基于贝叶斯PHMM模型的油价与股市关系变点诊断研究

发布时间：2018-02-25 07:20

本文关键词： 面板数据贝叶斯分析隐马尔科夫模型变点诊断时变异质性　出处：《湖南大学》2014年硕士论文　论文类型：学位论文

【摘要】：变化是金融市场永恒的主题,政策变化、金融危机、制度变迁等事件性冲击常对金融市场产生非常大的影响,可能引起金融变量关系发生结构性变化。在金融经济研究中,如果不考虑这些事件变点的冲击对金融变量关系的影响,就容易使决策产生偏差,引起对金融变量关系进行错误评估。大T面板数据背景下,平稳数据过程并不能保证变量关系不存在结构变化,而且截面个体之间不可观测异质性非时变假设难以成立,利用传统面板数据模型对金融变量关系变点进行研究难以解决上述问题,因此针对面板数据截面个体之间不可观测异质性非时变假设引起的删失变量偏差以及推断无效问题,本文通过引入隐马尔科夫方法对固定效应和随机效应面板数据模型进行时变异质性以及变量关系结构变化进行建模,利用贝叶斯理论对模型参数进行统计推断、设计MCMC抽样算法、定义贝叶斯因子解决模型选择(即变点诊断)问题,构建了贝叶斯固定效应和随机效应面板数据隐马尔科夫模型,在刻画时变异质性的同时,捕捉金融系统环境中可能存在的变点,考察其对金融变量关系的非系统性及系统性影响,并以油价与股市关系变点为研究对象,选取国际原油价格指数、13个行业股指数据以及中国利率数据,考察了金融系统环境变化对油价与股市关系的非系统性及系统性影响。研究结果表明：国际原油价格与股市之间存在长期稳定关系,但这种长期稳定关系会受到金融系统环境的冲击,其中,13个行业受系统环境影响可分为非系统性以及系统性影响,每个行业的个体变点均在3-5个,但共同变点只有3个；而且大T面板数据背景下,如果考虑行业间不可观测异质性的时变性,可以发现,系统环境中的不同事件将使油价与股市关系在不同时间段中表现为不同的相关关系,验证了贝叶斯固定效应和随机效应面板数据隐马尔科夫模型的有效性。
[Abstract]:Change is the eternal theme of the financial market policy change, financial crisis, institutional change impact events often have a very big impact on financial markets, financial variables may cause structural changes in the financial economy. In the study, if you do not consider these events change point of the impact on the financial variables, it is easy to make the deviation caused by the wrong decision, evaluation of financial variables. T panel data context, stable data process does not guarantee the variable relationship does not exist between the individual and the change of structure, section of unobservable heterogeneity while the hypothesis is not tenable, using traditional panel data model is studied to solve the problem of the relationship between financial variables change point, so the panel data can not observe heterogeneity between individual non section variable by the assumption of the censored variable bias and invalid inference question In this paper, by introducing the hidden Markov method of qualitative variation and variable relationship between Microstructure Modeling of fixed and random effects panel data model, the statistical inference of the model parameters by using Bayesian theory, the design of MCMC sampling algorithm, the definition of Bayes factor to solve the model selection (i.e. the change point problem, constructs a Bayesian diagnosis) fixed and random effect panel data of hidden Markov model, to describe the variation of nature at the same time, there may be a change point to capture financial system environment, the system and the non systemic effects of financial variables, and the relationship between oil prices and the stock market changes as the research object, selects the international crude oil price index. 13 industry index data and China rate data, investigated and non systemic effects of financial system environment changes on the relationship between oil and stocks.
The results show that there is a long-term stable relationship between international oil price and the stock market, but the long-term stable relationship will be affected by the environmental impact of the financial system, among them, 13 industries affected by the environmental impact system can be divided into systematic and non systemic effects, each individual industry change point in all 3-5, but only the common change point 3; and T panel data context, if not considering the time-varying unobserved heterogeneity can be between industries can be found in different event system environment will make the relationship between oil prices and the stock market performance in different periods for different relationship, verify the validity of Bayesian fixed effect and random effect panel data the hidden Markov model.

【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：F832.51;F764.1;F224

【参考文献】