SA bug

[svilen]

I found strange bug while using tramoseats for seasonal adjustment.
Please download attachments.
If you open the program in windows explorer (by clicking on it) then it produces nice sa results for both ser1 and ser2. Now, while it is still on screen, if you try to run it inside Eviews by click on the Run button you will see that for the ser1 the sa series is shown as NA.
It is interesting that in ser2 there is a missing observation and if you fill it then the above bug doesn't happen.

[EViews Gareth]

I cannot replicate it. If I run the program, it produces NAs for ser1.

[svilen]

I cannot replicate it. If I run the program, it produces NAs for ser1.

That was my point ser1 has seasonality but eviews says sometimes it hasn't sometimes it has!
Let me put some screenshots here:

[svilen]

I cannot replicate it. If I run the program, it produces NAs for ser1.

Gareth, could you replicate it or should I provide more info?

[EViews Gareth]

Ha, I can replicate it. Weird. As you probably know, we don't actually write the Trammo/Seats routines, but I'll see if we can figure anything out.

[EViews Jason]

We are working on a fix. The TramoSeats library is not reinitializing on subsequent runs for ser1. For the time being, you will have to restart EViews everytime.

[EViews Gareth]

To expand on Jason's post somewhat... Multiple runs of Tramo/Seats in the same running instance of EViews can cause a problem. This is due to an issue in the Tramo/Seats libraries that we call. EViews opens those libraries when it starts up and closes them when it exits. We're going to change it so that EViews opens and closes them every time you run Tramo/Seats, which will make things infidecimally less efficient, but should fix the issue.

In the meantime, as a possible work around you'll have to restart EViews every time you want to seasonally adjust a 2nd series. Obviously when you're running a program like yours, this is not feasible. Hopefully our patch will be out soon.

[svilen]

Hm, I don't feel like I wont the change that you suggest. If it is going take more time for sa procedure it will make some of programs that I use very slow (sa on +50 series). I would prefer to live with it in exchange for current speed. And it seems to me like exception that appears only when second series has missing observation (fill ser2 and check the result).
Maybe there is another solution. As of June'10 there is a new version of Tramo/Seats published here: http://www.bde.es/servicio/software/programase.htm As I could see your dll libraries are 10 years old so you could try with the new ones.

[EViews Gareth]

It will probably only slow things up by fractions of a second, even if doing it many times.

Moving to the new Tramo/Seats DLLs is something we're already looking in to, but it is a major job (since they have changed a lot), so it is doubtful it would be released until EViews 8.

[EViews Jason]

svilen,

A new patch is now available. FYI TRAMO/SEAT processing has slowed roughly by 18% but it will depend on your dataset. The good news is the returned results will be consistent.

[svilen]

Thank you Jason.
I'll test it.

[svilen]

It's working fine. Could you please fix the issue in EV6 as well.

[EViews Jason]

Support for versions of EViews 6 and earlier is limited and unfortunately the workaround will not be put in.

[EViews Glenn]

I might add that the original bug is in the Tramo/Seats code which we have no control over and the fix posted to EViews 7 is a workaround that required a fair bit of effort to implement. Given the fact that EViews 7 has been in release for over a year, we are only providing limited support for Version 6.

[donihue]

Further to this, my attempts to use TRAMO in EViews 7.1 have been frustrated by the repeated error message shown.

I am guessing that this has something to do with where EViews is installed (in my case in C:\Programmes\EViews7) but have no idea what to do to make it work.

Can you help?

Regards
Donihue