Warning only appears when rerunning program

[bareab87]

Hi all,
I am quite new to EViews (v9), and I am trying to work on/adjust a program given to me by someone else.

According to this link: http://forums.eviews.com/viewtopic.php?f=5&t=893 my question is "nonsensical", but here it goes anyway...

I have adjusted the program to my own data and specified a system I want to be estimated. Everytime I start Eviews and open and run the program, the estimation works fine (enough...). But when I rerun the program, I get a warning on singular covariance, and hence no standard errors etc. If I close Eviews completely and open and run the program again, there is no warning, and I get the standard errors (and different coefficients).

So I thought the issue had to be that Eviews does not "clear" its memory, since the problem only arises after rerunning the program, but according to the link, memory is cleared when rerunning a program.

Any idea what the problem could be? Why do I get different estimations and warnings when rerunning a program? This problem does not exist when I run the original program, it is only when I adjust it to use my own data and specify a different system the problem occurs.

Beginning of the code:

Code: Select all

close @all %path = @runpath cd %path %START = "1985q1" %END = "2015q4" wfcreate model q %start %end dbopen "xxx" fetch y1 y2 y3 x1 x2 x3 x4 smpl @all smpl 1985q1 @last

(And then the system is formulated)

Thanks!

[EViews Gareth]

We'd need to see the rest of the program and the database to have a good idea of what is going on.

[bareab87]

We'd need to see the rest of the program and the database to have a good idea of what is going on.

Since the code is not written by me, I am not comfortable with sharing the entire code, but I can see if I manage to generalize it a bit more (but then I'll probably generalize too much, so the problem is no longer visible in the code - as mentioned, the original code works fine, so it's clearly me who created the problem...). Do you have any idea on what could be the problem, what I should look for?

[EViews Gareth]

Two obvious and common causes for estimation results to differ: the underlying data are different, or the starting values used in estimation are different.

Third, although far far less common, possibility is that the estimation is actually very very very close to being singular, and sometimes random machine noise pushes it into the singular region, and sometimes pushes it to non-singular.

[bareab87]

Two obvious and common causes for estimation results to differ: the underlying data are different, or the starting values used in estimation are different.

Third, although far far less common, possibility is that the estimation is actually very very very close to being singular, and sometimes random machine noise pushes it into the singular region, and sometimes pushes it to non-singular.

Thank you, at least that gives me an idea on where I should look, I've been quite clueless until now.

What I'm trying to do is to estimate the short-run dynamics of a VECM-model, consisting of 3 equations (using system, append and FIML). If I remove one of the equations, the problem disappears (I get the same estimation results every time).

Also, if I keep all 3 equations, and restart Eviews so that the first time I run the program I get no warning on singularity, I get the message "convergence not achieved after 500 iterations".

Could this give a hint on whether the problem could be the database or different starting values? As long as I am opening the same database each time, how could that be the issue?

[EViews Gareth]

If the equation is not converging, all bets are off

[bareab87]

If the equation is not converging, all bets are off

Meaning....?

What problem should I focus on fixing first: that I don't get the same estimation results/I get a warning when rerunning a program, or that the equation is not converging if I restart Eviews. Or is the first a consequence of the latter?

I am running the same estimation in PcGive, that's why I'm so confused. I wouldn't even know about this problem if I hadn't decided to give it a go in Eviews.

[EViews Gareth]

Meaning that nothing with the equation estimates are valid if it has not converged.