The Study On Credit Risk Measurement Of Unlisted Real Estate Corporate Bonds Based On Modified BP-KMV Model

In recent years,with the development of bond market and expansion of issue size,the credit risk problem is increasingly more serious,credit risk has been always the important risk in bond market.But credit risk analysis,measurement and management is relatively lagging behind.Because it has been suffered from funding constraints and maturity pressure.Many real estate enterprises with high debt ratio have been exposed to high business risk and default risk.Nowadays,there are several mainstream credit risk measurement models,one of them is KMV model.The good predictive ability and other advantages make it become one of the best model.KMV model is based on Merton model,mainly used to measure the probability of default about listed companies.This model regards the liability of an enterprise as a call option,when the time comes,if the value of an enterprise's assets is less than the value of its liabilities,then the enterprise is going to default,and vice versa.This paper is about the measurement of unlisted real estate corporate bonds' credit risk based on BP nerve network and modified KMV model.Because KMV model is used for listed companies,this is not avaliable for most smes and unlisted companies.In order to use this model,it must use reasonable methods to estimate the market value and volatility of these companies.Meanwhile,KMV model is built on a large amount of data of the United States,but China's stock market is vastly different from that of the United States,it's necessary to rethink about the calculation formula of distance to default.First,introduce the theoretical basis of the probability of default rate of unlisted corporate bonds,including the main methods of corporate valuation,and the main approaches of credit risk models and compare them to KMV model.Then introduce BP nerve network and how to build the model.Based on the fact that coefficient in default distance function may not fit the situation in our country,I will use linear regression to find out proper coefficient.Next,in order to choose the proper indexes,use Pearson correlation to analyse.Finally,put the unlisted corporate data to the model and get the results,and compared them to the results of other two methods,analysed these results by quantitative analysis.We have the following conclusions:First,used Pearson correlation to screen training indicators in BP nerve network,and ended up with a BP nerve network that can predict the asset value and volatility of unlisted corporate very well,proved that Pearson correlation can screen the indicatorseffectively.Second,the results of model quite fit the actual situation,after compared the results with the results of unmodified BP-KMV model and PFM model,it turned out that the modified BP-KMV model is much more effective that the other two models,it proves that model is effective and it is possible to use it in the future.Third,as seen from the final results,most AA bonds and part of AA+ bonds have higher default probabilities,and for AAA bond,the default probabilities almost equal to zero,which consistent with the actual situation.