We propose a new dynamic Nelson–Siegel yield curve model in which two time-varying factor-specific decay parameters govern the slope and curvature factor loadings, and the factor shock variance–covariance (SV) follows a stochastic inverse Wishart process. The proposed model is compared with simpler specifications in terms of statistical and economic criteria to demonstrate the importance of jointly incorporating time-varying factor loadings and SV. We examine the out-of-sample yield curve density forecasting performance for statistical evaluation. The utility gain from the bond portfolio optimization of a Bayesian risk-averse investor measures the model’s economic value. Our out-of-sample experiment using United States monthly yield curve data indicates that the time-varying factor loadings and SV accommodate gradual structural changes in the yield curve dynamics around an unconventional monetary policy period, thereby improving the predictive accuracy and utility gain.