[Limdep Nlogit List] Panel SUR and restrictions

[William Greene]

Tomas. The adding up restrictions in a translog system of share equations
are imposed by solving out the constraints to begin with. In the example in
16.3.3, this is done (as it is generally) by normalizing all prices and cost
by one of the prices. In the KLEM system in 16.3.3, the 4 shares are made to
add to one by imposing the restriction a priori. Cost and each price variable
are divided by PM. Then, 3 of the four shares are included in the equation
system. The fourth one is one minus the sum of the other three. (There is a
typo on page E16-19. The third share equation in the system should be Se, not
Sk.) I do not see what you have in mind for TSCS; this is a different
model command entirely.
  I'm sorry I do not have any wisdom to add to the mix about your second
paragraph, something like a panel data application for SURE. There is none
built into LIMDEP. I have not seen it done elsewhere, though you seem to be
suggesting something in TSP.
/Bill Greene


----- Original Message -----
From: "Tomas Nilsson" <tknilsso at gmail.com>
To: limdep at limdep.itls.usyd.edu.au
Sent: Saturday, June 28, 2008 1:52:21 PM (GMT-0500) America/Bogota
Subject: [Limdep Nlogit List] Panel SUR and restrictions

Dear Limdep users,

What is the procedure for including an adding up restriction in the
SURE command when following the recommended MLE procedure as outlined
in, Chapter 16.3?

That is, according to the LIMDEP manual (and the error that is
subsequently generated) Chapter 16.3, it is not possible to introduce
the CLS command if the pattern command for the parameter matrix is
specified /because/ there is no accomodation for the equation lines.
Unforunately, I don't see how this can be perfomed even in the TSCS
regime, because the LHS is specified to be a dependent variable / and
not list of dependent variables.

And a second question, more a query on the appropriate note on
terminology would perhaps be in its place, given that I keep on
haraunging y'all on the list: I'm really looking for an approach to
fit a systems of  equations (linear, for now): one individual is
observed over several time periods; in each period this one individual
makes a series of multiple decisions; moreover the decisions are
correlated across product group and time in a particular manner;
furthermore, the data contains observations for multiple individuals,
observed over the same time period making similar choices. I thought
that this would call for a SUR(E) PANEL approach, am I correct? (not
according to the TSP examples I've came across to date).

ATB,

Tomas Nilson


---------- Forwarded message ----------
From: Tomas Nilsson <tknilsso at gmail.com>
Date: Thu, Jun 26, 2008 at 8:41 AM
Subject: Panel SUR and latent class
To: limdep at limdep.itls.usyd.edu.au


Dear all,

w.r.t. fitting a SURE/Zellner model to panel data... the PDS=t command
doesn't work... no response from the active participants in the list,
so I presume I can't see the forest because of all the trees... or
it's not as simple as 1-2-3... The systems format is attractive as
several cross-equation restrictions must be imposed.

So here is another ludicrous question: can one fit a SURE model but in
a latent class format; that is, estimate a set of share equations
where the final parameter estimates differ across classes but within
classes comply with cross-equation restrictions. The food demand
application I have in mind would be to let some goods be substitutes
for some consumers and complements for others. I like the idea of
"letting the data speak" rather than performing some ad-hoc
segmentation based on some arbitrary rules.

Can this be done within the context of not having to code the ML
function? It's probably pretty nasty analytically and testing for
statistical adequacy a living nightmare...  any thoughts, suggestions,
comments, critique greatly appreciated at this point in life...

Tomas




On Fri, Jun 20, 2008 at 2:56 PM, Tomas Nilsson <tknilsso at gmail.com> wrote:
> Greetings, Limdep users -
>
> what is the <canned> command for fitting a panel SURE (seemingly
> unrelated linear regression equations a la Zellner) in Limdep. I'm
> looking at R16 but still somewhat curious how to fit it to a panel
> (just add Pds=t?). How do one set up the data to perform such task (as
> in R6?)? To sound even more ignorant, can this model be fitted to an
> unbalanced dataset?
>
> Happy Solstice
>
> Tomas Nilsson
>
_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au


-- 
Professor William Greene
Department of Economics
Stern School of Business
New York University
44 West 4th St., Rm. 7-78
New York, NY   10012
http://www.stern.nyu.edu/~wgreene
212.998.0876