The Research On Prediction Model And Application Of Corporate Bonds Default Based On Neural Network

In recent years,as one of the most important ways of direct financing in China's capital market,the scale of bond financing has increased significantly,which has far exceeded the equity direct financing in the same period.However,since March 2014,when the "11 super day bonds" materially defaulted and broke the "rigid payment" in China's bond market,bond defaults have been emerging in an endless stream,and the "normalization" trend of bond defaults may occur in the future.Corporate bonds,as a variety of default bonds,have a certain representativeness in analyzing the specific reasons for default.With the development of emerging technologies,the prediction model based on neural network is expected to provide more ideas and research directions for bond default prediction.First of all,based on the existing literature of the influence factors of the corporate debt default,this paper sorted out and summarized 45 factors affecting the default of bonds are,which are divided into external factors(macro-economy,industry trend and policy influence),internal factors(financial factors,governance level and exceptional circumstances)and bonds features,and constructed a relatively complete theoretical logic system through qualitative analysis;Secondly,considered data availability and index repeatability,index screening was conducted through the random forest model to obtain the two groups of characteristic values ranking the top 10 for the t-1 period and t-2 period respectively,Thirdly,a BP neural network model was built to predict the default of 20 corporate bonds and 120 normal corporate bonds in the period of 2014-2018,and the predicted results were compared with the traditional regression model(Logistic model)to draw a conclusion.Finally,taking the default event of "14 Liyuan bond" as an example,proved that the BP neural network model can be applied to specific default cases and has certain reference value.Based on the above analysis,this paper comes to the following conclusions :(1)compared with the Logistic model,the BP neural network model for predicting corporate debt default has better prediction results;(2)the random forest model has a good dimensionality reduction function,and the eigenvalue input neural network model with sufficient influence degree is screened for empirical study,so as to improve the convergence rate and empirical efficiency of the model.(3)among many factors affecting the possibility of default of corporate bonds,the most important one is the internal financial index of the company.BP neural network has important practical significance in the prediction of corporate debt default.It can accurately predict the future default situation of the target corporate debt,help investors to screen investment targets and give early warning of default,in order to avoid the risk of bond default.