The RATS Software Forum

Hi Everyone,

I am using the code of Cushman David and Zha (1997). And I use a similar number of endogenous number of variables (7 domestic-3 foreing) but only 1-2 lags. In my dataste I have 240 observations.

My question is, given that I am using approximately 10-11 endogenous variables; Should I concern about the Over-dimensionality issue? Or how do I know when I should concern about it?

Thanks so much,

I appreciate it,

Fernando

How does a tight Bayesian Prior to estimate the standar errors relate to the number of variables in a SVAR?

lfc160030 wrote:Hi Everyone,

I am using the code of Cushman David and Zha (1997). And I use a similar number of endogenous number of variables (7 domestic-3 foreing) but only 1-2 lags. In my dataste I have 240 observations.

My question is, given that I am using approximately 10-11 endogenous variables; Should I concern about the Over-dimensionality issue? Or how do I know when I should concern about it?

Thanks so much,

I appreciate it,

Fernando

No. The small number of lags makes a big difference. The computational burden of a VAR that isn't a full symmetric model goes up with the number of coefficients cubed. 200-300 is doable, 1500-2000 probably isn't (or is, but it will take an overnight run). Note that if 200 coefficients takes about 2 minutes, 2000 would take a day and a half.