我国商业银行间同业拆借市场利率风险VaR度量实证研究

发布时间：2018-05-06 06:22

本文选题：同业拆借市场 + VaR　；参考：《西南财经大学》2012年硕士论文

【摘要】：在现代经济环境中,金融处于一个十分重要的位置,而商业银行在其中发挥着至关重要的作用,如何透过商业银行的视角对整体金融风险有一定程度的控制,不仅是商业银行关注的问题,更是监管当局和学术界研究的重点问题。在我国商业银行的整体风险管控中,利率风险预防和控制是首个突出并且应当重视的问题。由于国际上利率风险控制浪潮的推进,以及我国倡导利率市场化的步伐加快,中资银行在面临全球金融机构的挑战,如何对商业银行的风险进行有效的度量和管理研究,已经是商业银行管理面临的一大重要课题。商业银行利率风险研究是一个成熟而又新颖的课题,随着国际经济环境下对金融市场管制的逐步放松,出现了大规模的金融产品创新,使得金融市场波动日益频繁,金融机构由过去的市场资源化探索向内部管理和创新方式改革转变,在金融机构经营管理模式渐渐发生改变的同时,有关利率风险方面的预防和控制度量模型理论和实践也在不断探索与创新中。VaR模型在这一背景下发展起来并受到广泛认同,已成为西方各国公认的市场风险度量与管理工具之一。受当前世界经济环境和我国金融市场发展的影响,中国对商业银行利率风险度量的重视程度也在逐渐加强,对利率风险的关注和利率风险预防和控制的技术引入,甚至利率风险的体制问题都有所改善和进步。但是总的来看,随着利率市场化波动的愈发明显,我国商业银行的计量风险技术还不能满足市场风险正在逐步变大的形势,这与发达国家的商业银行风险度量水平还存在一定的距离。当前我国大部分商业银行少有运用比传统的风险度量方法更为精确的VaR模型法进行利率风险的度量,这与国际上通行的先进计量方法存在着较大差距,因此通过建立合适的VaR模型对我国商业银行利率风险进行度量和监管,具有重要的意义,因此,本文基于我国银行同业拆借市场中的隔夜拆借利率进行实证分析,希望在利率风险的理论和实践分析上能结果我国国情有所创新,能为我国从利率风险的理论、模型建立以及监管体系方面的监管扩展做出实际的学术贡献。在对利率风险相关课程的学习以及论文文献的研读中,了解到利率风险是指非预期的市场利率变化导致的对商业银行表内和表外头寸的影响,会对商业银行的资产价值及收益产生直接的影响。其中,在对利率风险度量的方法中,VaR作为一种新兴的模型正在被世界各地的风险监管当局和金融机构所推祟。VaR (Value at Risk)也叫在险价值,是一种利用统计方法计量风险价值的度量方法。现在,VaR模型己经成为了很多国家的金融风险管理标准,并且将其作为分析工具监管金融机构风险的重要工具,其动态监测和量化监管的特点受到金融监管当局和金融机构的认同与欢迎。本文第三章对商业银行利率风险的基本概念进行了理解和阐述,其中包括阐述商业银行利率风险的定义,以及对商业银行利率风险种类的了解。在对利率风险清晰定义的基础上,进一步深入了解商业银行利率风险的分类共有四种,包括重新定价风险、基准风险、收益率曲线风险和隐含期权风险,并对四种利率风险进行了详尽的定义和概念的诠释。对商业银行利率风险的清晰定义和了解有助于正确选择合适的风险度量工具,合理的规避风险。第二部分主要阐述商业银行利率风险度量方法的演进历程,正确认识和选择利率风险的度量方法是实现利率风险科学管理的重要前提,因此对商业银行利率风险度量工具的演变过程了解十分必要。在对传统的两种利率风险度量方法,即利率敏感性缺口分析和持续期模型的了解后,总结出静态的风险度量方法的优点在于简洁易懂,且易于操作和计算,但是都不能全面的反应账户中的综合交易情况。VaR相对于传统的方法,能更为全面的度量综合复杂的银行利率风险,因此受到广大风险监管者和金融机构的推广。总的来说,VaR方法可以分为参数法、非参数法以及半参数法。该部分中主要介绍了GARCH族模型和基于Risk Metrics的混合正态分布两种参数方法,以及历史模拟法和蒙特卡罗模拟法两种非参数方法。在对方法了解的基础上,拟选用GARCH族模型和基于Risk Metrics的混合正态分布两种参数方法对上海银行间同业拆借市场的隔夜拆借利率进行VaR度量的实证分析。第四章主要对上海银行间同业拆借市场的利率数据进行进行统计特征分析,由于SHIBOR比CHIBOR更为符合市场波动规律,市场化自由程度更广且隔夜拆借利率数据最为频繁,因此选择上海银行间同业拆借利率数据进行实证分析,并对其做收益率对数处理。在对SHIOBR对数收益率数据的统计特征分析中,可以得出以下结论：SHIBOR对数收益率数据并不服从正态分布,具有明显的尖峰厚尾特征,该样本是平稳的时间序列数据,并且存在自相关性。在对SHIBOR对数收益率的条件异方差检验中发现,样本序列的残差序列存在明显的ARCH效应,应该可以用于建立ARMA-GARCH模型。在第五章中,用GARCH族模型来对SHIBOR对数收益率数据进行利率风险度量的实证分析。在对GARCH族模型基本思想了解的基础上,首先确定了SHIBOR样本序列的自回归移动平均模型为AR(1),并得出结论：AR(1)基本满足平稳的要求,且不存在序列相关。在此基础上,选择了GARCH、T-GARCH、E-GARCH这三种GARCH族模型的形式来拟合模型的条件异方差,在对残差分布的选择中,不仅使用正态分布形式,还选择了T分布以及广义误差分布分别进行拟合,选择了18种GARCH模型形式对比其AIC值,其中GARCH(1,1)-T、GARCH(1,2)-T、 GARCH(1,1)-G、GARCH(1,2)-G、TGARCH(1,1)-G、TGARCH(1,2)-G EGARCH(1,1)-T、EGARCH (1,2)-T、EGARCH (1,1)-G、EGARCH (1,2)-G模型的AIC和SC值较小,然后对上述选择的模型进行参数显著性检验,剔除系数不显著及效果不优的模型,选择EGARCH(1,2)-G进行VaR值的计算。第三部分对EGARCH(1,2)-G的VaR值进行回测检验,该模型在95%和99%的置信水平下没有通过检验,说明模型对于风险的估计过于保守,或者可能因为数据样本量不够,但在99%的置信水平下拟合效果较好,从一定程度上能刻画SHIBOR收益率的尖峰厚尾特征。接下来,本文使用参数法中的另一个种方法—基于Risk Metrics的混合正态分布来对SHIBOR对数收益率数据进行拟合,在假定上海银行间同业拆借利率对数收益率服从双正态混合分布的前提下,取日收益率衰减因子为0.94,利用Matlab进行EM迭代,最后求得混合正态分布密度函数的参数估计值。通过标准化的SHIBOR对数收益率直方图可以发现,数据的尖峰厚尾信息特征较为明显,特别是尖峰信息特征,因此可以得出结论,混合正态分布相较于正态分布来说,拟合效果较好,在VaR值的计算上也能获得更高的精度。从混合正态分布VaR值的回测检验来看,在置信水平为99%、95%、90%三种情况下,LR统计量均小于临界值,模型通过回测检验,说明混合正态分布情况下,其实际失败天数与期望失败天数非常接近,该模型所测算的VaR值精确度较高,是一种值得推广的VaR度量方法。通过对GARCH族模型和基于Risk-Metrics的混合正态分布模型法的实证结果对比分析来看,首先,两种方法都是利用同一组数据进行实证分析,因此分析的结果可以直接进行对比,来判断方法不同所带来VaR值结果及其精准度不同的影响；其次,两种方法都是基于参数估计的基础得以实现,都属于参数法的范畴,可以对比其同种方法范畴下的不同之处,对方法的选择对比具有实际意义；最后,两种估计法在软件上能较为快速的实现,GARCH族模型通过Eviews软件来实现,而混合正态分布则需要通过Matlab进行编程计算,其计算速度并无差别,都能在软件中短时间显示结果。然而,两种实证分析方法的不同之处能体现在模型理论理解难易程度、软件实现难易程序、计算复杂程度、方法推广程度、计算结果精准度这五个方面的不同。最后,本文针对VaR模型的两种方法实证分析,得出以下结论：第一,在GARCH族模型的实证研究中,针对上海银行间同业拆借利率数据,经过一系列测算,最终选择EGARCH(1,2)-G模型为风险度量模型,利用EGARCH(1,2)-G拟合的的方差预测值对其分别求出每日动态VaR值。第二,在基于混合正态分布方法的拟合中,本文发现混合正态分布对金融时间序列尖峰厚尾特征的描述较为贴切,且能较为灵活的调整双正态分布各自拟合比例,最终得出较为稳定的参数估计值。第二,EGARCH(1,2)-G只有在99%的置信水平下通过Kupiec检验,而混合正态分布方法在三种置信水平下均通过检验。第四,对于极端风险情况的处理,GARCH族模型比混合正态分布模型更优,同时,混合正态分布拟合的每日VaR值更接近风险平均水平,因此对总体水平上的风险度量更为优化,是另一种值得考虑使用的风险度量方法。
[Abstract]:In the modern economic environment, finance is in a very important position, and commercial banks play a vital role in it. How to control the overall financial risk to a certain extent through the perspective of commercial banks is not only a concern of commercial banks, but also a key issue in the research of the regulatory authorities and academia. In the overall risk control of commercial banks, the prevention and control of interest rate risk is the first prominent and important problem. Because of the promotion of the wave of interest rate risk control in the world and the quickening pace of our country's advocacy of interest rate marketization, the Chinese banks are facing the challenges of global financial institutions and how to make the risk of commercial banks effective. The research on measurement and management has been an important subject in the management of commercial banks. The interest rate risk study of commercial banks is a mature and novel topic. With the gradual relaxation of the financial market regulation under the international economic environment, a large scale of financial product innovation has emerged, which makes the financial market fluctuating more and more frequently, and the financial institutions are becoming more and more frequent. From the past market resource exploration to the reform of internal management and innovation mode, while the management model of financial institutions is changing gradually, the theory and practice of the prevention and control measurement model of interest rate risk are also developed and widely recognized in this context in the continuous exploration and innovation of the.VaR model. It has been recognized as one of the tools of market risk measurement and management in western countries.
Influenced by the current world economic environment and the development of China's financial market, China's attention to the interest rate risk measurement of commercial banks is gradually strengthened. The interest rate risk, the introduction of interest rate risk prevention and control technology, and the institutional problems of interest rate risk are improved and progressed. However, in general, with the interest rate market As the fluctuation of the field becomes more and more obvious, the measurement risk technology of the commercial banks in our country can not meet the situation that the market risk is becoming bigger and bigger. There is a certain distance between the risk measurement of the commercial banks in the developed countries. At present, most of the commercial banks in our country have less precise VaR model using the risk measurement method more than traditional. There is a big gap between the measurement of interest rate risk and the advanced measurement methods prevailing in the world. Therefore, it is of great significance to measure and supervise the interest rate risk of commercial banks in China by establishing a suitable VaR model. Therefore, this paper is based on the empirical analysis of the overnight lending rate in the interbank lending market of China. It is hoped that the theoretical and practical analysis of interest rate risk can result in the innovation of our country's conditions, and can make a practical contribution to our country from the theory of interest rate risk, the establishment of model and supervision system.
In the study of interest rate risk related courses and the study of paper literature, it is understood that the interest rate risk is the effect of the unexpected market interest rate changes on the balance of the commercial bank's balance in the balance sheet and out of the table, which will have a direct impact on the value and income of the commercial banks. In the method of measuring the interest rate risk, VaR is used as a method. A new model is being called the value of.VaR (Value at Risk) by risk regulatory authorities and financial institutions all over the world. It is a measure of measuring the value of risk by statistical methods. Now, the VaR model has become a financial risk management standard in many countries and is used as an analytical tool to monitor the risk. The importance of dynamic monitoring and quantitative regulation is the recognition and welcome of financial regulatory authorities and financial institutions.
In the third chapter, the basic concepts of the interest rate risk of commercial banks are understood and expounded, including the definition of the interest rate risk of commercial banks and the understanding of the types of interest rate risks in commercial banks. On the basis of a clear definition of interest rate risk, there are four kinds of classification of interest rate risks in commercial banks, including the classification of interest risk. Re pricing risk, benchmark risk, yield curve risk and implied option risk, and a detailed definition and interpretation of four kinds of interest rate risks. A clear definition and understanding of the interest rate risk of commercial banks is helpful for the correct choice of appropriate risk measurement tools and reasonable avoidance of risk. The second part mainly describes commercial banks. It is necessary to understand and select the measure method of interest rate risk correctly to realize the scientific management of interest rate risk. Therefore, it is necessary to understand the evolution process of the interest rate risk measurement tool for commercial banks. In the traditional two interest rate risk measurement methods, that is, the interest rate sensitivity gap analysis and the analysis of the interest rate risk. After the understanding of the duration model, it is concluded that the advantages of the static risk measurement method are simple and easy to understand, and easy to operate and calculate, but the comprehensive transaction situation.VaR in the comprehensive response account can be more comprehensive to measure the complex bank interest rate risk compared with the traditional method, so it is subject to the risk supervisor. In general, the VaR method can be divided into parameter method, non parametric method and semi parametric method. In this part, we mainly introduce two parameter methods of GARCH model and mixed normal distribution based on Risk Metrics, and two non parametric methods of historical simulation method and Monte Carlo simulation method. The GARCH model and the mixed normal distribution based on Risk Metrics are selected to make an empirical analysis on the VaR measurement of the overnight lending rate of interbank interbank lending market in Shanghai.
The fourth chapter mainly analyzes the statistical characteristics of the interest rate data in the interbank lending market in Shanghai. Because SHIBOR is more in line with the law of market volatility than CHIBOR, the liberalization of the market is more extensive and the interest rate data is the most frequent. Therefore, the data of interbank lending rate in Shanghai is selected for empirical analysis, and it is done to it. In the analysis of the statistical characteristics of the SHIOBR log return data, we can draw the following conclusion: the SHIBOR logarithmic yield data does not obey the normal distribution, and has the obvious peak and thick tail features. The sample is a stationary time series data, and there is a self correlation. The condition of the SHIBOR logarithmic return rate is different. In the variance test, it is found that the residual sequence of the sample sequence has an obvious ARCH effect and should be used to establish the ARMA-GARCH model. In the fifth chapter, the GARCH model is used to carry out an empirical analysis of the rate risk measurement of the SHIBOR logarithmic return data. On the basis of the basic idea of the GARCH family model, the SHIBOR sample is first determined. The autoregressive moving average model of this sequence is AR (1), and draws the conclusion that AR (1) basically satisfies the requirement of stationary and does not have sequence correlation. On this basis, the conditional heteroscedasticity of the model is fitted with the form of three GARCH family models of GARCH, T-GARCH and E-GARCH, and not only the normal distribution is used in the selection of residual distribution, but also in the selection of residual distribution. The T distribution and the generalized error distribution are selected, and 18 GARCH models are selected to compare their AIC values, including GARCH (1,1) -T, GARCH (1,2) -T, GARCH (1,1). EGARCH (1,2) -G is selected to calculate the value of VaR by selecting the model that is not significant and the effect is not good. The third part of the model is to test the VaR value of EGARCH (1,2) -G, and the model is not tested under the confidence level of 95% and 99%, indicating that the model is too conservative for the risk estimation, or possible. Because the data sample is not enough, the fitting effect is better under the 99% confidence level, and it can describe the peak and thick tail characteristics of the SHIBOR yield to a certain extent. Next, this paper uses a hybrid normal distribution of Risk Metrics based on the mixed normal distribution of the parameter method to fit the data of the logarithmic return of the SHIBOR, in the assumption of Shanghai silver. Under the premise that the rate of return on interbank lending rate obeys the mixed distribution of double normal state, the attenuation factor of daily return rate is 0.94, and EM iteration is carried out by Matlab. Finally, the parameter estimation of the mixed normal distribution density function is obtained. Through the standardized SHIBOR logarithmic return rate direct graph, the information features of the peak and thick tail of the data are compared. For obvious, especially peak information characteristics, it can be concluded that the mixed normal distribution is better than the Yu Zheng state distribution, and can also obtain higher precision in the calculation of the VaR value. From the back test of the VaR value of the mixed normal distribution, the LR statistics are less than the critical in the 99%, 95%, and 90% confidence levels. The model shows that the actual failure days of the mixed normal distribution are very close to the expected number of failed days. The VaR value calculated by the model is high, and it is a kind of VaR measure worth popularizing. By comparing the empirical results of the GARCH model and the mixed normal distribution model based on Risk-Metrics. In the first place, first of all, the two methods use the same group of data for empirical analysis, so the results of the analysis can be compared directly to judge the effect of different VaR results and their accuracy. Secondly, the two methods are based on the base of parameter estimation, all of which belong to the category of parameter method. Compared with the same method category, the selection and contrast of the method has practical significance. At last, the two estimation methods can be realized more quickly on the software, the GARCH model is realized through the Eviews software, and the mixed normal distribution needs to be programmed by Matlab, and the calculation speed is not different and can be short in the software. Time shows the results. However, the difference between the two empirical methods can be reflected in the difficulty of understanding the model theory, the difficulty of software implementation, the complexity of calculation, the degree of popularization of the method and the accuracy of the calculation results. Finally, the following conclusions are drawn to the following conclusions for the two methods of the VaR model: the following conclusions: First, in the empirical study of the GARCH model, according to the interbank interbank interest rate data in Shanghai, after a series of calculations, the EGARCH (1,2) -G model is selected as the risk measurement model, and the daily dynamic VaR value is obtained by using the variance prediction value fitted by EGARCH (1,2) -G. Second, in the fitting based on the mixed normal distribution method, It is found that the mixed normal distribution is more appropriate for the description of the characteristics of the peak and thick tail of the financial time series, and it can be more flexible to adjust the fit ratio of the double normal distribution, and finally get a more stable parameter estimate. Second, EGARCH (1,2) -G is only through the Kupiec test under the confidence level of 99%, and the mixed normal distribution method is three. Fourth, the GARCH family model is better than the mixed normal distribution model for the treatment of extreme risk. At the same time, the daily VaR value of the mixed normal distribution is closer to the risk average, so the risk measurement on the overall level is better. It is another kind of risk measure worthy of consideration. Law.

【学位授予单位】：西南财经大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：F224;F832.33;F822

【参考文献】