Pricing the term structure with linear regressions: three different looks at the term premium

This paper reviews the Adrian, Crump, and Moench (2013) 3-step term structure estimation approach. We use a more up-to-date data set and extend the methodology by tackling the high persistence of yields using bootstrap bias correction and a coefficient restriction, in the first step of the Adrian et al. (2013) procedure. We find the two considered extensions _t the yield data to a similar degree as the authors' original model. In addition, a comparison of the forecasting performance of the two additional models reveals the original, non-altered model produces the most accurate 12-step ahead forecasts. Finally, we find that the term premium produced by the coefficient restricted model is the most economically justified term premium, as compared to the original and bias-corrected premia. The coefficient restricted model is therefore deemed more desirable to decompose the term premium. The unaltered model is preferred for forecasting purposes.