EViews.com

Hi everyone!
Few days ago I downloaded Eviews 8.1 in order to achieve the last chapter of my master thesis. I have never used it before and try to go on with it. I'am actually trying to implement a bi-varied Vector Autoregressive (VAR). It's a unrestricted VAR with two variables: the GDP growth which is yt= log_GDP(t)-log_GDP(t-1) and the inflation which is pt= log_GDP deflator(t)-log_GDPdeflator(t-1). I take the first log difference of the variables in order to make them stationnary. My purpose is to evaluate the impact of a supply shock and a demand shock on this two variables, knowing that shocks on the demand have no long-run effects on the GDP.

Here begins my issues, while I was discovering the program I'have heard about the long and short run restrictions. This is my point, first of all I have implented a model with short run restrictions by building two matrices assuming one with a digonal of 1's because of the univariated character of the variance of the forecast errors and then a second orthogonal one with a digonal of "NA" and 0 anywhere else because the covariance of my forecast errors are nul.

When I realised that I could force the long run response of a shock in the demand on the yt (growth of GDP) to be nul, I decided to do it by implementing a long run matrix with the response coefficient of y to the demand shock equals to 0 and others to "NA". However, as you know it's not possible to implement both short and long run restrictions so I am currently keeping the long run one, but I don't feel that satisfied with it. I'have got the feeling that I'am missing something in the process to write this kind of model. I am actually wondering what about the variance and covariance of my forescast errors if the the short run restrictions are suppoosed to fix this. I feel like what I did it's "too simple" and don't describe the reality I want my model to show me.

Perhaps, I don't really understand how to use these restrictions. If anybody could help me by telling me if I am wrong or not and If am, how could I get it, please? He would be blessed! :D

Hello,

On the practical side, you can use the EViews addin "SVARPatterns" to construct SVAR models with simultaneous short-run and long-run impulse response restrictions, although "short-run" is expressed differently in the addin than in EViews. Alternatively, EViews 10 (currently in beta) has this capability natively.

On the theoretical side, it sounds like you need to do some research on SVARs. One important factor is the number of restrictions (assumptions) you can impose. With a two variable VAR, if you wish to restrict the long-run responses then you can impose only a single such restriction (and no short-run restrictions) without risking over-constraining the model.