Advanced

Asymptotic Collinearity in CCE Estimation of Interactive Effects Models

Westerlund, Joakim LU and Petrova, Yana LU (2017) In Economic Modelling
Abstract
Researchers sometime fall into the dummy variable trap. A typical scenario in panel data is when wanting to estimate the effect of a regressor that is time invariant, such as sex or race, and accidentally including cross-section specific fixed effects. The problem here is that the fixed effects and the regressor are collinear, which causes the resulting pooled least squares estimator to break down. In interactive effects models such breakdowns can occur even if the regressors are not time invariant. The reason is that the interactive effects are flexible enough to generate a wide range of behaviours that are likely to be shared by the regressors. The current paper considers the challenging case when some of the regressors are... (More)
Researchers sometime fall into the dummy variable trap. A typical scenario in panel data is when wanting to estimate the effect of a regressor that is time invariant, such as sex or race, and accidentally including cross-section specific fixed effects. The problem here is that the fixed effects and the regressor are collinear, which causes the resulting pooled least squares estimator to break down. In interactive effects models such breakdowns can occur even if the regressors are not time invariant. The reason is that the interactive effects are flexible enough to generate a wide range of behaviours that are likely to be shared by the regressors. The current paper considers the challenging case when some of the regressors are asymptotically collinear with the interactive effects. The relevant asymptotic theory is developed and tested in small samples using both simulated and real data. (Less)
Please use this url to cite or link to this publication:
author
organization
publishing date
type
Contribution to journal
publication status
in press
subject
keywords
CCE estimation, Interactive effects, Asymptotic collinearity, Singular signal matrix, C12, C13, C33, C36
in
Economic Modelling
publisher
Elsevier
external identifiers
  • scopus:85027494176
ISSN
0264-9993
DOI
10.1016/j.econmod.2017.07.023
language
English
LU publication?
yes
id
114ccde1-df0e-460c-8e71-14d43fbfbcc4
date added to LUP
2017-09-04 06:52:15
date last changed
2017-09-17 09:52:49
@article{114ccde1-df0e-460c-8e71-14d43fbfbcc4,
  abstract     = {Researchers sometime fall into the dummy variable trap. A typical scenario in panel data is when wanting to estimate the effect of a regressor that is time invariant, such as sex or race, and accidentally including cross-section specific fixed effects. The problem here is that the fixed effects and the regressor are collinear, which causes the resulting pooled least squares estimator to break down. In interactive effects models such breakdowns can occur even if the regressors are not time invariant. The reason is that the interactive effects are flexible enough to generate a wide range of behaviours that are likely to be shared by the regressors. The current paper considers the challenging case when some of the regressors are asymptotically collinear with the interactive effects. The relevant asymptotic theory is developed and tested in small samples using both simulated and real data.},
  author       = {Westerlund, Joakim and Petrova, Yana},
  issn         = {0264-9993},
  keyword      = {CCE estimation,Interactive effects,Asymptotic collinearity,Singular signal matrix,C12,C13,C33,C36},
  language     = {eng},
  publisher    = {Elsevier},
  series       = {Economic Modelling},
  title        = {Asymptotic Collinearity in CCE Estimation of Interactive Effects Models},
  url          = {http://dx.doi.org/10.1016/j.econmod.2017.07.023},
  year         = {2017},
}