Robert C. Merton's article "On the Pricing of Corporate Debt: Risk Structure of Interest Rate" published in 1974 is the theoretical basis of modern credit default probability and recovery rate analysis. Its disadvantage is that it does not solve the actual observation problem of credit asset quality, and its application in empirical research is limited, which is also the focus of a lot of follow-up work after the birth of the model.
In view of the difficulties of Merton( 1974) model in the field of empirical application, several literatures try to provide flexible solutions. Crouhy and Galai( 1997) express Merton( 1974) model, which cannot be directly observed, as a function of credit default probability and recovery rate, thus simplifying the core of credit risk management to the observation and analysis of PD and LGD, which has great influence.
There are three methods to observe the LGD of quantitative finance Instruments (Liu Hongfeng, Yang Xiaoguang, 2003): market LGD (based on the market price of default bonds or tradable loans after the actual default event); Solve LGD (liquidation of LGD, the ratio of the estimated present value of a series of cash flows generated during liquidation and recovery to risk exposure); Implicit market LGD (LGD implied in the market, calculated by using the asset valuation model according to the spread and price of similar non-default bonds). In fact, there are many empirical studies based on bond secondary market or loan secondary market (such as personal housing mortgage securitization), but few empirical studies based on ordinary bank loans. One reason is the complexity of research methods, and the other is the non-disclosure of data.
1, American market research
Due to the availability of data, most of the current literature is concentrated in the American market.
Asarnow and Edwards (1995) use all economic losses after the default event to measure the expected loss of bank loans. Based on the default samples of Citibank 1970 to 1993, the calculated LIED is 34.79% and 12.75% respectively. An important finding of the study is that its distribution is "bimodal distribution", and the samples are concentrated at the high and low ends.
Carty and Lieberman( 1996) conducted an empirical study on 58 cases of default bank loans guaranteed by Moody's Company during the period of 1989- 1996. The results show that the average recovery rate is 765438 0%, the median is 77%, and the standard deviation is 32%. No "double model" was observed, but the recovery rate was obviously shifted to the high end.
Hamilton and Katy (1999) calculated the repayment rate of 159 bankruptcy cases by market method. The average repayment rate was 56.7%, the median repayment rate was 56%, and the standard deviation was 29.3%.
Gupton, Daniel Gates and Carty used 65,438+0,265,438+0 default loan samples in 2000. The results show that the average values of priority secured bank loans and priority unsecured loans are 69.5% and 52. 1% respectively, but the deviation from these averages is also significant in practical experience.
Gupton and Stein(2002) first put forward the LGD prediction model based on market value prediction, which is a multi-factor statistical model about American bonds, bank loans and LGD preferred shares.
Til Schuermann(2004) introduced the recovery distribution of all bonds and loans of Moody's Company 1970-2003, and explained the reason of bimodal distribution.
Michel A., M. Jocobs Jr, P. Varshey (2004) used the historical loan loss data of JPMorgan Chase during1982-1999 (* * * 3761defaulting customers) to study LGD. The average accounting LGD and economic LGD are at the same time, respectively. This study analyzes the mortgage LGD. Through the study of the sample from 1982 1 quarter to1999 quarter * * 1705, the average LGD of mortgage loan (1279 sample) is 27.7%, the standard deviation is 35.3%, and the unsecured loan is 45.3%.
2. Empirical research on LGD in other markets.
Hurt and Felsovayi of Citibank (1998) studied 1 149 bank loans in 27 countries in Latin America. The results show that the average default recovery rate is 68.2%, LGD is skewed, and macroeconomics and loan amount are one of the influencing factors of recovery rate.
La Porta et al. (2003) studied the PD and LGD of Mexican associated loans. In 1995- 1999, the average recovery rate of non-associated loans was 46%, and that of associated loans was 27%. The distribution shows that LGD deviates from the high end.
Xu Zhongmin, Taiwan Province Province (2004) conducted an empirical study on LGD based on the default information of bank borrowing enterprises from Taiwan Province Joint Collection Center Library 1996 to 2002. Taking the annual operating income of 5 million euros as the dividing standard, the average LGD of small enterprises (sample number 16454) below this standard is 75%, and the median is 88%, which is higher than that of standard large and medium-sized enterprises (sample).
Standard & Poor's Franks et al. (2004) used about 8,000 original data from Britain, France and Germany for research, and the data periods were 1993-2003 (France), 1996-2003 (Germany) and 1997-2003 (Germany). The data shows that the recovery rate in Britain is obviously higher than that in France and slightly higher than that in Germany. The distribution of recovery rate in France is obviously "bimodal distribution", while that in Britain and Germany is skewed.
Grunert and Weber(2005) studied the default data loss of 120 German companies 1992-2003. The data shows that the average recovery rate is 72.45% and the variance is 35.46%, and the distribution of recovery rate obviously deviates from the high end. The report also studied the impact of macro-economy, industry, loan conditions and tax policies.
All the above research reports only published the conclusive data after deep processing, the original data and model parameters were not published, and there was no LGD research report specifically for mortgage loans. Because the domestic corporate bond market is underdeveloped, the research on the data system of bank default loan recovery has not started for a long time. There are many theoretical introductions about loss given default in China, and there are few influential empirical data. Mainly includes:
1, relevant data of the four major asset management companies. The asset recovery data published by Huarong and other four major asset management companies can be used as indirect data to study domestic loans to loss given default. In 2004, the asset disposal results of the four major financial asset management companies in China were asset recovery rate of 26.60% and cash recovery rate of 20. 16%.
2. Other research. Zhang Haining (2004) took 19 1 credit projects of large commercial banks in China as samples (the time point was 1998) (involving loan principal of 26.629 billion yuan and interest of 7.708 billion yuan), showing that the average recovery rate was 33%, with the highest being 80% and the lowest being 0.
On May 28th, 2004, CCB auctioned the non-performing real estate assets with a book value of RMB 4 billion through international bidding. Citibank, Deutsche Bank, Lehman Brothers, JPMorgan Chase, Morgan Stanley and other institutions 15 participated in the bidding, and the final comprehensive capital recovery rate was 34.75%.