中文摘要：

随着计算机和互联网的快速发展,特别是在大数据时代,企业积累了大量有关企业经营、财务等相关数据,变量众多且关系纷繁复杂,如果利用传统的logistic回归建立企业信用风险预警模型往往效果不好。本文在充分考虑变量间的网络结构（Network）关系基础上,提出了网络结构Logistic模型,通过惩罚方法同时实现变量选择和参数估计。蒙特卡洛模拟表明网络结构Logistic模型要优于其他方法。最后,我们将其应用到我国企业信用风险预警中,充分考虑财务指标间的网络结构关系,科学地选择评估指标,构建更加适合我国国情的企业信用风险预警方法。

英文摘要：

With the rapid development of computer and the Internet,especially in the era of big data,some enterprises has accumulated a lot about their operation and finance data. Since the data is numerous and complicated,if we use the traditional logistic regression to build up the enterprise credit risk,the performance usually isn＇t good. In this paper,we propose network-logistic model based on considering the network relationship among variables,via penalized method to conduct variable selection and parameters estimation simultaneously. Simulation results show that network-logistic model performs better than other compared methods. Finally,we apply it to forecast enterprise＇s credit risk,under considering the network relationship between financial indicators,select significant variables and build up a suitable credit risk forecasting model for Chinese enterprises.