FMOLS - Error message

[bergeronsamuel]

Hi,

I want to run estimate using the FMOLS method. I previously runned Pedroni's cointegration test.

I select Quick / Estimate equation / Method : FMOLS

I put my three variables and I select Trend Specification : Constant (level), no additional trends, Method: FMOLS, Panel Method: Grouped.
Then I click on Long-run variances: Options button and select Lag specification: Auto-AIC, maximum lags : *, Kernel: Bartlett, Bandwith method: Newey-West Fixed, I click OK, and OK again.

I get the following error message:
"Unable to compute due to missing data in FMOLS estimation for cross-section #."

I tried it for my 3 different group files and always get the same error message but for a different cross-section #. I've looked again and again at my data and there is no missing data at all. I have a fully balanced panel.

The problem seems to come from the Lag specification option. If I select none instead of Auto AIC, then I can run the FMOLS regression, but all my panel unit root and cointegration test were done using that method so it would be consequent to use it for the FMOLS regression as well. I also tried with a fixed lag of 1,2 or 4 and always get the same error message.

Thanks for your help!

Samuel

[EViews Glenn]

Hard to say without seeing the data. You can post the workfile with a description of what you are trying to do and we'll take a quick look when we can. By the way, how many observations do you have in your panel and which panel cointegration test did you perform?

[bergeronsamuel]

Hi,

I have joined the file.

My panel data consist of 12 cross-section, each having 17 years of data for my 3 variables. I'm trying to run to cointegration test for the group called cointegration_gdp in the file.

I performed Pedroni panel cointegration test using the AIC automatic lag length selection with the Bartlett - Newey-West automatic criteria.

This is for my MSc. Dissertation and the draft of the results is expected asap.

Thanks for your help!

[bergeronsamuel]

Any news for me about the problem?

[EViews Glenn]

Sorry, I'm out of the office and haven't been able to look at this. I did just take a quick look but haven't been able to figure anything out from my quick look. I think I'll need to get back to the office to look more closely (which won't happen until the beginning of next week). Sorry for the delay. Thanks for being patient.

[EViews Glenn]

Just want to give you an update. It appears that the prewhitening VAR used in forming the long-run variance estimate in estimation is not stable for some of your cross-sections, so we are unable to recolor the long-run covariance estimate of the whitened data. The test procedure uses a different method for computing the prewhitening step and stability is not an issue for those estimates. Note that this only an issue for the combination prewhitening-kernel method of computing the long-run variances.

If you want to estimate using the same calculation that was used in the testing step, you should turn off the prewhitening in the test by setting the user-specified lag to zero, and leaving the lag specification for long-run variances calculation in estimation at "none".