Comparison of Optimization Algorithms in Term Structure Estimation

Fitting the term structure is associated with multiple possible issuesirrespective of the method used: some restrict the problem too much, someare far from being theoretically based and some involve nonlinear optimizationwith no guarantee of returning a global optimum and possibility of taking a longtime to compute without certainty about quality of results. In this paper I usevarious algorithms, starting points generation methods and two stopping rules tofind optimal parameters of Nelson-Siegel and Nelson-Siegel-Svenssonmodels and compare performance of models and algorithms in terms of accuracyand optimization time. In short the results are that for chosen algorithms startingpoints choice mostly doesn’t matter though one of the used method used a goodguess, additionally all algorithms perform with similar accuracy and a goodstopping rule greatly increases speed by sacrificing accuracy a little