I'm trying to estimate the following model: Labor demand=f(wages,lagged labor demand, GFCF, GDP, trade share, labor productivity, real exchange rate)... (1) Wages= f(labor demand,lagged wages, GFCF, GDP, trade share, labor productivity, real exchange rate)...(2)

When i run endogeneity test in stata and dumitrescu hurlin test in Eviews, they say that all regressors are endogenous. Is this something normal or I'm doing something wrong?
Eagerly waiting for a reply.

  • $\begingroup$ Are you asking if is possible if all the repressors you have are endogenous or if all the possible regressors in the properly specified model can be endogenous? $\endgroup$
    – BKay
    Jul 20 '18 at 13:52
  • $\begingroup$ @BKay... Yes, i am asking whether it's okay if all the independent variables turn out to be endogenous in my model or I'm doing something wrong? $\endgroup$ Jul 21 '18 at 15:09

In principle, there is nothing to exclude the case where all regressors are endogenous.

My issue is that the Hurlin-Dumitrescu test has to do with Granger-causality, and Granger-causality examines the relation between the regressors and the dependent variable, not the relation between the regressors and the error term. On the other hand, endogeneity refers to the existence of correlation of the regressors with the error term.

So it appears that the Hurlin-Dumitrescu test is of no use on testing for endogeneity. Do I miss something here?

  • $\begingroup$ Thanks alot for your response. I completely understood it. But, i have mentioned that i got the same results in stata when i test for endogeneity using xtivreg2 command. What is your say over it? Secondly,your first line says that it's okay if all regressors are endogenous? $\endgroup$ Jul 18 '18 at 17:09
  • $\begingroup$ Yes, there is no theoretical or other reason to say "they cannot be all endogenous". I am not familiar with what tests are included in the xtivreg2 command in stata. $\endgroup$ Jul 18 '18 at 17:15
  • $\begingroup$ Alright. Have you seen the case somewhere as well? Can you suggest any endogeneity test if you know either in stata or eviews? $\endgroup$ Jul 18 '18 at 19:03
  • $\begingroup$ Alecos: I am familar with endogeneity in the way you described it but then I looked here and there is a causality relation under the dynamic models section. en.wikipedia.org/wiki/Endogeneity_(econometrics). Still I'm not familar with stata so no idea if test described is useful. Also, there's something called the Wu Hausman test for endogeneity, but, as is the case with all tests, you need to be careful to make sure that it's appropriate for your problem. Sometimes these tests can be wolves in sheep's clothing. $\endgroup$
    – mark leeds
    Aug 18 '18 at 0:30
  • $\begingroup$ @markleeds The Hausman test does not really test for endogeneity. It just tests whether the OLS and the IV estimators have the same probability limit (that's the null hypothesis). If the null is rejected, what we learn is that they have different probability limits. Then, IF we accept (based on arguments unrelated to the test) that the IV estimator is consistent, then we "conclude" that the OLS is inconsistent and so that there exists endogeneity. But the test itself has nothing to do with testing endogeneity per se, never mind how it is presented in the papers that use it. $\endgroup$ Aug 18 '18 at 3:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.