Two time series if stationary means have constant mean and variance. If non-stationary at level, we need to check cointegration. Cointegration examine as if two series have constant co-variance over time. If so, long-run relation can be modeled through OLS (DOLS, FMOLS etc). In case of panel, concept is roughly same as in time series; however, some averaging or combined statistics are considered. Several tests have been proposed for panel cointegration like Pedroni (1999; 2004), Kao (1999) and a Fisher-type test using an underlying Johansen methodology (Maddala and Wu, 1999). The Fisher test is a simply the combined Johansen test (as for the time series). The Pedroni and Kao tests are based on Engle-Granger (1987) two-step (residual-based) cointegration tests. Pedroni proposes several tests for cointegration that allow for heterogeneous intercepts and trend coefficients across cross-sections.
So i would recommend Pedroni tests as its is comprehensive. But if you have a panel that has cross-sectional dependence then Westerlund (2007) panel cointegration tests will be preferred. Hope this is the answer.
Finally, the choice of model or estimates depends on the nature (behavior) of data. So comprehensive means a lot in econometrics.
Cheers, Jawad
I use panel co for parameters which is not stationary at level and stationary at first diff.
For Panel co I use pedrone of which three tests included. And in each test control the p value out of 11 outcome of 7 unit r test results. If the majority satisfied then tbe long run associationship is satisfied and there is no need to make other. But I do kao too. Fisher not necessarily.
Pedroni and Kao panel cointegration tests that extend the Engle-Granger two-step (residual-based) cointegration framework. Fisher’s cointegration test combines individual cross-sections. This method uses two ratio tests such as a trace test and maximum eigenvalue (max-eigen) test. This panel cointegration test may be robust than the conventional cointegration tests based on the Engle-Granger two step approach.
Yea. Pedroni is widely accepted for panel data regression analysis. To serve as a robustness check to that of Pedroni, I suggest another test especially Kao should be conducted.
Pedroni proposes several tests for cointegration that allow for heterogeneous intercepts and trend coefficients across cross-sections.
So i would recommend Pedroni tests as its is comprehensive. But if you have a panel that has cross-sectional dependence then panel cointegration tests will be preferred.
Finally, the choice of model or estimates depends on the nature of data. So it depends in econometrics.
Conventionally, Pedroni cointegration test is the most widely used in panel data regression analysis, because it takes care of cross-sectional dependence, especially where the countries have the same outlook ( either economical, socially, political etc) by allowing considerable heterogeneity. So in my opinion, Pedroni is recommended
I agree with the comments provided here! But one thing to put in mind is that, under presense of endogeneity, the panel cointegration faces some estimation challenges. Panel cointegration was mainly addressed in static models (fixed and random) models. But under dynamic models that handle endogeneity, panel cointegration is worse!
- Badges
- Science topic