Identification in linear dynamic panel data models

Neither the Dif(ference) moment conditions, see Arellano and Bond (1991), nor the Lev(el) moment conditions, see Arellano and Bover (1995) and Blundell and Bond (1998), identify the parameters of linear dynamic panel data models for all data generating processes for the initial observations that accord with them when the data is persistent. The combined Dif-Lev (Sys) moment conditions
do not always identify the parameters either when there are three time series observations but do so for larger numbers of time series observations. Thus the Sys moment conditions always identify the parameters when there are more than three time series observations. To determine the optimal GMM procedure for analyzing the parameters in linear dynamic panel data models, we construct
the power envelope and find that the KLM statistic from Kleibergen (2005) maximizes the rejection frequency under the worst case alternative hypothesis whilst always being size correct under the null hypothesis.