基于改进的自适应传播模型的农业风险区划分析  

Agriculture Risk Regionalization Analysis Based on Panel Data Clustering with Affinity Propagation

作　　者：谢远涛[1] 杨娟[2] 刘皓宇 XIE Yuan-tao YANG Juan LIU Hao-yu(School of Insurance and Economics, University of International Business and Economics, Beijing 100029, China Institute of Comprehensive Development, Chinese Academy of Science and Technology for Development, Beijing 100038, Chinas3. Deloitte China, Beijing 100738, China)

机构地区：[1]对外经济贸易大学保险学院,北京100029 [2]中国科学技术发展战略研究院,北京100038 [3]德勤中国,北京100738

出　　处：《统计与信息论坛》2017年第1期33-40,共8页Journal of Statistics and Information

基　　金：国家自然科学基金项目<风险信息共享背景下的个体风险评估研究>(71303045);对外经济贸易大学"优秀青年学者培育计划"(15YQ09);中国保险学会研究课题<生猪费率厘定系统分析:基于养猪风险管理因子与保险数据的研究>(15HX158);国家社会科学基金重大项目<巨灾保险的精算统计模型及其应用研究>(16ZDA052)

摘　　要：农业险定价中的核心问题是农业风险区划问题,为了体现农业区划中个体指标的动态发展特征,根据近邻传播改进自适应近邻传播聚类方法对数据进行优化,基于轮廓系数、归属度和吸引度得到最佳聚类中心和几何聚类中心,并将聚类转化为新数据集的聚类问题;选取代表性的棉花为例进行实证分析,通过计算生产、销售、收入、财政等指标进行棉花风险区划实例分析,计算最优棉花风险区划,结果表明对于具有动态特征的数据,本模型具有很好的有效性、实用性和解释性。Agriculture Risk Regionalization Analysis Based on Panel Data Clustering with Affinity Propagation XIE Yuan-tao1 , YANG Juan2, LIU Hao-yu3 （1. School of Insurance and Economics, University of International Business and Economics, Beijing 100029, China; 2. Institute of Comprehensive Development, Chinese Academy of Science and Technology for Development, Beijing 100038, Chinas3. Deloitte China, Beijing 100738, China） Abstract. Variables for individuals are developed with dynamic characteristics in many panel data sets when we deal with agriculture insurance pricing. In papers for panel data clustering, the similarity coefficients are computed by the numerical character, distribution character, and fluctuant character, but the clustering results cannot reflect the dynamic characteristics. This paper proposes the method to apply adaptive affinity propagation clustering （ad-AP）, which is improved from affinity propagation clustering, to optimize panel data set, and compute the best exemplars of each individual which constitute a new data set. Then panel data clustering analysis is transform into the new dataset clustering analysis. Experimental results on china agriculture insurance show the validity, practicability and interpretability of the design for panel data with dynamic characteristics. Key words：panel data clustering; affinity propagation （AP）; adaptive affinity propagation （ad-AP） ; cluster center

关 键 词：面板数据聚类 近邻传播 自适应近邻传播 聚类中心 

分 类 号：C81[社会学—统计学]