基于隔夜信息的中国股市波动率建模与预测研究

发布时间：2018-04-09 04:29

本文选题：隔夜信息　切入点：股市波动率　出处：《山东大学》2017年博士论文

【摘要】：股市波动率的建模与预测一直以来是金融经济学研究的重要内容。它对资产组合选择、金融资产及其衍生品定价、以及金融机构的风险管理都具有重要意义。20世纪80年代起,国内外学者提出了基于低频数据的GARCH类和SV类等模型对股市波动率进行估计和预测,很好地刻画了股市波动的集聚性和时变性特点。进入21世纪,基于高频数据的股市波动率的建模与预测成为新的研究趋势。在已实现波动率(RealizedVolatility,RV)的基础上,涌现了能够刻画股市波动长记忆性和异质特点的ARFIMA类和HAR类等经典模型。信息的传播和扩散是股市产生波动的内在原因。由于股市的交易期间持续时间短,导致股市在两个工作日之间的非交易时段内积累了大量信息,这就是所谓的隔夜信息。由于政策传导、经济走势和国际联动等因素,隔夜信息的产生涉及多个方面。因此,研究隔夜信息对中国股市波动预测的影响具有重要意义。以隔夜信息为新的切入点研究股市波动率的建模与预测是本文的核心内容,并具有十分重要的意义。在学术方面,拓展了对隔夜信息的界定和分类,并结合隔夜信息对股市波动率的影响这一特点对其进行建模和预测研究,丰富了金融波动率建模的理论空间。在理论方面,为政策制定者、信息披露者和股市管理者提供理论依据。本研究致力于使决策管理部门能够在保证调控目标和信息披露的前提下,清楚地认识到这些变动对股市波动造成的冲击,从而形成合理健全的制度体系,有效地降低对股票市场和金融系统带来的风险,维护金融市场稳定。在实践方面,正确认识隔夜信息对股市波动率的影响对于股市投资者做出正确判断有一定指导意义。股市波动并不是一种随机行为,而是受到隔夜信息等因素的影响而变化的。投资者的正确认识一方面可以减少市场的投机行为,另一方面有利于他们充分利用隔夜信息做出合理的投资决策。本文首先回顾了国内外学者在该领域的研究成果,确立了本文的研究方向和理论依据,并为模型建立打下了实证支撑。在此基础上,分别从理论和实证两个方面论证隔夜信息对股市波动率的影响。理论方面,界定了隔夜信息的内涵与分类,并通过信息与波动的相关理论、隔夜信息影响股市波动的微观基础以及隔夜信息影响股市波动的作用机理进行论证。实证方面,对各类隔夜信息、股市波动率及波动的隔夜表现和跳跃行为进行了度量,并通过格兰杰因果关系检验和中介效应分析两条路径来证明隔夜信息对股市波动的影响。最后,围绕本文的的核心,分别提出了三种基于隔夜信息的股市波动率建模方式,并与传统的波动率模型进行预测能力比较。其中,多因素-变系数模型和HAR-CJI模型是分别借助于隔夜信息影响股市波动的中介效应——隔夜表现和跳跃行为对现有的经典股市波动模型进行改进,将隔夜信息的影响考虑到波动率模型中。复合模型则是利用BP神经网络模型,将经典波动模型的估计结果与隔夜信息综合起来。通过对三种模型的实证检验发现,隔夜信息能够提升波动率模型的拟合效果和预测性能。相比较而言,前两者模型具有较好的理论解释能力,而后者则具有更好的预测效果。本文的研究结果体现在三个方面。首先,就隔夜信息对股市波动的影响来说,宏观政策指标类信息、国际市场类信息和上市公司信息披露水平对股市波动表现出不同的影响。具体表现在,基准利率、存款准备金率与采购经理指数等宏观政策指标类信息的变动,国际油价、伦敦金价与纳斯达克指数等国际市场类信息的利空表现,上市公司信息披露程度的提高和两个交易日之间的不连续对日内波动均有增大效应。同时,隔夜信息能够通过影响股市的隔夜表现和股价波动的跳跃行为,对股市日内波动率的预测起着重要作用。一方面,隔夜表现是各类隔夜信息影响股市波动的中介变量,而跳跃行为在部分隔夜信息对日内波动的影响中表现出一定程度的中介效应。其次,从基于隔夜信息的股市波动率模型构建方面来看,本文所提出多因素-变系数模型、HAR-CJI模型和以BP神经网络为基础的复合模型,在拟合效果和预测能力方面,比经典波动模型和神经网络非参数模型表现更好。最后,从新模型的预测能力上看,考虑隔夜信息提高了模型在对股市波动率变动方向和数值大小预测方面的精度,同时提高了非参数模型的稳定性。其中,对预测方向的改进主要表现在对股市波动正向变动的准确性上。基于隔夜信息的经典线性模型和神经网络模型在解释股市波动率的理论意义、预测方式以及预测效果上存在差别。
[Abstract]:Modeling and forecasting of the volatility of the stock market is always an important part of financial economics. Its choice of a portfolio of financial assets and derivatives pricing and risk management of financial institutions is of great significance to the.20 century since 80s, domestic and foreign scholars put forward based on the low frequency data of GARCH and SV models on the stock market volatility to estimate and forecast, can depict the agglomeration and time-varying characteristics of volatility in the stock market. In twenty-first Century, modeling and prediction of stock market volatility of high frequency data rate to become the new trend of research. Based on realized volatility (RealizedVolatility, RV) on the basis of the emergence of the stock market can describe the long memory of volatility and the heterogeneous characteristics of the ARFIMA and HAR classes, such as the classical model. The information dissemination and diffusion is the inherent reason of stock market volatility. The stock market transactions during the short duration of lead Non trading periods between two working days of the stock market accumulated a large amount of information, which is called the overnight information. Because of policy, economy and international linkage and other factors, produce overnight information involves many aspects. Therefore, effect of overnight information has important significance to predicting the fluctuation of the stock market. China to overnight information for the modeling and prediction of the starting point to study the stock market volatility of the new is the core content, and has very important significance. In the academic field, expand the definition and classification of overnight information, combined with the overnight information on the study of modeling and prediction of the effect of stock market volatility of the characteristics of the rich. The theory of financial volatility modeling. In theory, for policy makers, and provide a theoretical basis for information disclosure and stock management. This research is committed to making decision management to guarantee The premise of control objectives and information disclosure, clearly aware of these changes on the stock market volatility caused by the impact, so as to form a sound system, effectively reduce the risk of the stock market and the financial system, maintain the stability of financial markets. In practice, the correct understanding of the overnight information influence on the volatility of the stock market on the stock market investors make the right judgment has certain guiding significance. The fluctuation of the stock market is not a random behavior, but influenced by the factors such as the overnight information changes. A correct understanding of investors can reduce market speculation, on the other hand it helps them to make full use of the overnight information to make rational investment decisions. This paper reviews the domestic and foreign scholars in this field of research, established the research direction and theoretical basis, and gives empirical support in the model. On this basis, respectively from two aspects of theoretical and empirical demonstration of overnight information on the stock market volatility. The theory, classification and definition of overnight information, and through the relevant theories of information and volatility, the overnight information affects the micro foundation of stock market volatility and stock market volatility of the overnight information influence mechanism of empirical demonstration. For all kinds of information, overnight, overnight performance of the stock market volatility and volatility and jump behavior of measurement, and through the Grainger causality test and the mediating effect of the two paths to prove the effects of the overnight information on the stock market fluctuation. Finally, around the core of this paper, we proposed three different stock market volatility overnight information rate based on the modeling method, and the traditional volatility model forecast ability. Among them, multi factor variable coefficient model and HAR-CJI model respectively with the help of the overnight letter Intermediary: overnight performance and jump behavior of stock market fluctuations of interest effect to improve the classic stock market volatility model existing, will affect the overnight information considering the volatility model. Composite model is to use the BP neural network model, the estimation results of classical wave model and the overnight information together. By empirical test on the three the model found that the overnight information can enhance the volatility model fitting effect and prediction performance. By comparison, the model has good explanation ability, while the latter has a better prediction effect. The results reflected in three aspects. First, effects of overnight information on stock market volatility. The macro policy index information, the level of international market information and information disclosure of listed companies have different impact on the stock market volatility. Specific performance in the benchmark interest rate, deposit A reserve ratio and purchasing managers index of macro policy index information changes, the international oil prices, the price of gold in London and the bad performance of NASDAQ and other international market information, between the information disclosure of listed companies increased and two trading days is not continuous on internal dynamic increased effect. At the same time, the overnight information can be affected by the stock market performance and stock price volatility of the overnight jump behavior, which plays an important role in the prediction of stock market intraday volatility. On the one hand, overnight is all kinds of overnight information affect the intermediary variables of stock market volatility, and jump in some overnight information influence on the daily fluctuations in showing the mediating effect to a certain extent. Secondly based on the model, from the stock market volatility overnight information, this paper proposed the multi factors - varying coefficient model, HAR-CJI model and BP neural network based complex In the model, fitting effect and prediction ability, than the classical wave model and neural network non parametric model performs better. Finally, the forecasting ability of the new model from the point of view, considering the overnight information improve the model in prediction accuracy and value direction of change on the stock market volatility, and improve the stability of non parameter model. The improvement of prediction direction is mainly reflected in the accuracy of volatility on the stock market. The positive changes in the overnight information of classical linear model and neural network model in theory explain the volatility of the stock market based on different prediction methods and prediction results.

【学位授予单位】：山东大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：F832.51

【相似文献】