【作者】 桂文林；

【导师】 伍超标；

【作者基本信息】 暨南大学 ， 数量经济学， 2005， 硕士

【摘要】 过去几十年,金融市场的频繁动荡让人触目惊心,人们在致力于研究金融市场深层次运行机制的同时也加强了金融监管的力度。金融风险度量理论、资产组合理论和资本定价理论奠定了现代金融管理理论的基石。三者一脉相承、密不可分,后两者以前者为基础。金融风险度量的每一个问题的产生和有效解决都会对它们产生深远的影响。以此为基础,本研究综述了金融风险度量的主要问题和问题解决。其中包括金融资产回报分布的厚尾,极值分布的条件,分布的完整性以及组合资产回报极值分布等问题。本研究重在前两个问题深入总结和有效解决,用于中国证券市场(主要是股票市场)的实践也是其中不可抹煞的一笔。具体说来,在分块样本极大值理论模型和阀顶点(Peaks over threshold,POT)理论模型的两大框架下,文章侧重于模型参数的估计。前者既包括模型参数的极大似然估计还包括分位数的极大似然点估计及轮廓对数似然法的区间估计等。后者既包括基于广义帕雷托分布(GPD)参数的极大似然估计,又包括矩法估计等。在矩法估计中,本文结合形状参数矩法估计量1/(?)的实践,分析了实践中只计算一、二阶矩估计量的原因和条件,给出了二次Hall自助法计算分位数的算法流程图和随机数程序。实证分析,先用GARCH(1,1)模型拟合深圳成分指数的对数收益,对其标准化误差项拟合广义帕雷托分布(GPD)。在计算标准化误差项和对数收益率分位数的基础上,最后进行违背数的二项式检验,检验结果表明模型具有良好的估计预测效果。更多还原

【Abstract】 Past several decades, the multifarious turbulence of the financial market let people’s eyes shocking. People are concentrating on the research of the operation mechanism of the financial market in-depth, at the same time enhancing the dint degree of the direct finance management. Financial Risk Calculates Theory, Portfolio Theory and Asset Pricing Theory established the theoretical sill of management of modern finance. Three come down in one continuous line inseparably. Empress both regards the former as the foundation. The creation of each problem with valid solution of financial risk calculates will produce the profound influence to them. Regard this as the foundation, this research overview the main problem with solution including the "fat tail" distribution of financial asset return, the term of extreme value distribution, integrality of distribution and the extreme value distribution of the portfolio etc. The point is two ex- a problem thorough summary with valid solution. Used for the fulfillment research of the Chinese stock market is absolutely necessarily. Concretely, under the two frames of the theories model of the block sample maximum and POT(peaks over threshold), the article is laid emphasis in the estimate of the model parameter. The former since include maximum likelihood estimate of the model parameter to still include maximum likelihood point estimate of the model parameter and confidence intervals estimate of the profile log-likelihood etc. The latter since include maximum likelihood estimate of the GPD(generalized Pareto distribution ) model parameters to still include moment estimate etc. In moment estimate, we joined together the fulfillment of the moment estimate of the shape parameter1/a , analyzed the reason and the term of computing an and two rank moment estimates only in fulfillment. Finally, we present arithmetic flow chart of Hall twi-bootstrap to calculate the quantiles and the programme to generate random numbers. In the beginning, we intend to match logarithms rate of return of the composition index of Shenzhen withthe GARCH(1,1) model in the empirical research. Then we match the standardized residual with GPD. On the foundation of calculating the quantile of standardized residual and logarithms rate of return, we test the model with disobedient numbers which obey binomial distribution. The test results show the model the good effect of estimate and forecast.更多还原