作　　者： (蒋青嬗）; (韩兆洲）;

机构地区： 中山大学岭南学院,广东广州510275

出　　处： 《数理统计与管理》 2017年第5期821-832,共12页

摘　　要： 变量选择有助于简化模型,提高估计和预测的精度,但目前鲜有涉及面板半参数空间自回归模型变量选择的研究。本文在ALASSO的基础上提出了SSAR-ALASSO法,该法的核心在于惩罚函数的选择和目标函数的构建。SSAR-ALASSO在变量和参数的对应关系、惩罚函数的选择、特殊参数的取值区间以及适用模型等方面与ALASSO存在差异。模拟结果显示,SSAR-ALASSO法在变量选择的准确性和参数估计的精度两方面均表现良好,随着样本容量的增加表现效果更佳。本文在碳排放量影响因素实证中采用SSAR-ALASSO法对STIRPAT模型进行变量选择。研究结果表明人均财富、技术水平、产业结构、所有制结构和产业集聚显著影响碳排放量,城市化、对外开放、能源价格和环境政策对碳排放量无显著影响。 Variable selection can simplify models and improve the accuracy of estimation and prediction. Now there is little research on variable selection for semiparametric spatial autoregressive models. This article proposes a variable selection method called SSAR-ALASSO based on ALASSO. The key points of SSAR-ALASSO are the selection of penalty functions and the construction of objective function. SSAR- ALASSO and ALASSO are different on the corresponding relation of variables and parameters, penalty functions, the interval of special parameter and application. The numerical simulation shows that SSAR- ALASSO does well in variable selection and parameter estimation. With sample size increasing, the two aspects can do better. In the analysis for carbon emissions, we use SSAR-ALASSO to select STIRPAT model. The result shows that average wealth, technical level, industrial structure, structure of ownership and industrial agglomeration obviously affect carbon emissions, urbanization, opening-up, energy prices and environmental policy do not contribute significantly to carbon emissions.

关 键 词： 半参数空间自回归模型 变量选择 集群 惩罚 样条 模型