时间序列的局部影响分析

Local influence in time series analysis

目的研究时间序列模型中一次性探测所有异常点和强影响的方法。方法局部影响分析方法。结果时间序列模型中各数据点之间存在着一定的相关结构,这种相关结构使得异常点和强影响点产生的机理及相应的分析变得复杂。克服了数据删除对时间序列样本数据相依性的破坏和忽略。得出时间序列模型影响曲率的具体计算公式,从而可以一次性探测出所有的强影响点。最后给出了具体的数值实例,说明了文中结论的有效性。结论此方法可以一次性探测出所有的强影响点,与数据删除法相比,大大简化了计算量。

Aim To find an approach to detect all of the influence points at one time in time series model.Methods The local influence analysis of tench.Results The delete-points diagnostics procedure is not suitable for time series model due to the complicated dependence structure in the data.The curvature formula of local influenle can help us to detect all of the influence points in one time.In the end,a numerical example is given to illustrate the effectiveness of the conclusion.Conclusion This method can detect all strong influence points at one time,compared with data deleted,it greatly simplifyies the computation.