系统管理学报

系统管理学报 ›› 2022, Vol. 31 ›› Issue (2): 317-328.DOI: 10.3969/j.issn.1005-2542.2022.02.009

• 公司金融 • 上一篇    下一篇

面向特征因果分析的CFW-Boost企业财务风险预警模型

赵雪峰1，吴伟伟2，吴德林1，国旭3，时辉凝4   

  1. 1 哈尔滨工业大学（深圳）经济管理学院，广东 深圳 518006；2 哈尔滨工业大学 经济与管理学院，哈尔滨 150001；3 华为技术有限公司财经财务管理部，广东 深圳 523808；4 中国工商银行集约运营中心，广东 佛山 528010
  • 出版日期:2022-03-28 发布日期:2022-04-08
  • 作者简介:赵雪峰（1993-），男，博士。研究方向为创新管理、深度学习、机器学习、知识产权管理
  • 基金资助:
    国家自然科学基金资助项目(72072047)；国家社会科学基金资助项目(16AZD0006)；教育部人文社科项目（20YJC630090）；黑龙江省哲学社会科学研究规划项目(19GLB087)；中央高校基本科研业务费专项资金资助项目（IT.HSS.202102）；中央高校基本科研业务费专项资金资助项目（HIT.HSS.202139）；广东省基础与应用基础研究基金联合青年基金资助项目（2019A151510955）；深圳市基础研究专项（自然科学基金）面上项目（JCYJ2019806144607277）

CFW-Boost Model for Cause-and -effect Analysis in Enterprise Financial Risk Warning

ZAO Xuefeng1，WU Weiwei2，WU Delin1，GUO Xu3，SHI Huining4   

  1. 1 School of Management，Hrbin Institute of Technology (Shengzhen)，Shenzhen 518006，Guangdong，China；2 School of Management，Harbin Institute of Technology，Harbin 150001，China；3 Finance and Financial Management Department of Huawei Technologies Co.，Ltd，Shenzhen 523808，China；4 Foshon Operation Center of Industrial and commercial Bank of China，Foshan 528010，Guangdong，China
  • Online:2022-03-28 Published:2022-04-08

PDF

摘要： 当前多数模型一般以单类特征进行财务预警，缺乏多类特征为背景的预警分析，模型的预警准确率及鲁棒性也有待进一步提高。以财务指标及非财务指标构建多类财务特征的前提下，結合特征因果关系集成多棵CART树构建得到CFW-Boost，并利用其他预警模型，对比训练并实证分析CFW-Boost的表现，发现：CFW-Boost对比其他模型，准确率更高、预警表现更稳定；CFW-Boost模型通过特征因果分析降低特征维度，能够很好地避免特征冗余造成对模型鲁棒性的影响；CFW-Boost最优维度的数值最大，表示CFW-Boost相比于其他模型，在高维特征中优异性更强。本文提出的CFW-Boost经实证符合市场规律，可为企业及市场监督部门提供有益参考。

关键词: 财务预警, CFW-Boots, 特征因果关系, CART树

Abstract: At present? most models generally conduct early financial warning with low-dimensional features, which lacks early warning analysis with high dimensional features as the background, and the early warning accuracy and robustness of the model need to be further improved. On the premise of constructing multiple types of financial features with financial indicators and non-financial indicators, a CFW-Boost is obtained in combination with the causal relationship of features and integrating multiple, CART trees. The performance of CFW-Boost is empirically analyzed by using other early warning models to compare and train. The findings indicate that? compared with other models.CFW-Boost has a higher accuracy and a more stable warning performance. CFW-Boost reduces the feature dimension through feature causality analysis, which can well avoid the influence of feature redundancy on the robustness of the model. The value of the CFW-Boost optimal dimension is the largest, indicating that CFW-Boost has a stronger superiority in high-dimensional features than other models. The CFW-Boost proposed in this paper is empirically in line with the market law and can provide beneficial reference for enterprises and market supervision departments.

Key words: financial early-warning；CFW-Boost；characteristic causality；CART decision tree 

中图分类号: