基于时间相依Cox回归的动态财务预警模型及实证

doi:10.12005/orms.2020.0214

运筹与管理 ›› 2020, Vol. 29 ›› Issue (8): 177-185.DOI: 10.12005/orms.2020.0214

基于时间相依Cox回归的动态财务预警模型及实证

李鸿禧^1,2, 宋宇^3,4

1.中央国债登记结算有限责任公司博士后科研工作站,北京 100033;
2.中国人民银行金融研究所博士后流动站,北京 100800;
3.中国东方资产管理股份有限公司博士后科研工作站,北京 100034;
4.清华大学经济管理学院博士后流动站,北京 100084

收稿日期:2019-05-09 出版日期:2020-08-25
作者简介:李鸿禧(1988-),女,天津人,博士后工作站、博士后流动站博士后,博士,研究方向:财务评级、风险预警;宋宇(1988-),男,辽宁沈阳人,博士后科研工作站、博士后流动站博士后,博士,研究方向:公司金融、信用风险。
基金资助:
中国博士后科学基金资助项目(2019M650991,2020M670403);国家自然科学基金重点项目(71731003)

Dynamic Financial Early Warning Model Based on Time-Dependent Cox Regression and Empirical Study

LI Hong-xi^1,2, SONG Yu^3,4

1. China Central Depository & Clearing Co., Ltd., Beijing 100033, China;
2. Institute of Finance, The People's Bank of China, Beijing 100800, China;
3. China Orient Asset Management Co., Ltd., Beijing 100034, China;
4. Tsinghua University School of Economics Management, Beijing 100084, China

Received:2019-05-09 Online:2020-08-25

摘要/Abstract

摘要： 本文以中小企业为研究对象,从偿债能力、盈利能力等财务因素,加之公司治理、宏观环境等非财务因素出发,利用共线性检验和时间相依Cox回归构建动态财务预警模型,并与经典的Cox模型、logit模型进行对比分析。本研究的特色有二:一是通过时间相依Cox回归模型,构建随时间而变化的预警指标数据与财务危机之间的函数关系。利用偏似然估计、Breslow估计量分别拟合回归系数和基准危险强度,构建财务预警模型,预测企业在未来一段时间内每个时间点上的财务危机概率。相比于基于传统Cox模型的预警研究仅用一期的截面数据建模,本研究考虑了预警指标的动态变化对财务风险的影响,涵盖了更多的历史信息,达到提高预警精度的目的。二是考虑第一类错误“危机企业判为正常”与第二类错误“正常企业判为危机”给投资者造成的损失差异,衡量预警的“错误成本”,以错误成本最低为目标,反推出财务正常和财务危机之间的预警阈值,实现了对财务危机发生与否的提前预警功能。经过实证,本研究的财务预警模型精度较高,尤其对财务危机企业的正确识别率达到75%。相较于传统的Cox回归、logit模型,危机企业的正确识别率更高、错误成本更低。盈利能力、公司治理水平是对企业财务风险影响最为显著的因素。

关键词: 财务预警, 时间相依Cox回归, 预警阈值

Abstract: This paper takes SMEs as the research object, from the financial factors such as solvency and profitability, plus the non-financial factors such as corporate governance and macro environment, using the collinearity test and time-dependent Cox regression to construct a dynamic financial early warning model, and with the classic, the Cox model and logit model are compared and analyzed. This study has two features as follows. Firstly, the time-dependent Cox regression model is used to construct a functional relationship between the early warning indicator data and the financial crisis. Using the partial likelihood estimation and Breslow estimator to fit the regression coefficient and the baseline risk intensity respectively, a financial early warning model is constructed to predict the financial crisis probability of the enterprise at each time point in the future. Compared with the traditional Cox model based early warning research, only one-stage cross-sectional data modeling is used. This study considers the impact of dynamic changes of early warning indicators on financial risks, covers more historical information, and achieves the purpose of improving early warning accuracy. Secondly, this study considers the first type of error “crisis enterprises are judged as normal” and the second type of mistakes “normal enterprises are judged as crisis” to cause losses to investors, to measure the “wrong cost” of early warning, with the goal of the lowest cost of error. The early warning threshold between financial normality and financial crisis is introduced to realize the early warning function of the occurrence or failure of financial crisis. By empirical research, the financial early warning model of this study has high precision, especially for financial crisis enterprises, and the correct recognition rate is 78%. Compared with the traditional Cox regression and logit models, crisis companies have higher correct recognition rates and lower error costs. Profitability and corporate governance are the most significant factors affecting corporate financial risk.

Key words: financial early warning, time-dependent Cox regression, early warning threshold

中图分类号:

F830

李鸿禧, 宋宇. 基于时间相依Cox回归的动态财务预警模型及实证[J]. 运筹与管理, 2020, 29(8): 177-185.

LI Hong-xi, SONG Yu. Dynamic Financial Early Warning Model Based on Time-Dependent Cox Regression and Empirical Study[J]. Operations Research and Management Science, 2020, 29(8): 177-185.

参考文献

[1] 鲍新中.基于粒子群的K均值算法和粗糙集理论的财务预警[J].系统管理学报,2012,21(4):461-469.
[2] 顾海峰.信用突变下商业银行信用风险预警的实证研究——基于偏好熵权物元可拓模型的分析[J].审计与经济研究,2014,29(3):93-103.
[3] 谢赤,赵亦军,李为章.基于CFaR模型与Logistic回归的财务困境预警研究[J].财经理论与实践,2014(1):57-62.
[4] 孙晓琳.基于状态空间的财务危机动态预警模型在中国的实证研究[J].中国软科学,2013(4):140-147.
[5] 方匡南,等.基于网络结构Logistic模型的企业信用风险预警[J].统计研究,2016,33(4):50-55.
[6] Fabio Comelli. Comparing the performance of logit and probit early warning systems for currency crises in emerging market economies[J]. Imf Working Papers, 2014.
[7] Babecky J, Havránek, Tomá, Matěj, Jakub, et al. Banking, debt, and currency crises in developed countries: stylized facts and early warning indicators[J]. Journal of Financial Stability, 2014, 15: 1-17.
[8] 鲍新中,杨宜.基于聚类-粗糙集-神经网络的企业财务危机预警[J].系统管理学报,2013,22(3):358-365.
[9] 鲍新中,傅宏宇.基于变精度加权平均粗糙度决策树的财务预警研究[J].运筹与管理,2015,24(3):189-196.
[10] 黄超,韩婷婷,吴芃,等.基于双正交小波混合核KPCA-SVM财务危机预警研究[J].系统管理学报,2015,24(1):48-55.
[11] 叶焕倬,杨青,汪勇华.智能财务危机预警与原因诊断方法研究——基于自适应贝叶斯网络模型SABNM的探析[J].审计与经济研究,2013(3):74-82.
[12] 吴冲,刘佳明,郭志达.基于改进粒子群算法的模糊聚类-概率神经网络模型的企业财务危机预警模型研究[J].运筹与管理,2018,27(2):106-132.
[13] Koyuncugil A S, Ozgulbas N. Financial early warning system model and data mining application for risk detection[J]. Expert Systems with Applications, 2012, 39(6): 6238-6253.
[14] Sevim C, Oztekin A, Bali O, et al. Developing an early warning system to predict currency crises[J]. European Journal of Operational Research, 2014, 237(3): 1095-1104.
[15] 张亮,张玲玲,陈懿冰,等.基于信息融合的数据挖掘方法在公司财务预警中的应用[J].中国管理科学,2015,23(10):170-176.
[16] 符刚,曾萍,陈冠林.经济新常态下企业财务危机预警实证研究[J].财经科学,2016(9):88-99.
[17] Javier De Andrés, Lorca P, Juez F J D C, et al. Bankruptcy forecasting: a hybrid approach using fuzzy c-means clustering and multivariate adaptive regression splines(MARS)[J]. Expert Systems with Applications, 2011, 38(3): 1866-1875.
[18] Chen M Y. Predicting corporate financial distress based on integration of decision tree classification and logistic regression[J]. Expert Systems with Applications, 2011, 38(9): 11261-11272.
[19] Cox D. Regression models and life tables[J]. J Roy Stat Soc B, 1972, 34.
[20] 鲍新中,陶秋燕,傅宏宇.基于变量聚类和COX比例风险模型的企业财务预警研究[J].系统管理学报,2015(4):517-523.
[21] 卢永艳.上市公司财务困境风险的行业差异性研究[J].宏观经济研究,2012(3):80-84.
[22] Fisher L D, Lin D Y. Time-dependent covariates in the cox proportional-hazards regression model[J]. Annual Review of Public Health, 1999, 20(1): 145-157.
[23] 何晓群.多元统计分析[M].中国人民出版社,2012.
[24] Ali S, Liu B, Su J J. Does corporate governance quality affect default risk? the role of growth opportunities and stock liquidity[J]. International Review of Economics & Finance, 2018, 58: S1059056017307554.
[25] MSCI. MSCI ESG ratings methodology[R]. MSCI, 2018.
[26] 南开大学公司治理评价课题组.中国上市公司治理状况评价研究——来自2008年1127家上市公司的数据[J].管理世界,2010(1):142-151.

基于时间相依Cox回归的动态财务预警模型及实证

Dynamic Financial Early Warning Model Based on Time-Dependent Cox Regression and Empirical Study

PDF

可视化

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

编辑推荐

Metrics

[1]	张茂军, 李昊, 南江霞, 王国栋. 银行间与交易所国债市场的信息溢出效应研究[J]. 运筹与管理, 2021, 30(12): 165-171.
[2]	宗喆, 郑重阳, 王涧秋, 赵辉. 沪铜期货及其期权的量化定价实证研究[J]. 运筹与管理, 2021, 30(12): 179-184.
[3]	贾丽娜, 王芳云, 李慧瑛, 孙辉. 我国基金博彩型股票交易是博彩偏好的结果吗？——基于信息挖掘视角[J]. 运筹与管理, 2021, 30(12): 191-197.
[4]	吴可可, 余燕, 董大勇. 投资者注意的风险补偿对股市回报的预测[J]. 运筹与管理, 2021, 30(12): 198-203.
[5]	喻均林, 何瑞铧. 管理层业绩预告会影响定增价格吗？——基于中介效应的视角[J]. 运筹与管理, 2021, 30(12): 204-211.
[6]	程新生, 孙红艳, 贾晓玲. 机构投资者增持的经济后果:基于定向增发的实证研究[J]. 运筹与管理, 2021, 30(11): 135-141.
[7]	杨亚娟, 马如飞, 陈孔艳. 经济政策不确定性对“中-美”股市、“中国股市-黄金市场”的长期动态相关性影响研究———基于DCC-MIDAS模型[J]. 运筹与管理, 2021, 30(11): 142-146.
[8]	杨兴雨, 刘伟龙, 赵雪瑾, 张永. 考虑破产控制的多期模糊资产-负债组合优化模型[J]. 运筹与管理, 2021, 30(11): 147-154.
[9]	侯剑, 白雪斌, 周艳明. 供应链金融借贷市场中均衡解的存在性研究[J]. 运筹与管理, 2021, 30(10): 102-106.
[10]	朱怀念, 朱莹. 跳扩散模型下的非零和随机微分投资组合博弈[J]. 运筹与管理, 2021, 30(10): 183-190.
[11]	付帅帅, 陈伟达, 丁军飞, 王丹丹. 考虑不同融资金额需求的授权再制造融资策略研究[J]. 运筹与管理, 2021, 30(10): 213-219.
[12]	师苑, 王新华, 高红伟. 三级供应链企业间交叉持股时均衡定价及市场绩效研究[J]. 运筹与管理, 2021, 30(9): 145-151.
[13]	刘桂芳, 徐维军, 黄静龙, 赵琪. 考虑公司信息披露情绪的欧式脆弱期权定价[J]. 运筹与管理, 2021, 30(9): 164-171.
[14]	谭春萍, 秦学志, 王文华, 王麟. 基于权益再融资与策略性债务支付的公司财务困境纾解研究[J]. 运筹与管理, 2021, 30(8): 181-189.
[15]	迟国泰, 李哲. 基于动态调整的Copula-非线性分位数回归资产组合优化[J]. 运筹与管理, 2021, 30(6): 35-41.