Modelling and Forecasting the Yield Curve under Model uncertainty

This paper proposes a procedure to investigate the nature and persistence of the forces governing the yield curve and to use the extracted information for forecasting purposes. The latent factors of a model of the Nelson-Siegel type are directly linked to the maturity of the yields through the explicit description of the cross-sectional dynamics of the interest rates. The intertemporal dynamics of the factors is then modeled as driven by long-run forces giving rise to enduring effects, and by medium- and short-run forces producing transitory effects. These forces are re-constructed in real time with a dynamic filter whose embedded feedback control recursively corrects for model uncertainty, including additive and parameter uncertainty and possible equation misspecifications and approximations. This correction sensibly enhances the robustness of the estimates and the accuracy of the out-of-sample forecasts, both at short and long forecast horizons.