基于联动性和高频时变性的中国股市价格波动特征的实证研究

发布时间：2018-06-23 22:28

本文选题：波动特征 + 聚类分析　；参考：《浙江工商大学》2017年硕士论文

【摘要】：2014年下半年以来,股市波动的方差性堪比2008年的金融危机,在如此剧烈的波动情形下,对波动特征更为全面的研究尤为重要,更能为投资者和监管者应对短期波动提供有用的信息。本文围绕股市波动特征来进行研究,将波动特征分为微观层面的联动性特征和宏观层面的基于高频分量的时变性进行研究,旨在对波动特征进行更深入全面地研究。联动性的研究中,本文选用聚类分析方法,对上证50指数的内部成分股结构进行分析,旨在对比分析不同波动特征时期指数内部联动性特征。首先对经过筛选后的47个样本成分股的日收盘价数据进行处理,定义处理后的序列为收益—成交量序列,并对此序列根据四分位数进行符号化得到新的符号化序列,根据这一新的符号化序列按照定义划分的四个不同波动特征阶段分别进行聚类。从四个阶段的聚类群个数变化的角度和聚类得出的最小生成树图的分叉情况角度分析,行业的联动性随着波动的剧烈程度而逐渐减弱,并且在经历剧烈波动后,无法立刻恢复到波动前的结构特征。在行业联动性中,金融业的行业联动性最为稳定,反映出指数构造时金融业占比较大的合理性。除此之外,也有其他的聚类特征,如地域性和审计机构相似性。基于高频时变性的研究中,本文首先通过引用基于Huang(1998)提出的经验模态分解(EMD)基础上改进的集合经验模态分解(EEMD)的信号分解方法,通过对原始上证50信号序列加入白噪声进行分解,提取出高频分量并进行重构后作为研究对象,由于提取出的高频分量序列本身带有噪音性质,因此在进行一系列的检验后,通过建立拟合得到的ARMA(3,2)-GARCH(1,2)模型对提取出的高频分量进行回归分析。在回归分析中,首先对不考虑成交量增量因素的高频分量进行分析,发现未考虑成交量之前,指数过去的波动对于当前波动的影响为负,即过去的波动会减小当前波动的幅度;其次再对考虑了成交量增量因素的高频分量进行分析,发现指数过去的波动对于当前的波动变成了正向的,即过去的波动会加大当前的波动幅度。通过以上实证分析,并从行为金融学角度出发,得出以下几点结论:(1)行业的联动性在指数波动中占主导地位,尤其是金融业的联动性最为稳定,体现出指数构造时选股的合理性;(2)联动性随着波动的剧烈程度而逐渐增强,并且在极端波动情形下会打破行业限制;(3)指数的波动在短期内会因为信息披露、投资者情绪等因素对未来波动形成不同方向的影响,不仅反映出了股市中“羊群效应”等一些非理性现象的存在,也体现了“信息股市”的特点;(4)无论是联动性的变化以及指数短期波动特征,都体现股市波动“记忆性”。最后,从投资者和监管者的角度分别提出了建议。
[Abstract]:Since the second half of 2014, the variance of stock market volatility is comparable to that of the financial crisis in 2008. In such a volatile situation, a more comprehensive study of volatility characteristics is particularly important. More useful information for investors and regulators to deal with short-term volatility. This paper focuses on the volatility characteristics of the stock market. The volatility characteristics are divided into the micro level of the linkage feature and macro level based on the high frequency component of the time-varying study, aiming at a more in-depth and comprehensive study of volatility characteristics. In the study of linkage, this paper chooses the cluster analysis method to analyze the internal component structure of the Shanghai Stock Exchange 50 Index, aiming to compare and analyze the internal linkage characteristics of the index in different volatility periods. Firstly, the data of the daily closing price of 47 sample stocks are processed, and the processed sequence is defined as a profit-volume sequence, and a new symbolic sequence is obtained by symbolizing the sequence according to the quartile. According to this new symbolic sequence, four different fluctuation characteristic stages are divided according to the definition. From the point of view of the change of the number of clustering groups in four stages and the bifurcation of the minimum spanning tree graph obtained by clustering, the linkage of the industry is gradually weakened with the intensity of the fluctuation, and after experiencing the violent fluctuation, It is not possible to recover the structural characteristics immediately before the fluctuation. In the industry linkage, the financial industry is the most stable, which reflects the rationality of the financial industry when the index is constructed. In addition, there are other clustering features, such as regionalism and audit institution similarity. In the study of high frequency time-varying, the signal decomposition method of set empirical mode decomposition (EEMD) based on the empirical mode decomposition (EMD) proposed by Huang (1998) is introduced in this paper. By decomposing the original SSE 50 signal sequence with white noise, the high frequency component is extracted and reconstructed as the object of study. Because the extracted high frequency component sequence itself has the nature of noise, so after a series of tests, The regression analysis of the extracted high frequency components was carried out by establishing the fitting ARMA (3K2) -GARCH (1K2) model. In the regression analysis, the high frequency component which does not consider the increment factor of trading volume is first analyzed, and it is found that the influence of the past fluctuation of the index on the current fluctuation is negative, that is, the past fluctuation will reduce the amplitude of the current fluctuation. Secondly, by analyzing the high frequency component which takes into account the increment factor of trading volume, it is found that the past fluctuation of the index becomes positive to the current fluctuation, that is, the past fluctuation will increase the current fluctuation range. Through the above empirical analysis and from the point of view of behavioral finance, we draw the following conclusions: (1) the linkage of industry plays a dominant role in the index fluctuation, especially the linkage of financial industry is the most stable. It reflects the rationality of stock selection when the index is constructed; (2) the linkage increases gradually with the intensity of volatility and will break the industry restriction in extreme volatility; (3) the volatility of the index will be due to information disclosure in the short term. Investor sentiment and other factors affect the future fluctuations in different directions, which not only reflects the existence of some irrational phenomena such as "herd effect" in the stock market. It also reflects the characteristics of "information stock market". (4) both the linkage change and the short-term volatility of the index reflect the "memory" of the stock market volatility. Finally, from the perspective of investors and regulators, respectively, put forward suggestions.
【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.51

【参考文献】