Non-linear Gmm - Near singular matrix problem

[rhaos69]

Hi everyone!! It's the first time i write here because i have a very furstrating problem.....I want to replicate the results of a paper, where non linear gmm is used in panel data...
I want to estimate with quadratic Andreus spectral kernel the following equation : c(1)*con_g(1)^(-c(2))*inf(1)^(-1)*t_bill-1 with instruments : i con_g inf t_bill(-1).
I have created a new object system and write the following :

@inst i con_g inf t_bill(-1)
c(1)*con_g(1)^(-c(2))*inf(1)^(-1)*t_bill-1

where con_g is consumption growth rate, inf is inflation rate and t_bill is interest rate and i a vector of ones......

When i press estimate, i always get an error message of near singular matrix...... I've tried several things and saw that, whenever the term inf^(-1) was removed and also i was putting a value at the c database an estimation was made........But the problem is that i cannot estimate my initial equation because of the error....

I was wondering what is the fault?? Have i done something in writing the system or is the data that create a problem??

Thank you for your patience....

[EViews Gareth]

You can't use 0 as a starting value for c(1), since if c(1) is equal to zero, the value of c(2) is irrelevant (and thus you can get a singular matrix).

Try using 0.5 as a starting value for c(1).

[rhaos69]

I seen that when i change the value of c(1) (and c(2) sometimes) i get results. I just don't know if this is good for the estimation, because the paper i replicate doesn't write something about initial values.........unless it is considered that it doesn't have to be mention....

[startz]

I seen that when i change the value of c(1) (and c(2) sometimes) i get results. I just don't know if this is good for the estimation, because the paper i replicate doesn't write something about initial values.........unless it is considered that it doesn't have to be mention....

Indeed, papers generally do not mention initial values.