In this study we look into various ways to introduce time-varying parameters in the dynamic Nelson-Siegel model. We consider three extensions: (i) a time-varying loading parameter, (ii) time-varying volatility and (iii) a time-varying unconditional mean. For the introduction of a time-varying unconditional mean we consider specifications with- and without macroeconomic information. Furthermore, we combine the three extensions into a more sophisticated model. We perform an in- and out-of-sample analysis applied to zero coupon US government bond yields. We find clear evidence that the introduction of a time-varying loading parameter and time-varying volatility improves the in-sample fit, whereas evidence for the introduction of a time-varying unconditional mean is less pronounced. Furthermore, we find evidence that combining extensions improves the fit even further. In terms of predictive performance we find that time variation in the loading parameter and in the unconditional mean without the use of macroeconomic information generally improve the forecast performance on the complete yield curve, whereas the extensions of time-varying volatility only improves for the short-end of the curve. In addition we find that combining extensions leads to a model superior to all models with one specific extension and that macroeconomic information is only beneficial in periods of high volatility.

Additional Metadata
Keywords Dynamic Nelson-Siegel, Extended Kalman Filter, Time-Varying Parameters, Macroeconomic Information
Thesis Advisor Dijk, D.J.C. van
Persistent URL
Series Econometrie
Buitenhuis, J. (Jos). (2017, July 14). Time-Varying Parameters in the Dynamic Nelson-Siegel Model. Econometrie. Retrieved from