Weißbach, Rafael Ponyatovskyy, Vladyslav Zimmermann, Guido
Year of Publication:
Technical Report, Universität Dortmund, SFB 475 Komplexitätsreduktion in Multivariaten Datenstrukturen 2006,50
Due to their status as "the" benchmark yield for the world's largest government bond market and its importance for US monetary policy, the interest in a "good" forecast of the constant maturity yield of the 10-year U.S. Treasury bond ("T-bond yields") is immense. This paper assesses three univariate time series models for forecasting the yield of T-bonds: It shows that a simple SETAR model proves to be superior to the random walk and an ARMA model. However, dividing the sample of bond yields, dating from 1962 to 2005, into a training sample and a test sample reveals the forecast to be biased. A new bias-corrected version is developed and forecasts for March 2005 to February 2006 are presented. In addition to point estimates forecast limits are also given.
T-bond times series 10-year yield TAR model bias-correction non-linear time series