财务困境预测中的变量筛选——基于平均影响值的SVM方法  

Variable Selection of Financial Distress Prediction——The SVM Method Based on Mean Impact Value

作　　者：卢永艳[1] 王维国[2] 

机构地区：[1]东北财经大学管理科学与工程学院,辽宁大连116025 [2]东北财经大学数学与数量经济学院,辽宁大连116025

出　　处：《系统工程》2011年第8期73-78,共6页Systems Engineering

基　　金：教育部人文社会科学研究一般项目(10YJC790172)

摘　　要：在财务困境预测中,如何从大量备选指标中筛选出预警指标是一个重要环节。为了更有效地设计财务困境预测模型,本文将平均影响值方法应用于SVM回归来进行变量筛选,首先对训练集数据用SVM进行训练,然后分别增减每一自变量的10%来进行仿真,对两个仿真结果的差值按样本数平均,得出平均影响值;最后对各个自变量的平均影响值按绝对值大小排序,从而进行变量筛选。实证结果表明,该方法能够以较少的特征变量实现较高的分类精度,是切实有效的。How to select early warning variables from alternative indexes is an important part during the prediction of financial distress.In order to design a more effective model,the method of Mean Impact Value（MIV） is applied to SVM regression for variable selection.Firstly,the training set is trained with SVM model.Then every explanatory variable increased and decreased 10 percent respectively,two new training sets are obtained,which are used to simulate.The mean of the difference of the two simulation results is obtained by the number of samples.Finally,the variables are sorted based on their absolute values of Moves variables for variable selection.The empirical results show that the MIV method can give a higher classification accuracy by a fewer characteristics variables, so it is effective.

关 键 词：财务困境预测 变量筛选 平均影响值 SVM 

分 类 号：F272[经济管理—企业管理]