Something mysterious to me..

[DBel2012]

Hi,

I'm using EViews9, July 28, 2015 Build. Still waiting for my company to install the last patch!

I was wondering if you could help me understand something mysterious to me...

I've attached a workfile and a small program which estimates 3 equations.

I've noticed that the order of the appearance of the first two equations (epargnemensuel and epargneannuel) in the program file will not produce the same results for the third equation (epargnebilan.arch).

Is there something important missing in the third equation? Is it a question a starting values? Any ideas?

Thanks in advance for your help!

Danny

[EViews Gareth]

Usually I would jump to say starting values - since most estimation in EViews uses the C vector in the workfile for starting values, and each estimation overwrites the C vector, the order in which you estimate equations can impact their results as starting values change.

However for ARCH models the default is not to use the C vector, rather to use EViews' own supplied starting values. So in your case the starting values are not impacted by previous estimations.

I think in your case the equation simply is not converging well and is on a knife edge between two possible solutions. If you open up the equation and repeatedly estimate it, you can see it seems to jump from one set of solutions to another.

The warning messages on the equation output do indicate that convergence isn't necessarily achieved.

[DBel2012]

Thanks a lot Gareth for you help!