GMM in Eviews 5.1 and 6 gives different results

[Mir]

Dear Sir/Madame,

I am trying to replicate results in one econometrics textbook to learn about GMM estimation - the textbook example has been done with Eviews 4. When I try replicating the results using Eviews 5.1, they do not correspond (different number of iterations until convergence - despite the data and everything being the same). Surprisingly, with Eviews 6 the results are the same as in the textbook (and with version 4).

I do apologise if I am overlooking something obvious (I have just started to learn about GMM) but how is this possible? Is there some difference between the ways how different EViews versions carry out the estimation?

Thank you in advance for your help.

[EViews Gareth]

Could you post the workfile?

[Mir]

I have tried opening equation equsrfe and just pressing estimate - this yields the same result as upon opening of the equation (corresponding to authors results) with EViews 4 and 6 but when I re-estimate with 5.1 I get altered results.

Maybe it has to do with the initial estimates from which the programme starts the interations?

Thank you in advance for your help.

[Mir]

Also,

I have come across an empirical investigation usnig GMM and they detail 3 estimators - 2step GMM, iterative GMM (coefficients and the weighting matrix caluclated recursively) and "continuous updating GMM" (weighting matrix continuously altered in the process of estimation).

Which of these methods is closest to the one carried out by Eviews? Maybe this could explain the differences in results calculated by different versions of the software?

Thank you.

[EViews Gareth]

I get the same results in EViews 5.1 and 6, which are different from EViews 4. I presume that there was a change of some kind between EViews 5 and 6.

EViews does both 2step GMM and Iterative GMM, although in a single Equation object, you can only perform iterative GMM. If you want 2Step, you'll have to put your equation into a System object.

[Mir]

Thank you for doing the estimation.

I do not understand how is it possible then, that with 5.1 I get

USFF= C(2)*USFF(-1) +(1-C(2))*(C(1)+C(3)*USINFL(+12) +C(4)
*USGAP1)
Instrument list: C USGAP1(-1 TO -6) USGAP1(-9) USGAP1(-12)
USINFL(-1 TO -6) USINFL(-9) USINFL(-12) USFF(-1) USFF(-6)
USFF(-9) USFF(-12) DLPCM(-1 TO -6) DLPCM(-9) DLPCM(-12)


Coefficient Std. Error t-Statistic Prob.

C(2) 0.880423 0.015916 55.31815 0.0000
C(1) 2.606580 0.998248 2.611154 0.0098
C(3) 1.464063 0.262142 5.584991 0.0000
C(4) 0.444659 0.071428 6.225267 0.0000

and with 6 I get:

USFF= C(2)*USFF(-1) +(1-C(2))*(C(1)+C(3)*USINFL(+12) +C(4)*USGAP1)
Instrument list: C USGAP1(-1 TO -6) USGAP1(-9) USGAP1(-12) USINFL(
-1 TO -6) USINFL(-9) USINFL(-12) USFF(-1) USFF(-6) USFF(-9)
USFF(-12) DLPCM(-1 TO -6) DLPCM(-9) DLPCM(-12)

Variable Coefficient Std. Error t-Statistic Prob.

C(2) 0.929641 0.012593 73.82317 0.0000
C(1) 2.874756 0.991090 2.900601 0.0042
C(3) 1.730926 0.251889 6.871790 0.0000
C(4) 0.664965 0.100673 6.605195 0.0000

I just open the same equation and press estimate... ???

I would be very grateful for any suggestion where I might be going worng given you got the same results. Thank you very much.

[EViews Gareth]

Is your copy of EViews 6 up to date?

[startz]

You should also look at the starting values in the C vector before hitting the estimate button, to be sure they are the same in both cases.

[Mir]

I am not sure whether the 6 version is updated since it is a university software (I own 5.1).

The c vector is the same in both cases - I literally open a pre-estimated equation in a given file (thus c verctor is the same) just using the different versions of the software and press estimate and the results differ... ??? but thank you for the suggestions.

[startz]

I am not sure whether the 6 version is updated since it is a university software (I own 5.1).

The c vector is the same in both cases - I literally open a pre-estimated equation in a given file (thus c verctor is the same) just using the different versions of the software and press estimate and the results differ... ??? but thank you for the suggestions.

Do Help/About in both 5.1 and 6 and check the build date. Then compare these to the latest update dates on www.eviews.com.

[Mir]

My 5.1 version is updated but the university version 6 not (I cannot change this) - does this mean that out of the two sets of results the one I get with 5.1 is the correct one?

This would support the previous post that 5.1 and 6 yield the same results that differ from 4... Do the "same results" with 5.1 and 6 correspond to what I get with 5.1?

Thanks once again for helping me.

[EViews Gareth]

Yes, they do.

[Mir]

Thank you very much for doing the estimation and helping me - now I know that I can use my 5.1 version for the estimation.

Once more thank you.