Even if T = 17, panel data stationarity must be studied. Even for a few time periods, panel data may exhibit non-stationarity problems, including unit root. After testing for I(1) and I(2), you can employ either multiple imputation or a fixed-effects estimator to handle missing panel data. Stationarity must be checked for your Kaufman indicator-derived GOUV variable. The Kaufman indicators may have been aggregated to a range of -2.5 to +2.5, but your composite variable may still be non-stationary depending on how it was generated. For confirmation of regression results, you must check GOUV variable stationarity.

Thanks for your answer.Between the first generation test and the second generation test for stationarity. Wich generation will you suggest me for economic variables.

GOUV also is not an economic variable, but a composite indicator obtained by pca on the 6 Kaufman indicators of governance.

Should I test stationarity on a composite indicator ?

Syrile Sambou

For economic variables, I would recommend using second-generation unit root tests like the augmented Dickey-Fuller (ADF) test or the Phillips-Perron (PP) test, which can account for serial correlation and heteroskedasticity. On the other hand, even though GOUV is a composite indicator, stationarity MUST be tested because the composite indicator's stationarity is not guaranteed by the stationarity of its constituent indicators. You should note that non-stationary indicators may generate stationary composite indicators, or vice versa. Therefore, ensuring analytical validity, you MUST test for composite indicator stationarity.

Thanks

https://www.jstor.org/stable/23114967