Re: st: Quantile regression for panel data

[Gordon Hughes <G.A.Hughes@ed.ac.uk>]

The nature of your data is not entirely clear, but it seems as though 
what you have is a number of repeated cross-sections rather than a 
panel, though that should not cause too much difficulty.  Have you 
computed the period growth rates from data on, say, output in 2000, 
2002, 2004, ...?  In which case do you have (a) 4 growth rates (i.e. 
5 years of output data), or (b) only 3 growth rates (4 years of 
output data) for each firm?

You are going to lose at least 1 observation for each firm due to the 
lagged dependent variable.  In case (b) with 4 years of output data, 
that should leave you with a maximum of 2 observations for each firm 
assuming no missing data, whereas case (a) leaves you with 3 
observations for each firm assuming no missing data.

Next, how much missing data is there and what do you mean by 
including dummies to deal with fixed effects - is this a dummy for 
each firm?  In that case you would really use up a lot of degrees of freedom.

Your description implies that you have either ~ 5750 or ~ 4600 firms 
in your sample.  This would be a sizeable cross section if there 
isn't too much missing data, so maybe you should start by treating it 
as one or more cross sections and not worry about the repeated 
observations until you understand your data better.  Quantile 
regression ought to be able to handle a sample of this nature without 
difficulty (though bootstrapping the errors may be slow).

Gordon Hughes
g.a.hughes@ed.ac.uk

=========================

Dear Statalist,

I would like to estimate the growth rate of firms at different points in
time (4 points in an interval of 9 years). I have a pseudo-panel data set
with 23000 observations.
I've trying with quantile regression techniques (qreg and sqreg). I
included some dummy to lead with fiexd effects. My equation is:

growth= b1 b2 b3 growth(t-1) dummy1 dummy2 year

The problems are that many observations are dropped and few df remain. Do
you think it can be due to outliers? Or is it possible that I have
clustering problems?

I am very grateful for your advice.

Thanks a lot

Maria

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/