Comparing The Forecasting Accuracy Of Interest Rate Term Structure

Interest rate term structure is a core concept, not only in the economics and finance theories, but also in the pricing of fixed-income products. It is the basics of the positive bond management, and it can help government to determine macro economy policy. The research on forecasting interest rate term structure just started these years. It is the central issue of researchers and practitioners. There is hardly any literatures on this issue using chinese data. This thesis tries to do some systematic research on forecasting chinese interest rate term structure, based on the summarization of the latest foreign research works, using models based on Nelson-Siegel method.We begin with the linear model of three NS latent variables to forecast. To deal with the lack of short-term bond, we use inter-bank market repo rate as a substitute of short-term bond, after a lot of comparision. It is the sub-optimal choice under the data-constraint. When using the NS model to get the latent variables time series, one must ensure theλis time-invariant, so a properλis critical to our model. We transform the model into state-space form, using kalman-filter to get the QMLE estimation ofλ. This practice differs from that of existed chinese literature, and it avoids the ad hoc problem,helping to improve the forecast performance.According to the out-of-sample forecast result, when the forecast step lengthens, AR(1) model begins to outperform the UVAR, RVAR, Random Walk models. And the superiority seems to be more obvious in forecasting short-term rate than the longer ones. In order to find out whether adding macro variables helps to enhance forecasting performance, we choose 6 macro variables,...