Question about ARMA forecasting

For requesting general information about EViews, sharing your own tips and tricks, and information on EViews training or guides.

Moderators: EViews Gareth, EViews Moderator

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Question about ARMA forecasting

Postby startz » Mon Dec 16, 2019 10:38 am

I have a dumb question about dynamic versus static forecasting. I have estimated the following equation from 1953 through 2018

Code: Select all

ls spreaddemeaned ar(1) ar(2) ma(1)

If I forecast for 2019 from the equation using dynamic forecasting I get all zeros. If I forecast using structural forecasting I get a zero for 2019M01 but otherwise numbers that look like reasonable forecasts.

Can someone explain to me what's going on?

(And yes, I have read the section in EViews Illustrated and still don't get it.)
Attachments
three-month and 10-year governments.wf1
(38.16 KiB) Downloaded 347 times

EViews Glenn
EViews Developer
Posts: 2671
Joined: Wed Oct 15, 2008 9:17 am

Re: Question about ARMA forecasting

Postby EViews Glenn » Mon Dec 16, 2019 12:06 pm

Funny, I'm not seeing that behavior. Estimation from 1953 to 2018 gives estimates with 789 observations for ARMA-ML. Dynamic forecasting for 2019 gives values starting at -1.04 and ascending gradually to -0.55

As to the behavior that you reported...

I'm not saying that this is what happened, but note that if you estimated the equation for 2019 and then tried to forecast for 2019, you might see the behavior that you observed for the dynamic forecast.

The lagged residuals to initialize the ARMA forecast would be 0 since those were out-of-estimation-sample values with no presample residuals. If the estimation and forecast period are at the beginning of the workfile, EViews will try to adjust the start date of the forecast for you so that you get the required two presample residuals. If in the middle of the workfile, EViews is more respectful of the start of forecast period and does not adjust the start of the requested forecast period for you. In this case, adjusting the forecast period forward by 2 periods yields non-zero forecasts.

However, this may not be the explanation as one would still see zero values for structural forecasts in that period. In fact, unless my brain isn't working properly, I would expect all structural forecasts to be zero for your model whatever the samples.

Note also that while EViews Illustrated has a fine discussion of forecasting, some of the idiosyncracies of presample initialization of residuals and forecast samples are probably beyond the scope of that book.

startz
Non-normality and collinearity are NOT problems!
Posts: 3775
Joined: Wed Sep 17, 2008 2:25 pm

Re: Question about ARMA forecasting

Postby startz » Mon Dec 16, 2019 12:29 pm

Thanks Glenn. I am now getting the same forecasts that you report. Despite having gotten something different 6 times before I posted.


Return to “General Information and Tips and Tricks”

Who is online

Users browsing this forum: No registered users and 18 guests