线性回归模型参数变点的在线监测  

On-line monitoring parameter change in linear regression model

作　　者：陈占寿[1] 田铮[1,2] 丁明涛[1] 

机构地区：[1]西北工业大学应用数学系,西安710129 [2]中国科学院遥感研究院国家遥感科学重点实验室,北京100101

出　　处：《系统工程理论与实践》2010年第6期1047-1054,共8页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(60972150;10926197);西北工业大学科技创新基金(2007KJ01033)

摘　　要：在线监测金融时间序列并尽早发现变点,对减小金融风险有重要的意义.通过引进一个窗宽参数,对线性回归模型系数变点提出了一种改进的在线监测方法,并将此方法进一步用于模型方差变点的监测.在无变点的原假设下给出了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.模拟结果表明新方法比已有的方法势高且平均运行长度短,最后通过两组股票价格变点监测的实例,进一步说明了该方法的有效性.On-line monitoring financial time series and quickly finding change point are important to reduce financial risk.An improved procedure was proposed to monitor coefficients change in linear regression model by introducing a bandwidth parameter,furthermore,the procedure was used to monitor variance change of the model.The null distributions of the monitoring statistics are obtained,and some critical values are tabulated.Under the alternative hypothesis,the consistency of the procedures was also proved. Simulations indicate that the new procedures have higher power and shorter average run length than those in literatures.At last,the validity of the proposed procedures was demonstrated by empirical applications in two groups of stock data.

关 键 词：线性回归模型 变点监测 窗宽 平均运行长度 

分 类 号：O212.1[理学—概率论与数理统计]