Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Johannes Muck <Johannes.Muck@dice.uni-duesseldorf.de> |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: st: Standard error correction when using control function approach to endogeneity |
Date | Thu, 20 Jun 2013 11:25:12 +0200 |
Dear Austin, Thanks again for your reply. Yes, I had a look at both threads (and a couple of others dealing with the same question). However, until yesterday I only tried to implement the "stacking-approach" - however I was not sure whether stacking is appropriate for my problem since all examples with stacking on statalist involved two IV-regression with the _same_ dependent variable, whereas I have two different dependent variables. Following your suggestion, I implemented the 2-equations-GMM-approach suggested by Tirthankar Chakravarty yesterday. This works perfectly - now I have an estimate of the variance-covariance matrix of all coefficients and can perform tests on the coefficients from both equations. Since this workaround seems rather easy (once you know about it), I now wonder why -suest- cannot be used with -ivreg- / -ivregress- / -ivreg2- I know that Stata says it's because predict after the IV-commands does not allow for a score-option. But since it works with -gmm- it seems a bit odd that it doesn't work with -ivreg- /-ivreg2- and -suest- . Or is there a substantial difference between the common variance-covariance matrix estimated by -gmm- and the one I would hypothetically obtain by using -suest- ? Thanks again for your help! Best, Johannes -----Ursprüngliche Nachricht----- Von: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] Im Auftrag von Austin Nichols Gesendet: Mittwoch, 19. Juni 2013 18:41 An: statalist@hsphsun2.harvard.edu Betreff: Re: st: Standard error correction when using control function approach to endogeneity Johannes Muck <Johannes.Muck@dice.uni-duesseldorf.de> : did you read these? http://www.stata.com/statalist/archive/2009-11/msg01485.html http://www.stata.com/statalist/archive/2011-09/msg00284.html On Wed, Jun 19, 2013 at 5:40 AM, Johannes Muck <Johannes.Muck@dice.uni-duesseldorf.de> wrote: > Dear Austin, > > thank you very much for your answer. > > As far as point (1) is concerned, I know that I could also use the 2SLS > estimator by running ivregress or ivreg2. The reason for why I want to use > the control function approach instead (where fitted residuals rather than > fitted values are used in the second stage) is the following: I estimate a > system of two simultaneous equations and need to test whether a combination > of parameters from both equations is significantly different from zero (for > further details see my earlier post to the statalist: > http://www.stata.com/statalist/archive/2013-06/msg00566.html). > > By using the control function approach I circumvent the problem that > combining two IV-estimations with the -suest- command does not work because > I only use the -reg- command. However, before combining the two estimations, > I want to make sure that the standard errors of my coefficients are correct. > > Nevertheless, I am glad that bootstrapping the whole procedure would also do > the standard error correction. However, I guess it might be more elegant to > use the analytic standard error correction? > > Best, > > Johannes > > -----Ursprüngliche Nachricht----- > Von: owner-statalist@hsphsun2.harvard.edu > [mailto:owner-statalist@hsphsun2.harvard.edu] Im Auftrag von Austin Nichols > Gesendet: Dienstag, 18. Juni 2013 20:27 > An: statalist@hsphsun2.harvard.edu > Betreff: Re: st: Standard error correction when using control function > approach to endogeneity > > Johannes Muck <Johannes.Muck@dice.uni-duesseldorf.de>: > > 1) yes, just run these to get the same answers: > ivreg y1 x1 x2 (y2 = z1 z2) > ivregress 2sls y1 x1 x2 (y2 = z1 z2) > ssc inst ivreg2 > ivreg2 y1 x1 x2 (y2 = z1 z2) > > 2) yes, you can bootstrap the whole thing, but why would you? > > On Tue, Jun 18, 2013 at 11:31 AM, Johannes Muck > <Johannes.Muck@dice.uni-duesseldorf.de> wrote: >> Dear all, >> >> I am trying to fit a linear regression model with one endogenous variable >> using the control function approach (two stage residual inclusion > estimator) >> as described in Wooldridge (2010, pp. 126-129). >> >> More specifically, I estimate something like: >> >> (1) reg y2 x1 x2 z1 z2 >> (2) predict uhat, res >> (3) reg y1 y2 x1 x2 uhat >> >> where y1 is my dependent variable of interest, y2 is the endogenous >> variable, x1 and x2 are exogenous explanatory variables, and z1 and z2 are >> valid instruments for y2. >> >> Since the fitted residual from the first stage is included in the second >> stage regression as an additional regressor, the standard errors need be > to >> corrected. Wooldridge (2010, pp. 157-160) derives the formula for the >> corrected standard errors in his book in Appendix 6A, equation (6.58). >> >> Now my two questions are: >> >> (1) Has someone already implemented this standard error correction in > Stata >> or do I have to calculate equation (6.58) in Appendix 6A manually? >> >> (2) Could I also obtain a "standard error correction" by bootstrapping >> equations (1)-(3)? >> >> >> Any help is greatly appreciated. >> >> Best, >> >> Johannes Muck >> >> References: >> Wooldridge, J. M. (2010), Econometric Analysis of Cross Section and Panel >> Data, 2nd edition, MIT Press, Cambridge MA. >> * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/faqs/resources/statalist-faq/ * http://www.ats.ucla.edu/stat/stata/ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/faqs/resources/statalist-faq/ * http://www.ats.ucla.edu/stat/stata/