The RATS Software Forum

This is your own idea? You haven't described the behavior of y under the null. Nor is it at all clear what exactly you're testing.

But I described the behavior of z under the null in the code as below:
*
linreg(define=dzeq,noprint) dz
# constant dz{1 to 9}
set u = %resids
compute bstart=%regstart(),bend=%regend()
*
Then using bootstrap, the z sample is obtained under the null and then nonlinear deterministic component (which is called as Model D in Cook and Vougas(2006 and 2009)) is subtracted from the z sample. And finally, the KSS test is applied. This process was repeated 10000 times. Then the prob. value of the actual KSS t statistic is computed by using the distribution of 10000 t statistics. I’m not sure my program does the right coding. I can send You the program again.

Pointing to a line in the program doesn't describe your testing procedure. It looks like you're slapping together two completely unrelated tests. You need to figure out what your null is and what your alternative is.

Dear Tom Doan,
I’m trying to replicate Leybourne, Newbold and Vougas’s (1996) smooth transition unit root test (The LNV test) in RATS. According to the test result, there are important differences between the critical values of my test and the critical values of the original paper. I’d like to know whether I have done the right LNV coding. And if I am wrong, how can i fix the code. I will appreciate Your response.

H0: yt=yt-1+et
H1: yt=a1+a2*St(ϒ,τ)+νt, where νt is I(0)

Please find attaced the files.

This should t-c, not t-c*trend

frml glstar = %logistic(gamma*(t-c*trend),1.0)

Given that there *is* no break under the null, you may also need to do a more careful job of estimating the STR model. Cut down the number of draws to say 10 and take the NOPRINT's off and see if you get reasonable results.