统计研究

• 论文 • 上一篇    下一篇

广义嵌套空间模型变量选择研究——基于广义空间信息准则

张元庆 陶志鹏   

  • 出版日期:2017-09-15 发布日期:2017-09-20

Variable Selection for General Nesting Spatial Model----Based on the GSIC

Zhang Yuanqing & Tao Zhipeng   

  • Online:2017-09-15 Published:2017-09-20

摘要: 广义嵌套空间模型是一般形式的空间计量经济模型,本文提出了广义空间信息准则以解决该模型的变量选择问题。依据大样本性质的不同,将该准则分为两类——空间AIC类准则和空间BIC类准则。研究发现:空间AIC类准则能有效解决空间模型中变量的错选和漏选问题,但存在多选变量的倾向;而空间BIC类准则都能同时解决空间模型中变量的错选、漏选和多选问题,而且在特殊条件下能更有效解决错选和漏选问题,但往往需要更大的样本容量。Monte Carlo模拟结果印证了上述相关结论。最后,本文以城市对外资银行的吸引力为例,在给出测度指标的基础上,验证其空间相关性,并利用本文提出的方法对其影响因素进行变量选择。

关键词: 空间计量, 变量选择, 大样本性质, 地理区位

Abstract: General nesting spatial(GNS) model is the general form of spatial econometrics model. This paper studies variable selection problems of the model by raising the general spatial information criterion (GSIC). According to the asymptotic behavior, GSIC can be classified into two classes——spatial AIC-like and spatial BIC-like. It finds that spatial AIC-like criterions can solve the problem of selecting wrong variables or omitting variables, but trend to overfit, while spatial BIC-like criterions can solve all problem simultaneously and even do better in dealing with the problem of selecting wrong variables or omitting variables, but it usually need larger sample size. The Monte Carlo simulation also shows the above conclusion. In the end of the paper, it uses the ability of the city to attract foreign banks as the example. By giving measurement index, it demonstrates the spatial correlation, and uses the above methods to select factors of the above index.

Key words: Spatial Econometrics, Variable Selection, Large Sample Properties, Geographical