I think you should do the unit root test first. Click "view“ key on the top-left of the screen and then choose ”ADF test". As to the cointegration test, select "view Cointegration  Test" as well as assign parameters please.

Please refer to

http://www.adsr.com.ng/index.php/doc-man/doc_download/15-stationarity-cointegration-on-e-views.

STATIONARITY, COINTEGRATION AND ECM USING E-VIEWS

You can consult Eviews 8.1. You introduce the panel data and opening the box you can find cointegration panel tests. It is very easy. You can use RATS that is cheaper tan Eviews and very easy to use. Ican have helped you.

@Rajesh Sharma, kindly let me know why should we perform all of these three tests i.e. Pedroni, Kao, Fisher for Panel Cointegration Estimation?

Pedroni and Kao cointegration tests are residuals based taken from Engle Granger two step test. Both are one way cointegration. But Johansen Fisher cointegration test is a system based cointegration test for the whole panel set.

Hossain Academy invites you to see a video on panel unit root, Fisher Johansen panel cointegration test and finally panel vecm model below. 

https://youtu.be/nxlBU48FXis

conduct unit root test first, you may use LLC to ensure that all the variables are I(1). secondly conduct system approach to cointegration using Fisher Johansen approach and then estimate the restricted VECM to have the short run dynamic relation and the long run relation as well as the equilibrium parameter (ie the ECM). Finally conduct the Wald coefficient stability test. All these can be done using E-view. in case you have difficulty feel free to consult me on [email protected] or call +23460869048

I have used the Fisher Johansen approach, bu the resulting VECM has strong non-normality in the residuals no matter how many lags use. Also, impulse dummies don't help. What to do?

I have used the Johansen Cointegration test and have gotten 3 cointegrating equations. How do I proceed with this?

The suggestions from earlier respondents pretty much answer the question.

However, I'd like to add that you check the nature of the data and your expected inferences to make a choice. Else, you might be running from pillar to post after wasting a lot of time. Again, you can use a mixed-model/method for a more robust inferential analysis.

Gracia!

In non-panel Johansen cointegration VEC output of long run coefficients, the sign has to be switched. This doesn´t seem to be the case with Fisher Johansen coint test/VECM, right? When I run each country individually, I get negative signs, but when in panel mode, I get a positive sign. This is contradictory.