Forecasting with Models from a VAR

[brennan6738]

Can someone assist me with this? (Sorry, I could not figure out how to start a forum):

I can't seem to figure out how to get Eviews to forecast one period into the future. Here is basically what I do:

1. I do "Proc/Structure/Resize" and move the End Date one period forward.
2. I then open the VAR and change the Estimation Sample to one quarter forward. (Though the Estimation Output does not seem to reflect this).
3. I then select "Proc/Make Model" in the VAR. I use the full sample plus the additonal quarter that I resized to.
4. I then "Solve" the model, but I get an error that reads:
Solve terminated - Unable to compute due to missing data in "MUNICIPAL = @COEF(1) * MUNICIPAL(-1) + @COEF(2) * MUNICIPAL(-2).

Am I missing any steps here?

[EViews Gareth]

Sounds like you don't have values for MUNICIPAL for the periods prior to the forecast period.

[brennan6738]

Hi Gareth
I created two small models and attached them to illustrate my point a bit better. Each model has the same four variables and 30 observations. For the model "30 obs" I simply estimate the VAR with the sample set to 1 30. When I Solve the model I set the solution sample 1 30. I then get a message saying solve completed, but this only models to the 30th data point (I want to forecast the 31st data point). So for the "31 obs" model I try something different. First, I Proc/Structure & Resize and set the number of observations to 31. When I run the VAR I set the estimation sample 1 31. When I solve the model I set the solution sample 1 31. The message reads "Solve Terminated" and when I open the new series obersvation #31 reads N/A. When I solve and set the sample to 1 30 it says it completes the model, but the 31st observation in the new series still reads N/A. I think I just have the settings not quite right. Any suggestions? I definitely do not have balues missing prior to the forecast period. Thanks for such a fast response.

[EViews Gareth]

You don't have any values for X, Y, or Z for the forecast period.

[brennan6738]

OK. I think I see the issue. Basically I need to use lags to get the model to work. I still don't understand what is happening with a VAR though. When I change the order to four, it only seems to include four lags of the dependent variable. I thought the whole point of a VAR is when you spoecify the order it is including that many lags of dependen and indpendent variables. The output just seems to show the independent variables in one period though (and all the lagged coefficents for the dependent variable). Any suggestions?