Next steps after cross‑sectional dependence and mixed stationarity in panel data

[Rana Roy]

Hello everyone,

I am working with three panel datasets:

• 20 countries, 2000–2022
• 13 countries, 2000–2022
• 7 countries, 2000–2022

I tested for cross‑sectional dependence using xtcd and found evidence of CSD. I then applied second‑generation unit root tests and obtained a mix of I(0) and I(1) variables across the datasets.

My questions are:

1. What should be the next step in terms of model choice and estimation strategy given mixed integration orders and small N?
2. Should I proceed with CS‑ARDL/CS‑DL/CS‑ECM estimators, or consider pooled models, MG/PMG, or alternatives such as GMM, FE/RE?
3. How should I decide how many lags to include in ARDL/CS‑ARDL specifications? Is there a recommended approach (e.g. AIC, BIC, HQIC) for small panels?
4. Which Stata libraries/packages are most appropriate for this type of analysis (e.g. xtdcce2, xtpmg, or others)?

I would greatly appreciate guidance on how to handle mixed I(0)/I(1) variables in the presence of cross‑sectional dependence, and what estimation strategy would be most defensible for publication or thesis work.

Thank you very much for your time and advice.