Modeling and Forecasting the Brazilian Term Structure of Interest Rates by an Extended Nelson-siegel Class of Models: a Quantile Autoregression Approach

Introducing a five factor more flexible model this paper verifies the in-sample fitting and the out-of-sample forecasting performance of several extensions of the Nelson and Siegel (1987) parametric model which was reinterpreted by Diebold and Li (2006). We used different rules for fixing the parameters λ that govern the models ́ exponential components shapes, and predictions were made for different time horizons using different methods. We highlight the Quantile Autoregressive – QAR. The results showed that the five factor model presents a much better in-sample fitting, specially in the short and long term maturities of the term structure. Despite this, a greater predictive power is not guaranteed. It is also shown that, depending on the forecasting horizon, different values of the parameters λ can be optimally fixed. We also conclude that the Brazilian term-structure forecasts performed by the QAR estimated in the median are much more accurate than those performed by the autoregressive methods based on mean-regressions. It shows the robustness of the quantile regression models.