Pricing the term structure of government and corporate bonds with linear regressions

In this paper, I study a three-step linear regression model to forecast U.S. Treasury bond yields as well as corporate bond yields. Previous studies applied this model to government bonds which have virtually no risk of default, and found that they could predict yields very accurately. However, no study has applied the model to corporate bonds. I find that five pricing factors obtained using principal component analysis, provide the best forecast for Treasury and corporate bonds. The predicted yields from the model show significant bias for Treasury bonds with higher time-to-maturities, but after correcting this bias, the predicted yields have an almost perfect _t to the actual yields. For the corporate bonds, the model performs reasonably in periods where the economy is stable, but cannot be used for forecasting. In periods with high credit risk, the model does not predict or follow the yields correctly since the model predicts that yields go down instead of up. For larger time-to-maturities the model is accurate when reflecting the yields about the x-axis.