The RATS Software Forum

Dear Tom,

If I run CATS with export=VAR then I get the implied VAR model. How can I fit it to get all %... variables to do error bands such as bootstrap? I tried some options but it is not very convenient. Is there a proc get estimated VAR model (model with all given coeffs) then compute %sigma, %... as we estimate the var as usual?

Best,

Hi Tom,

In addition to question above, the VAR model exported from CATS if estimated using ESTIMATE would report different coeffs as they would have lost cointegrating information. So, if I do bootstrap I get the similar but not exact IRFs I've got from @VARIRF. Why it is so?

Best,

CATS is only going to do point estimates and the error correction terms are substituted out. The MC integration would have to be done in RATS using a SYSTEM with ECT terms as is done in the MONTEVECM.RPF example. The difference would be that the error correction equations would be coming out of CATS rather than out of JOHMLE---though JOHMLE will generally be fine once you've decided on the rank using CATS.