Study On The Influencing Factors Of Credit Spread Of Industrial Bond Based On Regression Model

With the increase of the scale of Chinese bond market,credit risk events also follow.In 2014,the substantial default of "11 chao-ri" bond broke the vicious circle of "rigid payment" in Chinese bond market.Since then,the number of defaulters increased in 2015 and 2016.And this number witnessed another surge in 2018,with 41 new defaulters in the whole year.In addition,the total amount of defaults in 2018 reached 119.85 billion yuan,more than the previous four years combined.From the perspective of macro economy,in the context of challenges to economic growth in 2018.Monetary policy will shift to a reasonable and sufficient level and the downward price of funds will open up space for the interest rate center to go down.Judging from the current policy setting,under the pressure of early deleveraging,the growth of Chinese overall debt scale slowed down slightly and the regulatory process gradually turned from deleveraging to stable leverage.In the context of frequent default events in Chinese bond market and tight financing environment for small and medium-sized enterprises,it is particularly important to study the influencing factors of credit spreads.Different from Lagrange interpolation method,Hermite interpolation method not only requires that the function value of the interpolation polynomial at a given node is the same as the original function value,but also requires that the first and multiple derivatives of the interpolation polynomial at the node are also equal to the corresponding derivative value of the interpolated function.This thesis uses Hermite interpolation method to calculate the risk-free interest rate of special maturity and estimates the credit spread of industrial bond under special maturity.Hermite interpolation method can not only meet the requirements of smoothness,stability and flexibility,but also more accurate than linear interpolation method.The methods of independent variable selection in the regression model can be divided into two categories:the classical subset regression method and the variable selection method based on penalty function.Among them,stepwise regression method is the most commonly used method in the former one.Although this method can avoid the disadvantages of forward regression method and backward regression method and ensure that the finally obtained regression subset is the optimal regression subset,this method has low stability and high computational complexity.The latter is more suitable for variable selection of high-dimensional data,which principle is to enhance the explanatory power of the model by compressing partial regression coefficients to 0.In addition,this kind of method takes ridge estimation as the cornerstone and gradually optimizes the properties of estimators from the earlier LASSO method to the SCAD method.This thesis takes the SW real estate development bond in the agreed range as sample bonds and uses the independent variable selecting methods including stepwise regression method,adaptive Lasso method and SCAD method to establish multivariate regression models.Several methods to establish the model of the similar fitting degree and the number of the independent variables are able to be compressed better.Contrast with the actual situation,this thesis finally selects the subset selected by adaptive Lasso method to build the regression model.According to the fitting effect of the independent variables,the total asset size,total debt/EBITDA,net cash flow generated by operating activities,return on equity and several indicators reflecting the operating capacity of enterprises have relatively high explanatory power for the credit spread of bonds among the financial analysis indicators.Principal component analysis(PCA)is a dimensionality reduction processing technique,which can transform a set of original correlated variables into another set of unrelated variables and the new variables are arranged in the order of decreasing variance.In this thesis,the data set of the variables selected by the adaptive Lasso method is used for PCA and five principal components should be retained after the gravel test and 100 times of parallel analysis.Among them,the cumulative contribution of the first four principal components reached 85%,therefore,the fifth principal component is abandoned.The former four principal components are subject to principal component regression.Finally,the principal component regression equation is obtained.