Empirical Study On Corporate Bond Yield Based On IPCA Model

Chinese bond market has entered a stage of rapid development in recent years.Corporate bonds also become an important part of the fixed income products.Chinese bond market is a dynamic nonlinear system,and accurate prediction of bond returns is an important research topic for both institutional and individual investors.In the past research,more and more factors have been mined and applied to asset pricing models,such as value,momentum,liquidity,volatility,earnings,etc.,and a wealth of econometric models and statistical means have emerged around the theme of income and pricing forecast.Our research objects are Chinese corporate bonds listed in the Shanghai and Shenzhen bond markets from January 1,2012 to December 31,2021.83 factors that contribute significantly to the prediction of bond returns are selected from the three aspects of bond,stock volume,price trading and corporate characteristics.The IPCA method is used to reduce the dimension and extract the principal component of the high-dimensional data.Firstly,we test the effectiveness of factors in risk prediction,and the good performance of IPCA in bond return prediction model is proved through inside and outside sample test,and compared with BBW,FF five-factor model and other classical factor models to verify the superiority of IPCA model.Secondly,we analyze the driving factors and economic significance of the five principal components extracted by IPCA method,tests the performance of IPCA in all market segments to verify its robustness,and analyzes the heterogeneity of different market segments.Finally,we combine the LSTM neural network to build IPCA-LSTM model to further deepen the bond return forecast model.The empirical results show that IPCA method has obvious advantages in factor dimension reduction,which is better than other models(MKT,BBW,FF),and shows strong stability in various industries and market segments.Moreover,combining IPCA and LSTM for dimension reduction can obtain higher prediction accuracy than single LSTM and PCA-LSTM model.