摘要
为尽早发现线性回归模型系数变点,文章提出了两种最小二乘残差平方和的比值型在线监测统计量,在无变点的原假设下给出了两个监测统计量的渐近分布及经验临界值表,在备择假设下证明了该方法的一致性.模拟结果显示该方法有较高的检验势和稍长的平均运行长度,但仍表明了该方法的稳健性.
Based on the least squares residues sum of squares,the paper propose two ratio-type detectors to sequentially monitor coefficient change quickly in linear regression model after a training period.It investigates the limit distribution of two statistics under the null hypothesis and the consistency under alternative assumption.Some critical values are tabulated.Simulations demonstrate that the proposed procedures have higher powers but longer ARLs.
作者
秦瑞兵
孙丽
宋冠仪
QIN Rui-bing;SUN Li;SONG Guan-yi(School of Mathematical Science,Shanxi University,Taiyuan 030006,China;School of Mathematics,Shandong University,Ji′nan 250100,China)
出处
《陕西科技大学学报》
CAS
2020年第1期175-179,共5页
Journal of Shaanxi University of Science & Technology
基金
国家自然科学基金项目(61807022)
山西省自然科学基金项目(201801D121005)
青海省自然科学基金项目(2019-ZJ-920)
关键词
变点
线性模型
在线监测
平均运行长度
change point
linear models
sequential tests
average run length
