估计VaR时,通常都假设超阈值满足独立同分布,但是实际中,超阈值往往是局部相关的。这样计算的结果就会使VaR估计值跟实际值相比较具有比较大的偏差。可以采用两种方法来消除超阈值的局部相关性:一是在POT模型中引入极值指标。实证分析的结论是——引入极值指标改善了POT模型估计VaR的有效性,同时也提高了POT模型估计的准确性。与此同时,实证分析还得出一个结论,在较低的置信水平上,即便在POT模型中引入极值指标,其所估计VaR仍是无效的;第二种方法是运用GARCH模型对收益率序列进行过滤处理。由于极值理论只在高的置信水平上有效,在低的置信水平上其可靠性不如一般的VaR估计方法,所以我们对GARCH模型过滤后的残差序列用极值理论与历史模拟法混合的方法来估计VaR值。经实证分析我们发现,这种方法估计的VaR在各置信水平上都有效,而且非常接近期望值。所以我们得出结论:用常规的方法直接估计我国创业板市场风险的VaR值,得不到在各个置信水平上都有效的VaR值。用GARCH滤波的历史模拟与极值理论的混合方法也可得到在各个置信水平都有效而准确的VaR估计。更多还原

【Abstract】 In recent years, with China’s accession to the WTO, the marketization of interest rate, capital account liberalization and the financial derivatives market is established, facing the market risk of financial assets will be increasingly prominent and complex.But along with the economic globalization and financial liberalization, competition and deregulation and financial innovation and technical progress factor, financial market efficiency is obviously improved, and the market dimensions expands quickly also, at the same time the volatility of financial market and the gradual increase in risk and increasingly apparent. The VaR method is currently the world’s widely used risk measure method. It is becoming the financial risk management of international standard. VaR method of measuring risk characteristics with quantitative, comprehensive, popular, because of this, many banks, financial institutions and regulatory bodies are widely used it. This method will be introduced into China’s securities market risk management is of great realistic significance.For Chinese securities market, there is no method at different confidence levels on efficient and accurate estimation of VaR. Based on the VaR estimation method on the basis of comparison and analysis, some of these methods are improved, and the gem ’s day yield data of the index as a sample, drawing on a wide range of domestic and foreign research results, the GEM market risk measure.The article uses the research technique mainly is empirical and comparative research methods. By using these two methods on Chinese stock market risk analysis, In order to gain the different confidence level VaR value, Then measure of the GEM market risks.If you want to compare the accuracy of estimation of VaR, we need to find a suitable model. This model can better fitting series of return rate distribution. Therefore, in this paper, We first analyze statistical characteristics and distribution of the Second board Market, and then select a more reasonable VaR estimation model based on the analysis of the conclusion. The conclusion of research is that through empirical analysis of normality, autocorrelation and ARCH effect, we found that the Chinese stock market returns series has thick tail peak, weak correlation, volatility clustering and other characteristics.Estimation of VaR value in many ways, this paper uses several commonly used estimation method to estimate VaR, such as normal method, GARCH method, historical simulation method and extreme value theory. And then the application of Kupiec failure returns test method on the VaR validity validity test.The results of empirical analysis shows In various confidence levels, using a simple average normal method to estimate VaR are invalid. Using extreme value theory to estimate VaR, It is invalid at high confidence level. While using the historical simulation method and GARCH model to estimate the VaR at lower confidence level is effectively, and invalid in the high confidence level. Therefore, apply these models directly to estimate the value of VaR and can’t get at various confidence levels are valid VaR values, so in this paper we from two aspects of the estimated VaR model is improved.When we use the POT model of extreme value theory to estimate VaR, usually assuming supra-threshold Meet the conditions of independent identical distribution, But in practice, the threshold is often topically related. The result of this calculation will make VaR estimation value and actual value compared with a relatively large deviation. VaR estimation value and the actual value with relatively large deviation. We can use two methods to eliminate the local correlation of exceeding the threshold. One is introduce extreme value index to the POT model. Introduction of extreme value index of improved POT model to estimate the effectiveness of VaR. At the same time, improved the accuracy of POT model estimation. It also leads to the conclusion, on the lower level of confidence, even in the POT model introduced in the extreme value index, the estimation of VaR is invalid. Second method is to use the GARCH model to deal with the return series. Because of the extreme value theory only in the high confidence level effectively, in a low confidence level on its reliability than general VaR estimation method of GARCH model, so we processed the residual sequence using extreme value theory and historical simulation method hybrid method to estimate VaR value. Through the empirical analysis we found, this method estimates the VaR in various confidence levels are effective, and very close to the expected value. So we concluded that ,In different confidence levels, using the general method to estimate the risk of Chinese stock market is effective value, Correction of simple average normal and EWMA, estimates of the conditional volatility, and then estimate the VaR, could be obtained at different confidence level on effective and accurate VaR value. Or use the GARCH processed historical simulation and theory and mixed methods can also get the effective and accurate VaR estimation at various confidence levels.更多还原