Bayesian Semiparametric Dynamic Nelson-Siegel Model

This paper proposes the Bayesian semiparametric dynamic Nelson-Siegel model, where the density of the yield curve factors and thereby the density of the yields are estimated along with other model parameters. This is accomplished by modeling the error distributions of the factors according to a Dirichlet process mixture. An efficient and computationally tractable algorithm is implemented to obtain Bayesian inference. The semiparametric structure of the factors enables us to capture various forms of non-normalities including fat tails, skewness and nonlinear dependence between factors using a unified approach. The potential of the proposed framework is examined using US bond yields data. The results show that the model can identify two different periods with distinct characteristics. While the relatively stable years of late 1980s and 1990s comprise the first period, the second period captures the years of severe recessions including the recessions of 1970s and 1980s and the recent recession of 2007-9 together with highly volatile periods of Federal Reserve’s monetary policy experiments in the first half of 1980s. Interestingly, results point out a nonlinear dependence structure between the factors contrasting existing evidence.