5

Given a data set with $N=2634$ and $T=92$, I want to estimate a dynamic model. My first though was to use a classic System GMM estimator, however digging through the literature it turned out that there might be problems with biased estimators when using GMM on data sets with large $T$. I read a couple of papers, mostly based on Alvarez and Arrelano (2003) and most claim that it is the relative size that matters, i.e., $T/N$. If $T/N \to 0$, then the bias should be negligible. However, I also found sources that claimed that the absolute length of $T$ was also relevant.

Also, it seems that there is no shortage of analyses on the asymptotic properties of different estimators for dynamic panel models with large $N$ and $T$, but the results seem contradictive and not coherent.

So my question is twofold:

  1. is it true that all what matters is $T$ in relation to $N$, or does the absolute length of $T$ matter as well?
  2. which estimator shows the best performance in practice for panels with large $N$ and $T$?
Stephan Kolassa
  • 95,027
  • 13
  • 197
  • 357
Gerald
  • 51
  • 3

1 Answers1

-2

Large T, Large N----> Pooled Mean Group, Pesarin and Shin..Check it out.

ibim3
  • 1
  • Welcome to Crossvalidated. Thank you for your answer. Please edit your first answer and delete the second one. – Ferdi Jan 03 '17 at 17:07
  • Interesting choice of papers...you state that, with unbalanced panel data, the Pesaran and Shin approach won't work. In that case, are there any options? – Mike Hunter Jan 03 '17 at 17:08
  • 3
    This is being automatically flagged as low quality, probably because it is so short. At present it is more of a comment than an answer by our standards. Can you expand on it? We can also turn it into a comment. – gung - Reinstate Monica Jan 03 '17 at 17:17