Forecasting Interest Rates using the (Rotated) Dynamic Nelson-Siegel Model with Shifting Endpoints

This paper has the goal to use the papers van Dijk et al. (2013) and Nyholm (2017) to improve the forecasting accuracy by combining the ideas of the Dynamic Nelson-Siegel model, Rotated Dynamic Nelson-Siegel model and shifting endpoints specifications. Forecasting the government bond yields is of practical importance for, for example, monetary policymakers and investors. We use yield and macroeconomic data. As macroeconomic data, we have inflation and economic growth measured with the consumer price index and industrial production. We consider the Dynamic Nelson-Siegel(DNS) model, the Rotated Dynamic Nelson-Siegel(RDNS) model and Random Walk processes. We expand the RDNS with macro-economic variables and we take account of shifting endpoints for DNS and RDNS. Additionally, we investigate structural breaks. We find that for the short horizon and short maturities the RDNS model, integrated with macroeconomic variables forecasts interest rates more accurately. For the short and middle horizons with long maturity and for the long horizon the DNS model under shifting endpoints with exponential smoothing forecast interest rates more accurately.