《Efficient Estimation for Structure Break Model》

浙江大学数学科学学院张荣茂教授莅临我院指导

2017年12月25日上午10点，应广州大学经济与统计学院和岭南统计科学研究中心的邀请，浙江大学数学科学学院张荣茂教授在行政东前座412会议室作了题为“Efficient Estimation for Structure Break Model”的讲座——暨“羊城讲坛”第二十七讲，旨在进一步提高年轻学者及研究生对研究的理解。此次讲座由李元教授主持，相关专业的师生参加了此次讲座。本报告将引入一种新的方法来估计具有多系统的结构断裂模型，并讨论它们的大样本性质。通过在一个sparsevariable选择环境中重构问题，提出了一种最小绝对收缩和选择算(LASSO)来估计一个具有未知数量的破断点的SBR模型，在那里可以有效地进行计算。我们证明了断点的个数和断点的位置可以被一致估计。我们还建立了近最优断点的收敛速度。此外，还将讨论一种改进版本，该版本包含了组套索和逐步回归变量选择技术及其对空间变化的扩展。仿真研究和实际数据分析 以说明所提方法的性能。

 

摘要：Structure break (SBR) model has attracted considerable attention in diverse areas such as signal processing, biological sciences,econometrics, environmental sciences,finance, hydrology, physics and population dynamics. However, the estimation of SBR  model constitutes a difficult task. In this talk, we will introduce a novel approach for estimating structure break (SBR) models with multiple-regime and discuss their large sample properties. By reframing the problem in a sparsevariable selection context, the group least absolute shrinkage and selection operator(LASSO) is proposed to estimate an SBR model with an unknown number of break locations,where the computation can be performed efficiently. We show that the number of break points and the location of the break points can be consistently estimated. We also establish near optimal the convergence rate of the break points. An improved version that incorporates group LASSO and stepwise regression variable selection technique and its extension to spatial change will also be discussed. Simulation studies and real data analysis are conducted to illustrate the performance of the proposed method.