EViews.com

Hi, everyone.

I am doing GMM for a forward looking Talor rule model.

Here is the equation I input in the equation box

fed-(1-c(4))*(5.97-(c(2)-1)*c(1)+c(2)*inf+c(3)*gap)-c(4)*fed(-1)

Here is the IV list:
c gap(-1) gap (-2) gap (-3) gap(-4) fed(-1) fed(-2) fed(-3) fed(-4) spd(-1) spd(-2) spd(-3) spd(-4) cpi(-1) cpi(-2) cpi(-3) cpi(-4)

I keep facing the error message of nearly singular matrix

Estimation worked fine for me:

equation e1.gmm fed-(1-c(4))*(5.97-(c(2)-1)*c(1)+c(2)*inf+c(3)*gap)-c(4)*fed(-1) @ c gap(-1) gap (-2) gap (-3) gap(-4) fed(-1) fed(-2) fed(-3) fed(-4) spd(-1) spd(-2) spd(-3) spd(-4) cpi(-1) cpi(-2) cpi(-3) cpi(-4)

Thank you for reading my question and working on my work file.
I tried again, it still does not work
I use QUICK-->equation
Select GMM
Input the equations and IV
Select HAC Quadratic Newey-West
Iteration 1000
It tells me that "nearly singular matrix"
Could you try the results for these three workfiles for me?


Thank you

It won't work in EViews 5.1

It should work in EViews 6.0, as long as that copy of EViews 6 is up to date.

So it means that Eview 6.0 in the lab may not be updated such that GMM could be run?

But when I was taking the class I used to use Eview 6.0 in the lab to run another GMM estimation.

Is there any other way of solving this?

You could try removing one of your instruments.

I tried manipulating with the IV many ways , up to reducing them to 4

But everytime it reports nearly singular matrix.

Thank you

I keep clicking on the OK button......it works after dozens of clicks....
OMG

It is possible, since your equation is quite non-linear, that changing the starting values will alleviate the problem. Although GMM always uses 2SLS to obtain starting values, in non-linear cases EViews requires starting values for the 2SLS step.

Perhaps your starting values changed between your 11th click and your 12th.

Firts of all, I apologise because I am new in this forum and I don't undestand exactly where to write. Anyway, I am sure that you are going to help me.

I am estimating a hybrid IS curve in GMM, using as instruments lags of the variables(dependent and independet). Well, the point is that if I use, for example, four lags of the variables (as instruments), I get some values for the coefficients, with the right signs for each variable. But, if I use, for example, six lags, I get again the right signs for the coefficients, but some of them change significantly. So, my questiong is how I can identify which set of instrument is the best for the estimation.

I know that I have to take into account the J-statistic provided by Eviews, but I don't know how I can compare the value of this test in the same spectification of the equation but with different set of instrument.

Thanks for your comment.