Non-linear GMM model estimation issue

[simonadit]

I'm trying to estimate a non-linear GMM model using the attached workfile, but can't get the system to run. I was initially receiving the error message 'No valid observations in equation', so tried tweaking the starting values until I no longer received the message. Unfortunately for all values I've tried for which I don't receive the 'No valid observations' error message, I recieve the 'Log of non-positive number' error message instead. I've had a look through previous solutions to problems of this type and tried to think of where my error may be coming from but I'm yet to figure it out.

Any suggestions that anyone has would be much appreciated. Thanks!

[EViews Gareth]

There appear to be many issues with that specification. First of all there isn't a series called ARMNEG in your workfile. Secondly you don't have anywhere near enough instruments.

[startz]

Are you missing the variable armneg?

[simonadit]

Sorry, I must have attached a slightly old copy of the workfile. I've included the up to date file with variable armneg.

[simonadit]

There appear to be many issues with that specification. First of all there isn't a series called ARMNEG in your workfile. Secondly you don't have anywhere near enough instruments.

It may well be an instrument issue, but the specification I'm basing my model on was estimated without instruments so if this is the issue, I don't know why it wasn't an issue before.

[startz]

I speculate that it's an EViews bug, or at least a rough edge. I think the system estimation is having trouble with 0^0. 0^0 works okay with the series evaluator but maybe not so well in systems.

[EViews Gareth]

For the current starting values, possibly, but the current starting values are not great. Though I'm not sure what starting values would perform better. Raising a series of nearly all zeros to a coefficient is going to be tricky.

And that's ignoring the instruments issue.

[simonadit]

I've updated my specification to try to avoid this issue and make it a bit cleaner by including an indicator function variable rather than having two separate variables. I'm now getting a mix of 'Overflow' and 'Near Singular Matrix' errors, depending on starting values.

Any suggestions on this? Or any suggestions on potential instruments to include that may help?

[simonadit]

Here's the updated workfile