论文部分内容阅读
Nelson-Siegel利率期限结构模型指数部分的衰减参数通常根据Diebold&Li的两步法、遗传算法或非线性最小二乘法估计,这会导致拟合误差偏大或估计结果不稳定。本文提出用遗传算法与最小二乘法交叉迭代的方法来改进Diebold&Li两步法对Nelson-Siegel利率期限结构模型参数的估计,并与Diebold&Li两步法和遗传算法进行实证比较。实证结果表明,用改进的两步法不仅能提高样本内模型的拟合优度,还能降低样本外模型的定价误差,特别是对于较长期限的债券数据,改进的两步法的模型估计效果明显好于Diebold&Li两步法和遗传算法。
The decay parameters for the exponential part of the Nelson-Siegel interest rate structure model are usually estimated from Diebold & Li’s two-step, genetic algorithm or nonlinear least-squares method, which leads to large fitting errors or unstable estimation results. In this paper, genetic algorithm and least squares cross-iterative method is proposed to improve the Diebold & Li two-step method to estimate the Nelson-Siegel interest rate structure model parameters, and empirical comparison with Diebold & Li two-step method and genetic algorithm. The empirical results show that the improved two-step method can not only improve the goodness of fit of the in-sample model, but also reduce the pricing error of the out-of-sample model. Especially for longer-term bond data, the improved two-step model estimation The effect is obviously better than Diebold & Li two-step method and genetic algorithm.