EViews.com

Hey,

my question is very easy. At the beginning I specify an equation. I want to do "real-time fitting", i.e. calculate the fitted values based on the estimation of data up to this point. Is there a way to reestimate an already existing equation (with a different sample) without specifying it again?

Let's say I have data from 1970 to 2010. I want to to "real-time fitting" for the period 1990 to 2010. In a loop I start with the estimation of the model based on data from 1970 to 1990 and then I save the fitted values for 1990. In the next step I do this again for 1991. As far as I see I have to specify the equation each time I want to estimate it. Is there a way to just reestimate it for a different sample (something like equation_name.update)? I've already looked in the command reference, but haven't found anything that helps.

Thanks in advance!

Best,
Thomas

No one?

Hello,

An equation object stores its own specification, which you can access to reestimate the equation without explicitly respecifying it:


You'd need to set the new sample before the above code, as the above implicitly uses the current workfile sample.

Just learned from EViews Gareth that there's a currently undocumented reestimate proc for equations. The following reestimates an equation with its original sample.


The following reestimates an equation with the provided <sample spec>.

That's awesome, thanks a lot! Currently I use @varlist to get the variables from the equation to estimate it again. Your solution is obviously far better!

I just tried it. Unfortunately it doesn't work. I estimated the equation for all years up to 1994. Then I tried to reestimate it for all years before 2000:


The result still says:


What did I do wrong?

Your example should work via:

It does, thanks a lot!

I have an additional question. If I set the sample in the "Estimation settings" (and not in the workfile), then the sample given to the reestimate procedure seems to be only a "subsample" of the one in the "Estimation settings". In other words, if there isn't the whole sample in the settings the reestimate procedure will not work "properly". Is there a command to change the sample in the settings? I just want to make sure that the whole sample is selected before doing the reestimate.

E.g. in the "Estimation settings" the sample "1970 2002" is used. The command


will then only use the sample up to 2002. I assume that the reestimate procedure will also update the sample used in the makeresid procedure?

You should be able to include @all in the reestimate sample, just as if you were using the smpl command:

Thanks a lot!