ARDL panel model in STATA

[Abbes HIRI]

Hi mates, I have a question about ARDL model in Stata
As I know The XTPMG command can calculate the ARDL (1 1). here is my command :
xtpmg d.lreerq d.lgdppcq d.nfagdpq d.ltotq d.lopenq d.lgovgdpq d.flo, lr(lreerq lgdppcq nfagdpq ltotq lopenq lgovgdp flo) ec(ECT) replace omg
After running this command I am not getting significant coefficients , so i want to run the ARDL(121111)
I generated a new variable which is the lagged (1) of the lgdppcq and named it lgdppcq_lag using the following command :
gen lgdppcq_lag = lgdppcq[_n-1]
Then I substituted the lgdppcq by lgdppcq_lag in the xtpmg command , as following :
xtpmg d.lreerq d.lgdppcq_lag d.nfagdpq d.ltotq d.lopenq d.lgovgdpq d.flo, lr(lreerq lgdppcq_lag nfagdpq ltotq lopenq lgovgdp flo) ec(ECT) replace omg
MY QUESTION is :
Am I running an ARDL(121111) model by this command ?
waiting your answer, and thank you very much in advance ...


Regards

[Jesse Wursten]

I don't know what an ARDL(xxxx) model implies exactly, but I'd like to point out that gen x = x [_n-1] is not a good way to generate lagged variables. Rather, I suggest you use L.x in the regression command and let Stata do its thing, or generate it yourself as L_x = L.x. The problem with the _n-1 method is that for the second panel, this will use the last value of the first panel as its first observation. Same for the 3rd and so on. Additionally, if your data are not equally spaced (i.e. you have gaps), this will also produce wrong results.

A second note, the xtpmg command does not calculate a "standard" panel ARDL model. Rather, its default (pmg) will calculate separate ardl models for each panel unit (e.g. N regressions), but constrain the long run coefficients of the variables in lr() to be the same across all panels. If you want to calculate the more standard pooled ARDL model, simply use reg.

[River Huang]

Please try:

xtpmg d.lreerq d(1/2).lgdppcq d.nfagdpq d.ltotq d.lopenq d.lgovgdpq d.flo, lr(l.lreerq lgdppcq nfagdpq ltotq lopenq lgovgdp flo) ec(ECT)

[Abbes HIRI]

Dear Mr.River Huang and Mr.Jesse Wursten and all, here i am showing you the results I got from couple of regressions I have done, please check the word file and tell me how to interpret the results and with model or which regression i should run, thanks very much
waiting your Feedback ASAP

[Carlo Lazzaro]

Abbes:
your request is not consistent with the way this forum works.
As far as I can see, these are your first posts on this forum and you might have not read the FAQ yet.
However, I take the liberty to express some considerations, that you can take in due count or not, as you like, being an unrequested advice.
As everybody is free to reply or not to to any query, it is not in the spirit of this forum to urge people to check what you did and report to you ASAP.
Otherwise, as per FAQ, you should conctact the contributors you prefer off the list (i.e., via a private message) and make an arrangement with them for a consultancy.

[Abbes HIRI]

Please try:

xtpmg d.lreerq d(1/2).lgdppcq d.nfagdpq d.ltotq d.lopenq d.lgovgdpq d.flo, lr(l.lreerq lgdppcq nfagdpq ltotq lopenq lgovgdp flo) ec(ECT)

Thank you for your Help, Appreciated

[Abbes HIRI]

I don't know what an ARDL(xxxx) model implies exactly, but I'd like to point out that gen x = x [_n-1] is not a good way to generate lagged variables. Rather, I suggest you use L.x in the regression command and let Stata do its thing, or generate it yourself as L_x = L.x. The problem with the _n-1 method is that for the second panel, this will use the last value of the first panel as its first observation. Same for the 3rd and so on. Additionally, if your data are not equally spaced (i.e. you have gaps), this will also produce wrong results.

A second note, the xtpmg command does not calculate a "standard" panel ARDL model. Rather, its default (pmg) will calculate separate ardl models for each panel unit (e.g. N regressions), but constrain the long run coefficients of the variables in lr() to be the same across all panels. If you want to calculate the more standard pooled ARDL model, simply use reg.

Thanks Jesse for your reply and I shared with you the results I hv got , thank you so much

[Abbes HIRI]

Abbes:
your request is not consistent with the way this forum works.
As far as I can see, these are your first posts on this forum and you might have not read the FAQ yet.
However, I take the liberty to express some considerations, that you can take in due count or not, as you like, being an unrequested advice.
As everybody is free to reply or not to to any query, it is not in the spirit of this forum to urge people to check what you did and report to you ASAP.
Otherwise, as per FAQ, you should conctact the contributors you prefer off the list (i.e., via a private message) and make an arrangement with them for a consultancy.

Thank you , have nice day

[Muhammad Azam]

To the best of my knowledge ARDL can not be run on panel data in STATA. XTPMG or PMG wouldnt help there. Am I right ?

[Rafael Noriller]

Please

For an ARDL Panel with lags determined using Criterion (SIC) how to estimate?

Considering: xtpmg aa dollar, lr (aa dollar) and ec (ect) replace pmg.

thank you!

[Ngozi ADELEYE]

Please

For an ARDL Panel with lags determined using Criterion (SIC) how to estimate?

Considering: xtpmg aa dollar, lr (aa dollar) and ec (ect) replace pmg.

thank you!

Hi Rafael, how many countries are in your data? and in case you have not gotten any response, you can modify this code (given by a Statalister on this forum):

forval i = 1/10{
ardl aa dollar if (country_code==`i'), maxlag(3 3)
matrix list e(lags)
di
}

Eventually you will receive for each country a summary line like this (this is just an illustration, I cannot predict what will be the lag numbers in your case)

e(lags)[1,3]
aa dollar
r1 3 1

Then, you choose for each of both variables, the maximal lag length over all countries (suppose 10 as the number of countries). Lets assume that you will receive the following lag lengths:
aa = 2; dollar = 3

Then you specify the xtpmg command as:
xtpmg d(1).aa d(1/3).dollar, lr(l.credit rate gdp) ec(ECT) replace pmg

Hope it helps.

[GwanSeon Kim]

Hi Rafael, how many countries are in your data? and in case you have not gotten any response, you can modify this code (given by a Statalister on this forum):

forval i = 1/10{
ardl aa dollar if (country_code==`i'), maxlag(3 3)
matrix list e(lags)
di
}

Eventually you will receive for each country a summary line like this (this is just an illustration, I cannot predict what will be the lag numbers in your case)

e(lags)[1,3]
aa dollar
r1 3 1

Then, you choose for each of both variables, the maximal lag length over all countries (suppose 10 as the number of countries). Lets assume that you will receive the following lag lengths:
aa = 2; dollar = 3

Then you specify the xtpmg command as:
xtpmg d(1).aa d(1/3).dollar, lr(l.credit rate gdp) ec(ECT) replace pmg

Hope it helps.

Hi Ngozi Adeleye and Rafael,

I am using panel ARDL, and I am stuck with number of lag selection.

I am using the code what Ngozi mentioned, but I still do not follow your explanation.

Once I loop 10 countries, I have following results:

Country1: e(lags)[1,3]
aa dollar
r1 3 1

Country2: e(lags)[1,3]
aa dollar
r1 2 0

Country3: e(lags)[1,3]
aa dollar
r1 2 2

Country4: e(lags)[1,3]
aa dollar
r1 1 1

Country5: e(lags)[1,3]
aa dollar
r1 1 2

Country6: e(lags)[1,3]
aa dollar
r1 3 3

Country7: e(lags)[1,3]
aa dollar
r1 0 1

Country8: e(lags)[1,3]
aa dollar
r1 1 3

Country9: e(lags)[1,3]
aa dollar
r1 0 2

Country10: e(lags)[1,3]
aa dollar
r1 2 1

If I get the result above, how can I select my lag for AA and Dollar??
Then how to set up the code to run the xtpmg command??

Please help me.

Best regards,