Modeling Panel Data in The Presence of Autocorrelation, Heteroscedasticity and Collinearity: A Monte Carlo Study
No Thumbnail Available
Date
2015-06-04
Journal Title
Journal ISSN
Volume Title
Publisher
Library and Publications Committee, University of Ilorin, Nigeria
Abstract
In this work, panel data that were characterized by features of autocorrelation, heteroscedasticity and collinearity were modelled using four estimation methods: Pooling (OLS), First-Differenced (FD), Between (BTW) and Feasible Generalized Least Squares (FGLS). Panel data like other aspects of econometrics, exploits regression analysis as one of the statistical tools to formulate, illustrate and appraise models. The regression analysis requires some assumptions which, if violated, results to one problem or the other. In such situation, the Pooling method of estimation
which is the naive approach remains linear, unbiased and asymptotically normally distributed but might not be efficient as the estimates of the parameters might become indeterminate, the associated confidence intervals may be too wide and the standard errors might become infinitely large. Monte-Carlo studies were carried out at different sample sizes and time periods, varying degrees of heteroscedasticity and levels of autocorrelation and collinearity. The results from this work showed that in small sample situations, irrespective of number of time periods, FGLS is preferable when heteroscedasticity is severe regardless of levels of autocorrelation and
multicollinearity. But when heteroscedasticity is low or mild with moderate autocorrelation level, both FD and FGLS are preferred, while BTW performs better only when there is no autocorrelation and low degree of heteroscedasticity. However, in large samples with little time periods, both FD and BTW could be used when there is no autocorrelation and low degree of heteroscedasticity, while FGLS is preferred if otherwise.
Description
Keywords
Panel data, Autocorrelation, Collinearity, Heteroscedasticity
Citation
Garba et. al. (2015). Modeling Panel Data in The Presence of Autocorrelation, Heteroscedasticity and Collinearity: A Monte Carlo Study