Explanatory and Predictive Analysis of Corporate Bond Indices Returns

Abstract

This paper studies the behavior of corporate bond indices. We find that a 2-factor model with unobservable factors is adequate in capturing the variation of corporate bond portfolio returns, however we cannot identify any linear regression model with observable variables that would be able to do so. Including several economic variables as regressors, like the Fama-French factors, liquidity factors, and allowing for time-varying coefficients did not save the model. Principal component analysis revealed that the residuals from our regressions are highly correlated, with the first factor explaining around 86 percent of the remaining variation. Furthermore, there is a big difference in the R2's, and the coefficients of the factors between the high and low-grade bonds. We view our findings as evidence in favor of segmentation between the Treasury, equity, and corporate bond markets, as well as evidence of segmentation within the corporate bond market. In addition, not only are corporate bond excess returns predictable, but the big missing factor is also predictable. This is particularly important for investing and hedging with corporate bonds.