Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: Dynamic panel data with very short panels

From   "Sebastian van Baal" <>
To   <>
Subject   st: Dynamic panel data with very short panels
Date   Thu, 18 Feb 2010 17:31:03 +0100

Dear Stata List Users: 
I have a problem involving dynamic panel data and I hope that you can point
me in the right direction. My panels are very short and therefore, I am
unsure if I can use the typically recommended approaches. 
I am trying to estimate a model of the following type: 
y[it] = ay[i,t-1] + bx[it]
The index i corresponds to about 2,000 subject sessions (my data are
experimental) and t (= 1,2,...T) are discrete points in time. The horizon T
differs by subject session, so my panel is unbalanced. There is serial
correlation in the residuals within subject sessions. 
The model appears to be a standard problem often dealt with in the
literature on panel data analysis. Due to the lagged dependend variable on
the right-hand side, simple estimators are regarded as inconsistent, so an
alternative approach is needed. In the literature, IV-estimators (and
GMM-estimators) are mentioned. In Stata, the corresponding commands are
xtabond, xtdpdsys and xtdpd. 
However, my panels are very short: Most T's are 3 or less. Hence, the IV-
and GMM-estimators appear to be problematic, since with these, the first few
observations are lost because instruments are constructed based on lags. In
order to test my model, I need all observations -- because the beginning of
the process is what is most important. 
Did I missunderstand something? Can I still use, e.g., xtdpd to estimate the
model? Is there an alternative approach? 
I will be very grateful for any hints you can give me. 
Sebastian van Baal
University of Cologne, Germany

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index