关于半参数神经网络GARCH模型族在中国股市的进一步研究

发布时间：2018-01-16 00:01

本文关键词：关于半参数神经网络GARCH模型族在中国股市的进一步研究　出处：《重庆大学》2014年硕士论文　论文类型：学位论文

【摘要】：在现代金融理论中,有两个金融市场的基本活动，它们分别是风险评估和资产定价，在对这两个基本活动的实践中我们都会遇到一个度量指标，也就是我们经常说的资产收益波动率。在投资组合的安排中，以及金融风险管理的理论和实践中，它是对资产收益不确定性的衡量，代表了资产的风险状况，因此能够准确的测量，乃至预测金融资产的波动率是金融市场研究的重要问题之一。在对于资产波动率的测量方法的研究中，国内外的专家学者，多年来已经产生了大量的文献著作，和实践结果，并取得了显著的研究成就。事实上，Engle于1982年首次提出的自回归条件异方差理论（即ARCH模型）就如同里程碑一样开启了异方差模型的大门，这一模型方法便被广而传之，随后的研究者们为了进一步刻画金融资产收益率的观测特性—包括：尖峰厚尾，聚集性，持续性等，，于是不断对ARCH模型进行拓展与改进，相继提出了GARCH、EGARCH、TGARCH、GJRGARCH、NAGARCH、APARCH等模型，其中最著名的莫过于Bollerslev在1986年提出的GARCH模型，即广义自回归条件异方差模型，它是对ARCH模型的一种最常见的推广形式，以至于我们将之后提出EGARCH等模型，统称为参数GARCH模型。可是，参数模型有其最大的弊端，也就是需要具体参数形式以及各种条件假设，因此也就造成了模型的局限性。研究者们为了突破这些假设限制，于是提出了另一类新的方法，即非参数GARCH模型。但是，研究者们同时也发现，非参数方法虽然避免了假设限制，却也存在其自身的诸多缺陷，例如，过度依赖大样本数据，维数灾，以及缺乏解释性等问题。因此在近几年的研究中，许多研究者尝试使用半参数GARCH模型来测量金融资产的波动率，这类模型不仅拥有非参数模型较好的拟合和预测能力，同时兼具传统参数模型的解释能力，可以较好测量金融资产的波动率变化。通常，半参数GARCH模型是由GARCH模型和另一种非参数方法杂交而成，例如：小波分析、局部多项式回归，Copula函数等，而本文对于中国股市的波动率研究正是基于神经网络算法和GARCH类模型族混合的半参数方法。本文在已有模型的基础上使用一种新的神经网络算法与GARCH模型结合，并总结和对比了这一类半参数GARCH模型族与传统参数GARCH模型族在中国市场上分析和预测能力的差别，且得到了较好的结果。
[Abstract]:In modern financial theory, there are two basic activities of financial market, they are risk assessment and asset pricing, and we will encounter a measurement index in the practice of these two basic activities. In the arrangement of investment portfolio, and in the theory and practice of financial risk management, it is a measure of uncertainty of asset return, representing the risk situation of assets. Therefore, accurate measurement and even prediction of volatility of financial assets is one of the important problems in financial market research. In the study of measurement methods of asset volatility, experts and scholars at home and abroad have produced a large number of literature, and practical results over the years, and have made significant research achievements. The autoregressive conditional heteroscedasticity theory (ARCH model), first proposed by Engle in 1982, opened the gate of heteroscedasticity model as a milestone, and this model method has been widely spread. In order to further depict the observational characteristics of the financial asset yield, including the peak and thick tail, aggregation, persistence and so on, the researchers continue to expand and improve the ARCH model. The models of GARCHN EGARCHN TGARCHN GJRGARCHN NAGARCHN APARCH and so on have been put forward one after another. The most famous one is the generalized autoregressive conditional heteroscedasticity model proposed by Bollerslev in 1986. It is the most common extension of the ARCH model, so we will later put forward the EGARCH and other models, collectively known as the parametric GARCH model. However, the parametric model has its biggest drawbacks. In other words, the specific parameter form and various conditional assumptions are needed, which leads to the limitation of the model. In order to break through these hypotheses, researchers have proposed another kind of new method. However, the researchers also found that although the non-parametric method avoids the hypothetical limitation, it also has its own defects, such as over-dependence on large sample data and dimensionality disaster. Therefore, in recent years, many researchers try to use semi-parametric GARCH model to measure the volatility of financial assets. This kind of model not only has the better ability of fitting and forecasting, but also has the interpretation ability of the traditional parametric model, which can measure the volatility of financial assets. Usually the semi-parametric GARCH model is a hybrid of GARCH model and another nonparametric method such as wavelet analysis local polynomial regression Copula function and so on. In this paper, the volatility of Chinese stock market is based on the hybrid semi-parametric method of neural network algorithm and GARCH class model family. This paper uses a new neural network algorithm and GA on the basis of existing models. RCH model combination. The differences between the semi-parametric GARCH model family and the traditional parametric GARCH model family in Chinese market are summarized and compared, and good results are obtained.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F832.51;TP183

【共引文献】