Estimating ultra long-term interest rates with raise regression

Abstract Accurate estimation of ultra-long-term interest rates is essential for financial regulators, life insurance companies, and pension funds. The Nelson-Siegel model and its extension, the Svensson model, are widely used thanks to their parsimony and rich economic intuition. The level parameter in both models is a direct indicator of ultra-long-term rates. However, these models are subject to high nonlinearity when estimated as nonlinear models, or multicollinearity when estimated as linear models. As a result, estimated interest rates can be unstable, which undermines their practical use. In this paper, we employ raise regression to alleviate the estimation issue. Our results demonstrate superior accuracy compared to existing methods.