空间权重选择的模拟研究——基于贝叶斯方法  

Simulation Study on Selection of Spatial Weight—Based on Bayesian Method

作　　者：范真 王西贝 Fan Zhen;Wang Xibei(Institute of International Economy,University of International Business and Economics,Beijing 100029,China;The People’s Bank of China Beijing Branch,Beijing 100045,China)

机构地区：[1]对外经济贸易大学国际经济研究院,北京100029 [2]中国人民银行北京市分行,北京100045

出　　处：《统计与决策》2024年第10期52-57,共6页Statistics & Decision

基　　金：国家社会科学基金资助项目(22BJY160);国家自然科学基金青年科学基金项目(72303031)。

摘　　要：文章在空间杜宾模型下讨论贝叶斯后验模型概率方法对空间权重矩阵的选择效果,并利用蒙特卡洛模拟验证了该方法的有效性。结果表明:贝叶斯后验模型概率方法对权重判别的准确性较高,且随样本量和时间的增加而提高,正确判别率与空间相关系数的关系呈现“U”型特征;在对具有较强相关性的权重选择中,其判别能力下降,但效果仍优于ML法;在对有不同属性的静态空间权重的判别中,贝叶斯后验模型概率可快速达到100%;对于动态和静态空间权重的选择问题,该方法在存在较大正向空间相关性的情况下具有良好功效。This paper discusses the effect of Bayesian posterior model probability method on the selection of spatial weight matrix under the spatial Durbin model,and then uses Monte Carlo simulation to verify the effectiveness of the method.The results go as the following:The Bayesian posterior model probability method has high accuracy in weight discrimination,and it improves with the increase of sample size and time,and the relationship between the correct discrimination rate and spatial correlation coef-ficient is U-shaped.In the weight selection with strong correlation,the discrimination ability of the method is decreased,but the effect is still better than ML method.In the static spatial weight discrimination of different attributes,the Bayesian posterior model probability can quickly reach 100%.For the selection of dynamic and static spatial weight,the method has a good performance in the case of large and positive spatial correlation.

关 键 词：贝叶斯后验模型概率 蒙特卡洛模拟 空间权重矩阵 

分 类 号：O212.8[理学—概率论与数理统计]