Pooled Mean Group estimation and singular matrix

[hk976]

Hello,
I am doing PMG estimation with my panel data and i divide the sample data into 3 sub-samples.
However, i can run a regression for the first period and get the result but not the two other periods. I got the error of near singular Matrix.
I tried VECM or VAR, but it did not happen, though. Could you help me out please?
FYI, I do not have any dummy variable. My variables are bond rate, GDP growth rate, real exchange rate, current account-to-GDP ratio, and debt-to-GDP ratio.
Thank you for consideration.

Sam

[EViews Gareth]

We'd need to see the workfile.

[hk976]

Attached is the file concerned.
Equation specification is
spread gdp rex debt ca (spread is dependent variable and others are independent ones)
and there are three periods: 2000Q1 - 2008Q2, 2008Q3 - 2012Q2, and 2012Q3 - 2015Q4
PMG estimation of the last two periods shows singular matrix error sign.
Thank you for consideration

Sam

[EViews Gareth]

That workfile appears to be corrupted. Make sure your copy of EViews is up to date.

[hk976]

Could you try this again?
I am using up-to-date student version of eviews 9.5.
If it is not working, let me send you my excel file to your personal email.
Thank you so much

[startz]

Can you give the details of the estimation command?

[EViews Gareth]

I don't think you have enough observations to run with the default 4 lags in those latter samples. You'll need to reduce the number of lags (or use longer samples).

[hk976]

This is the estimation command: spread gdp rex debt ca
I have 9 countries (panel data) but still is it not enough data?
And, is that why the error pops up?? It does not seem related to "singular matrix".
Thank you for consideration

[EViews Gareth]

No, you don't have enough observations, and not having enough observations will always result in a singular matrix.

[startz]

Note that pmg automatically inserts lags. That's why it's easy to run out of observations.

[hk976]

I divided the 9 countries into two panel groups (4 countries vs 5 countries)
I got the estimation results of all three periods of the first group.
However, I got the result of only first period of the second group the error pops up for the other two periods although I set the lag at the minimum and the second group has more observations than the first one.
I guess it is not just about the observations. (btw, I checked the correlation between variables and all are below 0.5)

[EViews Gareth]

It is almost certainly due to number of observations.