EXTRAPOLATED TRENDS VERSUS ENERGY MODEL PROJECTIONS – GLOBAL DISTRIBUTION DYNAMICS DERIVED FROM REGIONAL KAYA DECOMPOSITION

摘要

OverviewThe development of economic growth, energy demand, and environmental quality are key inputs to design policyoptions for a global sustainable development. Future scenarios produced by energy system models and integratedassessment models can provide useful insights into future pathways but they cannot be validated. We apply themethod of distribution dynamics to historical data as an alternative projection method and as a way to analyze andtest scenario output produced for the Global Energy Assessment Report, GEA (2012), the World Energy Outlook2015, as well as for the Global Energy and Climate Outlook, Labat et al. (2015). More specifically, we look into theevolution of cross-country/regional patterns appearing in the factors of the Kaya decomposition.The paper is structured in the following way. First, we briefly review the method of distribution dynamics and weshow how the distribution functions we have selected are related to Kaya’s identity. The main section thancompares the trend analysis of historical data with the projected patterns for the global distributions of the Kayafactors. To this end, an ergodic distribution is determined to estimate the behavior in the far future. A summaryand outlook concludes the paper.MethodsWe generalize the Kaya decomposition of the global CO_2 emissions to account for the regional distribution of thedrivers by representing the emissions as CO_2(t) = ƩiCO_(2,i)(t) = P0P(t) ×Ʃie_i(t) × f_i(t)×g_i(t) × p_i(t)where for each region I we have carbon dioxide emissions per final energy e_i , final energy per GDP f_i , GDPper population g_i , and p_i being the share of regional population in world population POP . Followingideas from density estimation theory, this expression is reinterpreted as the population-weighted expectation value of a trivariate joint probability function f CO_2(t) =P0P(t)∫dxdydzxyzf(x,y,z;t)which is given by a kernel density estimate asf(x,y,z)=1/h_xh_yh_zƩi_(pi)K(x-e_i/h_x)K(y-f_i/h_y)K(z-g_i/h_z)Here, K is the so called kernel function, which is assumed to be normalized with vanishing first moment. For asmooth estimation of the unknown distribution function f of the sampled data, a proper choice of the bandwidthparameter h_x ,h_y , h_z is essential, see Silverman (1986). To this end, we adapt data-driven bandwidth selectionrules given in the literature to our situation, cf. Goerlich Gisbert (2003) and Wang, Wang (2007). In principle, theprobability function f can by projected into the future by a Markov process, where the transition function ismodelled from historical data. In this way, trends in the evolution of the distribution function are propogated toobtain a prediction for the individual distribution of GDP per capita etc. and the joint function f. In contrast totraditional methods, where only mean values and/or moments of the distribution are extrapolated, such anapproach estimates the distribution function as a whole. In particular, the propagation into the far future can serveas a means to study if regional disparities converge or not. These distributions can be compared with kernel densityestimates for regional data generated by baseline scenarios in energy system models. In a first step, we implementthis approach for the univariate regional distribution of carbon intensity, final energy intensity and per capita GDP.Having these quantities at our disposal, we propagate the trivariate joint distribution function, which is a difficulttask in itself. We use a copula representation of distribution functions to connect the trivariate distribution to its univariate marginal distributions and to bivariate distribution functions, see Trivedi (2007). As a result, we obtain aprediction for carbon emissions extrapolating trends from a time period given by historical data.ResultsThe added value of our paper is three-fold: first, we analyze projected energy transformation pathways focusing onthe development of regional disparities in the world. This adds to the discussion of WEO, GEA and GECO results,because these studies look more into global results or that of specific regions. Second, we suggest a method that canbe used to analyze the development of projected regional disparities in contrast to what has been observed inhistory. This can support testing models which are per-se impossible to validate ex-ante (c. f. Schwanitz (2013)).Third, we contribute to the growing literature on distribution dynamics by widening its application beyond thestudy of income or final intensity distributions. We also propose a new bandwidth selection rule for weightedsamples.ConclusionsWe find that the projected future economic development of baseline scenarios is biased towards convergencewhereas energy demand patterns across world regions unfold smoothly from historical data and are close to thealternative projection that we derive from the ergodic distribution. The development of carbon intensity of finalenergy is strongly prescribed by regional frameworks. This global distribution seems unlikely to converge.