Studies On Credit Spreads In China's Bond Market

In recent years,the rapid development of China's bond market strongly supported the real economy.However,along with the expanding market size,the default of credit bonds was becoming more and more serious.Besides,the impacts of exchange rates on the market were increasingly indispensable,as RMB bonds became more attractive to foreign capital under the opening-up policy.Considering these changes,this thesis conducted an in-depth study on the credit spreads in China.China's bond market has witnessed rapid development in recent years,with the stock of bonds reaching 114.31 trillion yuan by the end of 2020.In 2020,the total amount of bonds issued reached 57.3 trillion yuan,an increase of 26.51% over the previous year,accounting for 56.39% of the GDP in that year,which strongly supported the real economy even though the COVID-19 epdemic hit the economy hard.As the size of the bond market increased,bond defaults became more severe.By the end of 2020,740 bonds issued by 181 issuers defaulted,from privately raised company bonds to public offering bonds,from private enterprises to the state-owned enterprises,from private company to listed companies,from default interest to the principal defaults,and the frequency,scope and severity of credit events further intensified.The credit risk of bonds have been attracting more and more attention from the society.Besides,since the launch of the "Bond Connect",foreign investors' holdings of RMB bonds have been growing at an average annual rate of 40 percent.As China's bonds have been officially included in the world's three major mainstream bond indexes,and the launch of Liquid China Credit Index,RMB bonds have been becoming more attractive for global investors.Therefore,the RMB exchange rate,an important factor affecting foreign investment in China's financial assets,would also have an increasingly prominent impact on the bond market.Moreover,in the information age,the factors that affected the bond market trend were increasing,and the decision-making difficulty of the bond market participants and regulatory subjects was increasing.Therefore,there was an urgent need for a more accurate grasp of the bond market trend.Under the above background,it was of great theoretical significance and practical value to study the credit spread,which was the basis of bond pricing and risk management,reflecting the cost and risk of bonds.First,this paper obtained the monthly unbalanced panel data from January 2010 to May 2020,which were made up of listed companies' corporate financing and bond issuing data,as well as stock and bond market index data.From the research perspective of credit spread decomposition,this thesis utilized fixed-effect model and generalized moment estimation for two-step systems to discuss the impact of stock and bond market systematic risk on credit spreads.It also examined the explanatory power of the model.Second,based on investor heterogeneity,this thesis also incorporated foreign exchange intervention,exchange rate,and credit spread into an open economic framework,building an endogenous dynamic system framework with considerations of the above factors.The time-varying parameter structural vector autoregressive model with stochastic volatility(SV-TVP-SVAR)was also adopted to empirically study the dynamic relationship between exchange rates and credit spreads.Third,the author further explored ways to predict credit spreads more accurately by comparing different models and data sets.To achieve this goal,the prediction data set was selected according to the above research results and a deep-gated recurrent neural network model(1)was built with deep learning algorithms,which were applied to make prediction of credit spread index in China's bond market.It also explored whether introducing attention mechanism into the deep learning model or adding Internet search behavior data into the predictive feature data set can further improve the predictive ability of the model.The results were compared with those from the benchmark models using multiple forecasting technologies.After digging into the above three aspects,this thesis concluded as follows:(1)Systematic risk in the stock market had a significant impact on corporate bond credit spreads,but the explanatory power was very limited(less than 5% on average).Systematic risk in the bond market had a strong explanatory power for corporate bond credit spreads,explaining 39% of credit spreads on average.(2)Systematic risks in the stock market and bond market were sensitive to the bond credit rating,residual maturity,and nature of company property rights.They were not sensitive to the economic region of the bond issuing company.The systematic risk premium was negatively correlated with bond rating and state-owned nature while positively correlated with residual maturity.(3)Central bank foreign exchange intervention and changes in the RMB exchange rate are linked with changes in credit spreads in a non-linear way.The non-linear linkage effects had strong time-varying characteristics.Besides,the linkage effects with changes in bond credit spreads of different maturities and ratings were asymmetric.Specifically,changes in the RMB exchange rate had the greatest impact on the credit spreads of medium-term bonds and low-rated bonds.Moreover,the linkage effect between the exchange rate fluctuation and the credit spread variation had a lag.And the effect of foreign exchange intervention from central bank was declining in stabilizing RMB exchange rate.(4)Intelligent algorithms such as deep learning and traditional machine learning performed better than non-intelligent algorithms such as traditional classic financial time series.Deep learning models with gated unit mechanisms had better prediction accuracy and higher prediction stability.From the absolute performance of the prediction,the deep gated recurrent neural networks was an effective model to predict the credit spread in China's bond market.Self-attention mechanism could only play a role when the prediction ability of deep learning model itself was not strong.In this case,the introduction of attention mechanism could improve the prediction ability of the model to some extent.When the predictive ability of the deep learning modelwas relatively strong,the introduction of the attention mechanism was a kind of interference,which would significantly weaken the predictive ability of the model itself.That is,the attention mechanism could only play the role of help,but could not play the effect of icing on the cake.Bond market was a market dominated by institutional investors,but Internet search index measured public attention,the information content contained in the index was not enough to affect the credit spread,and it was weak to improve the predictive ability of deep learning models.This research provided empirical evidence for the linkage effects of China's bond market with both the stock market and the foreign exchange market.It also helped to deepen the understanding of the "credit spread puzzle" in the bond market.Furthermore,it could also facilitate the decision-making process of investors' bond valuation and regulators' risk monitoring and control.