摘要
CUSUM型变点估计量中的调节参数,理论上一般假定其取值范围为(0,1),但在实际数据的变点估计时,不同的取值往往得到相异的结果.基于跳跃度变点模型,利用蒙特卡罗方法,研究了调节参数的取值对变点估计结果的影响.模拟结果发现:在跳跃度较大时,无论变点真实位置如何,变点估计值基本不受调节参数取值的影响;但在跳跃度较小时,调节参数的取值对变点估计结果有显著影响;特别在变点真实位置靠近序列其中任意一个端点时,调节参数取值为0.5时,估计的效果最好;当变点真实位置靠近中间时,调节参数的取值越小,估计效果越好.在模拟和实例分析基础上,提出了基于数据驱动的调节参数的选取方法,使得CUSUM型变点估计量更具稳健性.
Generally,the range of tuning parameters in CUSUM type change-point estimation statistic is assumed to be(0,1)in theory.But the different values of tuning parameters often lead to the different estimation results in application.Here Monte Carlo method was used to study the influence of tuning parameters on the change-point estimation based on the jump change-point model.It was found that when the jump is large,the change-point estimate is not affected by the value of tuning parameters no matter where the true location of the change-point is.However,the value of tuning parameter has a significant effect on the change-point estimate when the jump is small.Especially,when the true location of change-point is close to one of the two trails,best estimation is obtained with the tuning parameter at 0.5.When the true location of change-point is near the center of sequence,it was observed that the smaller the tuning parameter,the better the estimation.On the basis of simulation and applications,a data-driven method was proposed to select appropriate tuning parameters from a set of possible values,which makes the CUSUM type change-point estimator more robust.
作者
谭常春
江敏
TAN Changchun;JIANG Min(School of Economics,Hefei University of Technology,Hefei 230009,China)
基金
国家社会科学基金一般项目(16BTJ023)资助
关键词
变点
CUSUM型估计量
调节参数
change-point
cumulative sum estimator
tuning parameter
