Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: using IV estimation with spatial econometrics


From   "Henrique Neder" <[email protected]>
To   <[email protected]>
Subject   st: using IV estimation with spatial econometrics
Date   Tue, 28 Feb 2012 14:52:46 -0300

Dear Stata  list members 

I am making some estimates of spatial econometric models aiming to evaluate
the impact of a particular credit program oriented to small farmers in
Brazil. I have data in aggregated municipal level that are some economic and
social  indicators (the response variables, for example, difference in rural
poverty rates, difference in Gini index, number of occupied people by
farms), the indicator of credit (considered the primary causal variable in
the models) and some control variables. There are concerns for endogeneity
of the causal variable, both by reasons of reverse causality as for reasons
of possible existence of correlation between this variable and unobservable
variables. The strategy adopted is to reduce or eliminate the endogeneity
bias by using cross section models of instrumental variables. One approach
is the spivreg module usage, which in my view, focuses on the endogeneity of
spatially lagged dependent variable and the endogeneity of other regressors
in the right side of the equation. The second approach is the prior
generation of the spatial lag of the dependent variable and later use of
ivreg2 command. This command automatically perform several tests and save
their results: 1) sub-identification and weak  identification test, 2) a
redundancy test of a excluded instruments sub-set, 3) over-identification
restrictions test; 4) Exogeneity /orthogonality of suspected instruments
test; 5) Test of one or more endogenous regressors on the estimation
equation. In fact, I have more confidence in the estimate of causal variable
parameter if all the results of these tests ensure the proper identification
of the model. But only a few of several questions arise here: 1) some of the
models (for some dependent variables) fail to prove the endogeneity of the
regressors tested. This means that I abandon the IV-GMM estimation and stick
with the first approach only? Unhappily, with this (spreg and spivreg
command) I can?t perform tests. 2) In other cases, the test results seem to
contradict each other: for example, in the sub-identification and weak
identification test, the null is rejected and in the over-identifying
restrictions testing the instruments are checked valid. This means that in
this case I should get other more appropriate instruments? For the second
approach some excluded instruments are the spatial lags of the control
variables. What are the guarantees that these instruments are good and
sufficient for my application? I think that they only treat the endogeneity
of lagged space dependent variable.  Like in spivreg command I need more
instruments for implement with the ivreg2 command. Is there inappropriate to
use this command in conjunction with spatial econometrics estimation with
IV? Has anybody any good reference for this and other correlated questions?
I would be grateful if someone made a comment about it.


Thanks in advance
Henrique Dantas Neder
Professor at the Federal University of Uberlândia - Brazil


Henrique Neder
Prof. Associado - Instituto de Economia
Universidade Federal de Uberlândia
Tel.:  (34) 32394157  Cel: (34) 91216600  



*
*   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/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index