Optimal lag length for ARDL in panel

[Anne Tholen]

Hello everyone,

after having searched several helpfiles and threads here, I unfortunately couldn't find a solution to my problem. Therefore I'm posting my question here.

I have a panel dataset with cross-section identifier cid and yearly data (xtset cid year, yearly).

In order to conduct an ARDL analysis and then extend that to an ECM with xtpmg, I want to find out the ideal lag length, preferably using the Akaike information criterion, Schwarz Bayesian criterion or Hannan and Quinn. Unfortunately it seems like varsoc, which would usually report all of those, is not available for panel data.

Is there a panel alternative to varsoc or alternatively, an extension to xtpmg that reports the optimal lag length?

Best regards,
Anne

[Anne Tholen]

I would highly appreciate your help, since this is for a replication paper I have to hand in by Sunday and I didn't find a solution to check the author's choice of lag order yet :/

[Muhammad Azam]

I would also like to know if there is some automatic lag selection option for ARDL (Panel data) in STATA

[Anat Tchetchik]

Hi Muhammad and Ann,
Since I could bot find a command that obtains an optimal ARDL structure for panel data (If you find out something meanwhile, I would be happy to learn about that too)
What I did was to obtain an optimal lag structure for each country separately, and then to use the maximum number of lags obtained for each var. in the -xtpmg- command.
For this purpose I have used the -ardl- command by Kripfganz, S. and D. Schneider and run (using loop) the command for each country. I think it is preferable to specify to the -ardl-command up to 3 lags for each (non stationary) variable.
This is the command I used:

forval i = 1/50 { ardl Y x1, x2, x3, x4 if (country_code==`i'), maxlag(3 3 3 3 3) matrix list e(lags) di } I really hope it helps..(: Anat

[Ngozi ADELEYE]

Hi Anat,
I came across this now having searched every material on how to choose optimal lag length for my panel ARDL and I am using the PMG model too. Thank you very much, it works!!!!