空间计量模型变量选择方法及其应用

发布时间：2018-03-09 23:27

本文选题：变量选择　切入点：空间计量　出处：《上海师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：空间计量经济学作为分析空间经济数据的主流方法,经过三十年的发展,目前已成为计量经济分析重要的组成部分;变量选择方法一直都是计量经济学研究的热点问题,上述两者都沿着各自的轨迹发展,很少出现交叉。然而在信息技术高速发展的今天,我们能收集到的时空数据越来越丰富,但由于空间相关性的引入,使得Gauss-Markov假设不再成立,基于经典线性模型构建的变量选择不再适用,因而在进行空间计量建模时,如何选择解释变量,构造最优的模型是一个亟待解决的问题。本文将变量选择的方法引入空间计量模型,提出了一系列空间计量模型的变量选择方法,并对其有效性进行了论证,还将理论研究的结果应用于影响股票收益率的财务因素变量选择研究。在理论研究部分,本文首先基于空间自回归模型(SAR模型),在模型残差服从正态分布的条件下,分别利用K-L信息量和贝叶斯方法,将经典线性模型的AIC准则和BIC准则推广到空间模型,提出并论证了基于SAR模型的空间AIC准则(SAIC准则)和空间BIC准则(SBIC准则),并证明了在一定条件下,基于SAR模型的SAIC准则和SBIC准则在其变量选择中具有一致性。其次,本文将上述方法进一步推广到更为一般化的空间计量经济模型,空间自相关误差自相关模型(SARAR模型),也证明了在一定条件下,基于SARAR模型的SAIC准则和SBIC准则在其变量选择中具有一致性。再次,本文放松对残差的假设,在仅仅假设残差独立同分布的基础上,基于SARAR模型构造广义空间信息准则(SGIC准则),将SAIC准则和SBIC准则纳入统一的分析框架,通过SGIC准则大样本性质的不同,将空间信息准则分为空间AIC类准则和空间BIC类准则。在理论推导的同时,本文还设计计算机仿真实验,利用Monte Carlo模拟对空间模型变量选择的有限样本性质进行研究,研究发现:针对于空间数据,相较于经典线性模型模型的变量选择方法,本文提出的方法在空间模型的变量选择中更加有效。在实证应用部分,本文将理论研究的得到空间计量模型变量选择方法应用于影响股票收益率的财务指标变量选择研究。本文首先分析股票市场空间效应来源的基础上,构造板块-金融空间权重矩阵;其次,本文利用Moran’s I检验和不含解释变量的空间自回归模型对股票收益率的空间效应进行测算,发现我国股票收益率具有显著的空间效应;再次,本文利用理论研究部分得到的方法对影响股票收益率的财务指标进行变量选择,从选择结果中,可以看出反映公司盈利能力和发展能力的财务指标对股票收益率的影响最大,反映公司偿债能力的财务指标次之,反映公司运营能力的财务指标最小;最后,本文进行稳健性检验,发现本文得到的结论是稳健的。
[Abstract]:Spatial econometrics, as the mainstream method of analyzing spatial economic data, has become an important part of econometric analysis after 30 years of development, and variable selection method has always been a hot topic in econometrics research. However, with the rapid development of information technology, we can collect more and more space-time data, but because of the introduction of spatial correlation, the Gauss-Markov hypothesis is no longer true. The variable selection based on the classical linear model is no longer applicable, so how to choose the explanatory variable in the spatial metrological modeling, It is an urgent problem to construct the optimal model. In this paper, the method of variable selection is introduced into the spatial metrology model, and a series of variable selection methods of spatial metrology model are put forward, and its validity is proved. The results of the theoretical study are also applied to the selection of financial factors that affect the stock returns. In the theoretical research part, firstly, based on the spatial autoregressive model and SAR model, under the condition that the residual error of the model is normal distribution, By using K-L information and Bayesian method, the AIC criterion and BIC criterion of classical linear model are extended to spatial model, respectively. In this paper, the spatial AIC criterion based on SAR model and the spatial BIC criterion are proposed and proved. It is proved that the SAIC criterion and SBIC criterion based on SAR model are consistent in their variable selection under certain conditions. In this paper, the above method is further extended to the more general spatial econometric model, the spatial autocorrelation error autocorrelation model and the SARAR model are also proved under certain conditions. The SAIC criterion and SBIC criterion based on SARAR model are consistent in the selection of variables. Thirdly, the assumption of residual error is relaxed in this paper. Based on the SARAR model, the generalized spatial information criterion is constructed. The SAIC criterion and the SBIC criterion are brought into the unified analysis framework, and the large sample properties of the SGIC criterion are different. The spatial information criterion is divided into spatial AIC class criterion and spatial BIC class criterion. At the same time, the computer simulation experiment is designed, and the finite sample property of spatial model variable selection is studied by Monte Carlo simulation. It is found that the method proposed in this paper is more effective in the selection of spatial models than in the classical linear models. In the part of empirical application, it is found that the proposed method is more effective in the selection of variables in spatial models than in the classical linear models. In this paper, the method of variable selection of spatial econometric model is applied to the selection of financial index variables that affect the stock return rate. Firstly, the source of spatial effect of stock market is analyzed in this paper. Secondly, this paper uses Moran's I test and spatial autoregressive model without explanatory variables to measure the spatial effect of stock yield. Thirdly, this paper makes use of the method of theoretical research to select the financial index which affects the stock return rate, and from the selection result, It can be seen that the financial indicators reflecting the profitability and development ability of the company have the greatest impact on the stock return rate, followed by the financial indicators reflecting the solvency of the company, and the financial indicators reflecting the operating ability of the company are the least; finally, In this paper, we test the robustness and find that the conclusion obtained in this paper is robust.
【学位授予单位】：上海师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F224;F832.51;F275

【参考文献】