GMM instruments (Dynamic Panel Data Model Wizard)

[Max]

Hello everyone.

I want to employ GMM estimator for the following model
RISK = RISK(-1) + SIZE + PROFIT + DEBT + CASH + MARGIN + INTEREST

I use Eviews 8. I have an unbalanced panel dataset with small T and large N (T=15, N=360).All variables are stationnary I have tested variables for Granger Casuality and found 3 bidirectionnal relationships : RISK <=> SIZE, RISK <=> INTEREST, RISK <=> DEBT
So after some reading I get that I should use the SYSTEM GMM estimator developped by Blundell and Bond (1998). However, after going throught this forum I understand that such estimator cannot be implemented on Eviews. Is it still the case?
That being said, I still would like to estimate a straight GMM for my personnal knowledge, even though it does not have great properties in small T large N samples.
But I have a few questions about the Dynamic Panel Data Wizard. Indeed despite going through Eveiws helps and examples I still don't get the instruments definitions in step 4 and 5
I first select Method "GMM" and then open the Dynamic Panel Data Wizard
Step 1 : Specify dependent variable. my dependent variable is RISK, and I choose to use 1 lag (I have annual data)
Step 2 : Specifiy regressors. I write the whole right-handiside of my equation : SIZE + PROFIT + DEBT + CASH + MARGIN + INTEREST and include dummy variables for periods because it makes sense in my model
Step3: Select transformation method. I choose differences (Arellano and Bond approach) and I do not transform period dummy variables . What is the rationale behind choosing to transform dummies or not?
Step 4: Specifiy GMM level instruments. @DYN((RISK),-3) is already written. Should I add the regressors SIZE + PROFIT + DEBT + CASH + MARGIN + INTEREST? All of them or only the exogenous ones? I And if yes, should I use 1 lag as well ? (I understood that you should not use endoegnous variables as instruments, since it is the whole point of GMM to get rid of their influence)
And by the way what means @LEV(@SYSPER) in the output?
Step 5. Specify regular instruments. This is where I get mixed up. Is it here that I must add regressors (all of them or exogenous ones?). Can I use their lags here? And as in step 2, how to decide if I want to transform them or not?
Step 6 is okay

So put it simply: what is the difference between instruments that you set in step 4 versus step 5? And then in step 5 how to choose if you want to transform the instrument or not?

PS: I would rather use lagged variables as instruments, I m not really keen to look for an ad-hoc instrument outside my database


Thanks a lot

[emahembe]

I have a similar question. Has this question been answered?

[emahembe]

I have a similar question. Has this question been answered?

[Max]

I would be really grateful if someone could just give me some idea about the following:
"Step 5. Specify regular instruments: list all other instruments in one of the following edit boxes, depending on whether you wish to apply the transformation to the instrument". How to choose betweeen transforming (differences) my instruments or not?
Thanks a lot

[brinde]

I've chosen to bump this thread instead of making a new one, as I am wondering pretty much the same.

Many thanks in advance to anyone who can finally settle these questions.