汪红霞,罗学洪,林金官,唐 星.误差分布未知下时空模型的自适应非参数估计[J].数学年刊A辑,2021,42(2):125~148
误差分布未知下时空模型的自适应非参数估计
Adaptive Nonparametric Estimation of Spatio-Temporal Models with Unknown Error Distributions
Received:September 07, 2020  Revised:November 24, 2020
DOI:10.16205/j.cnki.cama.2021.0011
中文关键词:  时空模型  核密度估计  局部多项式方法  局部极大似然方法
英文关键词:Spatio-temporal model  Kernel density estimation, Local polynomial method  Local maximum likelihood method
基金项目:国家自然科学基金(No.,11831008, No.,11971235),国家社会科学基金(No.,17CTJ016)和江苏省研究生科研与实践创新计划项目(No.,KYCX19_1526, No.,KYCX20_1677)
Author NameAffiliation
WANG Hongxia School of Statistics and Mathematics, Nanjing Audit University, Nanjing 211815, China. 
LUO Xuehong School of Statistics and Mathematics, Nanjing Audit University, Nanjing 211815, China. 
LIN Jinguan Corresponding author. School of Statistics and Mathematics, Nanjing Audit University, Nanjing 211815, China. 
TANG Xing School of Statistics and Mathematics, Nanjing Audit University, Nanjing 211815, China. 
Hits: 1034
Download times: 1438
中文摘要:
      极大似然估计作为参数估计中较为有效的一种估计方法,在误差分布未知下无法进行, 另一方面, 时空数据经常含有奇异点或来自重尾分布,此时基于最小二乘的估计方法效果欠佳.考虑时空异质性和相关性,针对误差分布未知的时空模型,本文提出基于核密度估计的自适应非参数估计方法.在较弱的条件下证明了该估计量和已知误差分布下的局部极大似然估计量是渐近等效,比基于最小二乘的局部多项式估计量有效. 模拟和实证都验证了该方法对于有限样本的有效性, 尤其奇异点的存在,该方法在边界的拟合效果显著优于基于最小二乘的方法.
英文摘要:
      Maximum likelihood is an effective method for parameter estimation, but it can’t work with unknown error distribution. On the other hand, when the data contain outliers or come from population with heavy-tailed distributions, which appear very often in spatiotemporal data, the estimation methods based on least-squares method will not perform well.Considering the spatio-temporal heterogeneity and correlation, in this paper, the authors propose an adaptive nonparametric estimator for spatio-temporal model with unknown error distributions based on kernel density estimation. Asymptotic theory properties show that the proposed estimator is as asymptotically efficient as local maximum likelihood estimator with known error distributions, and more efficient than local polynomial estimator based on least-square method under some mild conditions. Simulation and case study are conducted to investigate the finite-sample performance of our procedure, especially for the existence of singularities, for boundary points, the performance of our procedure is significantly better than the method based on least squares.
View Full Text  View/Add Comment  Download reader
Close

Organizer:The Ministry of Education of China Sponsor:Fudan University Address:220 Handan Road, Fudan University, Shanghai, China E-mail:edcam@fudan.edu.cn
Designed by Beijing E-Tiller Co.,Ltd.