The Stochastic Factor-Augmented Nelson-Siegel Model

Accurately modeling and forecasting the term structure of interest rates is relevant for both academics and practitioners in the industry. The dynamic Nelson-Siegel model is suitable for this. Although the amount of research on the in-sample fit of the Nelson-Siegel model and its extensions is substantial, the number of studies examining the out-of-sample performance of these models is relatively little, particularly for the nonlinear class of Nelson-Siegel models. For this reason, the focus of this thesis is twofold. First, I examine the predictive performance of various extensions in the Nelson-Siegel framework relative to the standard dynamic Nelson-Siegel model. Second, I study the differences between the extended Kalman filter and the unscented Kalman filter in the context of nonlinear Nelson- Siegel models. I find that the results are maturity- and subsample-dependent. The greatest gain in predictive accuracy is found for the stochastic factor augmented Nelson-Siegel model, for which the improvement over the standard model attains values of a 28% decrease in the RMSPE when the yields are relatively volatile. Furthermore, the findings indicate that the use of the unscented Kalman filter rather than the extended Kalman filter for fitting nonlinear Nelson-Siegel models is beneficial for both ends of the yield curve and could have a positive impact on the accuracy in the predictive framework in some cases.