[Trevor Lee]

Thanks very much JanDitzen I've read your stata journal entries and descriptions, but I can't find further information on how the IV option works within xtdcce2? Does it just follow the normal 2-stage procedure - for a single endogenous regressor, obtain the predicted value of the said variable using the instrument/other exogenous covariates, and use this predicted value in place of the endogenous regressor in the ARDL?

[JanDitzen]

Thanks very much JanDitzen I've read your stata journal entries and descriptions, but I can't find further information on how the IV option works within xtdcce2? Does it just follow the normal 2-stage procedure - for a single endogenous regressor, obtain the predicted value of the said variable using the instrument/other exogenous covariates, and use this predicted value in place of the endogenous regressor in the ARDL?

Yes, it is a simple two-stage approach. Please bear in mind that the combination cross-section dependence and IV regressions was never published beyond working paper (Neal (2015), link), so there is no well founded theoretical foundation!

[Trevor Lee]

Yes, it is a simple two-stage approach. Please bear in mind that the combination cross-section dependence and IV regressions was never published beyond a working paper (Neal (2015), link), so there is no well founded theoretical foundation!

Great, that paper is very helpful. But yes, I'll keep in mind the tenuous theoretical foundations. On a separate note, I was wondering how where panel cointegration testing generally fits within the CCE-ARDL framework? Some papers seem to test for it before estimating ARDLs, whereas the illustration you give with estimating a Solow model (https://www.stata.com/symposiums/eco...n21_Ditzen.pdf) doesn't use it.

[JanDitzen]

You should do cointegration tests before you estimate the long run model. My talk at the Econ Symposium last year was tailored to panels with large N and large T and topics beyond the "classical" time series or panel data econometrics. Therefore I did not discuss cointegration in more detail.

[Trevor Lee]

You should do cointegration tests before you estimate the long run model. My talk at the Econ Symposium last year was tailored to panels with large N and large T and topics beyond the "classical" time series or panel data econometrics. Therefore I did not discuss cointegration in more detail.

Sorry to re-up this, but is co-integration a necessary condition for using the CCE-ARDL methods discussed in your talk? My basic understanding is that co-integration tests for long-run relationships between I(1) variables, hence I would think it necessary to find some evidence for co-integration before using an ARDL to estimate long-run effects. But this paper (https://link.springer.com/content/pd...pdf?pdf=button), which seems very respectable, uses the absence of co-integrating relationships to justify the ARDL methods employed (CCEMG/CCEPMG). Elsewhere I've seen positive results from co-integration tests used to justify further investigation with ARDLs.
Thanks in advance for addressing the confusion

[S. M. Woahid Murad]

Hi JanDitzen What is the minimum required time for CCEMG for each panel? Can I apply xtdcce2 if the time period (t) is less than 10 for a few countries and 80 for others? I am working on quarterly data, and I am covering 187 countries, i.e., p=187. Thanks in advance.

[JanDitzen]

There is no clear guidance on what is the smallest number of time periods. In general I would make sure that you have a sufficiently large degree of freedom if you would estimate the model with just the specific time series. Balanced vs. unbalanced data is not such a problem, xtdcce2 can deal with it.

[S. M. Woahid Murad]

Hi JanDitzen , Thank you so much for your quick response.