A Dynamic Estimation Of Chinese Term Structure Of Interest Rate

Finding a accurate, reasonable and effective method, which can dynamically estimates and forecast term structure of interest rates by market price information of Treasuries, is the foundational work for asset pricing, interest rate risk management and the portfolio allocation; besides, it can provide information on the operation of monetary policy for the central banks and make an important reference for the national macro-control targets.The traditional models of term structure of interest rates have their own weakness. The Static model is estimated based on a point of time, which does not reflect the dynamic changes of term structure; while the Dynamic model can reflect dynamic changes with strong theoretical foundation, but its efficiency of forecasting is not satisfactory. Diebold and Li (2006) proposed the dynamic Nelson-Siegel (DNS) model, which was AR(1) process based on Nelson-Siegel model. Now DNS model is widely used by scholars from various countries. Their experience shows that the DNS model forecast well and the macroeconomic variables can be added for macro-financial research. However, the DNS model is not well grounded in theory.Christensen, Diebold and Rudebusch (2007,2009) combined the DNS dynamic model and the no-arbitrage model, and proposed a series of hybrid models based on NS, which called the dynamic arbitrage-free NS models, such as AFDNS, AFGNS and so on. These combine the best of both models. They maintain the arbitrage-free theoretical restrictions of the canonical affine models, which make it forecast well and can be easily estimated. These were well proved by empirical data of U.S. Treasury.In this paper, we use arbitrage-free models of the dynamic NS, such as AFDNS and AFGNS models to estimate the term structure of interest rates in China. Then we compared the capability of estimation and prediction with these models based on the dynamic Nelson-Siegel hybrid model. We also compared the efficiency of estimation and prediction with arbitrage-free models and the DNS that rule out opportunities for riskless arbitrage. And we finally find the term structure model that fits for China.Using monthly data (January 2002-March 2010) from the inter-bank Treasury market of China, we compared AFDNS, AFGNS and DNS through quantitative analysis techniques. We found that by adding a no-arbitrage restriction, the out of sample of forecast power in short-term (1 month and 3 months) and in sample fitted capability of the AFDNS model is stronger than DNS model that without no-arbitrage conditions. The five factor AFGNS model has a better in-sample dynamic estimation capability than the other two models. And its prediction capability is better than tDNS and AFDNS in general, especially in 6 months and 12 months forecast.This thesis is divided into six chapters. Chapter 1 is the overview of the meaning and content of this article. The dynamic estimation of term structure is crucial for the pricing of asset and their derivatives; forecasting the interest rate; portfolio allocation; and guiding in the macroeconomic.Chapter 2 is an overview of available relevant literature. Sorting out the methods in the term structure of interest rate estimating, we have a brief review of the basic bootstrapping, static model, dynamic model and the hybrid model and then analyze the advantages and disadvantages of each model.Chapter 3 is the introduction of the hybrid models and the estimation method that we will use, such as DNS, AFDNS, AFGNS model. These models will be unified in the framework of the state space model that can be estimated and forecasted with the Kalman filter.In chapter 4, we dynamically estimated the hybrid models using Kalman filter. First, we described the data we chose; second, bootstrapped spot rates using Fama-Bliss; third, described initial value selection method; finally analyzed and compared the estimation efficiency of models, finding AFGNS has the highest precision in estimating the term structure of China.Chapter 5 is the comparison of these models' sample forecasting abilities. Our empirical data show that the NS models with no-arbitrage condition take a better performance on short-term forecasting. And prediction accuracy of the AFGNS model with five factors is above AFDNS model with three factors.Chapter 6 is the summary of whole thesis. Firstly, summarize the results three former chapters of the study, and put forward our recommendations; then presents the shortcomings and improved direction of this paper.The innovation of this paper is:firstly, it is the first time to join a dynamic no-arbitrage conditions to NS model for dynamic estimating term structure of interest rates of China; secondly, it is the first time to use five-factor AFGNS model to predict the term structure of interest rates, and compare the estimation and forecasting capabilities of the NS models with and without no-arbitrage conditions; thirdly, estimates and forecast term structure with these three models, using one-step state space model Kalman filter method.