The yield curve contains a lot of important information for asset pricing, financial risk management, portfolio allocation, monetary policy implementation, and so on. Despite the prevalence of dynamic Nelson-Siegel (DNS) models in capturing yield curve dynamics, comparatively little attention has been paid to forecasting the yield curve with high-frequency daily data. In this thesis, I explore the DNS models with different specifications using both U.S. and Chinese treasury bonds' monthly and daily data. Different estimation methods are employed, including a one-step DNS that applies a Kalman filter and a two-step DNS approach. Based on these models, comparisons of the characteristics and phenomena in both markets are made as well. I find that the in-sample fit of the DNS models is excellent with small root mean squared errors. I find strong evidence of the effects of variation in the slope factor on monetary policy variables and evidence for a reverse influence is not as strong. Forecasts with high-frequency data have higher predictive power for bonds with different maturities in both U.S. and Chinese markets. The inverted yield curve is a good indicator of macro risks in the U.S. financial market, but it has little correlation with economic recessions in the Chinese market.