The RATS Software Forum

Dear Tom,

Are you aware of RATS codes on a VAR with an interaction term? Like in this example

http://www.norges-bank.no/pages/95985/N ... 013_17.pdf

where the responses of the endogenous variables to a shock depend on the (exogenous) level of uncertainty.

Thanks for your help

Peter

Given how they are doing that, it's not particularly difficult. The X x Y variables are just "data" and they're using a flat prior, so the Monte Carlo integration for the coefficients is just standard. What's non-standard is the generation of the IRF's, but they avoid the problem of computing IRF's in a non-linear system by doing the calculations with just two fixed value settings for X (the empirical 10%-ile and the empirical 90%-ile). With that, the model, for purposes of computing the IRF, becomes a standard VAR. You can draw the coefficients from the augmented model, then build a coefficient matrix for the corresponding reduced model and generate the IRF's using the latter.