A dynamic Nelson–Siegel model for treasury bond yields with irregular mixed-frequency macroeconomic information

This study proposes the DNS-IMF-TVL-GARCH model incorporating mixed-frequency macroeconomic data, time-varying factor loadings, and GARCH-type volatility for Treasury bond yield dynamics. The model effectively captures complex nonlinear relationships between dynamic Nelson–Siegel factors and macroeconomic variables. Furthermore, we develop a new estimation strategy integrating the unscented Kalman filter with the bootstrap method to extract latent high-frequency macro factors while accounting for time-varying residual volatility. Empirical results demonstrate superior in-sample model fit and out-of-sample forecasting accuracy across horizons. Economic evaluations confirm that the proposed model outperforms other variants, delivering the highest Sharpe ratio and certainty equivalent return in dynamic portfolio allocation. Moreover, we document a strong correlation between the money supply (M2) and the time-varying loading parameter λ . We also find that foreign exchange noise is partially absorbed by the residual structure, thereby improving the model’s capacity to capture bond yield dynamics.