(The Estimation of Chinese Term Structure of Interest Rate and Treasury Bonds Pricing)

Abstract

Chinese Abstract: 基于上交所国债数据，沿袭Bliss（1997）的研究方法，并加以改进和扩展，本文考察了五种静态利率期限结构模型在样本内拟合和样本外预测上的表现，与Bliss（1997）的结果不同，我们发现三次平滑样条方法更适合用来估计中国上交所国债的利率期限结构。在此基础上，通过对各模型的拟合误差和预测误差的分析，我们发现，利率期限结构并不是决定国债定价的唯一系统性因素，上交所国债的定价还需考虑国债的其他特征因素，投资者在交易国债时会将息票金额的大小和债券价格的溢价程度考虑到定价中来,这一点与Bliss（1997）的结论相同。作为本文结论的一个应用以及对Bliss（1997）方法的扩展，我们发现息票金额低的投资组合和溢价因子高的投资组合分别高于息票金额高的投资组合和溢价因子低的投资组合，本文的结论对国债投资有一定的指导意义。English Abstract: The term structure of interest rate is a concept central to economic and financial theory and the pricing of interest rate contingent claims. An estimate of the term structure of interest rate is the necessary starting point for applying them to pricing and hedging interest rate derivatives. Since September 06, 2013 when treasury bond futures was launched by China Financial Futures Exchange, it had become more important to learn about the relation between the estimate of the term structure and the treasury bond pricing which is crucial to the pricing of treasury bond futures. To study their relationship, using Treasury bonds data from Shanghai Stock Exchange and methods proposed by Bliss (1997), this paper first compares five distinct models for estimating the term structure in China. Combining in-sample performance and out-of-sample performance, we find that the smoothed cubic spline method does best overall, which is different from Bliss (1997) who studied the USA treasury data.However, is the term structure of interest rate the only factor that determines the pricing of treasury bonds? To exploit this question, we examine the behavior of fitted-price errors of individual bonds from period to the next and across different models. Specifically, we have done three following studies:In the first place, we test the transition matrix for fitted-price errors classified as positive, zero, or negative. If the errors are indeed noise, the transition probabilities should reflect the frequency of positive, zero, and negative errors in the overall population. In this test, we find that the fitted-price errors for specific bonds are correlated through time.In the second place, we test the coincidence matrices of positive, zero, and negative errors across estimation models, that is, we examine the probability that a positive (zero, or negative) fitted-price error produced by one model will also have a positive (zero, or negative) fitted-price error under the other model. In this test, we find that the fitted-price errors are related across models.In the third place, we regress the fitted-price errors against variables that may be relevant to the bond pricing process. In this test, we identify coupon rates and Price Premium might be relevant to the pricing of Treasury bonds.All the three tests have the same results with Bliss (1997), meaning that the term structure of interest rate is not the only pricing element in the price of treasury bond market in China.Finally, as an application of our finding and an extension to the Bliss (1997), we exploit its implication for bond portfolio investment. We find that small-coupon portfolio and high-price-premium portfolio have better return than large-coupon portfolio and low-price-premium portfolio respectively.