Detection of Change-Points in Nonparametric Regression
Detection of Change-Points in Nonparametric Regression
复制标题
非参数回归中变化点的检测
DOI：
--
复制
发表时间：
1993
期刊：
影响因子：
--
通讯作者：
P. Speckman
中科院分区：
文献类型：
--
作者：
P. Speckman
非参数回归模型考虑检测一个或多个变化点的问题。在这些模型中，变化点被认为是平均响应函数或其导数之一的不连续性。提出了一种使用基于核平滑的半参数变点估计量的检验统计量，并在无变化假设下推导了其渐近分布。应用到两个重要的情况，即响应函数在变化点处存在跳跃的情况和一阶导数存在变化或跳跃的情况。后一种情况模拟了平均响应中的尖点。
The problem of detecting one or more change-points is considered in non-parametric regression models. In these models, a change-point is deened to be a discontinuity in the mean response function or one of its derivatives. A test statistic is proposed which uses a semiparametric change-point estimator based on kernel smoothing, and its asymptotic distribution is derived under the hypothesis of no change. Application is made to two important cases, the case where there is a jump in the response function at the change-point and the case where there is a change or jump in the rst derivative. The latter case models a cusp in the mean response.