Forecasting the term structure interest rate of government bond yields

In this paper, I introduce extensions on the dynamic NelsonSiegel (DNS) framework for forecasting the yield-curve. These extensions regard time-varying volatility and including macroeconomic variables in the estimation of the beta’s, which determine the forecasts of the yield-curve. I make forecast with all different models on different forecasting horizons, using both an extending out-of sample window and a moving out-of sample window. I assess the forecasts by their root mean squared error and I compare the forecasts of the different models against the DNS framework with the Diebold-Mariano test for equal prediction accuracy. The results show that the models which capture time-varying volatility produce better forecasts in almost all circumstances. Moreover, using a moving out-of sample window lead to better forecasts as well.