Rolling regression - is this what I want?

[VictorX]

Hi all,

I have been using Excel to run linear regressions, and I am very new to EViews. I am wondering if something I have been doing in Excel is what people call "Rolling Regression" - for every data point, I would compute coefficients using that data point and all prior data points before it, such as:

2012 Jan (compute coefficients by running regression from 2001 Jan to 2012 Jan)
2011 Dec (compute coefficients by running regression from 2001 Jan to 2011 Dec)
2011 Nov (compute coefficients by running regression from 2001 Jan to 2011 Nov)
2011 Oct (compute coefficients by running regression from 2001 Jan to 2011 Oct)
2011 Sept (compute coefficients by running regression from 2001 Jan to 2011 Sept)
.....
2001 Jan (compute coefficients by running regression from 2001 Jan to 2011 Jan)

I read "EViews Gareth's list of useful threads and FAQs", and it seems Rolling Regression is possible via programming.
* Is "Rolling Regression" what I have been doing in Excel?

Thanks

Victor

[EViews Gareth]

Yes, but with a fixed starting point and rolling end point. If memory serves me
correctly, the rolling regression add-in handles this case.

[VictorX]

Thanks Gareth,

I got it to work! This is very robust - I anchored the start date according to the manual and set the step size to 1 (increment observations by 1 per each rolling regression).

Is there a way to change the View > Actual,Fitted,Residual graph to use the new rolling regression? This is still tied to the original full regression.

Or do I have to manually compute and graph the Y levels from the rolling coefficients?

Thanks!

Victor

[VictorX]

Actually I think I figured it out.

Once I ran the rolling regression, the updated rolling coefficients are saved in the workfile for further use.

I can then go to View > Show > NewFit = rollcoefs_a * a + rollcoefs_b * b + rollcoefs_c * c

This NewFit will be saved as a new object, and I will be able to graph that object as the new fit under rolling regressions.

The process is somewhat manual, and I will need to generate a new residual variable, but it is workable.


Victor