EViews.com

Hard to know.

Don't include the constant term.

I have added code in the program to save the S.E. of regression also in the results output. Its working fine.

Hello. Running Eviews 9

I believe there is a small bug in the Advanced Rolling Regression with Anchored Start. When the sample is different from the full sample the plug-in will still store estimated coefficients (and other elements) as if the starting point was the first observation in the workfile. I think that the following fix is necessary in advroll.prg at line 268 and after:



becomes



Thanks a lot!

Updated.

I'm trying to get the confidence interval in the rolling coefficient graph (and other graphs as well). Does anyone know how to get it? I able to get it with an older version (I guess). But it seems like I couldn't get the confidence interval anymore after I switched to Eviews 9 and re-downloaded the addins. Would greatly appreciate for your help!

EViews Gareth wrote:Assuming you have a simple linear regression, you'll have to write out the algebra in a series expression manually. Something like:

Code: Select all

series fcast = coef1 + coef2*x1 + coef3*x2


Hi. Can the rolling regression be applied to ARDL bound testing approach (or any other cointegration approaches) in EViews? Eagerly waiting for your response.

which methods is suitable to detect multicollinearity under rolling regression? Nither VIF nor correlation plot isn't working for rollings. Any idea please?

Dear Gareth,
I have downloaded a rolling add_in.
I did it, it seems working but when i checking results the rolling estimate is just an exact copy of the one with out rolling? what could be the problem and why?

Dear Gareth,

Could you please confirm that the version of the add-in "roll" on the website is actually the most updated version, as it is dated 2010?

Many thanks and kind regards,

Francesco

It is.

Many thanks Gareth.

I run the roll add-in on a single OLS equation with monthly data from 1998m01 to 2014m09, i.e. y=c(1) + c(2)*x + c(3)*z + e, on a window of 36 observations and with 1-step size. Therefore I was expecting the first value of the parameters' estimates, e.g. c(1)_hat, c(2)_hat and so on, to be reported on 2001m01, but instead they are reported on 2000m12, which I believe is incorrect.

Could you please let me know if this is correct and a deliberate coding decision, or I need to fix it in the code (if so, grateful if you could assist me).

Many thanks and regards,

Francesco

If your window size is 36, the first regression will run from 1998m1 to 2000m12 (36month), and will thus report its results in the last observation, which is 2000m12.

Many thanks Gareth for the helpful clarification!
Kind regards,
Francesco