Bond Risk Premia And Volatility Decomposition

This paper decomposes the bond realized volatility into two parts:the realized bi-power variation, which is the continuous part, and the realized jump size which is the discontinuous part. We find that adding a rolling estimate of the mean realized bi-power variation—identified from high-frequency bond returns using the bi-power variation technique, to the regression of Jonathan Wright and Hao Zhou (2009), who augment a regression of excess bond returns on the term structure of forward rates and the realized jump mean, will significantly upgrade the predictive ability of the regression. This result is consistent with the setting of an unspanned risk factor in which the conditional distribution of excess bond returns is affected by a state variable that does not lie in the span of the term structure of yields or forward rates. In out-ofsample forecasting exercises, inclusion of the bi-power variation mean can reduce the root mean square prediction error by up to27percent on the basis of Jonathan Wright and Hao Zhou (2009). What is more, we find that the bi-power variation mean drives the innovations of current term structure while the impact of realized jump mean on bond yields is very limited. This result suggests that the jump mean is more likely to be a proxy of ’pure USV’, while the bi-power variation mean is, to a great extent, a ’partial USV’, which may serve as a bridge connecting the bonds and the fixed income derivatives.