多指标面板数据融合聚类分析

针对多指标面板数据的样品分类和历史时期划分问题,从多元统计分析理论角度提出一个多指标面板数据的融合聚类分析方法。该方法改进了多指标面板数据的因子分析和系统聚类方法,依据Fisher有序聚类理论,构造了Frobenius范数形式的离差平方和函数,提出了多指标面板数据的有序聚类方法。实证结果表明,该方法能够满足系统分析的统一性要求,保证指标之间的不相关;能够克服时间维度上均值处理造成的偏误,信息损失较少;能够解决面板数据有序聚类的问题;弥补了单一分析的片面性和局限性。

Fusion Clustering Analysis of Multivariate Panel Data

In the light of the sample classification and historical period division of multivariable panel data,it presents a fusion clustering method of multivariate panel data based on multivariate analysis theory.It put forwards a factor analysis and improves the hierarchical clustering method for multivariable panel data.Based on the Fisher ordinal clustering theory,it put forwards a kind of multivariate panel data ordinal clustering method by using Frobenius norm to construct square deviation function.The empirical research illustrates that this method is valid.Factor analysis could reach the requirement of unity.Hierarchical clustering could avoid the information loss caused by taking the mean value on the time dimension.Ordinal clustering could overcome the restriction of single indicator.In a word,the fusion clustering makes up the limitation of the above each method.