Nelson-Siegel Model and Multicollinearity

Nelson-Siegel model is used for important decision making about monetary policy, among others. Numerous researchers are aware of the potential multicollinearity in the Nelson-Siegel model that can lead to unstable estimations and signs contrary to expectations if the model is estimated by ordinary least squares (OLS). Some authors have proposed fixing the shape parameter to avoid multicollinearity problems, but that change can lead to extremely smooth time series. On the other hand, other authors have proposed estimating the Nelson-Siegel model with the ridge regression that is traditionally applied to estimate models with collinearity as an alternative to OLS. For a correct application of the ridge regression, data should be standardized which can make difficult the interpretation of the estimated model. Also, the inference in ridge regression is controversial. Alternatively, this work proposes the application of the raise regression to mitigate multicollinearity in Nelson-Siegel model. This methodology can be applied with the original data and maintains the global characteristics of the original model. The contribution of this paper is illustrated with two different empirical examples for American and European treasuries.