"Missing Value found in covariance/correlation matrix"

[ross76]

Hello,

I am now trying to run a PCA analysis on Eviews. The group on wich I am trying to run the PCA has over 800 series.
When I try to get the PCA's table this error message pop up "Missing Value found in covariance/correlation matrix".

I though this error could be caused by too many NAs or "0" in my sample.
However I did the test with series containing a lot of NA's on a smaller sample and it works just fine.

It seems the rejected series have a big succession of Zeros. I cannot spot the rejected series since I have too many and I cannot test them one by one.

Is there any rule stating on which aspects Eviews accepts or rejects series when running a PCA?

() I am using Eviews 6, and I am a beginner.

[EViews Glenn]

Do you have a series with zero variance?

[ross76]

Thanks very much, I deleted the series with zero variance and it works perfectly.

Cheers,

[ross76]

Hello again,

I in fact still have a problem, even when all the series have a variance different to zero, the message error pop up again on some of my sample.

So the PCA doesn't work with series with zero variance, but it has to be something else.

On Eview 6 the message is still the same "missing value found in the covariance correlation matrix"
On Eview 5 the message is "Insufficient number of observations".

I get a covariance matrix full of NA's/

I have no idea how many observation minimum have to be in the serie so Eview can take it into account (if this is the problem).

Do you have any clue what the problem is?

Thanks again

[EViews Glenn]

It's hard to diagnose this without seeing the data. You should send the workfile with the data and the specific examples where you get this to support@eviews.com. Please include your serial number. Thanks.