利率市场化背景下商业银行利率风险的测度
发布时间：2018-07-20 14:55
【摘要】：中国利率市场化改革开始于20世纪90年代中期,最近一次具有突破性的改革举措是2013年7月20日金融机构贷款利率的放开,这标志着我国利率市场化改革进入了实质性的阶段。利率实行市场化后,将会加剧商业银行之间的竞争、加大经营和管理受利率波动的影响,进而商业银行面临的最主要市场风险将会是利率风险,因此商业银行应当重视识别与防范利率风险,然而识别与防范利率风险最关键的就是对利率风险进行有效测度,本文基于利率市场化背景下,探究商业银行利率风险的有效测度方法,为我国商业银行利率风险的测度与管理提供一定的理论依据。为此,本文分别构建了两种不同的利率风险测度模型,通过比较分析选出较好的一种作为利率风险测度模型。第一种模型是在构建不同GARCH类参数、非参数的VaR模型基础上得到的APGARCH-GED-MC非参数VaR模型。通过比较随机变量分别服从正态、t和GED分布的GARCH(1,1)、EGARCH(1,1)和APGARCH(1,1)这9个模型,选出最优的APGARCH(1,1)-GED模型进行样本外预测,将得到的条件标准差σ作为MC模拟中利率波动的标准差,模拟出不同分布下的利率,将预测的利率差作为利率风险的VaR值。由Kupiec失败率检验法可知APGARCH-GED-MC VaR模型较参数法和非参数法计算的利率风险VaR值与实际收益率的差异小,失败率较GARCH类参数、非参数的VaR模型有所降低。第二种模型是从VaR定义(VaR是一定置信水平下的分位数)、分位数回归可应用在金融风险领域以及在构建第一种模型时发现收益率服从何种分布形式对计算VaR的影响较大角度出发,采用分位数回归的方法构建测度利率风险的VaR模型。分别构建两类分位数回归VaR模型：分位数回归的非递归和递归VaR模型。其中分位数回归的非递归VaR模型记为：QR.APGARCH-GED模型(非对称的PGARCH-GED分位数回归VaR模型),模型的解释变量σt+1,和σt+12由APGARCH(1,1)-GED进行样本外预测得到；分位数回归的递归VaR模型记为：AAVS-CAViaR模型(不对称绝对值、斜率的条件分位数自回归),该模型将VaR的自相关性和收益率的不对称性综合考虑在解释变量中。仍采用Kupiec失败率检验法对这两个基于分位数回归的VaR模型进行回测检验,发现QR.APGARCH-GED VaR模型计算的利率风险VaR值不仅对实际收益率有较好的覆盖性,而且降低了预测的失败率。综上所述,本文是从两种不同的角度分别构建利率风险测度的VaR模型。首先,通过Kupiec失败率检验发现在构建的第一种VaR模型过程中,由PGARCH-GED-MC VaR模型预测的利率计算的利率风险VaR值与实际收益率较GARCH类参数、非参数的VaR模型接近,失败率也有所降低,因此选择APGARCH-GED-MC VaR模型作为本文第一种测度利率风险的VaR模型；其次,在构建分位数回归VaR模型过程中,根据QR.APGARCH-GED VaR模型预测计算的利率风险VaR值对实际收益率有较好的区间覆盖,失败率比AAVS-CAViaR模型低,因此选择QR.APGARCH-GED VaR模型作为本文构建的第二种测度利率风险的VaR模型；最后,通过比较APGARCH-GED-MC VaR模型与QR.APGARCH-GED VaR模型的失败天数,发现根据QR.APGARCH-GED VaR模型计算的VaR值,不仅大大降低了模型预测的失败天数,而且对实际收益率有较好的区间覆盖,在实际收益率大幅波动的某些时间段,计算的利率风险VaR值也能随之剧烈波动。基于此本文最终选择QR.APGARCH-GED VaR模型作为当前利率市场化水平下商业银行利率风险度量模型。该模型不仅为我国现阶段商业银行利率风险的测度与管理提供了理论依据,同时为市场化程度更高阶段的利率风险测度提供了参考。
[Abstract]:The reform of China's interest rate marketization began in the middle of the 1990s. The recent breakthrough of the reform was the release of the loan interest rate of the financial institutions in July 20, 2013, which marks the substantive stage of the reform of the interest rate marketization in China. After the interest rate is marketed, the competition between commercial banks will be intensified and the operation will be increased. The most important market risk faced by commercial banks will be interest rate risk. Therefore, commercial banks should pay attention to the recognition and prevention of interest rate risk. However, the key to identify and prevent interest rate risk is to measure the interest rate risk effectively. Based on the background of interest rate marketization, this paper explores commercial banks. The effective measure method of bank interest rate risk provides a certain theoretical basis for the measurement and management of interest rate risk of commercial banks in China. This paper constructs two different interest rate risk measurement models respectively, and selects a better model of interest rate risk measurement by comparison and analysis. The first model is to construct different GARCH classes. The APGARCH-GED-MC nonparametric VaR model is obtained on the basis of the non parametric VaR model. By comparing the 9 models of normal, t and GED distribution GARCH (1,1), EGARCH (1,1) and APGARCH (1,1), the optimal APGARCH model is selected for the external prediction, and the obtained conditional standard difference Sigma is used as the interest rate in the simulation. The standard deviation of the fluctuation is used to simulate the interest rate under the different distribution, and the interest rate difference is predicted as the VaR value of the interest rate risk. The Kupiec failure rate test shows that the difference between the VaR value of the APGARCH-GED-MC VaR model and the actual rate of return calculated by the parameter method and the non parameter method is smaller than the GARCH parameter and the non parameter VaR model. The second model is from the VaR definition (VaR is the quantile under a certain confidence level). The quantile regression can be applied to the financial risk field and in the construction of the first model of the distribution of the rate of return on the calculation of the impact of the VaR, the method of quantile regression is used to construct the VaR model for measuring the risk of interest rate. Two types of quantile regression VaR models are constructed, respectively, the non recursive and recursive VaR models of quantile regression. The non recursive VaR model of quantile regression is recorded as the QR.APGARCH-GED model (asymmetric PGARCH-GED quantile regression VaR model), the explanatory variable of the model, sigma t+1, and the sigma t+12 from APGARCH (1,1) -GED. The recursive VaR model of quantile regression is recorded as the AAVS-CAViaR model (asymmetric absolute value, the conditional quantile autoregression of the slope). The model considers the autocorrelation of VaR and the asymmetry of the rate of return in the explanatory variables. The Kupiec failure rate test is still used to retest the two VaR models based on the quantile regression. It is found that the interest rate risk VaR calculated by the QR.APGARCH-GED VaR model not only has a good coverage of the actual rate of return, but also reduces the failure rate of the forecast. In summary, this paper constructs the VaR model of the interest rate risk measurement from two different angles. First, the first VaR model is found by the Kupiec failure rate test. In the process, the interest rate risk VaR calculated by the PGARCH-GED-MC VaR model is close to the GARCH class parameter, the non parametric VaR model and the failure rate. Therefore, the APGARCH-GED-MC VaR model is selected as the first VaR model of the interest rate risk measurement in this paper. Secondly, the quantile regression VaR model is constructed. In the process, the interest rate risk VaR calculated according to the QR.APGARCH-GED VaR model has a better interval coverage and the failure rate is lower than that of the AAVS-CAViaR model. Therefore, the QR.APGARCH-GED VaR model is selected as the second VaR model to measure the interest rate risk of this paper. Finally, the APGARCH-GED-MC VaR model and QR are compared by comparing the APGARCH-GED-MC VaR model and QR. The number of failure days of the.APGARCH-GED VaR model shows that the VaR value calculated by the QR.APGARCH-GED VaR model not only greatly reduces the number of failure days of the model prediction, but also has a better interval coverage for the actual rate of return, and the calculated interest rate risk VaR can also fluctuate sharply in some time periods of the substantial fluctuation of real returns. Based on this In this paper, the QR.APGARCH-GED VaR model is selected as the current interest rate risk measurement model of commercial banks under the current market level of interest rate. This model not only provides a theoretical basis for the measurement and management of interest rate risk of commercial banks in China at present, but also provides a reference for the measurement of interest rate risk in a higher stage of marketization.
【学位授予单位】：南京财经大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F832.33
本文编号：2133894
[Abstract]:The reform of China's interest rate marketization began in the middle of the 1990s. The recent breakthrough of the reform was the release of the loan interest rate of the financial institutions in July 20, 2013, which marks the substantive stage of the reform of the interest rate marketization in China. After the interest rate is marketed, the competition between commercial banks will be intensified and the operation will be increased. The most important market risk faced by commercial banks will be interest rate risk. Therefore, commercial banks should pay attention to the recognition and prevention of interest rate risk. However, the key to identify and prevent interest rate risk is to measure the interest rate risk effectively. Based on the background of interest rate marketization, this paper explores commercial banks. The effective measure method of bank interest rate risk provides a certain theoretical basis for the measurement and management of interest rate risk of commercial banks in China. This paper constructs two different interest rate risk measurement models respectively, and selects a better model of interest rate risk measurement by comparison and analysis. The first model is to construct different GARCH classes. The APGARCH-GED-MC nonparametric VaR model is obtained on the basis of the non parametric VaR model. By comparing the 9 models of normal, t and GED distribution GARCH (1,1), EGARCH (1,1) and APGARCH (1,1), the optimal APGARCH model is selected for the external prediction, and the obtained conditional standard difference Sigma is used as the interest rate in the simulation. The standard deviation of the fluctuation is used to simulate the interest rate under the different distribution, and the interest rate difference is predicted as the VaR value of the interest rate risk. The Kupiec failure rate test shows that the difference between the VaR value of the APGARCH-GED-MC VaR model and the actual rate of return calculated by the parameter method and the non parameter method is smaller than the GARCH parameter and the non parameter VaR model. The second model is from the VaR definition (VaR is the quantile under a certain confidence level). The quantile regression can be applied to the financial risk field and in the construction of the first model of the distribution of the rate of return on the calculation of the impact of the VaR, the method of quantile regression is used to construct the VaR model for measuring the risk of interest rate. Two types of quantile regression VaR models are constructed, respectively, the non recursive and recursive VaR models of quantile regression. The non recursive VaR model of quantile regression is recorded as the QR.APGARCH-GED model (asymmetric PGARCH-GED quantile regression VaR model), the explanatory variable of the model, sigma t+1, and the sigma t+12 from APGARCH (1,1) -GED. The recursive VaR model of quantile regression is recorded as the AAVS-CAViaR model (asymmetric absolute value, the conditional quantile autoregression of the slope). The model considers the autocorrelation of VaR and the asymmetry of the rate of return in the explanatory variables. The Kupiec failure rate test is still used to retest the two VaR models based on the quantile regression. It is found that the interest rate risk VaR calculated by the QR.APGARCH-GED VaR model not only has a good coverage of the actual rate of return, but also reduces the failure rate of the forecast. In summary, this paper constructs the VaR model of the interest rate risk measurement from two different angles. First, the first VaR model is found by the Kupiec failure rate test. In the process, the interest rate risk VaR calculated by the PGARCH-GED-MC VaR model is close to the GARCH class parameter, the non parametric VaR model and the failure rate. Therefore, the APGARCH-GED-MC VaR model is selected as the first VaR model of the interest rate risk measurement in this paper. Secondly, the quantile regression VaR model is constructed. In the process, the interest rate risk VaR calculated according to the QR.APGARCH-GED VaR model has a better interval coverage and the failure rate is lower than that of the AAVS-CAViaR model. Therefore, the QR.APGARCH-GED VaR model is selected as the second VaR model to measure the interest rate risk of this paper. Finally, the APGARCH-GED-MC VaR model and QR are compared by comparing the APGARCH-GED-MC VaR model and QR. The number of failure days of the.APGARCH-GED VaR model shows that the VaR value calculated by the QR.APGARCH-GED VaR model not only greatly reduces the number of failure days of the model prediction, but also has a better interval coverage for the actual rate of return, and the calculated interest rate risk VaR can also fluctuate sharply in some time periods of the substantial fluctuation of real returns. Based on this In this paper, the QR.APGARCH-GED VaR model is selected as the current interest rate risk measurement model of commercial banks under the current market level of interest rate. This model not only provides a theoretical basis for the measurement and management of interest rate risk of commercial banks in China at present, but also provides a reference for the measurement of interest rate risk in a higher stage of marketization.
【学位授予单位】：南京财经大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F832.33
【参考文献】
相关期刊论文 前4条
1 牛锡明;我国商业银行贷款风险度管理的理论研究[J];经济研究;1998年03期
2 吴涛;我国利率市场化改革中的金融安全问题分析[J];金融研究;2002年11期
3 施正可,涂三勤;VaR模型在我国证券市场的实证分析——基于t分布的RiskMetrics法[J];开发研究;2004年01期
4 叶五一;张明;缪柏其;;基于尾部指数回归方法的CVaR估计以及实证研究[J];统计研究;2012年11期
相关博士学位论文 前1条
1 解其昌;分位数回归方法及其在金融市场风险价值预测中的应用[D];西南财经大学;2012年
相关硕士学位论文 前1条
1 李雨芹;分位数回归与风险度量及应用[D];吉林大学;2009年
，本文编号：2133894
本文链接：https://www.wllwen.com/guanlilunwen/huobilw/2133894.html
