Nelson-Siegel model (NS model) and 2 extended NS models were compared by using daily interbank government bond data. Based on the grouping of bonds according to the residual term to maturity, the empirical research pr...Nelson-Siegel model (NS model) and 2 extended NS models were compared by using daily interbank government bond data. Based on the grouping of bonds according to the residual term to maturity, the empirical research proceeded with in-sample and out-of-sample tests. The results show that the 3 models are almost equivalent in estimating interbank term structure of interest rates. Within the term to maturities between 0 and 7 years, the gap of the absolute errors of the 3 models between in-sample and out-of-sample is smaller than 0.2 Yuan, and the absolute values of the in-sample and out-of-sample errors are smaller than 0.1 Yuan, so the estimation is credible. Within the term to maturities between 7 and 20 years, the gap of the absolute errors of the 3 models between in-sample and out-of-sample is larger than 0.4 Yuan, and the absolute values of the in-sample and out-of-sample errors are larger than 1.0 Yuan, so the estimation is incredible.展开更多
文摘Nelson-Siegel model (NS model) and 2 extended NS models were compared by using daily interbank government bond data. Based on the grouping of bonds according to the residual term to maturity, the empirical research proceeded with in-sample and out-of-sample tests. The results show that the 3 models are almost equivalent in estimating interbank term structure of interest rates. Within the term to maturities between 0 and 7 years, the gap of the absolute errors of the 3 models between in-sample and out-of-sample is smaller than 0.2 Yuan, and the absolute values of the in-sample and out-of-sample errors are smaller than 0.1 Yuan, so the estimation is credible. Within the term to maturities between 7 and 20 years, the gap of the absolute errors of the 3 models between in-sample and out-of-sample is larger than 0.4 Yuan, and the absolute values of the in-sample and out-of-sample errors are larger than 1.0 Yuan, so the estimation is incredible.
