ARDL, Long-run response, and hypothesis tests

[jhusein]

Hi Tom,

I'm estimating quiet few ARDL models and the example/program distriblag.rpf was very helpful. Assuming I'm doing it right and

After extracting the long-run response for example of Y to X1 via

compute arpoly=%eqnlagpoly(0,Y)
compute dlpoly=%eqnlagpoly(0,X1)
compute ardlpoly=%polydiv(dlpoly,arpoly,#)

I'm trying to test first if the long-run response is ZERO and second if the Long-run response is unity.
I appreciate if you (or anyone's) help with the proper test command(s) in RATS to do this? Thanks.
Jamal

[TomDoan]

You can use SUMMARIZE to do that. (You have to write the function in terms of the %BETA coefficients). See, for instance, the Harvey textbook example harveyp393.rpf.

<edit>

Upon further review, you can actually do this directly if all the calculations are done in one go:

summarize(numerical) %polyvalue(%eqnlagpoly(0,shortrate),1.0)/%polyvalue(%eqnlagpoly(0,longrate),1.0)

(The %eqnlagpoly functions are the ones that implicitly use the %BETA,so they can't be computed separately from the SUMMARIZE instruction).

[jhusein]

You can use SUMMARIZE to do that. (You have to write the function in terms of the %BETA coefficients). See, for instance, the Harvey textbook example harveyp393.rpf.

<edit>

Upon further review, you can actually do this directly if all the calculations are done in one go:

summarize(numerical) %polyvalue(%eqnlagpoly(0,shortrate),1.0)/%polyvalue(%eqnlagpoly(0,longrate),1.0)

(The %eqnlagpoly functions are the ones that implicitly use the %BETA,so they can't be computed separately from the SUMMARIZE instruction).

Many thanks and Yes, all calculations are done in one go.
It works for testing long run response is zero.
Can "summarize" be modified to test if long run response is unity though?
Thanks.

[TomDoan]

You can do a test for a long-run value of 1 with

summarize(numerical) %polyvalue(%eqnlagpoly(0,shortrate),1.0)/%polyvalue(%eqnlagpoly(0,longrate),1.0)-1.0