基于误判代价加权的Logistic财务预警模型研究  

Prediction of Financial DistressUsing Weighted Logistics Regression with Misclassification Cost

作　　者：何胜美[1] 方茂扬[1] 王响[1] HE Sheng-mei FANG Mao-yang WANG Xiang(Zhaoqing campus Guangdong University of Finance, Zhaoqing, 526040, Chin)

机构地区：[1]广东金融学院肇庆校区,广东肇庆526040

出　　处：《经济数学》2017年第2期37-42,共6页Journal of Quantitative Economics

基　　金：广东金融学院创新强校工程金融数据挖掘和量化投资创新团队项目(20170305)

摘　　要：1:1样本配比的财务预警模型的系数和概率估计是有偏的,全市场公司的样本数据又高度不平衡.为克服两类样本不平衡给预警模型带来的影响,引入公司误判代价分析,以ST公司误判代价为权重,通过最小化加权的对数似然损失函数,建立误判代价加权的Logistic回归财务预警模型.实证结果表明,误判代价加权的Logistic回归模型具有较好的预警效果,2007年的训练样本上正常公司和ST公司的识别率为89.43%和93.33%,2008年测试样本上两类公司的识别率分别为:92.1%和95.83%.The estimated coefficients and probability are biased in Prediction of Financial Distress with traditional 1 ： 1 sample ratio,and the sample data based on the whole market is highly imbalanced.So in order to overcome the influence of imbalance, the misclassification cost of two kinds of companies was analyzed.Taking misclassification cost of ST Company as the weight, minimizing the weighted log likelihood loss function, a weighted Logistic regression model was used in Prediction of Financial Distress.The empirical results show that the weighted Logistic regression model has perfect effect.The recognition rate between normal financial company and ST company on training data in 2007 year was 89.43 % and 93.33 %, respectively; while the recognition rate of the two types of company is 92.1% and 95.83%, respectively, in the independent test sample in 2008 year.

关 键 词：数理经济学 财务预警模型 加权Logistic回归 不平衡数据 

分 类 号：F061.5[经济管理—政治经济学]