Abstract: With the deep development of Chinese interest rate marketization and the diversification of the interest rate derivatives,how to predict interest rate effectively,as one of the kernel variables in securities pricing, has become a key work. The no-arbitrage DNS model is developed on the basis of DNS model, and theoretical studies show that it is a constrained affine Gauss process. With the term structure of yields implied in monthly bonds price from 2005 to 2012 in Shanghai Stock Exchange (SSE), the no-arbitrage DNS model and the DNS model are studied empirically using the method of quasi-maximum likelihood. Are the results show that the no-arbitrage DNS model keeps the no arbitrage condition of affine term structure model and good empirical results of DNS model; and also it keeps the forecast advantage of DNS model, because prediction errors of all term structures in both two models are within 1.5 basis points; and also the characteristics of constant adjustment item show that the value is increasing with the extension of maturities.

Key words: no-arbitrage DNS model, DNS model, term structure, affine factor model