摘要
POT极值模型参数的准确估计是计算金融资产回报厚尾分布市场风险的关键.由n阶概率加权矩得到参数的二项式回归估计,而将参数的零、一阶概率加权矩估计予以推广.极大似然估计中,将极大化似然函转化为二元函数无条件极值问题,其他参数估计方法的结果作为迭代的初始值,通过它们的似然函数值和极大似然函数值的比较以及迭代次数判断方法的优劣.实证研究表明:参数的零、一阶概率加权矩估计较接近于真值,随着阶数的提高,二项式回归参数估计的误差很大.参数的极大似然估计优于非线性回归估计优于零、一阶概率加权矩估计.在此基础上计算上证A股指数VaR值.
The POT Extreme model parameters accurate estimation is the key to calculatemarket risk of the financial property returns obeying the thick tail distribution. This paper obtains parameter binomial regression estimate by the n step probability-weighted moments, thus the parameter estimation of zero, one step probability-weighted moments has been promoted. In ML estimation, so making likelihood function maximal into the dual function unconditional extreme problems. To judge the method rank, other parameter estimation results as the initial value of iterative, We compare its likelihood function value to ML function value and the number of iterative. Empirical studies show： The parameters estimates of zero and one probability-weighted moments closer to true value, with the increased number of steps, binomial regression parameter estimation error great. The maximum likelihood estimator of parameters is superior nonlinear regression estimates than zero, one probability-weighted moments estimates. Calculate VaR of Shanghai A stock index on this foundation.
出处
《数学的实践与认识》
CSCD
北大核心
2008年第19期66-73,共8页
Mathematics in Practice and Theory
基金
惠州学院科研基金(C206.0107)
