some variables are related to each other (energy consumption and co2 emission) in theory but data is non-stationary and we can't reject null hypothesis that says no co-integration.
The solution may be to use a dynamic panel data model and to estimate by GMM panel. Then you must to check it with the systems of Arellano-Bover (1995) and Blundell-Bond (1998)
4) Smith RP (2001) - Estimation and inference of nonstationary panel time series data, Department of Economics, Birkbeck College, London (prepared for the RC33 Conference, Cologne, october 2000)
Generally to estimate dynamic panel data that contain lagged dependent variables, is to consider the lagged dependent variables as jointly determined variables in short panels. You should use instrumental variable-type estimator. Possibilities are 2SLS and Arellano-Bond..
The solution to the problem is to transform the time series data so that it becomes stationary. If the non-stationary process is a random walk with or without a drift, it is transformed to stationary process by differencing.