9:05 a.m.
Consider the following model: Y=b0+b1*X1+b2*X2+b3*X1*X2, where X1 is endogenous and X2 is
independent. The Instrumental Variable for X1 is Z1. How should I handle the model and
compute the results?
9:55 a.m.
Am 17.04.2023 um 11:05 schrieb elfsog(a)yahoo.gr:
Consider the following model: Y=b0+b1*X1+b2*X2+b3*X1*X2, where X1 is
endogenous and X2 is independent. The Instrumental Variable for X1 is Z1. How should I
handle the model and compute the results?
First of all, from a gretl-specific perspective, I'm not aware of any
tool in gretl that focuses on this nonlinear-in-variables IV setup.
However, on a general econometric level, this is still a model which is
linear in the parameters. I'd expect the product X1*X2 to be endogenous
(= correlated with the true disturbances) as well, although I believe
that this is not always clear and in general I expect that it depends on
the joint distribution. But if Z1 is known to be a valid instrument for
X1, my pragmatic approach would be to simply compute Z1*X2 and use that
as an additional instrument in a traditional linear IV setup. (I guess
it would also be possible to try other nonlinear transformations of Z1,
see the pretty large literature on non-linear instruments.)
If non-normal distributions are involved, another quite recent approach
might be one based on (nonparametric) control functions, see this one:
https://arxiv.org/abs/2207.09246 (Breitung, Mayer & Wied).
cheers
sven
8:54 a.m.
Am 18.04.2023 um 20:44 schrieb elfsog(a)yahoo.gr:
What if the product Z1*X2 is correlated with the exogenous variable
X2?
This is moving away from a question about gretl, but likely they will be
correlated indeed. So what? Valid instruments can be correlated among
each other (short from being perfectly collinear or close to that).
cheers
sven
6:43 p.m.
If X1 is a binary variable, s still 2SLS an appropriate method to correct for endogeneity?
Would it be possible to run a two-stage probit, or the bivariate two-stage model, in
Gretl?
7:46 p.m.
Am 20.04.2023 um 20:43 schrieb elfsog(a)yahoo.gr:
If X1 is a binary variable, s still 2SLS an appropriate method to
correct for endogeneity? Would it be possible to run a two-stage probit, or the bivariate
two-stage model, in Gretl?
Please be more explicit with your questions. For example, what do you
mean by "two-stage probit", given that you didn't tell us your model has
a binary dependent variable. Use formulas to explain what you plan or
are trying to do. Then we can tell you whether (or how easily) it can be
done in gretl.
thanks
sven
8:33 p.m.
The model is Y=b0+b1*X1+b2*X2+b3*X1*X2+e
Y is an ordinal six grade Likert variable,
X1 and X2 are binary variables. X1 is endogenous and X2 is exogenous.
Z1 is the Instrumental Variable for X1 and is an ordinal six grade Likert type variable.
4:31 p.m.
To be more specific:
I am trying to estimate a model with two predictors and an interaction feature using an
IV. Which would be the most appropriate approach, and how could I implement (the
following) model in Grelt?
The model is Y=b0+b1*X1+b2*X2+b3*X1*X2+e
Y is an ordinal six grade Likert variable,
X1 and X2 are binary variables. X1 is endogenous and X2 is exogenous.
Z1 is the Instrumental Variable for X1 and is an ordinal six grade Likert type variable.
2:47 p.m.
Am 21.04.2023 um 18:31 schrieb elfsog(a)yahoo.gr:
If Y is ordinal then a natural additional question would be whether
using an ordered probit might be (more) needed.
In that sense I would tend to recommend an explicit two-stage approach,
regressing the endogenous RHS terms on the set of valid instruments and
continue with the fitted values from there, doing the ordinal probit in
the second stage. A standard ordered probit routine is "of course"
available in gretl. Just mark the LHS variable as discrete and then
gretl will automatically apply the ordered probit estimation.
But I'm no expert for setups such as these, so I certainly also
recommend reading up on the textbook econometrics on doing an IV ordered
probit estimation. To my knowledge doing ordered probit with IVs is not
directly available in gretl, but I may be missing something.
(The HIP addon for gretl does IV-probit, but not IV-ordered-probit, if I
understand correctly.)
cheers
sven
5:16 p.m.
Can I add to Sven's answer that estimating the reduced form for the
endogenous variable allows one to test that the coefficient on the
instrument is non-zero? Otherwise, the system is not identified. (weak
instruments?)
John C Frain
3 Aranleigh Park
Rathfarnham
Dublin 14
Ireland
www.tcd.ie/Economics/staff/frainj/home.html
https://jcfrain.wordpress.com/
https://jcfraincv19.wordpress.com/
mailto:frainj@tcd.ie
mailto:frainj@gmail.com
On Sat, 22 Apr 2023 at 15:47, Sven Schreiber <sven.schreiber(a)fu-berlin.de>
wrote:
714
days inactive
719
days old
10 comments
3 participants
participants (3)
-
elfsog＠yahoo.gr -
John C Frain -
Sven Schreiber