기계학습을 이용한 회사채 발행금리 예측

This study analyzes the corporate bond yield spread predictability with machine learning methods (LASSO, Ridge, Elastic net). We consider comprehensive input variables such as bond characteristics, macroeconomic factors, bond and stock market, and financial factors of issuers. The major empirical findings are as follows. First, we find that credit ratings, credit spreads, interest rate spread, base rate, and GDP are critical factors in corporate bond yield spreads. In addition, the optimal prediction method varies depending on the model type, the number of variables, and the sample size. Second, we find that the predictive model with the general corporate bond yield spreads could be helpful in predicting ESG bond initial yield rates. Overall, our findings show that machine learning for predicting corporate bond yield spread is more valuable than the traditional OLS method. Our evidence provides a new perspective in determining corporate bond rates and practical implications for improving useful information to the bond issuer or investors.