Econometrics in Panel data?

[beliavsky at aol.com]

Cameron Laird wrote:
> In article <mub6j3-sne.ln1 at lairds.us>, I counseled:
> >In article <44606dd5$0$3285$5a62ac22 at per-qv1-newsreader-01.iinet.net.au>,
> >DeepBlue  <123 at 321.com> wrote:
> >>so are you saying that Python is not an appropriate language for doing
> >>econometrics stuff?
> >>
> >>
> >>Dennis Lee Bieber wrote:
> >>> On Tue, 09 May 2006 05:58:10 +0800, DeepBlue <123 at 321.com> declaimed the
> >>> following in comp.lang.python:
> >>>
> >>>> Hi all,
> >>>>
> >>>> I am new to Python. Just wondering can Python able to do econometric
> >>>> regression in either Time-series or pooled (panel) data? As well as test
> >>>> for hetero, autocorrelation, or endogeneity?
> 			.
> 			.
> 			.
> >There is not, however, a readily-accessible library targeted
> >for this sort of work.  If I had the opportunity to work in
> >econometrics now, I'd think seriously about R, Lisp, and
> >Mathematica, and see what's available among the functional
> >languages, along with Python.
>
> Smalltalk, too; I'd throw it in the mix.  Much serious econometrics
> has been done with Fortran, but I have no enthusiasm for pursuing
> that direction, mostly because I think too much of the computing
> world is going in a different one.

There are many statistics packages and programming languages used in
econometrics and in general, so most of the computing world is going in
a different
"direction", no matter which package or language you choose.

Enough programmers still use Fortran that major hardware vendors such
as Intel, IBM, and Sun are actively maintaining their Fortran 95
compilers and adding features from Fortran 2003. G95 is free, available
almost everywhere that gcc is, and good enough for production use IMO.

The recent book
http://methcenter.psu.edu/newbooks/fortranbook/thebook.html
Developing Statistical Software in Fortran 95
by David R. Lemmon and Joseph L. Schafer
Spriger (2005)

discusses how to build statistical software components in Fortran that
can be used in statistical packages.

The IMSL and NAG software libraries have extensive statistical
functionality and are available in Fortran, among other languages.

It is important for a programming language used for econometrics to
conveniently handle multidimensional arrays, and here Fortran outshines
C, C++, and Java (NumPy is good, though).

I am a quantitative financial analyst who implements econometrics
algorithms. Data sets are getting bigger -- use of intraday data is now
common -- and the CPU time for many algorithms scales as the N^2 or
N^3, where N is the number of observations. Speed still matters.