Smooth Dynamic Factor Analysis with an Application to the U.S. Term Structure of Interest Rates

We consider the dynamic factor model and show how smoothness restrictions can be imposed on the factor loadings. Cubic spline functions are used to introduce smoothness in factor loadings. We develop statistical procedures based on Wald, Lagrange multiplier and likelihood ratio tests for this purpose. A Monte Carlo study is presented to show that our procedures are successful in identifying smooth loading structures from small sample panels. We illustrate the methodology by analyzing the U.S. term structure of interest rates. An empirical study is carried out using a monthly time series panel of unsmoothed Fama-Bliss zero yields for treasuries of different maturities between 1970 and 2009. Dynamic factor models with and without smooth loadings are compared with dynamic models based on Nelson-Siegel and cubic spline yield curves. All models can be regarded as special cases of the dynamic factor model. We carry out statistical hypothesis tests and compare information criteria to verify whether the restrictions imposed by the models are supported by the data. Out-of-sample forecast evidence is also given. Our main conclusion is that smoothness restrictions on loadings of the dynamic factor model for the term structure can be supported by our panel of U.S. interest rates and can lead to more accurate forecasts.