Message error when dummy used as instruments in GMM

[wolly77]

Dear guys, I write you about a problem encountered in GMM estimation. I am using this estimation technique to estimate a Neo Keynesian Phillips cruve for US. The problem is the following. I have some dummies in the equation and then I want to use it as additional instruments. But when I insert these impulse dummies in the instrument lists, an error message appear in Eviews "Singular matrix". If, instead, I estimate the same model with IV technique, I obtain plausible results and the estimation runs.
I send you my database in Eviews. In the equation called "pc_gmm" I have dropped the dummy du2006q4 in the instrument lists because the error message appears. What is the problem? I think that should be a bug in Eviews.

Thank you.

P.S: my Eviews version is 7.1.

[EViews Gareth]

I'm not convinced this is a bug.

As you know, the weighting matrix (White) for GMM is calculated as inv(Zu'Zu) where Z is the instrument matrix weighted by the residuals. For the initial estimate, the residuals are taken from Two Stage Least Squares.

We can, therefore, easily manually create the weighting matrix with the following code:

Code: Select all

group insts infl(-1) infl(-2) infl(-3) infl(-4)  y(-1) y(-2) y(-3)  rulc_gg(-1)   rulc_gg(-2) rulc_gg(-3)  rulc_gg(-4)   du2008q4 c
pc_gmm.tsls infl infl_exp infl(-1) infl(-2) infl(-3)  y   du2008q4 du2006q4 du1986q2 c @ insts

pc_gmm.makeresid u

group instsu
for !i=1 to insts.@count-1
   %sername = insts.@seriesname(!i)
   instsu.add u*{%sername}
next
'handle constant
instsu.add u

matrix zu = @convert(instsu)

matrix w = @inverse(@transpose(zu)*zu)


Note that EViews errors on computing inv(Zu'Zu), since the matrix is close to singular. I tried putting that matrix into Matlab and Matlab also states that it is close to singular.

[wolly77]

Thank you Gareth for your reply.