Conditional impulses

[luching]

Hi Tom, I am trying to generate conditional impulses from a VAR. Essentially, impose that the responses of a variable are set to zeros while computing the set of impulses from the VAR system. I looked at the "condition" procedure. If I can add the shocks, I can do a work around from the conditional forecasts to get the conditional impulses. Is there any way to do that? Please help.

[TomDoan]

Those are two completely different animals. In conditional forecasting, an N variable system over H periods has NH total shock components, so you can have quite a few constraints and still have many ways to hit them---the point of conditional forecasting is to find the most probable of the ways of making the constraints happen. In an N variable system, an IRF is generally just an analysis of the behavior of period 0 shocks, so there are just N shock components. Unless you're using some type of restricted VAR (with constraints on the lags), you can't have more than N distinct constraints, and if you do anywhere near N, you are so severely constraining the shape of the shocks that it's unlikely to be interesting.

As long as you understand what's possible, the @IRFRESTRICT procedure can be used to help solve for an impulse vector which meets multi-step zero restrictions.

[luching]

Thanks. I also thought about tracing out the impulses directly from the model equations. That is, shock at time t=0 to get responses of all variables at t=0. Overwrite the response of the variable of choice by setting it as zero. Then, use that to trace out responses of all variables at time t=1. Wouldn't this work?

[TomDoan]

The short answer is no. There is no justification for doing that.