The Engle Granger approach suggests that we check the regression residuals stationarity with ADF test and if the residuals are stationary, even if not all other model variables are, we can say there is Cointegration.

I was thinking, in the event of a Panel Data model, instead of testing each panel's residuals for stationarity, could we apply a Panel Data Unit Root test, such as Im-Pesaran-Shin test (IPS) ?

I understand that the Null Hypothesis of IPS for example is different than the one of ADF test, but do you believe it could help give an overall impression if there exists cointegration ?

Since we have Panel Data, if the ADF for every single panel fails to reject the Null for just a small amount of total panels of the model, I think we would penalize the model to say there is no cointegration. In that case, a Panel Data specific Unit Root test, might be more "forgiving" if I may say.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.