摘要
以中国上市公司为研究对象,以因财务状况异常而被特别处理作为上市公司陷入财务困境的标志,采用交叉验证技术建立各种统计和神经网络模型,并在独立的预测样本集上进行比较.实验结果表明统计和神经网络模型都能有效地进行财务困境预测,对于提前两年预测,统计模型优于神经网络模型,而对于提前三年预测,神经网络模型优于统计模型.实验结果也表明了在与训练集同一财务年度区间内的测试集上不能正确估计模型的性能.
This study has the Chinese listed finn as its research object. Using the special treatment （ST） received by listed firms due to their abnormal financial performance as the indicator of financial distress, this study firstly constructs statistic and neural network models by cross validation techniques. Then, with the independent prediction data set, the study compares the performances of statistic models with those of neural network models. It has been shown that the statistic models are superior to neural network models in accuracy for two years prior to special treatment, and viceversa for three years prior to special treatment. It has also been shown that the results of models＇ performances, which are got on the test set, are not reliable.
出处
《系统工程理论与实践》
EI
CSCD
北大核心
2005年第9期29-35,共7页
Systems Engineering-Theory & Practice
关键词
财务困境预测
交叉验证
统计模型
神经网络模型
financial distress prediction
cross validation
statistic model
neural network model
