Phillips and Sul technique

[Ilaria Villani]

Dear all,

I am using the Phillips and Sul (2007) technique for convergence test and club identification for my PhD research. My problem is that sometimes, for one of the clubs identified (with the psecta and default options), the TStat in the output table is below the threshold value of -1.65, whereas if I use the adjusted method (Schnurbus, 2017) with the same dataset, therefore including the 'adjust' command specification, the results are ok. On the contrary, when using a different dataset, the same problem (with the TStat value below -1.65 for one of the clubs) happens when using the adjusted method, while the other method is ok.

How can I handle the issue? Why does it happen? Should I use some specific options or just discard the anomalous clustering obtained and choose the alternative method? How can I motivate this in my PhD research?

Thank you very much in advance for your reply, I really hope you can help me!

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions. You don't provide a full cite for Phillips and Sul. You don't even tell us specifically what estimation routine you used. Did you code this yourself or what?

That you get slightly different results with different methods is not surprising, but also not pleasant. Many object to significance tests because they impose an arbitrary hurdle and treat parameters must above the hurdle one way and just below a different way. I would suggest reporting both sets of results.

[Ilaria Villani]

Dear Phil,

Thank you very much and sorry for my late reply. I am sorry, indeed I did not write my question properly and with the correct reference.

I am using the method explained here: https://sites.google.com/site/kerryd...me/stata-files with Stata 15. My panel dataset includes 28 individuals and 16 years.

As explained in "Econometric convergence test and club clustering using Stata" by Kerui Du (which you may find at the link I provided above) and also in "Transition Modeling and Econometric Convergence Tests" by Phillips and Sul (2007), Volume 75, Issue 6, November 2007, in case the TStat shown in the output happens to be below the threshold value of -1.65, no club should be formed. Instead, in some cases, I get clubs even if the Stat value is below -1.65 .

The commands I am using are these (can be found in the manuscript by Kerui Du):

xtset id year
gen lnpgdp=ln(pgdp)
pfilter lnpgdp, method(hp) trend(lnpgdp2) smooth(400)
logtreg lnpgdp2, kq(0.333)
psecta lnpgdp2, name(country) kq(0.333) gen(club)
imergeclub lnpgdp2, name(country) kq(0.333) club(club) gen(finalclub)

which correspond to the first of the two methods proposed for the PS technique. The issue with TStat happens, in some cases, with the psecta command, and in others with the imergeclub command.

I hope I have been clearer now, thank you very much again!