The RATS Software Forum

Hi,

I would like to estimate a panel VAR which has 60 observations for 3 variables across 5 groups. I think I have found the code here to do the usual Eakin-Newey-Rosen type basic estimation of this.

However, I would then like to get IRFs for each group and each variable, as well as the aggregate model. So, the response of 3 variables to 3 shocks in group 1, in group 2 etc and then the aggregate across groups. What is the easiest way of doing this please?

furthermore, I would then like to extend the analysis to allow for time-varying coefficients. Would this be possible in rats?

finally, I may want to apply sign restrictions to the model. Again, would this be possible, at least for the basic model without time-varying coefficients?

any helps much appreciated - thanks!

MM72 wrote:Hi,

I would like to estimate a panel VAR which has 60 observations for 3 variables across 5 groups. I think I have found the code here to do the usual Eakin-Newey-Rosen type basic estimation of this.

However, I would then like to get IRFs for each group and each variable, as well as the aggregate model. So, the response of 3 variables to 3 shocks in group 1, in group 2 etc and then the aggregate across groups. What is the easiest way of doing this please?

The lag coefficients are homogeneous, so the IRF's are the same for each individual.

MM72 wrote: furthermore, I would then like to extend the analysis to allow for time-varying coefficients. Would this be possible in rats?

Time-varying how?

MM72 wrote: finally, I may want to apply sign restrictions to the model. Again, would this be possible, at least for the basic model without time-varying coefficients?

Not easily, at least basing this on the Holtz-Eakin,... method. Because it uses instrumental variables (and thus isn't likelihood-based), there's no simple way to generate draws as there is for a regular VAR.

Are you basing your questions on a particular paper, or is this a new concept?

TomDoan wrote:

MM72 wrote:Hi,

I would like to estimate a panel VAR which has 60 observations for 3 variables across 5 groups. I think I have found the code here to do the usual Eakin-Newey-Rosen type basic estimation of this.

However, I would then like to get IRFs for each group and each variable, as well as the aggregate model. So, the response of 3 variables to 3 shocks in group 1, in group 2 etc and then the aggregate across groups. What is the easiest way of doing this please?

The lag coefficients are homogeneous, so the IRF's are the same for each individual.

so it is of course...given that for my model, T is quite large and N is small, is there a way of doing a panel VAR with Rats where the coefficients are heterogeneous? because the N sectors present 5 sectors of the economy, I would expect the IRFs to be heterogeneous and I know their (weighted) sum would need to be the whole-economy response. Wondering if there is a way of accommodating all this?

MM72 wrote: furthermore, I would then like to extend the analysis to allow for time-varying coefficients. Would this be possible in rats?

Time-varying how?

so that the parameters would be allowed to vary in time, and IRFs would look different across time as well as across sectors (this would be interesting to see because of the potential effects of the financial crisis period).

MM72 wrote: finally, I may want to apply sign restrictions to the model. Again, would this be possible, at least for the basic model without time-varying coefficients?

Not easily, at least basing this on the Holtz-Eakin,... method. Because it uses instrumental variables (and thus isn't likelihood-based), there's no simple way to generate draws as there is for a regular VAR.

Are you basing your questions on a particular paper, or is this a new concept?

No, this would be a new concept, so I realize this may not work.

thanks again!