Summary: A two-factor model of stochastic volatility is established. A two-stage semi-parameter method is applied to estimate constant coefficient parameters of this model. Moreover, kernel estimator method is developed to estimate the long-term mean value function, by this method the consistency of the two-stage method and the asymptotic normality of parameters are obtained. The empirical results show that the likelihood function can be improved in the long-term mean value model rather than the constant coefficient model. Also, the model provides a good explanation for the effective policies implemented by the central bank and the government. Besides, the industries can use the above model for valuing interest-rate-derivative securities without increasing the dimension.