Prediction of Term Structure with Potentially Misspecified Macro-Finance Models near the Zero Lower Bound

In this paper, we study the forecasting performances of the affine term structure model (ATSM) and the quadratic term structure model (QTSM) with macro-finance features under the zero interest rate policy of Japan. As both the two models can be potentially misspecified, we adopt the pptimal pooling prediction scheme following the recent work by Geweke and Amisano (2011). We find that the QTSM provides a more realistic statistical description when bond yields are close to the zero lower bound. The ATSM gives a good fit to the macroeconomic variables and bond yields simultaneously, however, it predicts a large probability of negative interest rates and hence is not appropriate for the forecasting of bond yields. The Markov-switching prediction pool dominates individual models as well as the static and dynamic pools. Our results suggest that both of the ATSM and QTSM macro-finance models are potentially misspecified and one should use a combination of the two models for the prediction of future bond yields during different time periods. Our analysis sheds light on the macro-finance modeling using US data amid the Federal Reserve’s zero interest rate policy since December 2008.