空间计量模型的选择、估计及其应用

发布时间：2018-04-23 06:55

本文选题：空间计量模型 + 空间权重矩阵　；参考：《江西财经大学》2015年博士论文

【摘要】：空间计量模型的研究得到越来越被广泛的重视。在现实世界中,观测值实际上存在独立和非独立两种可能,传统的计量理论是建立在独立观测值假定基础之上的,而地理区域空间之间及其经济现象之间的空间依赖性的存在打破了经典计量经济学模型关于样本相互独立的基本假设。这时要准确的提取这些数据的空间关系,恰当的描述和运用空间特性对空间交互作用进行研究尤为必要。事实上,空间单元上的某种经济现象或某一属性总是与相邻空间单元上的同现象或属性相关,我们必须通过空间计量模型来分析这种关系。将空间效应纳入计量经济模型分析框架,将会面临着三个重要问题。一是如何合理地将空间效应引入既有的计量经济模型,或者根据空间效应的特殊性构造特定的计量经济模型；二是如何在具体的分析中恰当地选择空间计量模型以及特定的空间计量经济模型如何估计；三是如何利用空间计量模型进行规范的实证分析。第一个问题涉及到空间权重矩阵的合理设定。第二个问题涉及到空间计量模型的选择方法与估计方法。第三个问题实际上是前两个问题的解决方法与实证分析的结合。基于上面三个问题,本文从空间计量最为重要的研究对象——空间权重矩阵出发,将经典计量方法和MCMC方法相结合,分析了空间计量模型的选择与带未知异方差的广义空间模型的有效估计方法,并给出了实证分析上的应用。本文的研究结论如下：(1)空间权重矩阵是连接理论分析上的空间计量经济模型与真实世界中空间效应的纽带。能否构建并选择恰当的空间权重矩阵直接关系到模型的最终估计结果和解释力。不同的空间权重矩阵反映的是研究对象背后不同的经济学原理与视角,同时也对应着研究者对于空间效应的不同认识。空间权重矩阵的错误选择将会严重地干扰空间计量的各种检验分析,从而对空间计量模型的进一步研究产生很大影响。(2)空间计量模型选择是空间计量分析的重要研究课题。Moran指数检验、LM检验、似然函数、三大信息准则、贝叶斯后验概率、马尔可夫链蒙特卡罗方法对空间计量模型选择有很大的差异。通过模拟分析表明：在扩充的空间计量模型簇中进行模型选择时,基于OLS残差的Moran指数与LM检验均存在较大的局限性,对数似然值最大原则缺少区分度,LM检验只针对SEM和SAR模型的区分有效,信息准则对大多数模型有效,但是也会出现误选。而当给出恰当的M-H算法时,充分利用了似然函数和先验信息的MCMC方法,具有更高的检验效度,特别是在较大的样本条件下得到了较准确的判断。并且它对不同阶空间邻接矩阵的空间计量模型的选择也非常有效。(3)空间异方差问题也是空间计量分析中的一个重要问题。空间单元大小以及其它的经济特征上的差异,常常会导致空间异方差问题。对于广义空间模型包含异方差时,估计方法相对复杂,本文给出了三种不同的估计方法。第一种方法是将异方差形式参数化,来克服自由度的不足,使用ML估计进行实现。而针对异方差形式未知时,分别采用了基于2SLS的迭代GMM估计和更加直接的MCMC抽样方法加以解决,特别是MCMC方法表现得更加优美。蒙特卡罗模拟表明,给定异方差形式条件下,ML估计通过异方差参数化的方法依然可以获得较好的估计效果。而异方差形式未知的情况下,另外两种方法随着样本数的增大时也可以与ML的估计结果趋于一致。(4)结合空间权重矩阵的分析、空间计量模型的选择、带未知异方差的广义空间计量模型的估计和方向性距离函数GML超效率模型,对资源环境约束下我国省际全要素能源效率问题进行研究。从实证的过程得出如下结论：在做全要素能源效率的测度时应该考虑资源与环境的约束,这样得出的结果才能更加符合我国的实际情况。进而,在做省际全要素能源效率的影响因素分析时,应当考虑空间效应的影响,忽略空间效应的影响将会得出有偏误的估计。特别是在考虑空间效应时,还需要根据研究的空间计量理论与方法选择恰当的空间权重矩阵和合适的空间计量模型,以及使用合理的模型估计方法。只有把这些过程合理地整合在一起,才能构成全要素能源效率分析的一个完整框架体系。从实证分析的结果得出如下结论：资源环境约束下我国省际全要素能源效率持续走低,趋势不容乐观；资源和环境约束的条件下过多地依赖煤炭资源将会大大降低我国的能源效率,煤炭消费所带来的负面影响确实不容忽视；“污染天堂假说”在我国是成立的；服务业的比重增加是有利于能源效率总体上的提高；外资企业相对国内来说会采用更加先进的能源技术,且对国内企业存在正向溢出效应,对我国的能源效率存在正面影响。本文的整个研究具有一定的理论价值和实践价值。首先,本文首次比较系统地研究了空间权重矩阵,且通过图形作了一定程度的可视化分析,完善了空间权重矩阵的体系化认识。在空间计量模型选择方法的分析中给出了很少被使用而又很有效的MCMC方法。在空间计量模型的估计中,又研究了一种较为常见的空间计量模型——广义空间计量模型,且在考虑带未知异方差条件下给出了它的MCMC有效估计方法。最后都对抽象的理论分析都进行了蒙特卡罗模拟,给出了方法的有效性对比。这些都对空间计量经济的实证分析提供了重要的参考。正是基于此,本文最后结合理论分析进行了实证研究,整个实证过程做到了规范严谨。
[Abstract]:In the real world, there are two possibilities of independent and non independent observational values in the real world. The traditional econometric theory is based on the assumption of independent observational values, and the existence of spatial dependence between geographical space and its economic image has broken the classic. The basic assumption that the econometric model is independent of each other is necessary. It is necessary to accurately extract the spatial relations of these data, to describe and apply the spatial characteristics properly to study the spatial interaction. In fact, some economic phenomena or some attributes on the space unit are always the same as those on the adjacent space units. We have to analyze this relationship through the spatial econometric model. It will be faced with three important problems to incorporate the spatial effect into the econometric model analysis framework. One is how to rationally introduce the spatial effect into the existing econometric model, or to construct a specific econometric model according to the particularity of the spatial effect. The two is how to choose the spatial econometric model properly and how to estimate the specific spatial econometric model in the specific analysis. Three is how to use the spatial econometric model to carry out the normative empirical analysis. The first problem involves the rational setting of the spatial weight matrix. The second questions involve the selection of the spatial econometric model. In fact, the third problem is the combination of the solution method and the empirical analysis of the first two problems. Based on the above three problems, this paper, based on the space weighting matrix, combines the classical measurement method and the MCMC method, and analyzes the selection and unknowns of the spatial econometric model. The effective estimation method of the generalized spatial model of heteroscedasticity is given and the application of the empirical analysis is given. The conclusions of this paper are as follows: (1) the spatial weight matrix is the link between the spatial econometric model and the real world space effect in the theoretical analysis. The different spatial weight matrix reflects the different economic principles and perspectives behind the research object, and also corresponds to the different understanding of the spatial effect. The error selection of the spatial weight matrix will seriously interfere with the various tests and analyses of the space measure, thus to the space measurement. The further research of the model has great influence. (2) the selection of spatial econometric model is an important research topic of spatial econometric analysis:.Moran index test, LM test, likelihood function, three information criterion, Bayesian posteriori probability, Markov chain Monte Carlo method, which has great difference in the selection of spatial calculation model. The Moran index and LM test based on the OLS residuals have great limitations in the selection of the extended spatial econometric model clusters. The maximum principle of the logarithmic likelihood is lacking, and the LM test is only effective for the SEM and SAR models. When the M-H algorithm takes full advantage of the MCMC method of likelihood function and prior information, it has a higher test validity, especially in the larger sample condition. And it is also very effective for the selection of spatial econometric models of different order space adjacency matrices. (3) spatial heteroscedasticity is also a spatial econometric problem. An important problem in the analysis is that the space element size and the difference in other economic characteristics often lead to the spatial heteroscedasticity problem. For the generalized spatial model including the heteroscedasticity, the estimation method is relatively complex. In this paper, three different estimation methods are given. The first method is to parameterize the form of Heteroscedasticity to overcome the degree of freedom. The ML estimation is used to implement it. When the heteroscedasticity is unknown, the iterative GMM estimation based on 2SLS and the more direct MCMC sampling are used respectively, especially the MCMC method is more graceful. The Monte Carlo simulation shows that the ML estimation is parameterized by the heteroscedasticity under the condition of the given heteroscedasticity. Better estimation results can still be obtained. In the case of unknown variance, the other two methods can also be consistent with the estimated results of ML as the number of samples increase. (4) combining the analysis of the spatial weight matrix, the selection of the spatial econometric model, the estimation of the generalized spatial econometric model with the unknown ISO difference and the direction distance. The GML super efficiency model is used to study the energy efficiency of China's inter provincial total factor under the constraints of resource and environment. The following conclusions are drawn from the empirical process: the constraints of resources and environment should be considered when measuring the total factor energy efficiency, so that the results can be more consistent with the actual situation in China. In the analysis of factors affecting all factor energy efficiency, the effect of space effect should be considered. Ignoring the effect of space effect, there will be an error estimation. Especially when considering the spatial effect, the appropriate spatial weight matrix and suitable spatial measurement model should be selected according to the spatial measurement theory and method. The rational model estimation method. Only by integrating these processes properly can we form a complete framework for the analysis of all factors energy efficiency. From the results of empirical analysis, the following conclusions are drawn: under the constraints of resource and environment, the energy efficiency of China's inter provincial total factor is low, the trend is not optimistic; resources and environmental constraints are restricted. The excessive dependence of coal resources under the conditions will greatly reduce the energy efficiency of our country. The negative effects of coal consumption can not be ignored; the hypothesis of "pollution paradise" is established in China; the increase in the proportion of service industry is beneficial to the overall improvement of energy efficiency; foreign enterprises will adopt more first than domestic. The forward energy technology has positive spillover effect on domestic enterprises and has a positive impact on energy efficiency in China. The whole study of this paper has some theoretical and practical value. First, this paper systematically studies the spatial weight matrix for the first time, and has made a certain degree of visualization analysis through the graphics, and improved the space. The systematic understanding of the weight matrix. In the analysis of the selection method of the spatial econometric model, the MCMC method which is rarely used but very effective is given. In the estimation of the spatial econometric model, a more common spatial econometric model, the generalized spatial econometric model, is also studied, and it is given under the condition of considering the unknown heteroscedasticity. MCMC effective estimation method. Finally, the Monte Carlo simulation of abstract theoretical analysis is carried out, and the effectiveness of the method is compared. These all provide important reference for the empirical analysis of the spatial econometrics.

【学位授予单位】：江西财经大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：F224.0

【参考文献】