EViews.com

Hello there people of Eviews,

I think there is a bug in Eviews9 where, if you estimate a bunch of ARDL equations in a workfile, the coefficients for the each ARDL equation will be the same when viewed in Cointegrating and Long Run form, see below.



I think Eviews9 might just be displaying the coefficients from the most recently estimated ARDL equation?
When I view the Bounds Tests for each of these equations, the coefficients are different from the Cointegrating and Long Run form.

Build date of your copy of EViews (Help->About EViews)?

Hi there Gareth,

My build date is Apr 20 2015 build

I think my system administrators have disabled the update Eviews feature on my desktop, so it is probably out of date.

If this issue has already been fixed then please ignore my post.

Cheers

Andrew Lau

I would thoroughly recommend updating. You can try doing it manually:
http://www.eviews.com/download/download.html

Hello,

I think the bug of long-run form is still there. I am using Eviews 9.5 (updated September 2016). I found that the long-run coefficients calculated from bounds test are different from those from cointegrating and long run form. Please see pictures below.




For example: From bounds test (pic1), the long-run effect of RISK on LONG = -(0.442182/-0.163614)=-2.702592688. However, the result from cointegrating and long run form (pic 2) indicates that number is 0.196087, so much different from bounds tests. I used HQC information criteria in this example. When I changed to AIC, the bug disappears.

Could you please help me to fix that bug?
Thank you so much!

For ngantran:

The regressions you presented are not the same. There are two variables in the second model that are not included in the first one. So the models are not fully comparable.
p.s. Sorry, I was, probably, wrong if these two variables has zero lags. Thus, you question is relevant.

For Gareth:

It seems that there are some problems with ARDL.

1. When the lags are fixed then all is ok, but when the lags are selected by info criteria then the coeffifents in ARDL Bounds Test regression and Cointegrating Form are somewhat different. They have to be the same.
2. There is also another issue: fixed regressors should not be inclided in the long run equation. For this reason, current ARDL bounds test prosedure cannot replicate example from PSS(2001). One can do this only manually in Eviews.

It should be great if a new version of Eviews permits fully replication of PSS(2001). You can include an appropriate data file as an example.

Best regards,
Igor

Hi, thanks for your reply.

There is a case with zero lags in two variables. I realise that in this case, if you compute the long-run estimation based on bounds tests from Eviews, it will return different results from cointegration and long-run form. This error does not occur in the case with minimum 1 lags.

Could you please explain why as both of models, in theory, should return the same result?

Thanks.

You're right, there is a bug in the case that zero lags are selected. We'll look into fixing it (might take a while).

Any update on how the bug fixing is going? Is the bug still there?

Still working on it, but it shouldn't be too much longer.