Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

RE: st: xtdpd and dgmmiv: restricting the LDV instruments


From   "Mogues, Tewodaj (IFPRI)" <T.MOGUES@CGIAR.ORG>
To   <statalist@hsphsun2.harvard.edu>
Subject   RE: st: xtdpd and dgmmiv: restricting the LDV instruments
Date   Wed, 25 Jun 2008 15:52:48 -0400

Dear David,

Thanks, that pretty much answers my question. I didn't realise that the
error message was because I didn't specify the starting lag as well.
Interestingly, it will allow you to leave the ending lag unspecified,
but not the starting lag. In other words:
	xtdpd L(0/1).n w k , dgmmiv(n, lagrange(2 .)) div(w k)

(or the above with the "." replaced by any number greater than 1) can be
estimated, as can of course

	xtdpd L(0/1).n w k , dgmmiv(n) div(w k)

whereas

	xtdpd L(0/1).n w k , dgmmiv(n, lagrange(. 6)) div(w k)

(or any other number or "." instead of the "6") will give an error
message. I hadn't realised this "asymmetry" in specification rules, so
thanks for clarifying.

By the way, I was initially motivated to use -xtdpd- instead of
-xtdpdsys- because of two things I thought should be feasible in
-xtdpdsys- but I didn't see how to make work. I was wondering whether
they are simply not feasible or I was doing something wrong. These are: 

a) -xtdpdsys- doesn't seem to allow for the specification of instruments
that are external to the model, in the way that for example -xtabond-
allows through the subcommand -inst()-, or -xtdpd- by including the
instruments in the subcommand -div()- but not in the variable list of
regressors. Is it correct that external instruments are not an option in
-xtdpdsys-?

b) The subcommand for predetermined and endogenous variables in
-xtdpdsys-, i.e. -pre()- and -end()-, do not accommodate time-series
operators, although one may in many cases want to include as a regressor
just the first lag of a variable (and not both the contemporaneous
variable plus its first lag). The only solution here seems to be to
first create a new variable, eg. 
	gen L_x = L.x
and then use -pre(L_x)-, since -pre(L.x)- won't work. This however can
be rather tedious when trying out various lags and other time series
operators in various models. Is there no straightforward way to deal
with this other than creating new variables or using xtdpd? By the way,
this problem appears to also obtain in -xtdpdsys- as well as -xtabond-.

Best,
Tewodaj
=============================================================



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index