[This is alternative, blind friendly, text version of the webpage.] Go to the grafic version.
Continue reading or go to the main menu. Other possibilities: Menu; Sections; Go to Search; Footer.
We evaluate forecasts of the term structure of Slovak government bond yields at horizons ranging from one to 36 months in the out-of-sample period from January 2009 through December 2016. We split the forecasting exercise into two steps. First, we forecast the term structure of German government bond yields using a dynamic Nelson-Siegel (DNS) model and a three-factor model of the yield curve with economically motivated factors which we call a “trend–cycle” model. In the second step, we forecast the term structure of spreads to German government bond yields. This “sum-of-the-parts” forecasting model delivers forecasts that outperform the random walk benchmark at most forecasting horizons. A combination of the trend-cycle model for forecasting German government bond yields and an AR(1) model for spreads delivers the best results at horizons up to one year. At longer horizons, the trend-cycle model combined with the random walk spread forecast is the best forecasting model. The main source of the improvement in forecasting performance relative to the random walk benchmark is the negative correlation between credit spread forecast errors and forecast errors on German government bond yields, i.e. negative surprises to the future path of the policy rate tend to coincide with positive surprises to sovereign credit spreads. We show that the forecasting performance of the dynamic Nelson-Siegel model deteriorates significantly in periods when short-term interest rates are at or close to the effective lower bound.
End of content.
You are here:
Continue the menu:
Continue reading or go to the main menu. Other possibilities: Conetent; Sections; Go to Search; Footer.