Growth and environmental degradation in MENA countries: methodological issues and empirical evidence

ABSTRACT

A large number of studies on the validity of the environmental Kuznets curve for MENA countries have been conducted, producing (as expected) a mixed bag of results. Several econometric issues are considered with reference to estimates of the EKC for Algeria, Egypt, Jordan and Tunisia. These issues include the order of the polynomial, the validity of the log–log specification, cointegration and spurious correlation, missing variables, and the sensitivity and fragility of the results. It is concluded that the most serious issue is the sensitivity of the results to model specification and other factors, which is not considered in the MENA studies of the EKC. Robust results are produced with respect to the order of polynomial, estimation method and measurement of the income variable.

Acknowledgements

I am grateful to the editor of this journal and two anonymous referees for useful comments. An earlier version of this paper was presented at 24th conference of the Economic Research Forum, which was held in Cairo in July 2018. I am grateful to the participants for their remarks and to Kabbashi Suliman, who acted as a discussant, for detailed written comments on the paper.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1 The sample ends in 2014 because of the unavailability of more recent observations on CO2 emissions, at least not for all of the four countries. Since the empirical work on the paper was conducted initially, one more observation has become available. However, it is unlikely that the inclusion of this observation in the sample will change the results qualitatively as turning points can be observed within the sample period between 2008 and 2011.

2 Otherwise the relation may be monotonic or U-shaped, depending on the signs of the coefficients, implying either that pollution rises with the level of income or that it falls initially and rises subsequently.

3 The turning points reported in Table 1 do not coincide with those shown in Figure 1 because of differences in estimation methods. The turning points shown in Figure 1 are based on an OLS fit of a quadratic function. The turning points reported in Table 1 are based on the coefficients estimated by FMOLS.

4 The MEKC is obtained by augmenting the EKC with a variable or variables that represent a wider concept of development rather than pure economic growth, including well-being and sustainability of the development process. For example, the EKC can be modified by adding the United Nations’ human development index as an explanatory variable.

5 The Z statistic is based on transformation of the multiple correlation coefficient (R, which is the square root of the coefficient of determination) as half the log of the ratio of one plus R over one minus R. The Z statistic is calculated as the ratio of the difference of the two transformed variables divided by the standard deviation of the difference.

6 It is not obvious why a log-log specification is needed to deal with the possibility of negative values for measures of degradation, since negative values will never appear in actual data for the very reason that production always produces waste. It could be that Stern (2003) is concerned about the possibility of negative forecast values, but again the log-log specification does nothing to change the situation. A forecast in log with a positive value may be translated into a negative value of the underlying measure of degradation.

7 The choice between multiplicative and additive fixed effects should be an empirical issue.

8 One problem with the log-log specification is that it implies constant elasticities, which is highly restrictive and unrealistic. At the very least, the log-log specification should be a testable hypothesis rather than a functional form that is imposed in an arbitrary manner. The validity of the log-log specification should be demonstrated rather than assumed.

9 It is a rather strange claim that economic meaningfulness can be determined by a statistical test, be it cointegration or otherwise. Economic meaningfulness can be determined only by economic theory, or at least sound intuition. Unfortunately, it seems that we have chosen to ignore common sense and rely on numbers coming out of a computer to determine economic meaningfulness. Kennedy (2002) expresses a similar view eloquently by suggesting that rule # 1 is ‘use common sense and economic theory’ and raising alarm by warning: ‘never state the test proves/demonstrates the data are I(1), I(0) or I(anything)’.

10 Moosa (2017a) warns of the hazard of using cointegration testing to detect spurious correlation, providing evidence indicating that this procedure may lead us to believe that NASA is responsible for suicide and that the consumption of margarine leads to divorce.

11 In a multiple regression equation containing ‘control variables’, the coefficient on the variable of interest tells us how the dependent variable responds to changes in the variable of interest when the ‘control variables’ are held constant. In a complex world, we cannot control for everything—after all, this is economics, not physics and the EKC is not exactly Boyle’s law. If we cannot control for everything, we might as well use a two-variable regression (as in the EKC), which tells us how the dependent variable responds to changes in the variable of interest when everything else changes simultaneously. Novel works in economics that are based on this approach have produced, inter alia, the Phillips curve, Okun’s law and Wagner’s law. Unfortunately, this has become a thing of the past with the advent of ‘stir-fry regressions’ containing two dozen ‘control variables’.