中国信用债券违约回收率分析Analysis of Default Recovery Rate for Chinese Credit Bonds
李向真,刘士达,王浩
摘要(Abstract)：
本文使用2014年至2021年违约兑付数据研究我国债券违约回收率。债券违约后一年(截至2021年末)的回收率均值为8%(11%),呈非对称两极分布，即集中于0%与100%的两侧。从不同时间窗口看，违约债券的“黄金回收窗口”为一年。存量债券占资产比例是影响违约回收率的重要因素。发债企业是否为国企、是否上市；债券有无担保人、是否在银行间市场发行；宏观变量中的短期国债利率水平、长短期国债利差等显著影响违约回收率。在违约回收率预测方面，机器学习模型明显优于线性回归模型。其中，集成学习类模型的预测效果最佳。
关键词(KeyWords)： 违约回收率;信用风险;机器学习
基金项目(Foundation): 清华自主科研基金的资助(项目号2023THZWJC20)的资助
作者(Author): 李向真,刘士达,王浩
参考文献(References)：
- 陈光忠，唐小我，倪得兵.2010.银行违约损失率特征研究[J].中国管理科学，18(2):19-24.Chen G Z,Tang X W,Ni D B.2010.The study on bank loss given default[J].Chinese Journal of Management Science,18(2):19-24.(in Chinese) 陈忠阳.2004.违约损失率(LGD)研究[J].国际金融研究，(5):49-57.Chen Z Y.2004.Research on LGD[J].Studies of International Finance,(5):49-57.(in Chinese) 代太山，陈敏，杨晓光.2008.不同担保类型之下违约损失率的结构特征：针对中国的实证[J].南方经济，(8):28-39.Dai T S,Chen M,Yang X G.2008.Structural characteristics of loss given default under credit guarantees:Empirical evidence for China[J].South China Journal of Economics,(8):28-39.(in Chinese) 何自力.2006.抵押贷款违约损失率研究[J].南方金融，(3):30-31,29.He Z L.2006.Research on loss given default rate of mortgage loan[J].South China Finance,(3):30-31,29.(in Chinese) 黄建忠，褚保金.2011.商业银行违约损失率的影响因素——基于江苏省某银行历史债项数据的分析[J].现代经济探讨，(8):51-55.Huang J Z,Chu B J.2011.The influencing factors of loss given default rate of commercial banks—Based on the historical debt data of a bank in Jiangsu Province[J].Modern Economic Research,(8):51-55.(in Chinese) 林青，郝帅.2017.国际评级机构回收率评级方法研究[J].债券，(12):56-61.Lin Q,Hao S.2017.Research on the recovery rate rating method of international rating agencies[J].China Bond,(12):56-61.(in Chinese) 林晚发，王雅炯，幸丽霞.2020.企业定性信息与债券信用评级：基于股权性质的分析[J].中国软科学，(4):123-131.Lin W F,Wang Y J,Xing L X.2020.Qualitative information of enterprises and credit rating of bonds:Evidence from the nature of stock right[J].China Soft Science,(4):123-131.(in Chinese) 刘逸凡，王志雄，孙海容.2020.中国违约债券的回收率问题探究[J].债券，(3):28-33.Liu Y F,Wang Z X,Sun H R.2020.Research on recovery rate of bond default in China[J].China Bond,(3):28-33.(in Chinese) 陆瑶，张叶青，黎波，等.2020.高管个人特征与公司业绩——基于机器学习的经验证据[J].管理科学学报，23(2):119-139.Lu Y,Zhang Y Q,Li B,et al.2020.Managerial individual characteristics and corporate performance:Evidence from a machine learning approach[J].Journal of Management Sciences in China,23(2):119-139.(in Chinese) 沈沛龙，崔婕.2005.内部评级法中违约损失率的度量方法研究[J].金融研究，(12):86-95.Shen P L,Cui J.2005.The measurement of LGD in IRB approach[J].Journal of Financial Research,(12):86-95.(in Chinese) 王浩，刘士达.2022.信用评级的中国实践——历史、现在与未来[M].北京：清华大学出版社.Wang H,Liu S D.2022.Credit rating and its practice in China-Past,present,and future[M].Beijing:Tsinghua University Press.(in Chinese) 杨军，程建，潘俊武.2009.违约损失率模型开发的理论分析和实证研究[J].国际金融研究，(6):71-78.Yang J,Cheng J,Pan J W.2009.A theoretical and empirical study on LGD models[J].Studies of International Finance,(6):71-78.(in Chinese) 周志华.2016.机器学习[M].北京：清华大学出版社.Zhou Z H.2016.Machine learning[M].Beijing:Tsinghua University Press.(in Chinese) Acharya V V,Bharath S T,Srinivasan A.2007.Does industry-wide distress affect defaulted firms?Evidence from creditor recoveries[J].Journal of Financial Economics,85(3):787-821. Altman E I,Kishore V M.1996.Almost everything you wanted to know about recoveries on defaulted bonds[J].Financial Analysts Journal,52(6):57-64. Amihud,Yakov,Kenneth Garbade,Marcel Kahan,2000.An institutional innovation to reduce agency costs of public corporate bonds[J.Journal of Applied Corporate Finance,(1):114-121. Amiram D.2011.Debt contracts and loss given default[D].Chapel Hill:The University of North Carolina at Chapel Hill. Asarnow E,Edwards D.1995.Measuring loss on defaulted bank loans:A 24-year study[J].Journal of Commercial Lending,77(7):11-23. Bastos J A.2010.Forecasting bank loans loss-given-default[J].Journal of Banking & Finance,34(10):2510-2517. Breiman L.2001.Random forests[J].Machine Learning,45(1):5-32. Carty L V,Lieberman D.1996.Defaulted bank loan recoveries[M].New York:Moody's Investors Service Inc.. Carty L V,Hamilton D T,Keenan S C,et al.1998.Bankrupt bank loan recoveries[M].New York:Moody's Investors Service,Inc.. Caselli S,Gatti S,Querci F.2008.The sensitivity of the loss given default rate to systematic risk:New empirical evidence on bank loans[J].Journal of Financial Services Research,34(1):1-34. Dalianes P C.1999.Credit risk pricing:Summary and current methodology[J].Quantitative Models in Finance. Dermine J,De Carvalho C N.2006.Bank loan losses-given-default:A case study[J].Journal of Banking & Finance,30(4):1219-1243. Duffie D,Singleton K J.1999.Modeling term structures of defaultable bonds[J].The Review of Financial Studies,12(4):687-720. Frye J.2000.Depressing recoveries[J].Risk London Risk Magazine Limited,13(1):108-111. Gilchrist S,Zakraj?ek E.2012.Credit spreads and business cycle fluctuations[J].American Economic Review,102(4):1692-1720. Gupton G M,Gates D,Carty L V.2000.Bank loan loss given default[M].New York:Moody's Investors Service Inc.. Gupton G M,Stein R M.2002.LossCalcTM:Model for predicting loss given default (LGD)[M].New York:Moody's KMV Company. Han C,Jang Y.2013.Effects of debt collection practices on loss given default[J].Journal of Banking & Finance,37(1):21-31. Jankowitsch R,Nagler F,Subrahmanyam M G.2014.The determinants of recovery rates in the us corporate bond market[J].Journal of Financial Economics,114(1):155-177. Jarrow R A,Lando D,Turnbull S M.1997.A Markov model for the term structure of credit risk spreads[J].The Review of Financial Studies,10(2):481-523. Leland H E,Toft K B.1996.Optimal capital structure,endogenous bankruptcy,and the term structure of credit spreads[J].The Journal of Finance,51(3):987-1019. Longstaff F A,Schwartz E S.1995.A simple approach to valuing risky fixed and floating rate debt[J].The Journal of Finance,50(3):789-819. Loterman G,Brown I,Martens D,et al.2012.Benchmarking regression algorithms for loss given default modeling[J].International Journal of Forecasting,28(1):161-170. Nazemi A,Fabozzi F J.2018.Macroeconomic variable selection for creditor recovery rates[J].Journal of Banking & Finance,89:14-25. Qi M,Zhao X L.2011.Comparison of modeling methods for loss given default[J].Journal of Banking & Finance,35(11):2842-2855. Schuermann T.2004.What do we know about loss given default?[R].Working Paper. Strobl C,Boulesteix A L,Zeileis A,et al.2007.Bias in random forest variable importance measures:Illustrations,sources and a solution[J].BMC Bioinformatics,8:25. Stulz R M.2010.Credit default swaps and the credit crisis[J].Journal of Economic Perspectives,24(1):73-92. Thorburn K S.2000.Bankruptcy auctions:Costs,debt recovery,and firm survival[J].Journal of Financial Economics,58(3):337-368. Yao X,Crook J,Andreeva G.2015.Support vector regression for loss given default modelling[J].European Journal of Operational Research,240(2):528-538.
- ① 见王浩和刘士达(2022)。 ② 见林青和郝帅(2017)和刘逸凡等(2020)。 (1)在分布假设上,OLS线性模型会隐含正态分布的假设,其对回收率的解释力度实则较低;而机器学习模型多为非线性模型,对回收率的解释力度要明显高于OLS。在算法机制上,Strobl et al.(2007)指出,排序重要性更倾向于筛选连续变量和分类较多的离散变量,对0/1变量存在“歧视”,这也使得机器学习模型不易把债券特征(有无担保人、是否在银行间市场发行)和企业特征(是否为国企、是否为上市公司)变量作为有解释力度的重要变量。 (2)523支违约债券中有52只存在违约后的交易记录,占比仅为9.94%;另一方面,存在交易价格的债券中,26支债券交易记录小于2笔,占比50%。