Please use this identifier to cite or link to this item: https://hdl.handle.net/10419/62912 
Year of Publication: 
2004
Series/Report no.: 
Working Paper No. 523
Publisher: 
Queen Mary University of London, Department of Economics, London
Abstract: 
This paper considers the issue of bootstrap resampling in panel datasets. The availability of datasets with large temporal and cross sectional dimensions suggests the possibility of new resampling schemes. We suggest one possibility which has not been widely explored in the literature. It amounts to constructing bootstrap samples by resampling whole cross sectional units with replacement. In cases where the data do not exhibit cross sectional dependence but exhibit temporal dependence, such a resampling scheme is of great interest as it allows the application of i.i.d. bootstrap resampling rather than block bootstrap resampling. It is well known that the former enables superior approximation to distributions of statistics compared to the latter. We prove that the bootstrap based on cross sectional resampling provides asymptotic refinements. A Monte Carlo study illustrates the superior properties of the new resampling scheme compared to the block bootstrap.
Subjects: 
Bootstrap, Panel data
JEL: 
C32
C33
Document Type: 
Working Paper

Files in This Item:
File
Size
279.83 kB





Items in EconStor are protected by copyright, with all rights reserved, unless otherwise indicated.