2nd Generation Panel Cross Section ARDL Model in STATA for Cross Section Dependent & Dynamic Models

[@nikolaizaicev9297]

We need to use CD test for each variable independently? I see many of videos, where they use xtpmg+fe or re to generate residuals first, and then apply the CD test to those residuals. That has totaly confused me, since, I do not use linear fixed effects model but Panel ARDL+ECT and DFE with (2 2 2 2 2 ) optimal lags. Would I need to estimate the complete Panel ARDL model first with all the laged variables, run it, predict the residuals and then apply the CD test on those residuals? Or is it sufficient just to test each separate variable for CD like you did in your video, in my case? Thank you.

[@nomanarshed]

Both are correct methods the residual based and variable based. Actually of individual variables are independent then it can be assumed that residuals would be independent too.

[@aryaasvadishirejini8184]

Thank you for creating such a valuable content. I have a question, if the slopes(xthst command) were heterogeneous, can we use this model?

[@nomanarshed]

In that case you should run the Mean Group Estimate which provides country specific long and short run estimates. And if there is cross sectional dependence you can see its advanced version of CCMG or AMG

[@aryaasvadishirejini8184]

@nomanarshed Thank you so much. Please apologize for my rudeness, but I have another question: Does the AMG or model or the CC MG model cover heteroskedasity?

[@wendyaledon9110]

good day sir, in running for cointegration tests, are we going to use the level form or the stationarized data? also, in performing cointegration, does it need to be I(1) or can i use mix I(0) and I(1)? thank you in advance

[@nomanarshed]

This model can work on mixed order of integration, means you do not have to stationarize the data. And further cointegration tests have nothing to do with the order of the data, it checks if the set of variables are cointegated in long run.

[@mynameisjoejeans]

Great video, thank you. I have a strongly balanced panel with N=4, T=30. Do you know why I'm getting a conformability error when attempting to use xtdcce2?

[@nomanarshed]

In the about section you can find email. Please email screenshot of error i will see

[@mynameisjoejeans]

@@nomanarshed I've managed to fix it, thank you for all your help

[@kimsesizlerinsesi3567]

may you explain briefly how to determine lags variables in CS ARDL ? are we use akaike kriterio if yes. how to learn aic of model CS ARDL, which code?

[@nomanarshed]

There is no such code for panel data. You can add as many lags which can fit and stabilize the model.

[@necatib.2170]

why did you take cr_lags(0) ? pesaran recommned [T^(1/3)]. can we use crlags(0) ?

[@nomanarshed]

Lag order are not determined by the theory, they are selected on the basis of its model stabilization ability. Pesaran recommended lags may be used if the data is long enough to absorb these many lags.

[@eyuptanil4285]

@@nomanarshed thnk you, ı want to cite your this comment for my article, can you say me which pesaran article ?

[@dr.anusuyabiswas-ecoeconom7618]

@nomanarshed I received error while performing xtcd lCO2 (log carbon dioxide). I tested unit root xtunitroot ips, then employed xtreg on panel data with N-41 and T-26. afterwards trying to test cross sectional dependence but i found Error r(133). Plz help

[@dr.anusuyabiswas-ecoeconom7618]

xtcsd, pesaran unknown egen function group() r(133);

[@nomanarshed]

This test i think do not work on unbalanced panale data. Further there are further sub types of this test like pesaran frees etc. Check that.

[@dr.anusuyabiswas-ecoeconom7618]

@@nomanarshed I have strongly balanced data

[@Ganieirfan]

thanks for this video. I am getting this error when runnugn cross section depenendence test "__000001 contains all missing values r(498);" . I have an unbalanced panel with no missing values. Please help.

[@nomanarshed]

Some CSD tests do not work on unbalanced data.

[@lagabyousraassia5736]

Sir, I have a similar situation with conflicting results between the Westerlund and Pedroni tests. The Westerlund test indicate that there is no cointegration, while the Pedroni test indicates the presence of cointegration. (I have n=4, t=37, and the presence of CSD.) what i have to do? Thank you

[@nomanarshed]

First confirm if there is cross sectional dependence. If it is present the westerlund test is superior otherwise if it is not present you can proceed with the outcome of pedroni test.

[@nomanarshed]

For the case of presence of CSD have a look at demeaned perdroni and demeaned kao test if they can provide any evidence of cointegration. Or try changing lags in westerlund test

[@lagabyousraassia5736]

​@@nomanarshedthank you very much

[@mynameisjoejeans]

Is it possible to include a structural break in your ARDL model by treating it as an exogenous variable? The more specialised ARDL commands like N-ARDL and CS-ARDL don't seem to allow exog() from what I can find.

[@nomanarshed]

In the specialized model. If you add them in only short run equation they will become exogenous. If not then there is not much problem in adding in long run too.

[@mynameisjoejeans]

@@nomanarshed That worked, thank you so much. This is my last question I promise: how do you generate and analyse residuals from xtpmg2 and xtdcce2 commands? In previous ARDL models it was really simple, but I can't find anything on how to perform diagnostics using these newer commands, making my findings useless.

[@saqibsatti3978]

Very nice

[@nomanarshed]

Thanks

[@sopanigondwe7334]

what is the correct specification? specify the model in differenced form or only the dependent variable enters in differenced form? Can I run this model with all variables in levels as long as there is cointegration and n variable is I(2)?

[@nomanarshed]

The published papers are better guide to find specification. As per my knowledge cs-ardl is advanced form of panel ardl with cross sectional dependence. I am not sure if it is suitable for I(2)

[@sopanigondwe7334]

@@nomanarshed Sorry, I meant I(1) not 2. In your video here, you use the first difference of the dependent variable. Is there any econometric explanation for that? In other words, does it make a difference if one enters the dependent variable in level? Since an ardl is an unrestricted ecm, why do the other explanatory variables not enter in first difference? Any thoughts on these? I would appreciate.

[@zeeshankhawaja9769]

hello nauman kindly share more videos on panel data that how to select lag length while conducting panel unit root test and ardl panel data techniques including methods for cross sectional dependence.

[@nomanarshed]

I will share soon. Actually in these tests you provide the upper limit of lags then the method autometically selects optimal lag from them

[@necatib.2170]

@@nomanarshed its question , can you explain briefly how to determine lags variables in CS ARDL ?

[@LITERARYCLUB]

Pleass guidde when i run my model my all results getting insgnificnt only one was significant .but my data was stationray ​nd CD test was heterogeneous.i didn't understand what model we will apply .@@nomanarshed

[@rabiakhalid5611]

Thanks for this information, I am running cs ardl command in STATA but it is giving error (''option lr() is not allowed'')

[@nomanarshed]

This problem is sorted whe you use latest version of stata

[@kimsesizlerinsesi3567]

ı didnt understand why you add lag of same varieables inside lr(----) equiestions?

[@nomanarshed]

We have followed the same pattern which is suggested by the documentation of this model

[@eyuptanil4285]

ıf my all varieables I(1), can ı apply CS-ARDL ?

[@nomanarshed]

Yes. It is ARDL mode with capability to estimare when variables are dependent cross sectionally

[@necatib.2170]

can we add dummy variables in CS_ARDL?

[@nomanarshed]

Yes

[@lombechangala2803]

hello great video.i estimated xtdcce2 d.lpec, lr(L.d.lpec lutp lgdp lpg) lr_options(ardl) cr(lpec lutp lgdp lpg) cr_lags(0)fullsample and found that my error correction term in my regression is -1.02 which is greater than one. Also my variable of interest lutp has a coefficient greater than 100%.i also found that there was no cointegration in my model and cross sectional dependence is present. Please help on how to go about it. lpec represents log of primary energy consumption, lutp repesents log of urbanization, lgdp represents lgdp, lpg represents log of population growth

[@nomanarshed]

You should not take log to population growth. It might improve the outcome.

[@lombechangala2803]

@@nomanarshed hello,thank you very much for your reply.i have just tried than and there is an improvement in the prob>F to 0.04..but no change in coefficients

[@mariashabir5787]

I have a confusion in command , sir why did you put 2 independent variables and then you wrote (L(0/1).lter then 4th independent variable(L(0/2).idcps. I didn't understand those things can you please explain it

[@nomanarshed]

L(0/1).lter means i am using level of lter and first lag of lter and L(0/2).dcps means level, first and second lag is used for dcps while for others i did not add means no lags used for them. Further details can be explored from the help manual of this command

[@mariashabir5787]

@@nomanarshed Thank you, for explanation . It's mean we can choose lags and levels according to our models. Can you please make a one detail video on D-K model

[@nomanarshed]

Sure i will.make soon.

[@necatib.2170]

Ican select "cr_lag(0)" and if the probability value of the CD statistic is greater than 0.10 and F is less than 0.10, then I can use this model for my article

[@nomanarshed]

Try F test to have value less than 0.10 rest is permisible

[@sarahahmedchawsheen5455]

thanks for this informative video, can i use cs ardl method for N=7 and T=21 ?

[@nomanarshed]

Usually in this case we used first generation panel data models like Panel ARDL as the CD tests do not estimate Cross sectional dependence. If you are able to prove that there is cross sectional dependence you can use CS ARDL

[@sarahahmedchawsheen5455]

@@nomanarshed thanks for your replay, actually my data has csd for each variable except one independent variable, should I include this variable after cr in the command xtdcce2 ...... cr( variable), or not?

[@nomanarshed]

@@sarahahmedchawsheen5455 The cross sectional independent variables should not be added in the cr

[@firas1008]

Can u do it on eviews

[@nomanarshed]

This model is not available on eviews yet but you can manually try by adding cross sectional averages of variables as IV

[@younisahmed8751]

Does structural break test require any package I'm unable to run that command it's showing error

[@nomanarshed]

You have to install that command too

[@meenakshigautam9457]

for cross sectional dependence when i am using the command it is giving unrecognised

[@nomanarshed]

You might have to install it. Or there is syntax error in the command

[@akashayosha]

at 0:46 it shows that log of ter has missing values, what should be done for this?

[@nomanarshed]

kzbin.info/www/bejne/jqnQi32uo5eeiNk this method can be used to fill the gaps in data

[@AlishaMahajandse]

which is the constant term/ intercept for csardl?

[@nomanarshed]

It is varying across cross section. On average it is zero. You can view them using its appropriate codes

[@younisahmed8751]

Thank you so much Noman bhai

[@nomanarshed]

Welcome

[@kimsesizlerinsesi3567]

can ı select crlags 0 for my models??

[@nomanarshed]

Yes. Selection of lags are used to stabalize the model

[@mandeepthakur3305]

sir, if there is no long term relationship established by ARDL, do we need to use polynomial distributed lag model. if you have any idea. please direct

[@nomanarshed]

You can use simple lagged model which is also known as ARDL and the one in which the relation is not established is ARDL with cointegrating bounds. You can further learn about it from Walter Enders book on Applied Timeseries book

[@yousufkhan4367]

Assalamalaikum Noman bhai, I hope u r well when i m using this command in stata as u have mentioned in your video xtdcce2 d.im, lr[l.d.im y ghg] lr_option[ardl] cr[im y ghg] cr_[0] full sample it is saying option lr is incorrectly specified. can you help me to figure out this issue

[@nomanarshed]

Woa. You have to study the documentation of this command if this error is not dicussed then it might be reported to the author so that next version can rectify this error

[@kardbouazza1052]

Can I use test Bai and ng -panic

[@nomanarshed]

I think they are cointegration tests. You can use them

[@MyQamarali]

AssalamoAlikum Sir, I tried hard to run CS-ARDL or CS-DL but unfortunately I could not run it. I found this warning different time. "factor variables may not contain noninteger values" What does it mean?

[@nomanarshed]

See the data view any one of the selected variable may had non numeric values in the data like NA or null etc...

[@dr-muhammadayaz7265]

Could you please share the base paper for further demonstration?

[@nomanarshed]

Important articles are shared in the documentation of this test. Which can be seen via stata help command

[@dr-muhammadayaz7265]

@@nomanarshed thank you for your reply. Actually I am talking about the base paper (Ibrahem Saeed) on which you build the model.

[@nomanarshed]

It has not been completed yet

[@eyuptanil4285]

your f prob value is 0.13, can ı say this model biased ?

[@nomanarshed]

It says model is not fit.

[@eyuptanil4285]

@@nomanarshed should ıt be less than 0.05? another questions ı saw same videos they dont take first difference variables. why ?

[@nomanarshed]

@@eyuptanil4285 you should compare the model with the documentation if they are similar to the authors documented model then their model should be prefered

[@uroojmaqbool8934]

JazakAllah...

[@nomanarshed]

Welcome

[@uroojmaqbool8934]

Sir kindly constant term or error correction term ko identify kar dain iss model Mai ?

[@nomanarshed]

See the paper which developed this model

[@LITERARYCLUB]

Please give me all commands in comments thanks

[@yasmeensarwarabbasi]

Aoa sir May i have your email address please.i neend guidelines for the use of cs-ardl technique in my research project. As i am facing alot of issue n am also a new stata user

[@nomanarshed]

Woa you can contact at [e-mail]