企业债券违约风险识别的可解释机器学习模型研究

doi:10.3778/j.issn.1002-8331.2310-0298

摘要/Abstract

摘要： 在我国信用债违约风险不断积累的背景下，如何精准识别、高效预警企业债券违约风险成为学术界及实务界所重点关注的问题。为有效解决传统违约风险预警模型存在的预警性能不强、超参数优化目标单一以及模型可解释性较弱等关键问题，通过有机融合LightGBM、NSGA-II、SHAP等机器学习算法，构建了LightGBM-NSGA-II-SHAP企业债券违约风险预警模型，并通过实证分析检验了所提出模型的预警性能。研究结果表明，所提出模型的预警准确率达到85%以上，相比传统机器学习模型，所提出模型的预警性能更加优异。另外，通过SHAP算法可视化展示预警特征对于预警结果的影响，发现票面利率、固定资产净利润率、发行总额、应收账款周转率等是识别企业债券违约的关键特征。

关键词: 债券违约, 风险识别, 机器学习

Abstract: Against the backdrop of the gradually exposed credit bond default risk in China, how to accurately identify and efficiently warn of corporate bond default risk has become a key concern for both academia and practice. To effectively solve a series of key problems in the traditional credit risk warning model, such as insufficient warning performance, single optimization target of hyperparameters, and weak model interpretability, this study integrates machine learning algorithms such as LightGBM, NSGA-II, and SHAP to constructs a LightGBM-NSGA-II-SHAP for early warning of corporate bond default risk, and empirically analyzes and tests the warning performance of the proposed model. The research results show that the warning accuracy of the proposed model exceed 85%, and compared with traditional machine learning models, the warning performance of the proposed model in this study is more excellent. In addition, the impact of visualization of warning features on warning results is demonstrated through the SHAP algorithm, and it is found that coupon interest rate, profit margin on fixed assets, total issuance, and receivable turnover etc. are the key features for identifying corporate bond defaults.

Key words: bond default, risk identification, machine learning

邓尚昆, 宁宏, 刘宗华, 朱应可. 企业债券违约风险识别的可解释机器学习模型研究[J]. 计算机工程与应用, 2024, 60(12): 334-345.

DENG Shangkun, NING Hong, LIU Zonghua, ZHU Yingke. Interpretable Machine Learning Model for Default Risk Identification of Corporate Bonds[J]. Computer Engineering and Applications, 2024, 60(12): 334-345.

参考文献

[1] LIU P, LI Y. Comparative analysis of machine learning models for bond default forecasting based on financial data of Chinese listed companies[J]. BCP Business & Management, 2022, 34: 1151-1158.
[2] HUAN Z. On the effectiveness of graph statistics of shareholder relation network in predicting bond default risk[J]. Journal of Control Science and Engineering, 2022: 8401354.
[3] ALLEGRE J, RAYMOND H, RHARRABTI H. The impact of the European sovereign debt crisis on banks stocks. Some evidence of shift contagion in Europe[J]. Journal of Banking & Finance, 2017, 74: 24-37.
[4] 吴世农, 卢贤义. 我国上市公司财务困境的预测模型研究[J]. 经济研究, 2001(6): 46-55.
WU S N, LU X Y. A study of models for predicting financial distress in China’s listed companies[J]. Economic Research Journal, 2001(6) : 46-55.
[5] 丁志国, 丁垣竹, 金龙. 违约边界与效率缺口: 企业债务违约风险识别[J]. 中国工业经济, 2021(4): 175-192.
DING Z G, DING Y Z, JIN L. Default boundary and efficiency gap: debt default risk identification for enterprises[J]. China’s Industrial Economics, 2021(4): 175-192.
[6] ZEITUN R, TIAN G G. Does ownership affect a firm’s performance and default risk in Jordan? corporate governance[J]. The International Journal of Business in Society, 2007, 7(1): 66-82.
[7] 李萌, 王近. 内部控制质量与企业债务违约风险[J]. 国际金融研究, 2020(8): 77-86.
LI M, WANG J. Internal control quality and enterprises’ debt default risk[J]. Studies of International Finance, 2020(8): 77-86.
[8] 丁志国, 耿迎涛, 赵晶, 等. 上市公司财务困境时间效应的实证判别与理论猜想[J]. 会计研究, 2018(2): 62-68.
DING Z G, GENG Y T, ZHAO J, et al. An empirical study and theoretical conjecture on the time effect of financial distress of China’s listed companies[J]. Accounting Research, 2018 (2): 62-68.
[9] 刘晓, 周荣喜, 李玉茹. 基于Stacking算法集成的我国信用债违约预测[J]. 运筹与管理, 2023, 32(3): 163-170.
LIU X, ZHOU R X, LI Y R. Default prediction of credit bond in China based on stacking algorithm integrated model[J]. Operations Research and Management Science, 2023, 32(3): 163-170.
[10] SARKAR S. Can tax convexity be ignored in corporate financing decisions[J]. Journal of Banking & Finance, 2008, 32(7): 1310-1321.
[11] 郭斌, 戴小敏, 曾勇, 等. 我国企业危机预警模型研究——以财务与非财务因素构建[J]. 金融研究, 2006(2): 78-87.
GUO B, DAI X M, ZENG Y, et al. Warning model of Chinese enterprises: building financial and non-financial factors[J]. Journal of Financial Research, 2006(2): 78-87.
[12] 谢玲玲, 范龙振. 上市公司信用债违约风险评估[J]. 管理现代化, 2020, 40(2): 104-108.
XIE L L, FAN L Z. Assessment of listed companies’ debt securities[J]. Modernization of Management, 2020, 40(2): 104-108.
[13] 庞春潮, 罗苑玮. 基于投资者视角的信用类债券违约风险预警[J]. 统计与决策, 2023, 39(2): 152-157.
PANG C C, LUO Y W. Early warning of credit bond default risk from the investor’s perspective[J]. Statistics and Decision, 2023, 39(2): 152-157.
[15] ALTMAN E I. Financial ratios discriminant analysis and the prediction of corporate bankruptcy[J]. Journal of Finance, 1968, 23: 589-609.
[14] 郑煜, 吴世农. 基于财务信息和非财务信息的债券违约预警模型研究——Fisher模型与Logistic模型的实证分析与应用[J]. 财会月刊, 2023, 44(12): 22-29.
ZHENG Y, WU S N. A study on bond default early warning models based on financial and non-financial information—empirical analysis and application of fisher model and logistic model[J]. Finance and Accounting Monthly, 2023, 44(12): 22-29.
[16] MERTON R C. On the pricing of corporate debt: the risk structure of interest rates[J]. Journal of Finance, 1974, 29(2): 449-470.
[17] JARROW R A, TURNBULL S M. Pricing derivatives on financial securities subject to credit risk[J]. Journal of Finance, 1995, 50: 53-85.
[18] HUAN J Z, HUANG M. How much of the corporate-treasury yield spread is due to credit risk?[J]. Review of Asset Pricing Studies, 2012, 2(2): 153-202.
[19] DUFFIE D, SINGLETON K J. Modeling term structures of defaultable bonds[J]. Review of Financial Studies, 1999, 12: 687-720.
[20] 梁龙跃, 王浩竹. 基于图卷积神经网络的个人信用风险预测[J]. 计算机工程与应用, 2023, 59(17): 275-285.
LIANG L Y, WANG H Z. Personal credit risk prediction based on graph convolutional network[J]. Computer Engineering and Applications, 2023, 59(17): 275-285.
[21] 贾颖, 赵峰, 李博, 等. 贝叶斯优化的XGBoost信用风险评估模型[J]. 计算机工程与应用, 2023, 59(20): 283-294.
JIA Y, ZHAO F, LI B, et al. XGBoost optimized by Bayesian optimization for credit scoring[J]. Computer Engineering and Applications, 2023, 59(20): 283-294.
[22] 王玉龙, 周榴, 张涤霏. 企业债务违约风险预测——基于机器学习的视角[J]. 财政科学, 2022(6): 62-74.
WANG Y L, ZHOU L, ZHANG D F. Enterprise debt default risk prediction—based on the perspective of machine learning[J]. Fiscal Science, 2022(6): 62-74.
[23] KE G, MENG Q, FINLEY T, et al. LightGBM: a highly efficient gradient boosting decision tree[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS’17), Red Hook, NY, USA, 2017: 3149-3157.
[24] 马晓君, 沙靖岚, 牛雪琪. 基于LightGBM算法的P2P项目信用评级模的设计及应用[J]. 数量经济技术经济研究, 2018(5): 144-160.
MA X J, SHA J L, NIU X Q. Design and application of P2P project credit rating model based on LightGBM algorithm[J]. The Journal of Quantitative & Technical Economics, 2018(5): 144-160.
[25] FEURE M, HUTTER F. Hyperparameter optimization[M]// Automated machine learning. Cham: Springer, 2019: 3-33.
[26] MA Y, YUN W. Research progress of genetic algorithm[J]. Application Research of Computers, 2012, 29: 1201-1206.
[27] KENNEDY J, EBERHAR R. Particle swarm optimization[C]//Proceedings of IEEE International Conference on Neural Networks, 1995: 1942-1948.
[28] DEB K, AGRAWAL S, PRATAP A, et al. A fast and elitist multiobjective genetic algorithm: NSGA-II[J]. IEEE Transactions on Evolutionary Computation, 2002, 6: 182-197.
[29] LUNDBERG S M, LEE S. A unified approach to interpreting model predictions[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems (NIPS’17), Red Hook, NY, USA, 2017: 4768-4777.
[30] 徐舒玥, 曹艳华. 基于可解释机器学习的信用债违约研究[J]. 科学决策, 2023(5): 190-200.
XU S Y, CAO Y H. Credit debt default research based on explainable machine learning[J]. Scientific Decision Making, 2023(5): 190-200.
[31] LEI S, LIANG X, WANG X, et al. A short-term net load forecasting method based on two-stage feature selection and LightGBM with hyperparameter auto-tuning[C]//IEEE/IAS 59th Industrial and Commercial Power Systems Technical Conference, 2023: 1-6.
[32] TANG M, ZENG W, ZHAO R. Corporate credit risk rating model based on financial big data[J]. BCP Business & Management, 2023, 48: 33-42.
[33] DEMIR S, ?AHHIN E K. Predicting occurrence of liquefaction-induced lateral spreading using gradient boosting algorithms integrated with particle swarm optimization: PSO-XGBoost, PSO-LightGBM, and PSO-CatBoost[J]. Acta Geotechnica, 2023, 18: 3403-3419.
[34] 朱云伟, 黄海松, 魏建安. 基于GA-LightGBM的刀具磨损状态在线识别[J]. 组合机床与自动化加工技术, 2021(10): 83-87.
ZHU Y W, HUANG H S, WEI J A. Online tool wear status identification based on GA-LightGBM[J]. Modular Machine Tool & Automatic Manufacturing Technique, 2021(10): 83-87.
[35] GAO B, BALYAN V. Construction of a financial default risk prediction model based on the LightGBM algorithm[J]. Journal of Intelligent Systems, 2022, 31: 767-779.
[36] 周卫华, 翟晓风, 谭皓威. 基于XGBoost的上市公司财务舞弊预测模型研究[J]. 数量经济技术经济研究, 2022, 39(7): 176-196.
ZHOU W H, ZHAI X F, TAN H W. Research on financial frauds prediction model of Chinese public companies with XGBoost[J]. Journal of Quantitative & Technological Economics, 2022, 39(7): 176-196.
[37] 张宁静, 顾新, 杨铖. P2P校园贷款个人违约风险因素指标探析[J]. 财会月刊, 2018(6): 82-89.
ZHANG N J, GU X, YANG C. Analysis on risk factors of individual default of P2P campus loan[J]. Finance and Accounting Monthly, 2018(6): 82-89.
[38] 邹旺, 江伟, 冯俊杰, 等. 基于ANN和SVM的轴承剩余使用寿命预测[J]. 组合机床与自动化加工技术, 2021(1): 32-35.
ZOU W, JIANG W, FENG J J. Bearing remaining useful life prediction based on artificial neural network and support vector machine[J]. Modular Machine Tool & Automatic Manufacturing Technique, 2021(1): 32-35.
[39] LI G, SHI Y, ZHANG Z. P2P default risk prediction based on XGBoost, SVM and RF fusion model[C]//Proceedings of the 1st International Conference on Business, Economics, Management Science (BEMS 2019), 2019: 470-475.