The RATS Software Forum

The error bands for the IRF's are doable by computing and saving the IRF's inside the draw loop (rather than saving only the coefficients and doing the IRF's only at the average coefficients). However, it's not clear how you would ever present that information---the graphs would be an unreadable jumble. The whole point of doing the IRF's with panel data is to give an indication of the robustness across countries, which is probably more useful information than the single country error bands.

I completely agree with you Tom. How can we do variance decomposition in this model?

TomDoan wrote:The error bands for the IRF's are doable by computing and saving the IRF's inside the draw loop (rather than saving only the coefficients and doing the IRF's only at the average coefficients). However, it's not clear how you would ever present that information---the graphs would be an unreadable jumble. The whole point of doing the IRF's with panel data is to give an indication of the robustness across countries, which is probably more useful information than the single country error bands.

I please need some help with this code to obtain average coefficients, IFRS and confidence band for the entire sample rather than individual countries-which the code seems to be doing.
Thanks.
oz

The average is simple (you have all the responses, so you just need to average them). I'm not sure what the rest of that means.

TomDoan wrote:The average is simple (you have all the responses, so you just need to average them). I'm not sure what the rest of that means.

Thanks Tom. I guess as you suggested, averaging the individual estimate is straightforward.
However, I am not sure how to obtain the confidence bands for the (averaged) IFRs once the average is obtained.

Alternatively if there is another PVAR code in Rats which could be used for this purpose please let me know. I have purchased the course material but does not seem to include a PVAR which offers IFRs (for the whole sample).
Gratefully
oz

If you can compute the average for each draw, then you can compute error bands based upon those using the same techniques as would be used with any other IRF calculations. Note, however, that that is really defeating the purpose. The point of the shrinkage estimator is to allow them to be similar, but different. What's the point of averaging them? If you want them to be the same, then you would use a different estimator in the first place. The point estimate of the average has some value as a summary, but I fail to see what error bands on the average would accomplish.

TomDoan wrote:If you can compute the average for each draw, then you can compute error bands based upon those using the same techniques as would be used with any other IRF calculations. Note, however, that that is really defeating the purpose. The point of the shrinkage estimator is to allow them to be similar, but different. What's the point of averaging them? If you want them to be the same, then you would use a different estimator in the first place. The point estimate of the average has some value as a summary, but I fail to see what error bands on the average would accomplish.

Thanks Tom, I see your point.
oz