动态利率模型估计方法的一个实证检验

Empirical test of the estimation methods by dynamic models of the interest rate

在分析广义矩估计法和极大似然法原理和方法的基础上,采用上海证券交易所国债回购利率数据对这两种估计方法在动态利率模型估计上的实证效果进行检验.为使其检验结果具有可比性,运用蒙特卡罗法产生仿真回购利率路径,然后对利率模型参数进行二次估计,所得结果就可用于对比评价这两种估计方法.检验结果表明,两种估计方法都能够比较准确地估计长期平均利率θ值,同时都过高地估计均值回复参数;对于波动项参数σ2 和γ值,MLE法的估计均值和标准误差要优于GMM法;总体上MLE法的估计误差明显小于GMM法的误差.

By using the dynamic model of the interest rates, two different estimation methods based on Generalized Method of Moments and Maximum Likelihood Estimation were introduced to estimate the parameters of dynamic models from discretely sampled data. Both methods were evaluated over sets of data simulated by Monte-Carlo method and empirically tested over 4 set of daily data of the repurchasing rate in Shanghai Stock Exchange. The results show that both methods can estimate the long term of interest rate θ accurately and over-high mean-reversion parameter. For volatility parameters σ2 and γ, the mean and standard deviation of MLE are superior to that of GMM. In sum, MLE errors are less than GMM ones when being with dynamic models with diffusion process.