This paper proposes a panel vector error correction model investigation of a quadratic relationship linking CO2 emissions, GDP levels and electric power consumption. We find that two independent long-run relationships emerge from the data. Since the null of homogeneity across units with regard to long-run elasticities is strongly rejected, we proceed by clustering countries according to the signs of the estimated coefficients. The approach allows us to form three groups: in the first there is evidence of an optimistic scenario, where both CO2 emissions and electric power consumption are bound to decrease in the long-run. An optimistic scenario for emissions reduction is also provided in the second cluster where, however, the long-run relationship between income and electric power consumption shows an U-shaped pattern, instead. Finally, the third cluster can be associated with a much worrying scenario where per capita CO2 is expected to grow with income. A joint consideration of long-run parameters and causality links allows us to propose cluster-tailored policy suggestions.