CAJ | 학술논문

上市公司财务预警模型受到不同配对比例的下采样影响较大，2007—2008年上市公司财务数据的分析结果表明：配对比例过高，ST公司的识别率太低；配对比例过低，模型识别结果变异太大，结果不可靠；而现代统计学中针对不平衡数据的统计方法SMOTO方法和Bagging算法均能较好地克服样本比例不均衡的影响，上述数据的实证研究结果显示：基于上述两种方法的财务预警模型在测试集上对正常公司和ST公司都取得了较好的稳定识别率。
상시공사재무예경모형수도불동배대비례적하채양영향교대，2007—2008년상시공사재무수거적분석결과표명：배대비례과고，ST공사적식별솔태저；배대비례과저，모형식별결과변이태대，결과불가고；이현대통계학중침대불평형수거적통계방법SMOTO방법화Bagging산법균능교호지극복양본비례불균형적영향，상술수거적실증연구결과현시：기우상술량충방법적재무예경모형재측시집상대정상공사화ST공사도취득료교호적은정식별솔。
Financial distress early-warning models chosen by listed companies are affected significantly by different matching ratios. Through the analysis of the effects of the corporation financial data collected be-tween 2007 and 2008,the study finds that with higher matching ratios come lower identification rates among ST companies,while lower matching ratios seem to lead to greater variations in model identification and thus bring about unreliable results. In view of imbalanced data set,the SMOTO and Bagging algorithm methods are often applied in modern statistics aiming to minimize the effects of imbalanced sample proportion. The results of the above-mentioned empirical study show that the early-warning models based on the two meth-ods in the dataset test have achieved a steady recognition rate in normal and ST corporations respectively.