The Carbon Kuznets Curve: A Cloudy Picture Emitted by Bad Econometrics?

In this paper we discuss three important econometric problems with the estimation of Environmental Kuznets Curves, which we exemplify with the particular example of the Carbon Kuznets Curve (CKC). The Carbon Kuznets hypothesis postulates an inverse U-shaped relationship between per capita GDP and per capita CO2 emissions. All three problems occur in the presence of unit root nonstationary regressors in panels. Two of them are rather fundamental: First, the use of nonlinear transformations of integrated regressors in the Kuznets curve, which usually contains GDP and its square is problematic. This stems from the fact that nonlinear transformations of integrated processes are in general not integrated, which implies that (panel) unit root and cointegration techniques, widely used by now in the Kuznets curve literature, cannot be applied meaningfully in this context. Second, all methods applied up to now rest upon the assumption of cross-sectional independence. With a first application of factor model based methods that allow for cross-sectional dependence, we find evidence for nonstationary common factors in both the GDP and CO2 emissions series. Estimating the CKC on stationary de-factored data, we do not find support for an inverse U-shape. The third problem, abstaining at this point from the above two fundamental problems, is that the unit root and cointegration methods have been used too uncritically. In particular the notorious small sample problems of unit root and cointegration problems have been neglected. By applying various bootstrap algorithms and several estimators we show that a careful analysis should have lead researchers to interpret their results with more caution than commonly done, even when being unaware of the two problems stated aboveCarbon Kuznets Curve; panel data; unit roots; cointegration; cross-sectional dependence; nonlinear transformations of regressors