Bond Default Prediction Based On Heterogeneous Information Processing

As credit risk in the bond market broke out,bond defaults occur frequently.Based on objective data,using computer technology for default prediction has great significance to bond investors and practitioners.The traditional financial models are mostly for the analysis and predict of bonds issued by listed companies,and lack of early warning ability for a large number of non-listed companies.At the same time,there are a lot of relationship data and category data in the bond information.How to use these data reasonably is of great significance for bond default prediction.Therefore,based on the multi-source information of bonds and issuers as well as macroeconomic data,we use knowledge graph and deep learning techniques to predict bond defaults.On the basis of constructing the bond knowledge graph,we use the knowledge representation learning to vectorize the knowledge,and input these extracted vectors into the deep learning model to predict the bond default.The main research contents of this article contains the followings:Acquisition and preprocessing of bond information.There are many factors affecting the default of bonds.This paper mainly uses the public information of bonds and issuers as well as macroeconomic data to predict default.We obtain bond information through the data provider's API and web crawler.Screening bond and cleaning data to provide accurate data protection for the construction of bond knowledge graph and default prediction.The construction of the bond knowledge graph and the knowledge representation based on knowledge representation learning.In view of the characteristics of bond information including multiple relationships,we construct the bond knowledge graph,which can clearly express the complicated relationship in the bond market.Using the knowledge representation learning model to learn the semantic information and structural information of the knowledge graph,as a supplement to the prior knowledge of bond default.Using TransE,TransH and TransR model,we map the entities and relationships into the vector space,then the bonds can be represented by low-dimensional dense vectors.Research on bond default prediction based on deep learning model.Through the analysis of bond default events,we extract the relevant characteristics of bond default issues.In view of multi-category characteristics of bond data,we propose DeepFMKG model.The model inputs the vectorized representation of the bond knowledge graph as priori characteristic.Finally,we set up some experiments to evaluate the performance of the DeepFM-KG model on bond default issues.In the comparison of the evaluation results,the DeepFM-KG model achieves relatively good results.