Further results on bias in dynamic unbalanced panel data models with an application to firm R&D investment

Abstract

This article extends the LSDV bias-corrected estimator in (Bun and Carree, 2005) to unbalanced panels and discusses the analytic method of obtaining the solution. Using a Monte Carlo approach the article compares the performance of this estimator with three other available techniques for dynamic panel data models. Simulation reveals that LSDV-bc estimator is a good choice except for samples with small T, where it may be unpractical. The methodology is applied to examine the impact of internal and external R&D on labour productivity in an unbalanced panel of innovating firms.

Acknowledgement

I thank Martin Carree and for useful comments.

Notes

¹ We implemented a Fortran code for the LSDV-bc estimator, available upon request. For the additive LSDV bias corrected estimator we used – xtlsdvc – module for Stata discussed in Bruno (2005a,b) and for GMM routine -- xtabond2 -- written by David Roodman, Center for Global Development, Washington, DC. To generate the data, we used Stata 9.0 program – xtarsim – developed by G. Bruno, and described in Bruno, 2005a,b. We performed 10 000 replications with a fixed seed.

² GMM results are from the two-step estimator, which is more efficient than the one-step. The two-step estimates of the SEs tend to be downward biased (Arellano and Bond, 1991; Blundell and Bond, 1998). The SEs are corrected via a finite-sample correction to the two-step covariance matrix derived by Windmeijer (2005).