Singularity Errors

[CharlieEVIEWS]

Dear all,

Apologies for what may be a ridiculous question. However, it's been bothering me for some time, and may be important for further progress on a project.

When I attempt to undertake a Johansen test (option s (6)) using the attached workfile, I receive a 'Near Singular Matrix Error': please see VECM_group in the .wf (1 lag on the test). However, when I change the order of the variables and perform the same test, no error appears (vecm_group1). Further to this, a quick visual inspection of the plots('graph' in the .wf) indicates that none of the variables are (even close to being) exactly co-linear? This should give the Pi matrix a reduced form rank, and hence no singularity, if I am not mistaken? Apologies, my VECM theory may be somewhat lacking, but a brief explanation of what is going on in the underlying EViews code would be more than extremely helpful.

Best wishes, and thanks as always for all the of the help,

Charlie

[EViews Gareth]

One of the matrices that are computed during the test are probably very close to being singular. Machine inaccuracy means that in one order it fails the singularity test, and in another order it just passes.

[CharlieEVIEWS]

Thanks Gareth, this is more or less as I expected. If I may ask one more question (I am generating an extremely large number of Johansen tests/VECM tests in a computational framework for simulated series): is it true that if the matrix is truly singular (exactly co-linear series), as opposed to being close to singular, then the ordering wouldnt matter and an error would result? Do you have a quick way I can check for exact singularity in EViews? (I'm thinking something through the View --> Covariance Analysis window, but cant figure out exactly what it would be off the top of my head)

Best wishes,

Charlie

[EViews Gareth]

There is no way to check for exact singularity.