Calibration of the Nelson-Siegel (NS) model to adjust to sovereign yield curves has been found to be problematic since the model exhibits a correlation between its factors, and generates objective functions with local multiple optima. These problems are often disregarded in the Colombian market, enough importance is not being given to the calibration process. This study aims to evaluate two non-gradient based methods for solving the non-linear least squares problem: the differential evolution metaheuristic and an incremental search procedure on the non-linear parameters. Comparisons of the results are made in terms of the achieved fit, consistency (for the metaheuristic) and the shapes obtained for the curves. The same procedure is carried out on the Nelson-Siegel-Svensson (NSS) model, evaluating the advisability of its use in the local market.

Calibration of the Nelson-Siegel (NS) model to adjust to sovereign yield curves has been found to be problematic since the model exhibits a correlation between its factors, and generates objective functions with local multiple optima. These problems are often disregarded in the Colombian market, enough importance is not being given to the calibration process. This study aims to evaluate two non-gradient based methods for solving the non-linear least squares problem: the differential evolution metaheuristic and an incremental search procedure on the non-linear parameters. Comparisons of the results are made in terms of the achieved fit, consistency (for the metaheuristic) and the shapes obtained for the curves. The same procedure is carried out on the Nelson-Siegel-Svensson (NSS) model, evaluating the advisability of its use in the local market. Doi: https://doi.org/10.1016/j.jefas.2016.0...