摘要
根据中国资本市场的实际和样本数据特点,设计一套从样本准备到模型参数优化、再到模型比较的集成解决方案,对上市公司经营失败进行预警,通过实验分析参数调整和核函数选择对支持向量机建模的影响,寻求最优的支持向量机模型。实证结果表明,经营失败预警应用中,参数和核函数的选择对预警模型有较大影响,基于最优支持向量机模型的预测效果优于统计方法和神经网络方法,支持向量机适合中国上市公司分行业小样本的实际,特别处理事件作为经营失败样本切分标准对模型产生一定影响。
According to feature of Chinese capital market and samples, the paper designs an integrated scheme including sampling preparation, optimization of models and model comparison for business failure pre- diction of listed companies. The effects of parameter-adjusting and selection of kernel functions on model performance by simulation experiments is discussed. Then an optimized support vector machine model is build, Empirical results show that kernel function and parameters have effect on the performance of the support vector machine models. It is also suggested that the optimized support vector machine model for business failure pre- diction outperforms statistical methods and neural network, support vector machine is suitable as business prediction model for small samples of some industries of listed companies. Chinese “special treat” event as cut-off standard of business failure sample may have some effect on classification models.
出处
《管理科学》
CSSCI
2008年第1期115-120,F0003,共7页
Journal of Management Science
基金
教育部哲学社会科学研究重大课题攻关项目(07JZD0020)
教育部新世纪优秀人才支持计划(NCET-04-415)
上海市教育委员会科研创新项目(08ZS33)
关键词
经营失败预警
参数与核函数
最优支持向量机模型
人工智能
business failure prediction
model parameter and kernel function
the optimized support vector machine model
artificial intelligence