空间面板模型的识别、估计与应用研究

发布时间：2018-06-03 11:14

本文选题：空间面板模型 + 残差Bootstrap方法　；参考：《华中科技大学》2016年硕士论文

【摘要】：空间计量经济学是研究如何处理截面自相关问题并对其进行建模分析的理论,传统的计量分析假定截面单元独立同分布,这往往是不符合实际的。自Anselin(1988)的开创性文献以来,空间计量经济学迅速发展,已经成为了计量经济学领域的一个重要的分支学科,它也由单纯处理区域经济学的问题而扩展到了劳动经济学、教育经济学以及金融学等多个重要的领域。本文将以实证研究中广泛采用的空间面板模型为分析框架,在学习和理解基本的空间面板模型的设定与估计方法、空间自相关检验等问题的基础上,从空间面板模型的识别、应用、估计这三个角度各选择一个问题进行深入分析与探讨,力求完善现有的空间面板模型的分析框架,并尝试解决中国经济的现实问题。本文首先回顾了空间计量经济学的基本理论,介绍了利用空间模型分析问题的必要性、空间权重矩阵的建立、空间面板模型的设定,从截面空间模型开始介绍了空间模型估计的矩方法与极大似然方法,介绍了空间面板模型的估计方法及应注意的问题,简要梳理了空间相关性检验的步骤和应注意的问题。在识别部分,本文将Sargan(1964)提出的共同因子约束(COMFAC)检验引入了空间面板杜宾模型(SDM)的识别问题中,探讨了SDM模型与空间面板误差模型(SEM)模型的识别问题。通过仿真我们发现,基于渐近临界值的Wald检验虽然有着良好的检验功效,但却存在着较为严重的尺度扭曲。采用残差Bootstrap方法能够有效解决这一问题。在应用部分,本文详细收集并测算了中国分省域1997-2012年的碳排放数据,构建空间面板模型分析了产业结构、能源结构和技术因素对分省域人均碳排放的影响因素和空间效应,通过识别检验发现建立SEM模型是最优的。实证分析的结果充分表明,在碳减排过程中要加强区域合作,要特别重视改善能源消费结构。在估计部分,我们以在微观计量经济学中广泛存在的短面板数据结构为框架,分析了短动态面板SEM模型的估计问题。我们给出了针对这个模型的三步系统广义矩(GMM)估计方法,并与拟极大似然估计(QMLE)估计方法的有限样本性质进行了比较。通过仿真我们发现,两种估计方法在不同情形下的表现各有优劣,但在一般情形下,QMLE的有限样本表现更好。
[Abstract]:Spatial econometrics is a theory to study how to deal with cross-section autocorrelation and to model and analyze it. The traditional econometric analysis assumes that the cross-section units are distributed independently, which is often not in line with the reality. Since the pioneering literature of Anselin (1988), spatial econometrics has developed rapidly and has become an important branch of the econometrics field. It has also expanded from simply dealing with the problems of regional economics to labor economics. There are many important fields such as educational economics and finance. In this paper, the spatial panel model, which is widely used in the empirical research, is used as the analysis framework. Based on the study and understanding of the basic spatial panel model setting and estimation methods, spatial autocorrelation test and so on, the paper will identify the spatial panel model. It is estimated that each of the three angles should choose one problem for further analysis and discussion, and try to improve the existing analysis framework of spatial panel model and try to solve the real problems of China's economy. This paper first reviews the basic theory of spatial econometrics, introduces the necessity of using spatial model to analyze the problem, the establishment of spatial weight matrix, and the setting of spatial panel model. This paper introduces the moment method and maximum likelihood method of spatial model estimation, introduces the estimation method of spatial panel model and the problems that should be paid attention to, and briefly combs the steps of spatial correlation test and the problems that should be paid attention to. In the recognition part, the common factor constraint (COMFAC) test proposed by Sargan-1964 is introduced into the recognition problem of the spatial panel Dobbin model, and the recognition problem of the SDM model and the spatial panel error model is discussed. The simulation results show that although the Wald test based on asymptotic critical value has good performance, it has serious scale distortion. The residual Bootstrap method can effectively solve this problem. In the application part, the paper collects and calculates the carbon emission data in China from 1997 to 2012 in detail, and constructs a spatial panel model to analyze the influence factors and spatial effects of industrial structure, energy structure and technology factors on per capita carbon emissions. The identification test shows that the SEM model is optimal. The results of empirical analysis show that regional cooperation should be strengthened and energy consumption structure should be improved in the process of carbon reduction. In the estimation part, we analyze the estimation problem of the short dynamic panel SEM model based on the short panel data structure, which is widely used in microeconometrics. In this paper, we give the generalized moment GMMs estimation method for the three-step system, and compare the finite sample properties of the QMLE-based estimator with the quasi-maximum likelihood estimator. The simulation results show that the two estimation methods have their own advantages and disadvantages in different cases, but in general, the finite samples of QMLE are better.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F224

【参考文献】