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

st: re: do 2sls, ivreg, etc. check the rank condition for identification?

From   Kit Baum <>
Subject   st: re: do 2sls, ivreg, etc. check the rank condition for identification?
Date   Thu, 11 Oct 2007 20:16:20 -0400

In message statalist.0710/Author/article-299.html

I discussed the results of using my -checkreg3- routine, posted several days ago, and asserted that it properly classified certain simultaneous equations models as unidentified by the rank condition. David Drukker of StataCorp has kindly pointed out in private communications that the methodology I was applying was incorrect, and did not faithfully reflect (or even closely approximate) the logic that should be applied to diagnose a failure of the rank condition in a simultaneous system. I am grateful to David for taking the time to point out precisely how this calculation should be done, and apologize for my misstatements on the subject in response to his posting. With his help, I have rewritten -checkreg3- to faithfully reflect Jeff Wooldridge's logic (as cited in the help file) and produce the correct verdict for a simultaneous system. The corrected version of the routine is available from SSC or -adoupdate-.

A couple of additional points should be made along the lines of the original discussion of this issue as raised by Thomas Cornelissen. First, it is explicitly an issue of full-information or systems estimation. An equation flagged as failing the rank condition in a simultaneous systems estimation may well be estimable by limited- information (single-equation) methods such as -ivregress-. The proper test for the rank condition of an equation in a systems estimation depends not only on that equation's structure but on other equations in the system.
Second, although -reg3- will not flag a failure of the rank condition (for which -checkreg3- is useful), -reg3- with the -ireg3- option will fail to converge, signalling the problem. I was mistaken in asserting that this would not be the case; in several tests I have verified that David's statement on this aspect of the problem is absolutely correct.


Kit Baum, Boston College Economics and DIW Berlin
An Introduction to Modern Econometrics Using Stata:

* For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index