## Innovation generation

For technical support, tips and tricks, suggestions, or any other information regarding the EViews model object.

Moderators: EViews Gareth, EViews Moderator

aamaro
Posts: 1
Joined: Thu Mar 17, 2016 9:32 am

### Innovation generation

Hi all,

I am working at a stochastic model solving for var forecast on Eviews8. I have a Svar identification varcov matrix. How do I include that in the Innovation generation to substitute the Choleski?
Form Eviews help Innovation generation : "When normal random numbers are used, a set of independent random numbers are drawn from the standard normal distribution at each time period, then these numbers are scaled to match the desired variance-covariance structure of the system. In the general case, this involves multiplying the vector of random numbers by the Cholesky factor of the covariance matrix. If the matrix is diagonal, this reduces to multiplying each random number by its desired standard deviation.

Thanks,

Alessandro

EViews Glenn
EViews Developer
Posts: 2600
Joined: Wed Oct 15, 2008 9:17 am

### Re: Innovation generation

For model solution, EViews always uses the reduced form VAR results and the reduced form residual covariance matrix. Thus, if your SVAR is just identified, then the stochastic results will be the same; otherwise, as you point out, the overidentification restrictions will not be taken into account in the stochastic model simulation.

In the latter case, the only way to do what you want would be to estimate the equivalent of your SVAR using the system object using FIML and then to include this estimated system in the model. Fortunately, the just released EViews 9.5 allows you to estimate your FIML assuming diagonal variances, so you should be able to estimate the equivalent SVAR using a system object which you can then include in the model. You should then be set since the diagonal variances and SVAR restrictions are built into the system.

What I would do if I were you is to make a system from the VAR using Proc/Make System, and then to edit the specification to add the contemporaneous dependencies from your structural VAR. You can then estimate the coefficients of the system using FIML while imposing diagonal variance restrictions. Then you should be good to go.

jason83189
Posts: 4
Joined: Wed Mar 16, 2016 10:03 am

### Re: Innovation generation

Hi Glenn,

I have a bivariate VARX-MGARCH-MEAN model and try to shock one of my exogenous variables and examine the response of my endogenous variables.
I created two equations and merge them into a system.
I solved the model and get solutions for both of my endogenous variables.
I then create an unit shock to one of my exogenous variables as scenario 1, and solved the model again.
I subtract the scenario 1 results from the baseline results, and get the response.

My question is, is it possible to add error bands into my graph?

Thanks.