Abstract
In this article, we examine the usefulness of the bias-corrected first-difference (BCFD) estimator by Chowdhury (1987) from two angles: inference and testing. First, we compare the BCFD estimator with Bun and Carree's (2005) estimator and the GMM estimator in terms of accuracy of inference. Second, we propose to use the Hausman test based on these three estimators to test the null of no individual effects. Simulation results show that the BCFD estimator and the Hausman test based on the BCFD estimator perform better than the other two estimators.
Acknowledgements
I would like to thank Taku Yamamoto and Hiroaki Chigira for helpful comments. Financial support from the JSPS Fellowship is also gratefully acknowledge. All remaining errors are my own.
Notes
1 On these problems, see Blundell and Bond (Citation1998), Alvarez and Arellano (Citation2003), Bun and Kiviet (Citation2006) and Hayakawa (Citation2006a, Citationb, Citation2007).
2 Few studies have dealt with the issue of testing for the absence of individual effects in dynamic panel data models. To the best of our knowledge, only Holtz-Eakin (1988) addresses with this problem.
3
is Bun and Carree's estimator (Bun and Carree, Citation2005).
is a GMM estimator where the individual effects in the model are removed by the Helmert transformation and all past level instruments, i.e. (y
i
,1, … , y
i
,t−2), are used. A more detailed explanation of
is provided by Alvarez and Arellano (Citation2003).
4 Simulation results which show the presence of a bias in the estimation of the SE in are available from the author upon request.
5 We do not consider H bc and H gmm since their sizes are distorted.