Abstract:
We study the prediction of the German Treasury bond yields by using the dynamic Nelson-Siegel and dynamic Svensson models by extending Diebold and Li (2006) work. We try to improve the prediction by (i) using the relationship between macroeconomic and yield curve factors to explain the movement of the yield curve, and (ii) including an extra curvature factor (i.e. Svensson model) to provide higher flexibility for fitting the yield curve. The effect of the macroeconomic selection methods to the prediction is also studied. From this study, the cross relationship between macroeconomic and latent factors helps improve yield curve forecasts for medium horizons (6-12 months) at most maturities. At very short (1 month) and very long (60 months) forecast horizons, however, the models without macro variables are better. This indicates the time required for the changes in the macroeconomic variables to reflect in the yield curve movements, and how long the effect lasts. Focusing on the medium forecast horizons (6-12 months) where macroeconomic variables improve the forecast performance, we find that the method of selecting the macroeconomic variables is important and model dependent: the Nelson-Siegel model is better with the traditional approach and the Svensson model is better with the correlation-based approach. Comparing the two models, we find that the flexibility and statistical support from the Svensson model with the correlation-based approach leads to better forecasting results.
