Specification analysis of interest rates factors : an international perspective.

Summary This thesis concerns the modeling of the dynamics of international yield curves with the consideration of several dependency channels. Using a new database of international sovereign rates, we observe that the criterion of explained variability, proposed in the literature, is not able to select a better combination of factors describing the joint dynamics of the yield curves. We propose a new method of factor sectioning based on the likelihood maximization of a linear Gaussian state-space model with common and local factors. The associated identification problem is solved in a novel way. By estimating different combinations of countries, we select two global and three local factors with predictive power for the macroeconomic variables (economic activity and inflation rate) in each economy considered. Our method also allows us to detect hidden factors in bond yields. They are not visible through a classical principal component analysis of bond yields and they contribute to the prediction of the inflation rate and the growth rate of industrial production.