Max number of time series in factor analyis

[fboehlandt]

Hi everyone,

I am trying to 'make factors' for time series within a pre-specified group. The resulting error message is as follows:

'eviews specification too long for estimation/specification by dialogue'

Admittingly, the number of time series at just over 1600 would put most programs out of business. Does anybody know what the upper limit for factor analysis is? The error message let me to believe that triggering factor analysis from code might help but the program just crashes :( I might be able to bring down the number of series in the model but would like to know what is required. Thanks in advance for your input.

Best regards

Florian

p.s. please, this is merely a functional enquiry, not one concerned with the statistical implications of a massive model like the one mentioned above

[EViews Glenn]

It shouldn't crash. From a practical point of view we should handle things until we gracefully handle an error like running out of memory. Apparently, we're not doing the "graceful" part.

How many observations do you have? Is your workfile very large? If not, posting or sending to support@eviews.com should help is fix the crashing. Information about the exact operation you were performing, the amount of memory on your system and the version/date of EViews would be useful.

[fboehlandt]

Hi Gareth,

I include the message I have sent to eviews support. Just in case something catches your eye
Regards
Florian

Hi,

I have the following problem:

I have a large workfile with many timeseries which I openend as a group. With the group opened I am trying to open the dialogue for creating factors (Proc->make Factor). The action is aborted with ‘Specification too long for estimation/specification by dialogue. This led me to try execution by code. However, that results in Eviews crashing: ‘Eviews has stopped working’.

Additional information:
- # of timeseries > 1500
- # of observations per series = 60
Specifications:
- Eviews Version: 7.2 – Jun 15 2012 build
- System Specifications:
o Windows Vista Enterprise
o Intel Core i5 – 2520M CPU @ 2.50 GHz
o 4.00 GB RAM

I am assuming that ‘make factor’ may not work due to the discrepancy between the number of timeseries in the model versus the length of the series itself. However, I didn’t expect Eviews to crash on execution. Is there anything I can do to keep that from happening?

Regards



Florian

[fboehlandt]

Hi,
adding a few series at a time I was able to produce a more meaningful error message: 'the number of factors to extract is too large'. This occurs somewhere around 330 dataseries (each with 60 observations). Any thoughts on this?

regards

Florian