We investigate time-varying risk premia in Korean government bonds using an affine model with the five-factor specification. We find that the constructed five-factor model constructed generates nearly perfectly fitted yields and predicts expected excess returns of bonds with more precision than the four-factor model outside the U.S. More importantly, we find the statistically significant predictive power of the model for future bond returns using forward rates from both cross-sectional and time-series regressions firstly in the literature. The predictive power of the model varies in time for bonds with different maturities and reverts to the mean values for short and long maturities while showing a sign of momentum for intermediate maturities. This pattern remains salient for ten days. In addition to the simplicity in model construction using the three-step regression, our finding of the predictability enables fixed income investors to enhance short-term portfolio returns in a tactical strategy using forward rates.

Keywords: affine term structure model, fixed income asset pricing, return forecasting
JEL classification: G12 G15 C52 C53