Hausman Test - Panel ARDL

[Ronak Parikh]

Hi,

I'm currently running a Panel ARDL test, and I was trying to find the most efficient estimator for my model. I am slightly confused by the Hausman Test, in terms of how to operate it on Stata.
For example, if I write:

hausman mg pmg, sigmamore

and I get my p-value > 0.05, what does this mean? What order should I keep my variables in? I've been trying to choose between PMG and DFE, and when I input the code

hausman pmg DFE, sigmamore

I get a p-value of 0.9998. What does this mean?

Thanks!

Ronak

[Carlo Lazzaro]

Ronan:
the -hausman- null is that -re- estimator is the way to go.
That said, see https://stats.stackexchange.com/ques...l-bounds-steps

[Ronak Parikh]

Thank you very much - what is the significance of the order of the variables in the Hausman test? For example,

hausman pmg DFE, sigmamore

returns different results to

hausman DFE pmg, sigmamore

What is the difference between these two tests?

[Carlo Lazzaro]

Ronak:
the sequence (that should not be reversed) is:
1) the estimator that is consitent under both H0 and H1 but inefficient if H0 is not rejected;
2) the estimator that is consistent and efficient if the null is not rejected but inconsistent if H0 is rejected.
Therefore:

hausman 1) 2)

is the only way to go.

[Ronak Parikh]

Thank you very much, that's really helpful!

[Tasfaye Fikadu]

Thank you very much - what is the significance of the order of the variables in the Hausman test? For example,

hausman pmg DFE, sigmamore

returns different results to

hausman DFE pmg, sigmamore

What is the difference between these two tests?

The second one is appropriate

[Tasfaye Fikadu]

Ronak:
the sequence (that should not be reversed) is:
1) the estimator that is consistent under both H0 and H1 but inefficient if H0 is not rejected;
2) the estimator that is consistent and efficient if the null is not rejected but inconsistent if H0 is rejected.
Therefore:

hausman 1) 2)

is the only way to go.

Dear Carlo:

I really need your recommendation.

I want to test between the dynamic fixed effect (dfe) and the pooled mean group (pmg) :


hausman dfe pmg, sigmamore.

and then the result of chi2 is negative like:

warning: chi2 < 0 ==> model fitted on these data
fails to meet the asymptotic assumptions
of the Hausman test; see suest for a
generalized test.


I was trying to find an alternative test and what the suest generalized test means following the 'help suest', but I could not handle both; is there an alternative to hausman test?

[Carlo Lazzaro]

Tasfaye:
do you have singletons?

[Tasfaye Fikadu]

Tasfaye:
do you have singletons?

I don't know whether I have got your question or not,

I want to test both:


hausman dfe pmg, sigmamore.
hausman mg pmg, sigmamore.

and the results are similar [ negative chi2]

With Kind regards.

[Carlo Lazzaro]

Tasfaye:
see last example, -hausman- entry, Stata .pdf manual, where the negative -hausman- is considered as a strong evidence that the null cannot be rejected.

[Tasfaye Fikadu]

Tasfaye:
see last example, -hausman- entry, Stata .pdf manual, where the negative -hausman- is considered as a strong evidence that the null cannot be rejected.

Thank you very much !

[Clara Joan Joachim]

Hello, I want to do the same test to compare pmg, mg and dfe. I first tried PMG vs MG and the result shows "maximum number of iterations exceeded" for MG but worked fine for PMG. Secondly, PMG vs DFE, and same problem to DFE but worked fine for PMG. Therefore, it could not give me hausman test result. Anyone knows what is the problem here? Thank you.

[Clara Joan Joachim]

Hello, I managed to solve question above. However, my next question is, when I ran Hausman test, it gives a result but with this:

Note: the rank of the differenced variance matrix (7) does not equal the number of coefficients being tested (8); be sure this is what you expect, or there may be problems
computing the test. Examine the output of your estimators for anything unexpected and possibly consider scaling your variables so that the coefficients are on a similar
scale.

Please advise on what I should do. Thank you very much!

[Andrew Musau]

Can you show the output? Post this within CODE delimiters.

[Clara Joan Joachim]

Hello Andrew Musau my output is as follows

 hausman pmg DFE, sigmamore

Note: the rank of the differenced variance matrix (7) does not equal the number of coefficients being tested (8); be sure this is what you expect, or there may be problems
        computing the test.  Examine the output of your estimators for anything unexpected and possibly consider scaling your variables so that the coefficients are on a similar
        scale.

                 ---- Coefficients ----
             |      (b)          (B)            (b-B)     sqrt(diag(V_b-V_B))
             |      pmg          DFE         Difference       Std. err.
-------------+----------------------------------------------------------------
        tech |   -.0833375     .0755057       -.1588432        15.54078
         cli |   -.1534192    -.0051064       -.1483128        8.594972
         fcy |    .0002846     .0005092       -.0002246        .2382741
        nfcy |   -.0010914    -.0007804        -.000311        .0810243
          al |   -.0213634     .2129817       -.2343452        104.0759
          cl |    .0069398    -.0006649        .0076047        .8362317
       lngni |    .1171303     .0076487        .1094816        7.000661
         exp |    6.60e-12    -7.88e-12        1.45e-11        1.07e-09
------------------------------------------------------------------------------
                          b = Consistent under H0 and Ha; obtained from xtpmg.
           B = Inconsistent under Ha, efficient under H0; obtained from xtpmg.

Test of H0: Difference in coefficients not systematic

    chi2(7) = (b-B)'[(V_b-V_B)^(-1)](b-B)
            =   0.01
Prob > chi2 = 1.0000
(V_b-V_B is not positive definite)

.