稳健典型相关方法在金融分析中的应用比较研究  

Comparative Research of Robust Canonical Correlation Analysis in Financial Analysis

作　　者：何志锋 王斌会[2] HE Zhi-feng;WANG Bin-hui(School of Financial Mathematics and Statistics,Guangdong University of Finance,Guangzhou 510521,China;School of Management,Jinan University,Guangzhou 510632,China;School of Economics,Jinan University,Guangzhou 510632,China)

机构地区：[1]广东金融学院金融数学与统计学院,广东广州510521 [2]暨南大学管理学院,广东广州510632 [3]暨南大学经济学院,广东广州510632

出　　处：《数理统计与管理》2021年第5期901-913,共13页Journal of Applied Statistics and Management

基　　金：教育部人文社会科学研究一般项目(17YJC630090).

摘　　要：由于传统典型相关分析方法崩溃点为0,存在一个任意大的离群值将导致传统典型相关分析估计结果与实际不相符。在社会经济数据中分析数据往往存在一定的位移污染离群值,导致传统典型相关分析在不同范围内使用结果不稳定,典型变量缺乏经济意义等问题。针对这一现象,本文运用FAST-MCD法对总体协差阵进行稳徤估计构建出典型相关分析的稳健算法并进行模拟实验,同时对穗徤典型相关系数构造随机实验模拟检验临界值。最后利用2014年沪深300和创业板各股票财务指标和估值指标进行实证分析,从糢拟实验和实证研究上都说明了本文构建的方法具有稳健性。Because of the zero breakdown point of the traditional canonical correlation analysis(CCA),the estimation results of traditional CCA will not correspond to the fact even there is one outlier.In the social economic data,there are often some certain displacement pollution outliers,which will turn traditional CCA breakdown,as the estimation will be unstable and lack of economic significance.In view of this phenomenon,we improve the traditional CCA by robust covariance matrix base on FAST-MCD and describe the properties by simulation.Simultaneously,we simulate the testing critical value of robust canonical correlation coefficient.At last,we use financial indicators and valuation indicators of the stocks of CSI 300 index and GEM index to demonstrate difference between robust CCA and traditional CCA.All above illustrate that the method we proposed is robustness.

关 键 词：稳健典型相关分析 位移污染离群值 FAST-MCD 统计模拟 

分 类 号：O212[理学—概率论与数理统计]