迭代局部多项式国债收益率曲线模型研究

发布时间：2018-04-08 21:01

本文选题：国债收益率曲线　切入点：迭代局部多项式模型　出处：《西南财经大学》2012年博士论文

【摘要】：随着现代金融市场的发展,利率市场化进程的推进,以及国家宏观调控力度的加强,国债收益率曲线研究受到越来越多地重视。利用模型对国债收益率曲线的准确估计具有重要的意义：一方面可以对金融产品进行准确定价,确保金融市场运行的稳定有序；另一方面也可以帮助国家准确地把握宏观经济运行情况,实施有效的宏观经济调控。然而由于我国国债市场较西方发达国家发展较晚,发展不够完善,表现出附息国债占比较大,样本数据异常点较多等各种实际情况,导致国外先进的收益率曲线估计模型在我国的应用效果并不十分理想。目前,国内对收益率曲线的估计研究大多集中对国外模型的缝补和实证检验,从未根据我国市场特征提出合适的估计模型。于是何种收益率曲线模型更适合我国国债市场特征?能否根据我国国债市场特征提出合适的收益率曲线估计模型?这都是我国国债收益率曲线理论研究和实际应用中迫切需要解决的问题。有鉴于此,本文围绕以上问题展开,立足于我国国债市场的现实特点,在统计学中新近发展的局部多项式估计模型的基础上,通过理论研究和实证分析,提出了适合我国市场特征的迭代局部多项式国债收益率曲线静态估计模型、动态估计模型和宏观金融模型。其中,迭代局部多项式静态估计模型提高了对交易日当天收益率曲线的静态估计,而且具有良好的统计性质；迭代局部多项式动态估计模型不仅能够很好地刻画样本内收益率曲线的动态变化趋势,更重要的是还具有较强的样本外预测能力；迭代局部多项式宏观金融模型,将宏观经济变量和收益率曲线相结合,研究宏观经济对收益率水平的影响,对我国金融市场的稳定和发展,以及宏观调控的有效实施具有重要的现实意义。本文对国债收益率曲线模型的研究沿着“静态估计模型→动态估计模型→宏观金融模型”的研究思路,依次逐步展开,层层递进：首先,对我国国债市场的现状和现有静态模型进行分析和研究,并据此提出适合我国市场特征的迭代局部多项式静态估计模型；其次,在动态NS模型的理论框架下,将静态模型推广到动态。通过扩展,不仅增加模型对样本期内收益率曲线变动趋势的刻画效果,更重要的是提高了模型的预测能力；最后,将动态模型和宏观金融变量相结合,提出迭代局部多项式宏观金融模型。充分利用宏观经济变量和收益率曲线蕴含的信息,进一步提升模型的预测能力。具体内容如下：在静态估计模型方面,本文对目前我国国债市场的现状和现有静态估计模型进行了细致的梳理和分析,归纳出了我国债券市场的特点和现有静态估计模型运用时所存在的问题。在此基础上,通过对现有统计学中新近发展的局部多项式估计方法进行改进,提出了适合我国市场特征的迭代局部多项式静态估计模型。并通过蒙特卡洛模拟、实证分析等方法对新模型的拟合效果、统计性质进行了研究和比较。在动态估计模型方面,本文对现有动态估计模型,特别是近期提出的动态NS模型进行了分析和研究,发现传统动态估计方法虽然能够对收益率曲线动态变化规律进行较为准确地刻画,但是对未来收益率水平的预测能力表现很差。而新近提出的动态NS模型,通过两步估计方法,建立了收益率曲线模型和三个动态因子的——对应关系,从而有效地改善了动态模型的预测能力。但是在两步估计中第一步采用的是剥离息票静态估计方法,该方法在我国债券市场上的应用情况并不理想。因此本文基于迭代局部多项式静态估计模型,借助动态NS模型的框架提出迭代局部多项式动态NS模型。并利用上海证券交易所国债交易数据实证比较了迭代局部多项式动态NS模型对样本期内收益率曲线的拟合效果和对样本期外收益率水平的预测能力。同时,本文还对动态NS模型中三个关键动态因子的含义进行了探索和研究,通过理论分析认为三个动态因子既可以表示收益率曲线的期限特征(短期、中期和长期)也可以表示收益率曲线的形状特征(水平、斜率和曲线)。利用实证分析认为三个动态因子与期限特征相关性较弱,而与形状特征有较强的相关关系,表明三个动态因子分别通过刻画收益率曲线的水平、斜率和曲线特征来拟合收益率曲线。在宏观金融模型方面,本文首先对宏观经济运行情况进行分析,提炼出实体经济变量、货币经济变量、物价水平变量和利率市场变量。用脉冲响应函数研究宏观经济变量冲击对动态因子的影响。然后将宏观经济变量与动态因子相结合,提出估计收益率曲线的宏观金融模型。试图能够充分利用收益率曲线自身和宏观经济运行的相关信息有效地预测未来收益率水平。本文通过将理论剖析与实证检验相结合,取得以下主要实证结果：一是,在静态模型的研究中,分别从对特殊形状收益率曲线的拟合效果比较、蒙特卡洛模拟和实证研究三个方面比较迭代局部多项式模型和现有静态估计模型的拟合效果。结果表明迭代局部多项式模型拟合的收益率曲线对国债定价与实际交易价格之间的绝对误差和均方根误差最小,拟合效果最佳。另外,通过数理统计理论证明和蒙特卡洛模拟表明迭代局部多项式模型估计具有渐近正态分布的优良统计性质。从而为该模型的广泛应用打下坚实的基础。二是,在动态模型的研究中,通过选取上海交易所附息债券交易数据实证分析和比较模型的样本内拟合效果和样本外预测能力。结果表明迭代局部多项式动态NS模型不仅比其他动态模型具有更好的样本期内拟合效果,更重要的是具有更强的样本期外预测能力。另外,通过对迭代局部多项式动态NS模型中动态因子的现实含义研究表明,三个动态因子分别通过刻画收益率曲线的水平、斜率和曲率特征,来描述收益率曲线的动态变化过程。三是,在宏观金融模型的研究中,通过脉冲响应函数研究宏观经济冲击对三个动态因子的影响,结果表明宏观经济冲击对动态因子具有显著而持续的影响,特别是货币市场的冲击影响最为明显。另外,选取上海证券交易所附息国债交易数据实证检验了宏观金融模型的预测能力,结果表明加入宏观经济变量之后,宏观金融模型能够有效地利用宏观经济和收益率曲线自身的信息,提高模型的预测能力。论文通过理论考察和实证研究,在以下方面取得了部分创新成果：一是,提出迭代局部多项式静态估计模型,并对模型的统计性质进行探究。在梳理和分析我国债券市场的现实特征和现有收益率曲线静态估计方法的基础上,提出了适合我国市场特征的迭代局部多项式静态估计模型,通过蒙特卡洛模拟和实证分析证实了新模型具有更好的估计效果。并对模型的统计性质进行研究,表明模型具有渐近正态分布的统计性质。二是,提出迭代局部多项式动态NS模型,并对模型中三个动态因子的现实含义进行探究。在动态NS模型和两步估计的理论框架下,将迭代局部多项式静态估计推广到动态。该模型不仅具有较好的样本内拟合效果,更重要的是具有较强的样本外预测能力。同时还对模型中三个动态因子的现实含义进行研究。结果表明三个动态因子分别表示收益率曲线的水平、斜率和曲率特征。三是,提出迭代局部多项式宏观金融模型。在分析宏观经济变量冲击对动态因子的影响基础上,利用向量自回归模型将迭代局部多项式动态NS模型和宏观经济变量相结合提出宏观金融模型。研究表明加入宏观因子后的宏观金融模型能够更加充分地利用宏观经济和收益率曲线自身的信息,具有更强的预测能力。本研究虽然取得了一些有用成果,但由于理论和实际经济数据方面的限制,论文在对模型的使用的普遍性和深度研究上还存在不足,期望随着理论的发展和经济数据资料的丰富,在后续研究中不断改进和完善。
[Abstract]:With the development of modern financial markets, the liberalization of interest rate, and strengthen the national macro-control efforts, the Treasury yield curve is paid more and more attention. The research has important significance to accurately estimate yields curve using the model: on the one hand to accurate pricing of financial products, to ensure the financial market operation stable and orderly; on the other hand can also help countries accurately grasp the macroeconomic situation, the implementation of effective macroeconomic regulation. However, due to China's bond market than the western developed countries developed late, the development is not perfect, showing interest bearing bonds accounted for a larger, more variety of abnormal sample data of the actual situation, leading to the foreign advanced the yield curve estimation model is applied in our country is not very ideal. At present, for the estimation of the yield curve of domestic focus Sewing and empirical test of foreign models, according to the characteristics of China's market has never put forward appropriate estimation model. So what the yield curve model is more suitable for the characteristics of China's bond market? Whether according to the characteristics of China's bond market is proposed to estimate the appropriate model of the yield curve? This is an urgent need to solve the curve theory research and practical application the issue of China's bond yields.
In view of this, this article focuses on the above problems, based on the reality of China's bond market, the local polynomial newly developed in statistics based on the estimated model, through theoretical research and empirical analysis, we propose an iterative local polynomial yield curve for the Chinese market characteristic of static estimation model, model and macro finance model dynamic estimation. The iterative local polynomial static estimation model improves the static estimation of the yield curve on the day of the transaction, but also has good statistical properties; state estimation model can well describe the dynamic change trend of the yield curve in sample dynamic iterative local polynomial, but also has a strong sample local polynomial prediction ability; iterative macro finance model, the macroeconomic variables and the yield curve combination of macroeconomic research The effect of yield level is of great practical significance to the stability and development of China's financial market and the effective implementation of macro regulation.
Study on the curve model of this paper on the rate of bond yields along the "static estimation model, dynamic estimation model, research the macro financial model", in order to gradually expand, progressive layers: firstly, analyze and study the status quo of China's bond market and the existing static model, and puts forward the iterative local polynomial is suitable for the characteristics of China's market. Static estimation model; secondly, in the theoretical framework of dynamic NS model, the static model is extended to dynamic. By extension, not only increases the trend rate of return curve model for the sample period describe the effect, more important is to improve the prediction ability of the model; finally, the dynamic model and the combination of macro financial variables. An iterative local polynomial of macro financial model. Make full use of macroeconomic variables and the yield curve contains information, to further improve the prediction ability of the model. The contents are as follows:
In the static estimation model, this article on the current status of China's bond market and the existing static estimation model and analyzes in detail, summed up the existing problems when using the model of the characteristics of China's bond market and the existing static estimation. On this basis, the local polynomial of existing statistics in recent development estimation the improved method, we propose an iterative local polynomial for Chinese market characteristic and static estimation model. Through Monte Carlo simulation, the fitting effect of empirical analysis on the new model, the statistical properties were studied and compared.
In the dynamic estimation model, based on the existing dynamic estimation model, especially the dynamic NS model proposed recently for analysis and research, found that the traditional method of dynamic estimation can change rule of the dynamic curve yields are more accurately described, but the ability to predict performance of yield level is very poor. The future dynamic NS model the newly proposed method, through two step estimation, established the corresponding relationship of yield curve model and three dynamic factors, thus effectively improve the prediction ability of the model. But in the two step in the estimation of first step using the static estimation method of stripping the coupon and application of this method in the bond market in China is not ideal. So this paper based on local polynomial estimation iterative static model, dynamic model of the proposed framework using NS iterative local polynomial dynamic NS model and use. The Shanghai stock exchange bond trading data empirical comparison of iterative local polynomial dynamic NS model fitting effect to the yield curve in the sample period and the level of sample period yields the prediction ability.
At the same time, this article also has carried on the exploration and study of three key dynamic factor dynamic NS model in meaning, through theoretical analysis that the three dynamic factors can not only represent the term feature of the yield curve (short, medium and long term) can also represent the shape feature of the yield curve (level, slope and curve). The empirical analysis that three dynamic factors and duration characteristics of weak correlation, and have a strong correlation with the shape feature, show that the three dynamic factors respectively to describe the yield curve level by, slope and curve to fit the yield curve.
In the aspect of macro financial model, this paper firstly analyzes the macroeconomic situation, refine the real economic variables, monetary economic variables, price level variables and the interest rate market variables. Using the impulse response function of the impact of macroeconomic variables on the dynamic factor. Then the macro economic variables and dynamic factor combination, the paper estimates the macro the financial model of the yield curve. To make full use of the yield curve of its own information and macro-economy effectively predict the future yield level.
Through the combination of theoretical analysis and empirical test, the following main empirical results are obtained.
One is, in the study of static model, respectively, from the result of fitting curves of special shape yield comparison, Monte Carlo simulation and empirical research on the three aspects of a comparative iterative local polynomial model and the existing static estimation model fitting effect. Results show that the iterative local polynomial model fitting the yield curve of the minimum absolute error between the bond pricing and the actual transaction price and the RMS error, the best fitting effect. In addition, through the mathematical statistics theory proving and Monte Carlo simulation show that the iterative local polynomial estimation model is asymptotically normal distribution of excellent statistical properties. It is widely applied for the model to lay a solid foundation.
Two, study on the dynamic model, the ability to predict by selecting the Shanghai stock exchange coupon bond transaction data empirical analysis and comparison model in sample fitting and out of sample. The results show that the iterative local polynomial dynamic NS model not only has better than other dynamic models of the sample period fitting effect, more important is to have more samples period prediction ability. In addition, by studying the meaning of dynamic factor iterative local polynomial dynamic NS model shows that the three dynamic factors respectively to describe the yield curve of the level, slope and curvature characteristics to describe the dynamic changes of the yield curve.
Three, in the study of macro financial model, the impulse response function of the macroeconomic impact of the three dynamic factors. The results show that macroeconomic impact on the dynamic factor has significant and lasting influence, especially the influence of the money market the most obvious impact. In addition, select the Shanghai stock exchange treasury bond transaction data analysis to test the predictive ability of macro financial model, after the results showed that the addition of macroeconomic variables, macro financial model can effectively use the macro economy and the yield curve of their information, improve the prediction ability of the model.
Through theoretical investigation and empirical research, some innovative achievements have been achieved in the following aspects:
First, an iterative local polynomial static estimation model, and statistical properties of the model are explored. In combing and analyzing the realistic characteristics of China's bond market and the current yield curve static estimation method based on the proposed iterative local polynomial is suitable for the characteristics of China's market static estimation model, the new model has a better estimation effect confirmed by Monte Carlo simulation and empirical analysis. And study the statistical properties of the model show that the model has statistical properties of asymptotic normal distribution.
The two is an iterative local polynomial dynamic NS model, and the real meaning of the three dynamic factor model are explored. In the dynamic NS model and two step estimation theory, the iterative local polynomial estimation is extended to the dynamic static. The model not only has a better fitting effect in the sample, more important is the ability the prediction has a strong sample. At the same time the research meaning of the three dynamic factors in the model. The results show that the three dynamic factors respectively the yield curve level, slope and curvature characteristics.
The three is an iterative local polynomial model. In the macro financial impact analyzing the impact of macroeconomic variables on the dynamic factor, using vector autoregressive model iterative local polynomial dynamic NS model and macroeconomic variables is proposed based on the macro financial model. Research shows that adding the macro financial model of macro factors can make full use of the macro economy and the yield curve of their information, predictive ability is stronger.
This research has made some useful achievements, but because of the theory and the actual economic data limitations, based on the universality and depth of use of the model is not expected, with the development of the theory and economic data rich, constantly improve and perfect in the follow-up study.

【学位授予单位】：西南财经大学
【学位级别】：博士
【学位授予年份】：2012
【分类号】：F224;F812.5

【参考文献】