空间面板数据固定效应模型Bootstrap LM检验有效性研究

发布时间：2018-03-18 15:30

本文选题：Bootstrap抽样　切入点：空间相关性LM检验　出处：《华南理工大学》2016年硕士论文　论文类型：学位论文

【摘要】：空间经济计量分析中对具体模型空间相关性识别一般采用以误差项服从正态独立同分布和大样本为基础的渐近LM检验。然而,在实证研究中,通常样本量有限或者误差项不满足渐近LM检验假设条件,有可能出现空间相关性渐近LM检验存在偏差。目前,在误差项不满足正态分布或者小样本下截面数据空间计量模型的空间相关性渐近LM检验难题已经解决,而面板数据空间计量模型的渐近LM检验尚未解决。本文探究有效的Bootstrap方法,并构建Bootstrap LM检验统计量,通过Monte Carlo模拟实验,从水平扭曲和功效两个角度证明Bootstrap LM检验有效,为解决渐近LM检验难题提供一条有效途径,为从事实证研究的空间经济计量分析工作者提供有效的分析工具。本文分为三大部分:第一部分为绪论和文献综述,在阐述研究背景以及梳理相关文献基础上,明确研究目的、研究内容和研究的技术路线;第二部分为固定效应模型空间相关性检验Bootstrap LM检验研究思路及实验方案的设计;第三部分是对Bootstrap LM检验Monte Carlo模拟结果进行分析并得出研究结论。通过Monte Carlo模拟实验表明,当研究模型的误差项服从空间相关检验LM统计量渐近分布假设时,渐近LM检验和Bootstrap LM检验具备良好的有限样本性质,均为有效的统计量。当误差项存在异方差或者时间序列相关时,渐近LM检验存在严重的水平扭曲,致使渐近LM检验失效。Bootstrap LM在不损失渐近LM检验的功效下,能够有效地校正其的水平扭曲,是为更理想的检验方法。简言之,无论误差项是否满足正态独立同分布,Bootstrap LM检验较渐近LM检验是为更有效的检验统计量。
[Abstract]:Spatial econometric analysis to identify the specific model of spatial correlation is generally used to test the asymptotic LM error obeys normal distribution and independent sample basis. However, in the empirical study, usually limited sample size or the error does not satisfy asymptotic LM test assumptions, there may be spatial correlation asymptotic LM test deviation at present, the error does not meet the normal distribution or cross-sectional data of spatial econometric model under small sample asymptotic spatial correlation problems of LM testing has been solved, and the asymptotic LM test the spatial panel data model has not been solved. This paper explore Bootstrap effective method, and construct the Bootstrap LM test statistic by Monte Carlo simulation, from the two point level distortion and efficacy that Bootstrap LM test effectively, provide an effective way to solve the problem of asymptotic LM test, in fact Spatial econometric analysis of research workers to provide an effective analytical tool. This paper is divided into three parts: the first part is the introduction and literature review, in this research background and review of relevant literature on the basis of a clear research purpose, research contents and technical route; the second part is the design of fixed effect model spatial correlation test Bootstrap LM test research and experimental plan; the third part is on the Bootstrap LM test Monte Carlo simulation results were analyzed and conclusions. Through Monte Carlo simulation experiments show that the error of the model when the spatial correlation test LM obey the asymptotic distribution assumption, asymptotic LM test and Bootstrap LM test has good finite sample properties are. For the effective statistics. When the error heteroscedasticity or time series, the asymptotic LM test there is a serious water Flat twist, resulting in failure of.Bootstrap LM in the asymptotic LM test without loss of asymptotic LM test function, can effectively correct the level of distortion, is an ideal method for inspection. In short, regardless of whether the error satisfies the normal i.i.d. LM test is asymptotically Bootstrap, LM test is test statistic for more effective.

【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F224

【相似文献】