I have another problem now, though. All the regressions I try to run like this give me either a standard deviation of 0 for all the estimated parameters, or I get an error message that I have insufficient degrees of freedom. However, I have 8 orthogonality conditions in the example I am trying this out on, and only two parameters. The example is as follows:
# initializations go here
series e1 = 0
series e2 = 0
matrix W = I(8)
scalar alpha = 0
scalar beta = 0
  e1 = (alpha / beta) * CONSUMPTION_DET
  e2 = (1 / alpha) * GDP_DETREND
  orthog e1 ; IVs
  orthog e2 ; IVs
  weights W
  params alpha beta
end gmm
Where CONSUMPTION_DET, GDP_DETREND and R_DETREND are time series with 170 observations.
Why should there be a problem with the degrees of freedom?
Thanks again,
On Mon, Oct 26, 2009 at 9:09 PM, Allin Cottrell <> wrote:

On Mon, 26 Oct 2009, Gal Wettstein wrote:

> I have been trying to run a GMM regression which has two first
> order condition equations. Is this possible with Gretl? If it
> is, how can I do so?

Yes you can do it; you can specify as many orthogonality
conditions as you like.  Here's a trivial example in which gmm
wraps OLS done both forwards and backwards:

open data4-1

matrix b = {50, 0.1, 30, 6}
matrix W = I(4) ./ 5
list X1 = const sqft
list X2 = const price
series e1 = 0
series e2 = 0

 e1 = price - b[1] - b[2]*sqft
 e2 = sqft - b[3] - b[4]*price
 orthog e1 ; X1
 orthog e2 ; X2
 weights W
 params b
end gmm

Allin Cottrell
Gretl-users mailing list