Do methane emissions converge? Evidence from global panel data on production- and consumption-based emissions

Methane emissions are the second most important contributor to global warming. Knowledge about the dynamics of methane emissions facilitates the formulation of climate policies and the understanding of their consequences. We investigate whether methane emissions released from production and embodied in consumption converge within and across regions. Our estimates rely on global panel data on methane per capita and methane intensities over 1997–2014. We find that emissions converge within countries. The short half-lives show that the emissions of countries are close to their steady states. There is no evidence for international convergence of aggregate emissions. Yet, convergence of emissions across regions occurs in a number of economic sectors. Our results highlight the difficulties to achieve methane abatement in the medium run. The formulation of climate policies should take into account the sectoral specificity of the dynamics of methane emissions. Supplementary Information The online version supplementary material available at 10.1007/s00181-021-02162-9.


Variable
Description Source

Dependent variables
Growth of CH4 pc prod. First difference of the natural logarithm of production-based CO2 emissions per capita divided Fernández-Amador et al. (2020b) by the length of the period.
Growth of CH4 pc cons. First difference of the natural logarithm of consumption-based CO2 emissions per capita divided Fernández-Amador et al. (2020b) by the length of the period.
Growth of CH4 va prod. First difference of the natural logarithm of production-based CO2 emissions per value added Fernández-Amador et al. (2020b) divided by the length of the period.
Growth of CH4 va cons. First difference of the natural logarithm of consumption-based CO2 emissions per value added Fernández-Amador et al. (2020b) divided by the length of the period.

Control variables
Ln(CH4 pc prod.) Natural logarithm of production-based CH4 emissions per capita. Values for composite regions were obtained as GDP weighted averages. If data was missing for individual group members, group averages were used.  Note: * CI 90%, ** CI 95%, *** CI 99% where CI stands for the equal-tailed credible interval.

A.3 Detailed results for sectoral emissions
A.3.1 Individual-specific convergence Individual-specific convergence: CH 4 production per capita Note: * CI 90%, ** CI 95%, *** CI 99%. aiv is the strength of the correlation between the errors of the instrumental and the outcome equation. All variables but group dummies and income pc growth enter in lagged values. The half-life is calculated as ln(0.5)/ln(1 + β). The Bayesian R 2 is the mean of the R 2 computed for each draw q of the Markov chain (MC), R 2 Dq is the deviance measure associated with draw q in the MC (see Spiegelhalter, 2002, Gelman et al., 2004. Results are based on 3 MC with 50000 iterations each, after a burn-in of 25000.  Note: * CI 90%, ** CI 95%, *** CI 99%. Definitions and further information are provided in the note to Table A.5. Individual-specific convergence: CH 4 production per value added Note: * CI 90%, ** CI 95%, *** CI 99%. Definitions and further information are provided in the note to Table A.5.  Note: * CI 90%, ** CI 95%, *** CI 99%. Definitions and further information are provided in the note to Table A.5.  Note: * CI 90%, ** CI 95%, *** CI 99%. Definitions and further information are provided in the note to Table A.5.  Note: * CI 90%, ** CI 95%, *** CI 99%. Definitions and further information are provided in the note to Table A.5.  Note: * CI 90%, ** CI 95%, *** CI 99%. Definitions and further information are provided in the note to Table A.5.  Note: * CI 90%, ** CI 95%, *** CI 99%. Definitions and further information are provided in the note to Table A.5.

A.4 Robustness of the results for different values of ν
We performed several robustness checks to analyze the sensitivity of the results from economy-wide regressions to the degree of heteroscedasticity. These checks consisted in (i) specifying alternative values for the hyperprior of the hyperparameter ν (see Fernández-Amador et al., 2019, for details on the parameters), and (ii) imposing different values for ν exogenously. In the first experiment, alternative values of the hyperprior for ν did neither affect the estimated value of ν nor the results concerning the existence of convergence reported in the main text. Therefore, we conclude that the estimates of the degree of heteroscedasticity are rather robust.
In the second experiment, we exogenously fixed the parameter ν, such that ν = {10, 20, 40}. The results are summarized in Table A.13, where we report the estimated values of the convergence parameter, β, and the implied half-lives for convergence towards international steady states conditional on the control variables for economy-wide emissions. The results suggest that the estimates reported in the main text are robust to a wide range of degrees of heteroscedasticity. Specifically, the results remain robust, for most emission inventories, for values of ν ranging up to ν = 40. For production-based emissions per value added, the estimates of the heteroscedastic model are not qualitatively affected for ν = 10, but there is some (marginal) evidence for convergence for higher values of ν. In general, the speed of convergence is estimated to be faster as the degree of heteroscedasticity is restricted to be lower (as the value of ν increases). Note: * CI 90%, ** CI 95%, *** CI 99%.