条件异方差时变自回归模型参数的贝叶斯估计

发布时间：2018-03-30 18:04

本文选题：条件异方差　切入点：时变自回归模型　出处：《吉林大学》2017年硕士论文

【摘要】：本文主要研究了条件异方差时变自回归模型参数的贝叶斯估计,在时变自回归模型参数的贝叶斯估计研究基础上,扰动项的方差服从ARCH(1)模型,对模型的参数不施加服从随机游走过程之类的限制,而是假设不同时刻其系数的先后状态具有一定的相关性,并且时间间隔距离越短它的相关性表现越强烈.由于时间的不可重复性,通常我们只能获得单个样本条,在这种情形下采用贝叶斯估计方法对参数进行估计.由贝叶斯公式得到参数的后验分布,用后验均值估计参数;对于复杂的后验分布要计算满条件分布,如果满条件分布不是已知分布,则不能对其直接抽样,应用逐分量的Metropolis-Hastings抽样方法进行抽样,用样本均值估计参数,从而得到参数的贝叶斯估计.本文通过分析一个简单的条件异方差时变模型,对其进行数值模拟来展示这个模型系数的估计效果.对我国GDP增长率的实例分析表明通过贝叶斯估计方法得到模型参数的估计值,能够很好地揭示实际问题变化过程中存在的内在规律.
[Abstract]:Variable estimation of Bias parameters of autoregressive model this paper mainly studies the conditional heteroskedasticity, the time-varying autoregressive model of Bias parameter estimation based on perturbation variance with ARCH (1) model, the parameters of the model are applied to random walk process such restrictions, but the assumption that different moment coefficient has the state has a certain correlation, and the time interval the shorter the distance between the more intense. Because of the time can not be repeated, usually we can only obtain a single sample, the estimation method of the parameters are estimated by Bias in this case. Parameters are obtained by the Bias formula of the posterior distribution, posterior mean parameter estimation; for complex posterior distribution to calculate the full conditional distribution, if the full conditional distribution is not known for its distribution, not by direct sampling, component Metropolis-Has Tings sampling method, parameter estimation using the sample mean, in order to get the Bayesian estimation of parameters. Through the analysis of a simple conditional heteroskedasticity time-varying model of the numerical simulation to show the effect of the estimated model coefficients. Case analysis on China's GDP growth rate indicates that the Bayesian estimation method to estimate model the value of the parameter, can well reveal the inherent law of change process in practical problems.

【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F124;F224

【参考文献】