【作者】 谷伟；

【导师】 万建平；

【作者基本信息】 华中科技大学 ， 概率论与数理统计， 2005， 硕士

【摘要】 近年来,巴林银行、东南亚金融机构和长期资本管理公司等一系列因承担市场风险而发生巨额损失甚至倒闭的案例,使得无论金融机构还是监管当局都日益重视对市场风险的管理。为使风险管理体现客观性和科学性,市场风险管理多采用定量分析技术,大量运用数理统计模型来识别、度量和监测风险。VaR 模型正是这样一种定量分析工具,目前已受到业内人士的广泛认可,成为国内国际许多金融机构所采用的风险分析方法。VaR 的计量是在一定概率水平下,投资组合价值在一段时期内最多可能损失的金额。它比传统的风险测量技术,如到期日、持续期以及标准差等有更大的适应性和科学性。正因为这样,VaR 在金融风险控制、机构业绩评估以及金融监管等方面被广泛应用。传统的VaR 计算方法中一般假定市场因子的未来变化服从正态分布,而事实上,众多文献和大量的实证研究都说明金融数据的实际分布较正态分布具“厚尾细腰”的特征,即所谓的“厚尾”分布。从而传统的计算VaR 的方法往往会低估了实际损失值,特别是置信度较高时,当投资者进行某些高风险的投资时,如果低估了自己所面临的风险时,就可能因为较低的储备金而无法应付这种危险局面,从而可能引发破产危机。为解决正态分布对实际金融分布拟合失真的问题,本文分别引入“厚尾”分布中的t 分布和双曲分布进行VaR 建模,并结合实证分析,由t 分布和双曲分布所计算的VaR 值可避免上述不足,不会低估实际损失值,所以t 分布法和双曲分布法比传统计算VaR 的方法有较高可信度,同时说明t 分布和双曲分布在拟合金融数据的样本分布方面的性能与真实性较正态分布大为改善,因此使用它们对金融数据进行建模则会更贴近于实际。此外,VaR 虽然是金融市场风险测量的主流模型,而随着金融市场风险测量技术的发展,近年来我们渐渐发现了它的一些不足,与此同时出现了CVaR 等新的风险度量方法,在文中我们从一致性风险度量的角度比较了它和VaR 性质上的不同,发现它们各有其应用的优势。更多还原

【Abstract】 In recent years, a series of cases occured such that Barings Bank 、Southeast financial institutions and Long Term Capital Management companies and so on suffered huge losses even bankruptcy because of bearing the risk in market, which make not only the financial institutions but also the regulatory agencies increasingly pay attention to the management of market risk. In order to make risk management objective and scientific, it uses quantitative analysis technique, that is, largely uses mathematical statistics model to identify、measure and inspect ventures. VaR model is exactly such a technique, which currently has been widely accepted by this field of industry and adopted by a lot of financial institutions in the world. What VaR model measures is the most possible losses that the investment value suffers in certain period and under a given probability level. It is more practical and scientific than traditional risk measurement technology such as maturity、duration and standard error and so on. Just for this, VaR model is widely applied to financial risk control、financial supervision、the achievement evaluation of institutions and so on. This article primarily argues that the traditional VaR technique method is unpractical because it supposes the future changes of market factors follow the normal distribution, and a lot of articles and positive researches have suggested that financial data’s actual distribution has much thicker tail and thinner waist, that is the so-called “heavy tail”distribution. This traditional VaR method usually underestimates the actual losses, especially the case is that if the credit degree is much higher, when investors make some high-ventured investments, and if he underestimates the risk that he will have to face, he will not be able to deal with the dangerous situation, and may runs the risk of bankruptcy. In order to solve the problem that normal distribution has not the character of “heavy tail”, this article will respectively introduce the t distribution and hyperbolic distribution to VaR model, together with the positive research to suggest, the VaR numerical value that t distribution and hyperbolic distribution get doesn’t underestimate the actual losses, so t distribution and hyperbolic distribution are more reliable. At the same time, compared the sample distribution with the normal distribution 、t distribution and hyperbolic distribution,we can see that t distribution and hyperbolic distribution have greatly improved the capability and trueness of combining the distribution of the financial data sample, so the application of t distribution and hyperbolic distribution to set the model of financial data can be more consistent with the practice.With the development of risk measurement technology,we find some disadvantage of VaR and come into a new method of CVaR.In this article,we compare them through coherent risk measurement and find each of them has its superiority.更多还原