Page 1 of 1

Direct Forecasting with VAR Models

Posted: Mon Jan 05, 2009 3:07 pm
by Harvard
Dear colleagues,

I have created an var2 model by following command: var var2.ls 1 2 a(-10) b(-10).
With this model I want to forecast the 10th period in the future directly.
But when I use the command "makemodel", an error message appears: " a is not a variable in the model in ":var2" ".
Has anybody an idea, where my mistake is?
The error doesn't appear, when I use a var model like this: var var2.ls 1 2 a b

Greetings

Re: Direct Forecasting with VAR Models

Posted: Mon Jan 05, 2009 3:12 pm
by startz
Harvard wrote:Dear colleagues,

I have created an var2 model by following command: var var2.ls 1 2 a(-10) b(-10).
With this model I want to forecast the 10th period in the future directly.
But when I use the command "makemodel", an error message appears: " a is not a variable in the model in ":var2" ".
Has anybody an idea, where my mistake is?
The error doesn't appear, when I use a var model like this: var var2.ls 1 2 a b

Greetings


EViews is often unhappy with a lagged (or lead) variable on the left of an equation in which a variable is being generated. My speculation is that you are effectively getting a(-10) on the left.

Re: Direct Forecasting with VAR Models

Posted: Mon Jan 05, 2009 3:20 pm
by EViews Gareth
This is probably a bug, and we're looking into it.

In the meantime, is there any particular reason to specify the VAR that way, its really the same thing as:

var var2.ls 1 2 a b

but estimated over a different sample (just put the estimation sample 10 periods back).

Re: Direct Forecasting with VAR Models

Posted: Mon Jan 05, 2009 6:19 pm
by Harvard
O.K. Thanks for your help. I think for my calculations the following command will be sufficient: var var2.ls 10 11 a b

Greetings