[Sebastian Kripfganz]

The ARDL(1,1,0,1,1,0) model is a special case of the ARDL(2,1,0,1,1,0) model. Evidence for a structural break in the latter is inconsistent with evidence against a structural break in the former.

Even with the ARDL(1,1,0,1,1,0) model, there is still no evidence for a long-run relationship. Thus, the best model still appears to be an ARDL model in first differences. Unfortunately, the ardl command does not support time series factors in its variable lists. You would need to generate first differences of your variables and then run the ardl command (without the ec option) on those first differences. Or, without searching for optimal lags again, simply use the regress command:

regress D.y DL.y D.x1 D.x3 D.x4

Notice that an ARDL(2,1,0,1,1,0) model in the variables y x1 x2 x3 x4 x5 becomes an ARDL(1,0,0,0) model in D.y D.x1 D.x3 D.x4 once the long-run terms are removed.

[misso ouhibi]

I am so grateful to you really I appreciate your help so much.
Otherwise, I would like to ask you if there is possible to estimate the ARDL-ECM model when there is some data are missing (monthly data).

[Sebastian Kripfganz]

In principle, you can still estimate an ARDL-ECM model if there are gaps in your data, yes.

[misso ouhibi]

same case for testing the bound test? there is no problem?
Many thanks for your help

[misso ouhibi]

when i want to estimate the ardl model with maxlags (12)
there is a problem, i don't know how can i deal with this problem. Any help please?

ardl lnY lnX1 lnX2 lnX3 lnX4 lnX5,maxlags(12) aic

# of lag permutations (4455516) exceeds setting of 'maxcombs' (100000)

[Sebastian Kripfganz]

Try the following:

ardl lnY lnX1 lnX2 lnX3 lnX4 lnX5, maxlags(12) aic maxcombs(5000000) dots

[misso ouhibi]

ok firstly before asking you for this answer, I use this code :
and lnY lnX1 lnX2 lnX3 lnX4 lnX5, maxcombs (122222) dots then I found these results :

but when i use this code
ardl lnY lnX1 lnX2 lnX3 lnX4 lnX5, maxlags(12) aic maxcombs(5000000) dots i found this results

[misso ouhibi]

and when i use this code
ardl lnY lnX1 lnX2 lnX3 lnX4 lnX5, maxlags(12) aic maxcombs(5000000) dots i found this results

[Sebastian Kripfganz]

With your first code, you are using the default maximum lag order of 4. With your second code, you have set the maximum lag order equal to 12. Consequently, results differ.

[misso ouhibi]

Yes i understand thanks a lot sebastian for your help and for all really it's so interesting.

[misso ouhibi]

Hello sébastian,

Please I want to ask you about the ARDL model when I choose the maximum lags of 12 aic.

then ARDL model chooses automatically the number of lags of each variable: ARDL(12,12,0,5,0,9) regression: 12 for the dependent variable and the other for the independent variables.

I want to ask you about the number of maximum lags; I have 6 variables; monthly data and number of observations 263 with gaps.

[misso ouhibi]

Any help please?

[Sebastian Kripfganz]

It is not clear to me what exactly your question is.

[misso ouhibi]

I want to ask you for the appropriate maxlags to estimate the ARDL model in the case of 6 variables Monthly data and Number of observations 263 with gaps.
Because when I choose the maximum lags 12 aic, the ARDL model automatically selected a number of lags for each variable ARDL(12,12,0,5,0,9) regression: 12 for the dependent variable. Then I am stopping here because when I see the results, there are many p values of many lags that aren't significant.

[Sebastian Kripfganz]

I am unable to tell you what the best choice for the maximum lag order would be in your case. 12 does not seem unreasonable to me. For some regressors, the AIC might choose an optimal lag order which is well below the maximum lag order. This happened indeed for some of your variables. The AIC is not removing specific lags when they are statistically insignificant. It is thus normal that some statistically insignificant coefficients are remaining in the model.