量价混合信息GJR-GARCH模型下的上证指数量价关系分析与风险测度

发布时间：2018-05-15 01:21

本文选题：量价混合信息GJRGARCH + 量价关系　；参考：《西南财经大学》2014年硕士论文

【摘要】：量价关系研究是金融学术和实务分析研究中备受关注的命题。研究表明,金融资产的价格与成交量存在显著的统计相关性,而这种量价关系可以分为两个层次,一是资产价格的绝对值变化与成交量存在正相关关系,二是资产价格的波动率与成交量存在正相关关系。同时,量价理论中混合信息分布理论、信息不对称理论、意见分歧理论等细分理论用实证分析的方法从经济学、社会学、心理学的角度对市场表现出的量价关系进行了研究,开辟并验证了许多理解市场表现的新角度和新方式。作为金融市场技术面分析的重要内容,它关系到市场投资者对资本市场的理解和资产价格未来走势的判断,用价格辅以成交量表现对未来价格走势进行判断,进而帮助确定金融资产的买卖决策,对市场交易产生着直接的影响。在险价值(Value at Risk)对于风险测度与监控的重要性已在过去几十年全球金融市场的发展过程中得到了验证。伴随着全球经济金融一体化,以及金融衍生产品市场的迅速发展,暴露或隐藏于资本市场的风险及其对应的监控方法成为监管层、学术界和投资者日益关注的热点。在险价值的概念和测度模型在此环境中产生且逐步实现了成熟发展。《巴塞尔协议一》的补充文件《资本协议关于市场风险的补充规定》将资本金在险价值VaR模型纳为规范的测算模型方法,《巴塞尔协议二》则要求各银行在构建风控模型的基础上进行压力测试,以计算VaR用以监测金融资产的风险水平。 GARCH族模型是刻画具有时变方差性质的金融时间序列的有效手段。在其有效刻画金融时间序列的基础之上,能运用具有不同特征的子模型有针对性地进行量价关系和在险价值等拓展领域的深入研究。例如,运用经典的GJR-GARCH等非对称性模型能有效反映出金融市场的杠杆效应,在各类GARCH族模型中根据不同的样本特征也可以根据需要选择出最佳的在险价值测度模型。在阅读了相关文献的基础上,本文把能反应昨日收益率和成交量信息的量价混合信息虚拟变量引入到经典的GJR-GARCH模型中,依据所得实验模型的回归结果判定模型拟合样本序列的有效性。若实验模型拟合有效,则一方面进一步用实验模型对杠杆效应进行更为细化的结构分析,并结合经典的量价理论对上证指数进行量价分析。另一方面对实验模型进行在险价值的测度和检验实验,通过与其他经典模型测度在险价值的能力进行比较,考察实验模型在风险度量方面的能力。本文的结构和主要内容如下。首先,“绪论”章节对为何进行量价关系和在险价值的研究进行了选题背景和意义的说明,对GARCH族模型、量价关系理论和在险价值理论的相关文献进行了梳理和归纳,并提出了如上一段所述的研究思路。其次,“量价混合信息GJR-GARCH模型的构建与运用”章节对各类GARCH族的基本形式和VaR的测度与检验方法进行了详细回顾,并对GARCH族模型建模过程中事前检验和事后检验等建模步骤进行阐述,而后介绍了量价混合信息虚拟变量的构造方式,以及量价混合信息GJR-GARCH模型的建模思想与基本形式。再者,“样本数据的搜集与预处理”章节对本文样本的选择及选择的依据进行说明,介绍了样本预处理的目的和方法,描述了样本的统计特征,并对样本的自相关、异方差性质及样本分布进行了识别,为下一步的实证分析做好准备。最后,“实证结果与分析”章节记录了量价混合信息GJR-GARCH模型的实证建模过程,比较了实验模型和经典GARCH族模型的拟合结果,对实验模型拟合的有效性进行了判定。在此基础上用实验模型对杠杆效应进行更为细化的结构分析,进而结合经典的量价理论对上证指数进行量价分析,并比较了实验模型与经典GARCH族模型的风险测度能力。现将该章节得到的主要结果分为实验模型拟合的相关结果与上证指数量价关系分析的相关结果两部进行阐述。一方面,通过实证分析,可得到的量价混合信息、GJR-GARCH模型拟合的相关结果包括如下几方面。 1、量价混合信息GJR-GARCH模型在拟合效果上具有良好效果首先,从该模型的参数显著性检验、Box-Ljung检验、ARCH-LM检验、信号偏误检验(sign bias test)和皮尔森卡方Goodness-of-Fit检验等事前和事后检验来看,量价混合信息GJRGARCH模型都有着良好的表现。Nyblom参数稳定性检验所体现出的部分参数的稳定性表现欠佳,但这是所拟合得的各类GARCH族模型的共同特征。其次,GJR-GARCH模型的回归参数符合模型符号的参数定义。量价混合信息GJR-GARCH模型中α+β之和在所有用于参照的GARCH族模型之中最小,说明量价混合信启、GJR-GARCH模型对杠杆效应的刻画比其他参照模型能更有效地解释部分短期波动信息的持续性影响。最后,从信息准则统计量来看,量价混合信息虚拟变量GJR-GARCH模型的变量个数最多,但其AIC、BIC、SIC、HQIC等信息准则量是众多模型之中的最小值(HQIC值为次小值),且这也从另一方面说明了该模型在拟合效果上的良好性质。 2、量价混合信息GJR-GARCH模型能对杠杆效应进行细化分析从量价混合信息GJR-GARCH模型的回归结果来看,并非所有昨日均值方程负残差的出现均会使今日出现杠杆效应。所有昨日负残差相较非负残差对应的今日收益率的平均额外波动率之所以显著,是由于昨日出现大额负向收益率和中等额度负向收益率时所引发的今日相对大幅的额外波动,拉升了所有昨日负向残差对应的今日平均额外波动的水平。因此,在运用和理解杠杆效应模型时,应当注意杠杆效应是在平均意义上存在的。 3、基于量价混合信息GJR-GARCH模型和各类GARCH族模型进行的VaR测度和风险预测能力分析首先,量价混合信息GJR-GARCH模型虽然在估计的似然值、信息准则和其他统计检验方面的表现均要优于其他模型,但在风险测度方面并没有因此而体现出独到的优越性,在某些分布假设的某些显著性水平下甚至比部分模型的风险刻画能力来得弱。值得注意和思考的是,量价混合信息GJR-GARCH模型在t分布下估计的VR(Violation Rate)值普遍比其他模型都来得小,在正态分布的5%显著性水平下的VR值也是所有模型中较小的,一方面体现出它对VaR较为大胆的估计和与之相应而生的较强的风险警惕能力,另一方面小于1且偏离1较多的VR数值反应出它过度估计风险、风险管理成本较高的性质。其次,不同分布假设与不同显著性水平设定会对模型的在险价值测度效果产生影响。所用的样本收益率序列在正态分布假定的5%显著性水平下运用各类GARCH模型进行风险测度,均能取得良好的风险估计效果。而在student-t分布假定的5%显著性水平下运用各类GARCH模型进行风险测度,几乎不能取得准确的风险估计效果,所预测得的VaR值过度估计了风险。当显著性水平由5%变为1%时,正态分布假定下的各类GARCH模型所预测出的VaR值将大幅低估未来的风险,学生t分布假定下的各类GARCH模型高估了风险但部分模型能较为准确地对风险进行测度。因此,在评价一个模型的风险测度能力是否良好时,应当将分布与显著性水平与其评价紧密结合,而不能简单地以一种分布假定在一种特定的显著性水平的风险测度结果作为衡量不同模型风险测度能力的唯一标准。另一方面,基于量价混合信息GJR-GARCH模型的拟合结果,可得到的上证指数量价关系分析的相关结果如下所述。当昨日市场出现大跌幅、中等跌幅表现时,所蕴含的价格信息平均来看会对今日收益率的额外波动产生显著影响,且伴随不同换手率的量能信息对今日收益率的额外波动会产生不同程度的影响。具体体现在昨日大跌幅大换手率、大跌幅小换手率、大跌幅中换手率、中等跌幅大换手率、中等跌幅小换手率、中等跌幅中等换手率的量价表现对应的回归系数均在1%的显著性水平下拒绝了系数为零的原假设；当昨日市场出现小跌幅表现时,所蕴含的价格信息平均来看对今日收益率的额外波动所产生影响的显著性水平不高或者不显著,具体体现在昨日小跌幅大换手率、小跌幅中等换手率的量价表现对应的回归系数在10%的显著性水平下显著不为零,小跌幅小换手率的量价表现对应的回归系数不显著。不同量价信息会对今日的杠杆效应产生不同程度的影响。大跌幅大换手率和中等跌幅大换手率的昨日信息比未出现所有标的量价信息时的平均波动率高出额外0.9个百分点左右；大跌幅小换手率和大跌幅中等换手率的昨日信息比未出现所有标的量价信息时的平均波动率高出额外0.7个百分点左右；小跌幅大换手率、中跌幅小换手率、中跌幅中换手率的昨日信息比未出现所有标的量价信息时的平均波动率高出额外0.4到0.5个百分点左右。小跌幅小换手率和小跌幅中等换手率的昨日信息比未出现所有标的量价信息时的平均波动率高出额外0.1到0.2个百分点,且回归系数的显著性不强。可归纳出,昨日量价表现对今日收益率波动影响的两个特征。一是昨日跌幅等级越高,昨日的量价信息、对今日收益率波动的贡献度往往越大。二是昨日同级别的跌幅等级中,大换手率等级的信息量对今日收益率波动的贡献度最高,且远大于小换手率等级和中等换手率等级的信息量对今日收益率波动的贡献度；而昨日小换手率的成交量信息在对应大跌幅和中跌幅的价格信息时,所带来的今日收益率额外波动均比中等换手率的成交量信息所带来的额外波动要大,仅在小换手率小跌幅信息量出现时不会带来显著的额外波动。运用量价理论结合实证分析可对上述量价关系特征做出经济意义上的解析。综上,本论文具有如下特点。首先,本文在经典GJR-GARCH模型的基础上引入量价混合信息虚拟变量进行拟合效果试验,实验模型的拟合结果在统计检验和统计信息量上均表现良好,说明实验模型能对样本进行有效拟合。且与参照族的模型相比,实验模型在拟合效果上具有一定的优越性。其次,本文在GJR-GARCH模型中引入量价混合信息虚拟变量,实现了对非对称杠杆效应更为细致的刻画,加深了对杠杆效应的理解。再者,本文能运用经典的量价关系理论对实验模型的实证结果进行解析,从中捕获上证市场交易的量价关系特征,将统计建模知识与金融学理论相结合,丰富了实证分析的经济学内涵。最后,本文在不同的显著性水平和不同的分布条件下,对实验模型和传统模型的在险价值测度能力进行了比较,提出了基于实证结果的风险测度模型的评价标准。
[Abstract]:The study of the relationship between volume and price is a topic of concern in the study of financial academic and practical analysis. The study shows that there is a significant statistical correlation between the price of financial assets and the volume of trading, which can be divided into two levels, one is that the change of the absolute value of the asset price has a positive correlation with the volume of the transaction, and the two is the fluctuation of the asset price. There is a positive correlation between rate and volume. At the same time, the theory of mixed information distribution, information asymmetry theory, disagreement theory and other subdivision theories have studied the relationship between price and price from the perspective of economics, sociology and psychology in the theory of mixed information distribution, information asymmetry theory and Opinion Divergence theory, which opened and verified many understanding of market performance. As an important part of the technical analysis of the financial market, it relates to the market investor's understanding of the capital market and the judgment of the future trend of the asset price, the judgment of the future price trend with the volume of the price, and the determination of the future price trend, and then to help determine the decision of the sale of the financial assets, and direct the market transaction. Influence.
The importance of Value at Risk to risk measurement and monitoring has been verified in the development of global financial markets over the past few decades. With the global economic and financial integration and the rapid development of the financial derivatives market, the risk of exposure or hidden in the capital market and its corresponding monitoring methods have become a supervision. The concept and measurement model of the risk value produced and gradually achieved mature development in this environment. < the supplementary document of the Basel agreement > < the supplementary provisions on the market risk of the capital agreement > the method of calculating the capital in the VaR model of the value of the value of the risk, < Basel. Agreement two > requires banks to conduct stress tests on the basis of building wind control models to calculate the risk level of VaR to monitor financial assets.
The GARCH model is an effective means to describe the time series of time-varying variance properties. On the basis of its effective characterization of the financial time series, we can use the sub models with different characteristics to carry out the in-depth study of the quantity price relation and the expansion of the value of the risk value. For example, using the classical GJR-GARCH and other asymmetries. The sexual model can effectively reflect the leverage effect of the financial market. In the various GARCH models, the best value measurement model can be selected according to the different sample characteristics.
On the basis of reading related literature, this paper introduces the virtual variable of volume and price mixed information that can respond to yesterday's rate of return and volume information into the classic GJR-GARCH model. According to the regression results of the experimental model, the model fits the validity of the sample sequence. If the experimental model is fit, then the experiment is further used. The model makes a more detailed structural analysis on the leverage effect, and analyzes the quantity and price of the Shanghai stock index with the classic price theory. On the other hand, the experimental model is measured and tested on the risk value. The experimental model is compared with other classical models to measure the ability of the risk value, and the experimental model is examined in the risk measurement. Ability. The structure and main contents of this article are as follows.
First, the chapter of "Introduction" explains the background and significance of the research on the relationship between quantity and price and the value of the value of risk, combing and summarizing the GARCH model, the theory of quantity and price relation and the related literature of the value theory, and puts forward the research ideas as described in the previous paragraph.
Secondly, the chapter "construction and application of the GJR-GARCH model of volume and price mixed information" is a detailed review of the basic forms of various GARCH families and the measurement and inspection methods of VaR, and the modeling steps of the GARCH model modeling process, such as pre test and post inspection, and then the construction of the virtual variables of the mixed information of quantity and price is introduced. Method and the modeling idea and basic form of mixed information GJR-GARCH model.
Furthermore, the section of "sample data collection and preprocessing" explains the selection and selection of the sample, introduces the purpose and method of sample pretreatment, describes the statistical characteristics of the sample, and identifies the autocorrelation, heteroscedasticity and sample distribution of the sample, and is prepared for the next step of the empirical analysis.
Finally, the empirical results and analysis section records the empirical modeling process of the mixed information GJR-GARCH model of quantity and price, compares the experimental model and the classic GARCH model, and determines the validity of the experimental model fitting. On this basis, the experimental model is used to make a more detailed structural analysis of the bar effect. Combined with the classic price theory, the quantity and price of Shanghai stock index is analyzed, and the risk measurement ability of the experimental model and the classic GARCH model is compared. The main results obtained in this chapter are divided into two parts of the correlation results of the experimental model fitting and the relationship analysis of the Shanghai Stock index and price.
On the one hand, through empirical analysis, we can get mixed information of volume and price, and the relevant results of GJR-GARCH model fitting include the following aspects.
1, the mixed information GJR-GARCH model has a good effect on the fitting effect.
First, from the parameter significance test of the model, Box-Ljung test, ARCH-LM test, signal error test (sign bias test) and Pearson chi square Goodness-of-Fit test and other pre and post test, the volume and price mixed information GJRGARCH model has a good performance of the stability of the.Nyblom parameter stability test of the stability of some of the parameters of stability. Poor performance, but this is the common feature of the GARCH models that are fitted.
Secondly, the regression parameters of the GJR-GARCH model conforms to the parameter definition of the model symbol. The sum of alpha + beta in the mixed information GJR-GARCH model of the quantity and price is the smallest in all the GARCH models used for reference, indicating that the amount and price are mixed, the depiction of the leverage effect in the GJR-GARCH model is more effective than the other reference models to explain some of the short-term volatility information. The persistence effect.
Finally, according to the information standard statistics, the variable number of GJR-GARCH model of volume and price mixed information virtual variable is the most, but its AIC, BIC, SIC, HQIC and other information criteria are the minimum values of many models (HQIC value is the sub value), and this also illustrates the good properties of the model in the other aspect.
2, volume price mixed information GJR-GARCH model can refine the leverage effect.
From the regression results of the mixed information GJR-GARCH model, not all the negative residuals of yesterday's mean equation are all leveraged. The average extra volatility of all yesterday's negative residual difference compared to the non negative residual is significant, which is due to the large negative rate of return and the middle level of yesterday. The relatively large additional volatility caused by the degree of negative returns has raised the level of today's average extra fluctuation corresponding to the negative residual of yesterday. Therefore, when using and understanding the leverage effect model, it should be noted that the leverage effect exists on the average.
3, the VaR measure and risk prediction capability analysis based on the mixed price information GJR-GARCH model and all kinds of GARCH family models.
First, the mixed information GJR-GARCH model is superior to the other models in the estimated likelihood, information criteria and other statistical tests, but it does not reflect the unique superiority in the risk measurement. At some significant levels of some distribution assumptions, it is even better than the risk characterization of some models. The strength is weak.
It is worth noting and thinking that the VR (Violation Rate) value estimated by the mixed information GJR-GARCH model in the t distribution is generally smaller than that of the other models. The VR value under the 5% saliency level of the normal distribution is also the smaller of all the models. On the one hand, it embodies the stronger estimation of the VaR and the stronger wind corresponding to it. Risk alert ability, on the other hand, is less than 1 and deviates from 1 more VR values, reflecting its overestimation risk and higher risk management cost.
Secondly, different distribution assumptions and different levels of significant level setting will affect the value measurement effect of the model. The sample return sequence uses all kinds of GARCH models under the 5% saliency level of the normal distribution hypothesis to measure the risk, which can achieve a good risk estimation effect, while the Student-t distribution assumption is 5% significant. Using all kinds of GARCH models to measure risk at the level of sex, almost no accurate estimation of risk can be achieved. The predicted VaR value overestimates the risk.
When the significant level changes from 5% to 1%, the VaR values predicted by all kinds of GARCH models under the normal distribution assumption will significantly underestimate the risk of the future. The various GARCH models under the t distribution hypothesis of the students will overestimate the risk, but some of the models can measure the risk more accurately.
Therefore, when evaluating the risk measurement ability of a model, the distribution and significance level should be closely combined with its evaluation, and the risk measurement results of a specific level of risk can not be used as the only criterion to measure the risk measurement ability of different models.
On the other hand, based on the fitting results of the mixed GJR-GARCH model of volume and price, the related results of the Shanghai stock index volume and price relationship can be obtained as follows.
When there was a big drop in the market and a medium decline, the price information contained on average would have a significant impact on the extra volatility of today's rate of return, and the amount of energy information with different turnover rates would have different effects on the extra volatility of today's rate of return. The turnover rate, the turnover rate in the large decline, the medium drop large turnover rate, the medium drop small turnover rate, the medium decrease medium turnover rate, the corresponding regression coefficient at the 1% significant level all reject the original hypothesis that the coefficient is zero. The significant level of the impact of the rate of return volatility is not high or significant. It is embodied in yesterday's small turnover rate, and the corresponding regression coefficient of the small drop medium rate is not zero under the significant level of 10%, and the corresponding regression coefficient of the small exchange rate is not significant.
The information of different prices will have different degrees of influence on the leverage effect today. The information of yesterday's big and medium drop rate is about 0.9 percentage points higher than the average volatility of all the price information. The average volatility of all the marked price information is about 0.7 percentage points higher, and the small change rate, the small turnover rate, the exchange rate in the middle drop rate are 0.4 to 0.5 100 points higher than the average volatility when the price information is not presented. The average exchange rate of yesterday's information is 0.1 to 0.2 percentage points higher than the average volatility of all the standard volume of information, and the significance of the regression coefficient is not strong.
It can be concluded that yesterday's volume price performance has two characteristics on the impact of today's yield fluctuation. One is the higher the decline level of yesterday, the information of yesterday's volume and price, the greater the contribution to the fluctuation of the rate of return today. Two is the decline grade of yesterday's same level, the amount of information of the large turnover rate has the highest contribution to the fluctuation of the rate of return today. The contribution of the amount of information that is larger than the small turnover rate and the medium turnover rate to the fluctuation of the rate of return today; while the volume information of yesterday's small turnover rate is related to the price information of the large decline and the middle decline, and the additional volatility of today's rate of return is more than the medium turnover rate.

【学位授予单位】：西南财经大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F832.51

【参考文献】