WPS7742 Policy Research Working Paper 7742 Technology Strategies for Low-Carbon Economic Growth A General Equilibrium Assessment Ian Sue Wing Govinda Timilsina Development Research Group Environment and Energy Team July 2016 Policy Research Working Paper 7742 Abstract This paper investigates the potential for developing countries feasible in both countries, enabling reductions of up to 4 per- to mitigate greenhouse gas emissions without slowing their cent of baseline emissions while generating slight increases expected economic growth. A theoretical frame- work is in GDP (1 percent in Armenia and 0.2 percent in Geor- developed that unifies bottom-up marginal abatement cost gia). The results demonstrate how MAC curves can paint a curves and partial equilibrium techno-economic simulation misleading picture of the true potential for both abatement modeling with computational general equilibrium (CGE) and economic growth when technological improvements modeling. The framework is then applied to engineering operate within a system of general equilibrium interac- assessments of energy efficiency technology deployments tions, but also highlight how using their underlying data in Armenia and Georgia. The results facilitate incorpora- to identify technology options with high opportunity tion of bottom-up technology detail on energy-efficiency cost elasticities of productivity improvement can lead to improvements into a CGE simulation of the economy- more accurate assessments of the macroeconomic conse- wide economic costs and mitigation benefits of technology quences of technology strategies for low-carbon growth. deployment policies. Low-carbon growth trajectories are This paper is a product of the Environment and Energy Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at gtimilsina@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Technology Strategies for Low-Carbon Economic Growth: A General Equilibrium Assessment∗ Ian Sue Wing1 and Govinda Timilsina2 1 Dept. of Earth & Environment, Boston University 2 Development Research Group, World Bank JEL classification: C68, Q48, Q54, Q55, Q65 Keywords: Static Computable General Equilibrium, Green Growth and Sustainability, Low Carbon Development, Climate Change Mitigation, Energy Efficiency ∗ Comments by Jan Gaska and Mun Ho are gratefully acknowledged. The authors also acknowledge the financial support from Knowledge for Change (KCP) Trust Fund of the World Bank. 1 Introduction Mitigating global emissions of greenhouse gases (GHGs) to levels that avert the threat of dangerous climate change requires the active participation of developing nations. Projected 21st century economic growth and population expansion in non-OECD countries are asso- ciated with substantial increases in GHG emissions and radiative forcing of climate which cannot be completely offset by GHG abatement in the developed world—even if the latter countries stop emitting entirely (Jacoby et al, 1997). This state of affairs has led to increasing calls for industrializing countries to take on binding emission reduction commitments. Thus far, developing nations have generally been reluctant to accept abatement targets, citing re- source constraints that limit their ability to lift their burgeoning populations out of poverty, let alone incur the opportunity costs of curtailing emissions. These constraints, developing countries argue, mean that that they will only be able to pursue substantive emission re- ductions if the necessary technology and financial resources are provided by industrialized countries—an issue which is at the crux of ongoing international climate negotiations. A temporary compromise solution is for developing countries to identify and pursue activities that slow the growth rate of their emissions, but do not hinder their expected economic growth. Assessing the viability of such “low-carbon economic growth” strategies is the objective of this paper. As noted by Jacoby et al (1998), Steward and Weiner (2001) and Aldy et al (2003) among others, the goal is to allow for a transition path in which developing countries have the “headroom” to continue economic expansion over the next two or three decades, but with declining rates of increase in emissions. For low-carbon growth to be feasible there must be activities which avoid GHG emissions but also provide “win-win” economic benefits, and large-scale deployment of energy efficient devices and processes are of particular interest in this regard (Hallegate et al, 2011; Rozenberg et al, 2013). The latter investments are typically more costly compared with conventional technology, but they enable firms and households to use less fossil energy, simultaneously lowering emissions and generating offsetting savings on energy and other operational costs. Crucially, since a key benefit of these cost savings is improved productivity at the micro-level, sufficiently widespread deployment of efficiency-enhancing technology within a country could in principle generate an aggregate gross productivity gain. And if, in turn, the resulting stimulus to the growth of output at the macro level is large enough that it exceeds the economy-wide opportunity costs of additional expenditure on energy-efficiency investments, the result will be a net increase in GDP. How might such an auspicious outcome arise? Simple intuition suggests that the largest potential for win-wins should come from intensive investment in technology options which have a high elasticity of energy efficiency improvement with respect to the incremental cost of their acquisition. Engineering assessments (e.g., McKinsey & Co., 2010) are the key data source on technologies’ performance and direct cost characteristics which allow us to identify such options. But these assessments frequently uncover what appear to be large “energy efficiency gaps” in which technical potentials for efficiency remain unrealized due to factors such as imperfect information or institutional constraints, which cause economic actors to systematically overlook making energy efficiency investments that are privately 1 profitable.1 By contrast, microeconomic analyses contend that the actual magnitude of the energy efficiency gap is small because engineering studies suffer from problems of poten- tially substantial unobserved or mis-measured ancillary costs and/or benefits, uncertainty in projections of the net present value of future energy cost savings, and heterogeneity among agents in the returns to energy efficiency investments (Allcott and Greenstone, 2012). Policy makers thus face three challenges: (i) identifying putatively negative net cost—so- called “no-regret”—technology options whose savings exceed the corresponding investment cost premia at prevailing capital and energy prices, (ii) estimating these options’ market penetration potential and attendant fossil fuel/GHG emission savings based on economic as well as technical fundamentals, and (iii) assessing the implications for emissions, output and welfare at the level of the macroeconomy. Addressing these is a tall order, principally be- cause of the difficulty reconciling often incommensurate results of engineering and economic calculations across three disparate sets of analyses. This is the essence of the “bottom-up” versus “top-down” divide in the climate policy literature. The majority of analyses culminate at stage (i), developing marginal abatement cost (MAC) curves that rank, using engineering cost assessments and discounted cash flow anal- ysis, various technology options or activities based on their net costs and avoided GHG emissions. MAC curves are attractive because they are easy to construct and bring to light activities that are profitable from the perspective of private actors investing in an individual technology. However, such analyses often fail to adequately account for the true opportunity cost of investments in technologies. They also ignore the potential substitutability or com- plementarity among different technologies which determines the latter’s market penetration. Of concern are unforeseen interdependencies among mitigation options’ cost or technical potential that render the shape of the MAC curve endogenous, where the width or height of one segment on the curve can influence the width or height of other segments, and their joint marginal-cost ranking. Stage (ii) analyses are able to account for potential interdependencies through the use of partial equilibrium techno-economic process or activity analysis models. The latter incorpo- rate data on the cost and energy use characteristics of a broad array of discrete technology options, and solve for the minimum cost vector activity-specific capacities and associated emission reductions in one or more economic sectors. But these models omit interactions be- tween the specific technnologies and the broader economy—especially the effects of changes in demand on fuel prices and capital rental rates—and must often employ ad-hoc constraints to smooth activities’ time-paths of penetration or exogenously their technical potentials. It is here that the institutional or market barriers discussed above come in. The latter raise the issue of why a country might need to pursue low-carbon technology deployment policies in the first place—in other words, why the investment decisions of decentralized profit-maximizing firms would diverge from the command optimum solved for by the model, and how the resulting gap might be closed by policy interventions such as subsidies or mandates. Systematic accounting for the web of spillover effects is the strength of computable gen- eral equilibrium (CGE) simulations, which are the principal tool for conducing stage (iii) analyses of the macroeconomic costs of GHG mitigation policies. However, these top-down economic models struggle to incorporate the detailed characteristics of discrete technology 1 Allcott and Greenstone (2012) label these “investment inefficiencies”. 2 options within the framework of smooth cost and expenditure functions customarily used to represent the envelope of firms’ and households’ substitution possibilities. This chal- lenge has spawned the development of hybrid bottom-up/top-down modeling approaches which represent discrete mitigation options as production functions that generate GHG-free energy (abated emissions) from inputs of GHG-intensive fossil fuels (unabated emissions) and technology-specific abatement capital. Recent methodological advances by Kiulia and Rutherford (2013a,b) have developed procedures to derive the underlying abatement poten- tials and capital input requirements from MAC curves, but substantial calibration effort is required, and the resulting computational scheme does little to elucidate which options have the greatest potential for capital-fuel substitutability and market penetration, or the macroeconomic impacts of deploying different technology packages. In this study the specific question we address is deceptively simple: can developing countries pursue GHG mitigation without slowing down their expected economic growth, and if so does significant abatement result? To provide answers we first develop a theoretical framework that unifies stage (i), (ii) and (iii) analyses by consistently tying together the underpinnings of MAC curves, bottom-up techno-economic simulation modeling, and CGE modeling. We then apply this framework to bottom-up engineering assessments of technology deployment for improving energy efficiency in Armenia and Georgia, drawing on detailed data developed by Sikharulidze et al (2015) and Timilsina et al (2015). We build on our theoretical framework to develop a methodology for incorporating bottom-up technology detail on energy-efficiency improvements into top-down CGE model, which allows us to estimate the economy-wide economic costs and mitigation benefits of technology deployment policies in our two case-study countries. Surprisingly, we find that low-carbon growth trajectories are feasible in both Armenia and Georgia, and that it is possible for them to offset similar fractions of their baseline CO2 emissions (up to 4%) while enjoying slight increases in GDP, despite significant differences in the scope of technological opportunities for energy efficiency improvement. The major di- vergence in outcomes is the extent of the productivity and GDP increase over baseline levels, which can be as large as 1% in Armenia, but only 0.2% in Georgia. Our results demonstrate how the exogenous trajectories of efficiency penetration that underlie the construction of bottom-up MAC curves can paint a misleading picture of the true potential for both abate- ment and economic growth when these technological changes operate within a system of general equilibrium interactions. But they also highlight how the use of the data that un- derpins MAC curves—not the curves themselves—to identify energy efficiency improvement options with high cost elasticities of energy productivity can facilitate assessments of the macroeconomic consequences of technology strategies for low-carbon growth. The remainder of the paper is divided into four sections. Our theoretical framework is developed in section 2. Section 3 introduces our bottom-up Armenia and Georgia case stud- ies. The results of numerical experiments with the CGE model are presented and discussed in section 4. Section 5 discusses caveats to the analysis and future research directions, and offers concluding remarks. 3 2 Bottom-Up vs. Top-Down: A Unifying Framework 2.1 Marginal Abatement Cost Curves: Engineering Fundamentals Consider a vector of technology options, indexed by θ. For example, θ can indicate building shell technologies, lighting or space conditioning devices. For any such option let there be two technology variants, a conventional and a high-efficiency variety, the second of which we identify using an asterisk (*). The installed bases of each variety are x and x∗ units, and differences in their service lives ( , ∗ ) translate into distinct capital recovery factors (denoted ρ). Moreover, the cost of capital acquisition is typically lower for the conventional variety (h < h∗ ), which determines the value of the technology’s installed capital base as vθ = ρ[ θ ]hθ xθ + ρ[ ∗ ∗ ∗ θ ]hθ xθ (1) The key distinction between conventional and high-efficiency varieties is the intensity of their fuel use (φ > φ∗ ), with the technology’s total use of fuel type e given by fe,θ = φe,θ xθ + φ∗ ∗ e,θ xθ (2) Introducing t ∈ [0, T ] as an index of time, e as fuel-specific emission factors, and P F e,t as exogenous trajectories of fuel prices, eqs. (1) and (2) determine the technology’s instan- taneous operating cost and emissions as weighted sums of the stocks of conventional and high-efficiency varieties: cθ,t = cθ [xθ,t , x∗ θ,t ] = vθ,t + P F e,t fe,θ,t e = ρ[ θ ]hθ + P F e,t φe,θ xθ,t + ρ[ ∗ ∗ θ ]hθ + P F e,t φ∗ e,θ x∗ θ,t (3) e e zθ,t = zθ [xθ,t , x∗ θ,t ] ∗ = e φe,θ xθ,t + e φe,θ x∗ θ,t (4) e e Given a discount factor β t , the economic and environmental performance of a particular scenario of technology penetration xθ ,x∗ θ can be summarized by the present values of cost and emissions, respectively: T Cθ = β t cθ [xθ,t , x∗ θ,t ] (5) t=0 T Zθ = β t zθ [xθ,t , x∗ θ,t ] (6) t=0 In this bottom-up setting the customary representation of policy actions to abate emis- sions is an exogenous shift in the mix of capital toward increased penetration of the high- efficiency variety. Let the nomenclature in (1)-(6) identify some business-as-usual state of 4 the world without mitigation. Now introduce a tilde over a variable to indicate the coun- terfactual imposition of emission mitigation, with new technology trajectories (xθ , x∗ θ ) such ∗ ∗ that xθ ≤ xθ and xθ > xθ . The impact of the resulting sequence of instantaneous cost and emissions can be summarized in the same way as (5) and (6): Cθ = T t ∗ t=0 β cθ [xθ,t , xθ,t ] ≷ Cθ and Zθ = T t ∗ t=0 β zθ [xθ,t , xθ,t ] < Zθ . Constructing a MAC curve from the baseline and coun- terfactual versions of (5) and (6) is a straightforward procedure. For each technology option, abatement (A) and its average cost (τ ) are simply Aθ = Zθ − Zθ > 0 (7) τθ = (Cθ − Cθ )/(Zθ − Zθ ) ≷ 0, (8) and all that is necessary is to rank-order elements of θ from low to high values of τ and plot their coordinates in τ -A space. 2.2 No Regret Potentials and Capital-Fuel Substitution Our exposition renders transparent how negative-cost MAC curve segments arise. These are called “no-regret” options because they are no opportunity costs to pursuing the req- uisite investments—at least in this partial equilibrium setting. The high-efficiency variety’s larger acquisition costs increase capital charges (vθ > vθ ) at the same time as its lower fuel intensities reduce fuel expenditures (fe,θ < fe,θ ). When the latter outweighs the former, instantaneous total cost dips below its baseline level (cθ,t < cθ,t ); if such a divergence is suf- ficiently large and persistent the numerator of (8) will be negative. Such behavior emerges out of the parameter combinations in curly braces in eqs. (3) and (4), and their interaction with scenarios of energy-efficient technology penetration xθ ,x∗ θ versus xθ ,xθ . ∗ It should also be clear that the force driving the technology dynamics that give the MAC curve its shape is instantaneous substitution of capital for fuel. This is apparent from (3), which can be rearranged to yield cθ F vθ F zθ πθ = (1 − θ ) + θ (9) cθ vθ zθ πθ ≷1 >1 <1 ≷1 Here, F is the baseline share of fuel in the technology’s operating cost, π denotes the implicit average price of emissions (i.e., the total value of fuel use divided by the quantity of emissions), and the terms on the right-hand side are the (increasing) capital and (declining) fuel components of cost.2 Total costs decline if the second component exceeds the first, which occurs where the high-efficiency alternative’s capital cost premium relative to the conventional is small but its fuel use and emissions are much lower, and/or fuel constitutes a large share of total operating cost. Here, however, our interest is different. We focus not on overall cost, but on the ability of capital to substitute for energy, indicated by the elasticity d log zθ zθ − zθ vθ − vθ ξθ = ≈ (10) d log xθ zθ vθ 2 Note that it is always the case that π z < πz , making the second term < 1. 5 2.3 Beyond Scenarios: Endogenous Energy Efficiency Improve- ment Two features of foregoing analysis highlight the need for transformation of the results of bottom-up assessments in order to evaluate mitigation options’ market penetration poten- tial. Engineering cost calculations implicitly treat the quantity and character of output of the various options as constant, which may not be the case if high-efficiency alternatives are imperfect substitutes for conventional varieties. More problematic is that costs and abatement hinge on exogenous assumptions about the time-path of investment in the high- efficiency alternative. Ceteris paribus, a technology with a more (less) aggressive trajectory will exhibit larger (smaller) discounted costs and total emissions, but due to the geometric character of discounting the result will vary in a nonlinear fashion, with implications for mitigation options’ average cost ranking and the overall shape of the MAC curve. Following from these observations, endogenizing technologies’ abatement potential and cost requires one key additional ingredient: specifying output in the form of flows of services from the conventional and energy efficient varieties, indicated by s (s∗ ). It is simplest to assume that production is Leontief with a fixed capital coefficient κ (κ∗ ), and that both supplies are perfect substitutes that fulfill an assumed exogenous time-path of aggregate S service demand Dθ,t , which presumably increases with projected growth of the economy. In the counterfactual scenario which is of interest here, the resulting supply-demand balance constraint is S sθ,t + s∗ ∗ ∗ θ,t = xθ,t /κ + xθ,t /κ ≥ D θ,t (11) We may then solve for the capacity trajectories by minimizing system cost while satisfying demand: xθ,t , x∗ θ,t = arg minxθ,t ,x∗ θ,t Cθ (1)-(5), (11) (12) θ This capacity expansion problem is a linear program which is at the heart of virtually all bottom-up techno-economic models. An unpleasant feature of (12) is that capacity may exhibit unrealistic “bang-bang” be- havior, switching over completely from the conventional to the energy-efficient variety in the space of one or two time-steps. This stems from the fact that there is no meaningful con- straint on technology-specific capital, which in this framework is perfectly elastically supplied at the exogenous cost of capital acquisition. A common means of ensuring smooth market penetration dynamics is to augment the problem with ad-hoc expansion and decline con- straints, (γ, γ ) and (γ ∗ , γ ∗ ), based on technology options’ service lives, e.g. (cf Vogt-Schilb et al, 2014), γ [ θ ] ≤ xθ,t+1 /xθ,t ≤ γ [ θ ] (13a) γ ∗[ ∗ ∗ ∗ ∗ ∗ θ ] ≤ xθ,t+1 /xθ,t ≤ γ [ θ ] (13b) Yet, the question this remedy raises is what are the economic processes out of which these constraints arise. Evidence abounds that the true opportunity cost of energy efficiency investments—even those clearing a net profit—can be substantial (Anderson and Newell, 2004), and can increase as more and more output is forgone with progressive diversion of 6 capital away from conventional production and toward lower-return energy saving activities. These additional costs are the true hurdle that the savings from energy efficiency improve- ments must clear, and the best way to account for them is a general equilibrium framework. 2.4 General Equilibrium Implications The main thrust of the paper is to model the outcome of eq. (12) without explicitly repre- senting the procedural details of capacity adjustment. Accordingly, we develop a deliberate simple approach that does not distinguish conventional and energy efficient varieties, but instead seeks to bring the effect of penetration of the latter on efficiency improvement within the ambit of top-down production and cost functions. Each technology is modeled as a monolithic, constant returns to scale technology that produces a flow of services (with quantity QS and price P S ) from inputs of a vector of fuels and technology-specific capital (with quantities QFe and QX , and prices P Fe and P X , respectively). The top-down analogue of eq. (3) is then P Sθ QSθ = P Fe QFe,θ /ηθ + P Xθ QXθ (14) e where η is the overall improvement in energy efficiency experienced by each technology option as a consequence of increased efficiency penetration. We treat production in (14) as Leontief, which implies that the relationship between output and input quantities admits a dual unit cost function that is linear in output and input prices: QF1,θ QFe,θ QXθ QSθ = min ηθ , . . . , η θ , X λF 1,θ λF e,θ λθ ⇔ P Sθ ≤ λF X e,θ P Fe /ηθ + λθ P Xθ ⊥ QSθ (15) e All else equal, efficiency improvement lowers the unit cost of service production. Following the complementarity format of general equilibrium, each technology’s unit cost function ex- hibits complementary slackness (⊥) with its output quantity.3 The latter satisfies a demand for services (DS ) determined offstage in the rest of the economy S QSθ ≥ Dθ [P Sθ ] ⊥ P Sθ (16) 3 Formally, for a variable u and a function F (u), F (u) ⊥ u ⇔ u ≥ 0, F (u) ≥ 0, u · F (u) = 0. The economic logic of this formulation is easiest to grasp in the case of market clearance (cf eq. (17)), where, given a commodity with price p and supply and demand functions S (p) and D(p), S (p) − D(p) < 0 p→∞ Excess demand bids up prices to infinity S (p) − D(p) > 0 p=0 Excess supply bids down prices to zero S (p) − D(p) = 0 p>0 Supply-demand balance at strictly positive, finite prices 7 where each service’s supply-demand balance exhibits complementary slackness with the cor- responding technology options’ level of activity. Symmetrically, the supply of fuels (S F ), which for our purposes also originates offstage, satisfies rest-of-economy demand (DF ), and, in addition, aggregate fuel demand across discrete technology options, which declines with efficiency improvements: F F Se [P Fe ] ≥ De [P F e ] + λF e,θ QSθ /ηθ ⊥ P Fe (17) θ This expression shows that expansion in discrete technology options’ capacity is a source of additional demand for fuel, which, ceteris paribus, increases fuel prices. This effect is moderated by energy efficiency improvements. Supply-demand balance for technology specific capital is given by QXθ ≥ λX θ QSθ ⊥ P Xθ (18) In this model the policy lever is the exogenously-mandated unit cost of technology-specific capital, which determines the rate of improvement in technologies’ energy efficiency. The latter benefit comes at a cost, which we represent via a markup on conventional “jelly capital” (µ ≥ 0). The resulting zero-profit condition is complementary to the endogenously- determined quantity of technology capital: P Xθ = (1 + µθ )P K ⊥ QXθ (19) We take pains to emphasize that µθ is not a tax. Rather, it is meant to capture the impact of mandates to purchase energy-efficient technology capital which is more expensive than conventional (i.e., jelly) capital. The general equilibrium opportunity costs of technology- specific capital derive from the fact that the latter represents an additional claim on the aggregate capital endowment, and must therefore compete with aggregate conventional de- mands (DK ). The upshot is a rise in the economy-wide cost of capital (P K ) and a drag on overall economic activity: QK ≥ DK [P K ] + QXθ ⊥ PK (20) θ We close our model with three equations at the core of every CGE simulation, namely, the definition of nominal income, the unit expenditure function (E ) and the market clearance condition for the aggregate welfare good, which are complementary to the levels of income and welfare (M and W ) and the unit expenditure index or economy-wide price level (P ), respectively: M = P K QK + . . . ⊥M (21) P ≤ E [P Sθ ; P Fe ; P K ; . . .] ⊥W (22) W ≥ M/P ⊥P (23) The final elements are the linkage between energy efficiency improvement and increases in technologies’ capacity, which we specify using the analogue of eq. (10): ηθ = (1 + µθ )−ξθ (24) 8 as well as an expression for economy-wide emissions, obtained by substituting (16) into the right-hand side of (17): F Z= e De [P Fe ] + λF S e,θ Dθ [P Sθ ]/ηθ (25) e θ We apply our stylized model (15)-(25) to elucidate the macroeconomic implications of mandating accelerated energy-efficiency investment. With a fixed jelly capital endowment, growth in µθ induces a decline in the price of technology-specific service output (P Sθ ) and an increase in the prices of jelly capital (P K ). Lower values of P Sθ depress the unit expenditure index, while higher values of P K increase it, resulting in a change in the price level which is ambiguous. On the other hand, rising P K unequivocally boosts nominal income. If the former change falls short of the expansion in income then the economy will experience a welfare gain. Eq. (23) indicates that this outcome requires dW /W = dM/M − dP /P > 0, where dM K ≈ ψP K -µθ µθ > 0 (26) M θ and, plugging (19) into (15) and then substituting the result into (22), dP 1 F ≈ ψE -P Sθ (ψP Fe -ηθ − 1) (−ξθ )µθ P θ e ηθ e,θ + ψE -P Fe ψP Fe -ηθ (−ξθ )µθ θ e X 1 1 + θ ψE -P Sθ 1+µθ + ψ µθ P K -µθ µθ θ + ψE -P K ψP K -µθ µθ (27) θ in which K is capital’s aggregate income share, ψP K -µθ > 0 is the elasticity of the price F X of jelly capital to technology markups, e,θ and θ indicate technologies’ fuel and capi- tal cost shares, ψP Fe -ηθ < 0 is the is the elasticity of fuel prices to energy efficiency, and ψE -P K , ψE -P Sθ , ψE -P Fe > 0 denote the elasticities of the aggregate price level with respect to the prices of jelly capital, technology capital, and fuels. The first two terms in (27) are negative while the last two are positive. The second and fourth terms account for the effects of efficiency increase and markups on the price level transmitted through the prices of fuel and capital. The first and third terms trace the same effects through the channel of the price of technology-specific services. In particular, the first term captures efficiency’s direct productivity-enhancing effect as well as its indirect effect through lower fuel prices, which are jointly moderated by technologies’ fuel-service price ratio, and decline with the level of efficiency. The sign of dP /P is ambiguous, and depends on the value of the parameters, but a welfare gain is more likely if it tilts in a negative direction. Conditions under which this tends to occur include when technologies are initially relatively fuel intensive (λF X e,θ > λθ ) and/or highly inefficient (ηθ 1), their efficiency responds relatively elastically to markups or the portfolio of markups intensively 9 targets technologies with such elastic responses, and fuel prices decline relatively elastically in response to efficiency improvement.4 It also occurs if the economy is relatively capital- intensive ( K > ψE -P K ) and capital’s price is simultaneously relatively unresponsive to technology markups. A key implication of eq. (27)’s first term is that the win-win of gross welfare improvement with emission reductions becomes progressively difficult to sustain. As the level of energy efficiency rises dP /P inexorably shifts in the positive direction, sharply diminishing the macro-level productivity benefit of additional efficiency investment. Of course, even with a welfare gain our primary interest is in the magnitude of the associated climate benefit. After plugging (19) into (15) and then substituting the result into (25), the concomitant change in the economy’s emissions is dZ Z -F ≈ e e F -P F ψP Fe -η (−ξθ )µθ ψDe e θ Z θ e Z -S 1 F + e e,θ 2 e,θ ψDθ S -P S (ψP Fe -ηθ − 1) (−ξθ )µθ θ e ηθ θ Z -S 1 X 1 1 + e e,θ θ ψDθ S -P S 1+µθ + ψ µθ P K -µθ µθ θ e ηθ θ Z -S 1 − e e,θ (28) θ e ηθ Z−S Z−F were e and e,θ are the discrete technology and rest-of-economy shares of CO2 in each fuel’s contribution to aggregate emissions, and ψDe F -P F , ψD S -P S < 0 are the own-price e θ θ elasticities of aggregate (rest-of-economy) demand for fuels and technology-specific services. Interestingly, a decline in aggregate emissions is not assured: as in the case of eq. (27), the sign of dZ /Z depends on the parameter values, with the first two terms positive and the remainder negative. The latter respectively capture the drag on technologies’ service output and associated fuel use from the increase in the cost of capital, and the direct effect of efficiency improvement’s reductions in the fuel necessary to generate that output, which is the driver of abatement. The first term captures the offsetting intersectoral “rebound” effect of efficiency improvements on fuel use in the rest of the economy through the channel of fuel price changes.5 This is the energy conservation analogue of emission leakage: the narrower the sectoral scope of technology mandates, the greater the potential for their direct fuel savings to “leak out” via increases in unregulated consumption in response to lower energy prices (see, e.g., the discussion in Broberg et al, 2015). The second term captures the intra-sectoral rebound effect of the offsetting influence of the productivity boost provided by higher energy efficiency. No-regrets options’ lower unit production costs and associated output prices stimulate increased demand for technology-specific services, resulting in an expansion of output that blunts energy savings and abatement. 4 The latter might occur if the portfolio of technologies under consideration accounts for a substantial share of the economy’s energy use. 5 Rebound refers to a situation where an improvement in energy efficiency causes the effective price of energy services to decline, which creates incentives economic actors to behave in ways that at least partially offset the expected beneficial impact of the efficiency gain (Allan et al, 2007, 2009). 10 Eq. (28) suggests that economy-wide abatement is likely to arise where the characteristics of the economy allow the fourth term to dominate. This occurs when the demands for fuels and technologies’ outputs are both highly inelastic, fuel prices are relatively unresponsive to efficiency improvements, technology options subject to efficiency mandates constitute a substantial share of aggregate emissions, and opportunity costs are large. As before, the factors 1/ηθ suggest that abatement becomes progressively less feasible as the level of energy efficiency increases. 3 Application: Energy-Efficiency Improvements in Geor- gia and Armenia 3.1 Background Improving efficiency is central to energy policy in both Armenia and Georgia, which have developed national programs to implement energy efficiency measures in major energy con- suming sectors such as buildings, industries and transportation.6 Both countries depend on imports to meet domestic demand for petroleum and natural gas, which together account for more than two-thirds of their total primary energy supply, and are major sources of GHG emissions and local air pollution. Large-scale deployment of energy efficient technologies remains limited in the private sector and relies on an array of government supports, raising the question of how to allocate scarce funding across technologies and industries. Thus far, the basis for making these decisions has been engineering bottom-up analysis, in particular evaluation of technology options by Sikharulidze et al (2015) and Timilsina et al (2015). In the remainder of the paper we apply the framework developed in section 2 to introduce dis- crete technology options into a dynamic CGE model for the purpose of assessing the broader impacts on Armenian and Georgian economies. 3.2 Bottom-Up Analysis Table 1 summarizes the slate of technology options considered by Sikharulidze et al (2015) and Timilsina et al (2015) and gives their distribution across sectors and countries.7 The industries identified therein (Food Products, Chemicals, Rubber & Plastics, Iron & Steel, Non-metallic Minerals, Other Industry, Services and Private Dwellings) are referred to as “technology-rich” to distinguish them from other sectors in which discrete technology options are not represented. MAC curves developed using eqs. (1)-(8) are shown in panel A of Figure 1. Their striking feature is the overwhelming preponderance of no-regrets options, primarily 6 Armenia’s 2005 Renewable Energy and Energy Efficiency law and subsequent national program on energy saving and renewable energy led to the creation of the energy efficiency action plan, which includes measures such as new construction codes for building energy, provision of efficient public lighting, monitoring of building energy consumption and savings, and industrial heating systems improvement. Georgia is developing a low carbon strategy for economic development while promoting economic growth and prosperity, of which a key component is energy efficiency improvement. 7 Data consist of ( θ , ∗ ∗ ∗ ∗ θ ), (hθ , hθ ), (φe,θ , φe,θ ) and e,θ and assumptions about P F e , xθ , xθ and the discount rate (7.5%), which is used to calculate ρ. 11 Table 1: Technologies Considered in This Study Residential Commercial Industrial Private Services Food Chemicals, Iron Non- Other Dwellings Products Rubber & Steel Metallic Industry & Plastics Minerals Building Shell Windows G G Roof Insulation G G Wall Insulation G G Insulation (general) A ←−−−−−−−−−−−−−−−−− A∗ −−−−−−−−−−−−−−−−−→ Lighting A,G A,G Appliances Washing Machines G Refrigerators A,G Televisions A,G Air Conditioning A A Mechanical G G G G G High Temp. Processes G G G G G Discrete technology options and their encompassing sectors analyzed by Sikharulidze et al (2015) and Timilsina et al (2015) shown in Roman text, aggregation to technology classes and industry groups in the present study shown in italics. A = Armenia, G = Georgia. ∗ Different industries considered as a single aggregate. in the areas of residential appliances and insulation, which indicate potential for long-run discounted abatement in the amount of 0.67 MTCO2 in Armenia and 6 MTCO2 in Georgia at or below zero net cost. The underlying scenarios of energy-efficiency penetration assume trajectories of capital-energy substitution over the 2015-2034 policy horizon that are shown in Panel B. Importantly, a substantial share of the assumed total abatement emanates from improvements in technology options whose efficiency improvements can only be acquired at a significant capital cost premium (particularly residential lighting, and commercial lighting in Georgia). The key issue here is that, even controlling for the cost of abatement and the “profitability” of the options under consideration, the penetration of energy efficiency and the overall quantity of abatement both turn on assumptions about the size of the market for technologies’ services. We shall see that this question of market potential figures prominently in the subsequent numerical analysis in Section 4. Panel C provides details of technologies’ costs of fuel and capital inputs in the 2015 base year, along with capital-energy elasticities of substitution implied by the penetration of energy efficient technology relative to the baseline scenario.8 The magnitude of the latter reflect the steepness of the loci in panel B, and indicate the potential for larger energy efficency improvement at lower opportunity cost. 8 Elasticities are estimated from OLS fits of (10) to annual projections of technology costs and emissions under baseline and counterfactual scenarios. 12 Figure 1: Bottom-Up Analysis Results A. Marginal abatement cost curves Armenia Georgia Residential Televisions 100 Household Televisions 100 Residential Washing Machines Discounted Marginal Abatement Cost ($/tCO2e) Discounted Marginal Abatement Cost ($/tCO2e) Commercial Air Conditioning Residential Wall Insulation 50 50 Public Lighting Commercial Wall Insulation Household Air Conditioning 0 0 ‐50 ‐50 Residential Windows ‐100 Residential Roof Insulation ‐100 Commercial Windows ‐150 Household Insulation Commercial Roof Insulation ‐150 ‐200 Public Lighting Industrial Insulation Residential Refrigerators ‐200 ‐250 Household Refrigerators Residential Lighting Household Lighting Commercial Lighting ‐250 ‐300 0 100 200 300 400 500 600 700 800 0 1000 2000 3000 4000 5000 6000 7000 Discounted Potential ktCO2e saving Discounted Potential ktCO2e saving B. Capital-energy substitution Armenia Georgia 1 1 HH Building Shell HH Building Shell Fractional change in energy input from baseline ( ̃ / ) Fractional change in energy input from baseline ( ̃ / ) HH Lighting HH Lighting 0.9 0.9 HH Appliances HH Appliances 0.8 HH HVAC 0.8 Com. Bldg Shell Com. HVAC Com. Lighting 0.7 Com. Lighting 0.7 Ind. Mechanical Ind. Insulation Ind. High Temp. 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 1 2 3 4 5 6 1 3 5 7 9 11 13 Fractional change in capital input from baseline ( / ) Fractional change in capital input from baseline ( / ) C. Base-year technology input costs ($ M) and implied capital-energy elasticities Armenia Georgia Gas Coal Elec. Capital ξθ Gas Coal Elec. Capital ξθ Residential Bldg. Shell 376.1 79.0 -2.15 Bldg. Shell 300.7 88.6 -0.77 Lighting 97.0 2.5 -0.16 Lighting 63.7 4.6 -0.11 Appliances 114.5 73.4 -1.50 Appliances 73.7 111.1 -1.28 HVAC 22.4 9.8 -1.13 Commercial HVAC 8.0 3.9 -1.16 Bldg. Shell 57.2 8.4 -0.76 Lighting 9.1 8.6 -1.31 Lighting 15.6 9.0 -0.24 Industrial Insulation 93.7 0.4 -0.19 Mechanical 69.2 4.8 -0.07 High Temp. Proc. 57.1 66.3 38.1 -0.60 13 3.3 CGE Modeling We conduct a top-down analysis using a three-region, 14-sector recursive-dynamic CGE sim- ulation of the world economy. Consumers in each region are modeled as a representative agent who owns the factors of production (labor and jelly capital) and rents them out to producers in exchange for income which is used to finance consumption of commodities. Pro- ducers in each region’s sectors are modeled as a representative firm which combines inputs of intermediate commodities and factors to produce a sector-specific good which is sold to the households, other sectors, the government, and other regions. The government is mod- eled as a passive entity which produces a government good from purchases of commodity inputs that are financed by revenue collected from taxes on production, imports and exports. Inter-regional trade in commodities is modeled using the Armington formulation in which a region’s uses of a given commodity are a constant elasticity of substitution (CES) composite of domestic and imported varieties, and the region’s aggregate imports of each commodity are a CES composite of quantities exported by its trade partners. The dimensions of this static general equilibrium structure are summarized in Figure 2.A. Following the comple- mentarity format of equilibrium, the structure is posed algebraically as a square system of zero profit, market clearance and income balance conditions. The technical coefficients of the constituent cost and demand functions are numerically calibrated on the Global Trade Analysis Project (GTAP) version 8 database (Narayanan et al, 2012) using the MPSGE subsystem (Rutherford, 1999) for GAMS (Brooke et al, 1998), and the resulting numerical problem is expressed and solved as a mixed complementarity problem using the PATH solver (Dirkse and Ferris, 1995; Ferris and Munson, 2000). The static sub-model described above is embedded within a dynamic process which up- dates economies’ savings/investment demands, and endowments of labor, capital and nat- ural resources, over the 2015-2035 simulation horizon on a five-year time step. Regions’ labor endowments are assumed to increase exogenously with population, following projec- tions in World Bank (2015). Regions’ representative agents are assumed to exhibit constant marginal propensity to save and invest out of factor income, and investment drives inter- period jelly capital accumulation according to the standard perpetual inventory formulation. The model’s dynamics are tuned to produce baseline trajectories of GDP and CO2 emissions are consistent with DOE/EIA (2014) projections. We do this by adjusting the growth rates of exogenous aggregate labor productivity and autonomous energy efficiency improvement (AEEI) parameters. Additional modeling details are given in an appendix to the paper. Two challenges attend implementation of the general equilibrium model of section 2.4. The first is specifying the demand for technology-specific services in eq. (16). In technology- rich sectors (denoted by the index j ) discrete technology options’ demand for fuel and capital (QFe,θ,j and QXθ,j ) make up a portion of the total sectoral use of these inputs. From a modeling perspective, the key unknown is how precisely the services thus produced (QSθ,j ) combine with the broader set of inputs to generate the sector’s output. Our approach is to specify a “technology services composite” which is a constant elasticity of substitution (CES) aggregation of the outputs of the technologies in that sector, and allow this bundle of services to substitute for capital, labor energy and materials. As shown in Figure 2.B, the practical consequence is to split the target sectors into technology and non-technology components (indicated by the dashed and dotted areas), where, starting at the top of the 14 Figure 2: Structure of the CGE Model A. Regional and sectoral structure Regions Sectors Fuels (e) Technology-Rich (j ) Other Final Demand Armenia Coal* Food Products Agriculture Consumption Georgia Crude Oil Chemicals, Rubber & Plastics Rest of Economy Investment Rest of World Natural Gas* Iron & Steel Government Refined Fuels* Non-metallic Minerals Electricity Other Industry Services Private dwellings * Sectors whose output commodities are associated with CO2 emissions. B. Nesting Structure in technology-rich production sectors   Domestic    Production                Technology    Services  Energy + Materials    Composite  + Value Added                Discrete    Technology    Options  Energy  Value Added    ( )  Composite  + Materials            0        Technology   Fuel  Coal  Crude  Gas  Ref.  Elec.  Value‐  Intermediate    Capital     ( )  Oil  Fuel  Added  Materials  (  )              Armington  Armington  Armington    Energy  Energy  Labor  Capital  Non‐energy    Commodities  Commodities  Commodities                    Domestic  Imported  Domestic  Imported    Domestic  Imported    Inputs  Inputs  Inputs  Inputs  Inputs  Inputs    Technology    Component  Non‐Technology Component        Elasticities of substitution: D TS   Y σ = 0.2, σ = 0   .5, σ = 0.8, σ E = 0.5, σ N V A = 0.75, σ N = 1.5, σ V A = 0.93 − 1.43, σ DM = 1.9 − 11           15 production hierarchy, a composite of technology-specific services substitutes for the com- ponent of sectoral output produced from inputs of labor, capital and intermediate energy and material commodities. A key issue we confront in numerically calibrating this struc- ture is a pervasive lack of empirical estimates on the substitutability of technology services for other inputs to production. Absent evidence to the contrary, our assumption is that the relationship between technology services and sectoral output is complementary, though not strictly so. We therefore assign the elasticity of substitution between technology and non-technology components (σ D ) a default value of 0.2. Although the services produced by individual technology options are necessary inputs, it is plausible to assume that they are fungible to a limited degree, which we capture by setting the substitution elasticity σ T S to a default value of 0.5. Following eq. (15), each technology’s output is a Leontief combination of technology-specific capital and Armington energy commodities, with input coefficients determined by the benchmark cost shares in Table 1.C. We model the rest of the sector using a familiar nested CES formulation, in which a composite of residual intermediate energy inputs substitute for a composite of residual in- termediate material inputs and value added. One level lower in the hierarchy, the energy (materials) composite is a CES aggregation of intermediate (non-) fuel Armington commodi- ties, while the value-added composite is a Cobb-Douglas aggregation of labor and capital. One caution is in order regarding interpretation of Figure 2.B, however. It may be tempting to think of the dashed area as representing energy efficient technologies and the remainder of the sector as representing conventional technologies, but such an interpretation is not correct. Crucially, as in the analytical model in section 2.4, each technology option rep- resents an average of conventional and energy efficient varieties. The effect of low-carbon growth policies is to shift toward the latter, which results in an increase in the quantity of technology-specific capital and a decline in technologies’ use of energy. We implement this by scaling the coefficient on QXθ,j upward while simultaneously scaling the coefficient on QFe,θ,j downward.9 The second challenge is representing the opportunity cost of increased investment in energy efficient varieties of capital. Perhaps the closest CGE analogue of the capacity ex- pansion problem in eq. (12) is the forward-looking model of Timilsina and Landis (2014), in which distinct technology-specific capital stocks are used to model individual technologies’ capacities, and investment is allocated among the various options according to expectations of their future rates of return. But application of this formulation in the present setting is beyond our capability, for several reasons. Our model’s myopic character means that invest- ment decisions would need to be based on current as opposed to future prices. Another issue is the multiple-capital-stocks model’s lack of intra -temporal flexibility: technology supports have to be expressed either as mandated shares of investment in all technology options as a share of aggregate new capital formation, or as a vector of θ investment subsidies, otherwise their prescribed acceleration in targeted options’ investment and capital accumulation rules out substitution where some technologies expand endogenously at the expense of others. But by far the biggest hurdle is the algebraic strictures of CGE models’ computational im- 9 Of course other production structures are possible—e.g., where technology-specific and rest-of-sector inputs of capital substitute for one other within the same nest, and the same for fuel—but their limitation is that by separating inputs across sub-nests they do not allow for the production of technology services within the sector. 16 plementation, which dramatically complicates the specification of efficiency improvements as endogenous relationships between additional current investment (the intertemporal ana- logue of µ) and declines in the values of different technologies’ coefficients on energy inputs in subsequent periods (the intertemporal analogue of η ). For all of these reasons we opt for the simpler approach of specifying opportunity costs not through the investment channel, but through intra-period endogenous transformation of the economy-wide stock of jelly capital into quantities of technology-specific capital input, following eq. (19). In each period the economy is endowed with a fixed quantity of jelly capital which is capable of being re-allocated across the various non-energy sectors as the relative prices of their inputs and outputs change. By Shephard’s Lemma, the derivative of the cost function represented by Figure 2.B with respect to the price of technology-specific capital determines the unconditional demand for this input, as well as the extent of transfor- mation of jelly capital into technology capital that must generate the supply necessary for market clearance. It is this transformation process which mandates to improve energy effi- ciency are assumed to distort. Additional technology deployment policies such as subsidies may be introduced here as well, but for the purposes of this study we narrow the focus to mandates because their ultimate economic and environmental impacts are transparent, and uncontaminated by tax interaction effects associated with pre-existing distortions that tend to be more common in developing countries. The model is spun up from the GTAP database’s year-2008 benchmark to 2015 by scaling regions’ endowments of labor and capital to generate increases in GDP consistent with IMF (2014) growth forecasts. In the 2015 base year technology options are introduced as “backstop” activities by recalibrating the model to accommodate both their demands for capital and fuels given in Figure 1.C, and corresponding demands for their outputs by the remainder of their encompassing sectors, per Figure 2.A. The upshot of this modeling choice is that the levels of base-year demand and activity solved for by the model diverge from the bottom-up values in Figure 1.C, with some options exhibiting large increases (e.g., mechanical processes in Georgia’s food products sector: +42%) or decreases (appliances in private dwellings in Armenia: -66%), but most coming in at between 80% and 110% of their target values. From 2015 onward the model is simulated with discrete technologies active. In the baseline solution, technology options have no cost premium over jelly capital (µθ = 0), and capacity is allocated endogenously, in line with the capital intensities and relative rates of growth of the output of the various technology-rich sectors. Apropos the discussion in section 2.3, the model is not capable of selecting the optimal portfolio of energy-efficiency investments by endogenously adjusting the vector µθ in each period. Rather, the markups are prescribed by the analyst.10 We conduct several counterfactual numerical experiments in which we model technology mandates as an exogenously-imposed increase in technologies’ capital cost markups. Specifically, we assume that (1 + µθ ) increases from a value of unity in 2015 at 1%, 2%, 3%, 5%, 7% and 10% per year for all technologies simultaneously. In addition to treating technology options with equanimity, we also investigate the impact of 10 Note that if it were possible to solve for the optimal baseline portfolio, GDP improvement would be im- possible unless there were additional new technologies that were somehow only available in the counterfactual scenario. 17 mandates that are targeted to differentiate among technologies. We do this by simulating the penetration scenario that underpins the MAC curves in Figure 1.A, specifying (1 + µθ ) as evolving according to the trajectories in Figure 1.B. In each case, we draw inferences about the economic consequences of such a policy by comparing the resulting trajectories of economic variables against their counterparts in the baseline model solution. 4 Numerical Results 4.1 Baseline Projections We begin by giving a sense of the technologies’ importance in a general equilibrium set- ting. In our 2015 base year the levelized cost of discrete technology options makes up a significant proportion of total production costs in only a handful of industries: building shell technologies in private dwellings (Armenia 43%; Georgia 36%), high-temperature process technologies in non-metallic minerals and mechanical equipment in iron & steel (29% and 11%, respectively, both in Georgia), appliances in private dwellings (15% in both countries), with the remaining technologies each accounting for less than ten percent of the total cost of their encompassing sectors.11 All technologies exhibit quite stable output shares over the course of the baseline simulation. Regarding technologies’ environmental performance, half of the technologies use electric power as an energy source. This is significant because improvement in these options’ energy efficiency generates emission reductions only indirectly, by reducing the demand for—and, crucially, domestic production of—electric power generated from fossil fuels. The quantity of abatement thus produced depends on the extent to which the consequent reduction in electric power demand depresses electricity prices, inducing those parts of the economy out- side the scope of technology mandates to increase their consumption, and the CO2 emission intensity of power ultimately conserved as a result. Therefore, to account for abatement from electricity savings we impute emissions embodied in electric power using the average CO2 intensity of each country’s domestic generation.12 This accounting indicates that, overall, technologies account for 15%-17% of Armenia’s and Georgia’s 2015 emissions, amounting to some 860 thousand tons (kT) and 1.2 million tons (MT) of CO2 , respectively. Their contribution grows slightly over the course of the baseline simulation, by 2035 reaching 18%-20%, or 1.5 MT and 1.8 MT.13 And with a few notable exceptions, these totals are broadly distributed across the slate of technologies. In Armenia, the largest emitters are building shell technologies in the private dwellings and other industry sectors (3%-4% of aggregate CO2 ). In Georgia, they are mechanical tech- nologies in iron & steel (1.4% of total emissions), and high-temperature process technologies 11 Noteworthy among the rest are lighting in private dwellings (Armenia 9%; Georgia 6%) and high- temperature process technologies in Georgia’s iron & steel sector (7%). 12 Although the fraction of electric power consumption satisfied by imports, and how much CO2 is em- bodied in the kWhs generated by each electricity-exporting trade partner, bias this calculation, the fact that Armenia’s and Georgia’s electricity imports are each only 3% of production makes this a reasonable approximation. 13 This is generally in line with the expansion of projected aggregate emissions over 2015-2035 from 5.6 MT to 8 MT in Armenia and 6.9 MT to 8.8 MT in Georgia. 18 in iron & steel, non-metallic materials and food products (3%, 3.4% and 4%, respectively). These proportions are also stable over time. Energy-efficiency mandates tend to have larger beneficial spillover effects on sectors’ productivity, and larger abatement potential, when the technologies that they target make up a large fraction of sectors’ energy consumption. In Armenia, building shell technologies generate 25% of the CO2 emitted by industrial sectors, while in Georgia high-temperature process technologies account for more than 40% of emissions from iron & steel and 65% from food products and non-metallic materials. Discrete technologies account for all of the CO2 emitted by private dwellings, but countries vary in the distribution of this total across different options.14 In Armenia two-thirds of residential CO2 is generated by building shell technologies, with the remainder split evenly between lighting and appliances, By contrast, half of Georgia’s residential emissions emanate from appliances and a further 42% come from lighting. The temporal stability of these shares is lower, but they tend to vary over the simulation by only a few percentage points. As our theoretical analysis indicates, realizing the economic benefits essential to low- carbon growth strategies necessitates identification of technology options which possess both substantial abatement potential and elastic efficiency improvement responses to technology deployment mandates. Unfortunately, in this regard silver bullets may be rare on non- existent; here we see only two: residential building shell technologies in Armenia, and, to a lesser extent, industrial high-temperature process technologies in Georgia. The key implica- tion is that small quantities of abatement may have to accumulated across a broad portfolio of technologies, with attendant uncertainty in technology strategies’ ultimate economic and environmental outcomes. It is also likely that such potential efficiency benefits may be limited by rebound effects. The latter are more likely when the targeted technologies comprise a substantial share of the demand for energy commodities that are widely used across the economy. This is also apparent from the model’s solution. Electricity-using technologies accounted for 21% of the CO2 emitted by Armenia’s generators and 16% of similar emissions in Georgia. In both counties electricity consumption is dominated by residential technologies: building shell (Armenia 12%; Georgia 4.5%), and lighting and appliances (each 2%-3.5%). 4.2 Counterfactual Scenarios The macroeconomic implications of our counterfactual simulations are illustrated in Figure 3. The main result of the paper is in Panel A, which shows the percentage change in CO2 emissions relative to the baseline against the percentage change in real GDP from the baseline. Despite the two charts having different scales, the scenarios trace out qualitatively similar loci in the shape of a backwards “C” that begins at the top (0,0) in the year 2015 and ends at the bottom in 2035. The surprising result is that low carbon growth strategies are feasible for both Armenia and Georgia. Energy-efficiency investments’ productivity benefits outweigh their opportunity costs over an unexpectedly broad range, enabling both countries to reduce their emissions by up to 4% from baseline levels while enjoying higher real GDP. 14 This is by construction. In the benchmark social accounting matrix the private dwellings sector account does not record intermediate energy commodity purchases; these are tabulated in final consumption. 19 Figure 3: Energy Efficiency Penetration: Aggregate Impacts, 2015-2035 Armenia Georgia A. Real GDP and CO2 Emissions 0 0 -0.5 % Change in CO2 emissions from baseline % Change in CO2 emissions from baseline -1 -5 -1.5 -2 -2.5 -10 -3 -3.5 -15 -4 -4.5 -5 -20 -2 -1.5 -1 -0.5 0 0.5 1 1.5 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 % Change in Real GDP from baseline % Change in Real GDP from baseline 1% 2% 3% 5% 7% 10% Bottom-Up Analysis 1% 2% 3% 5% 7% 10% Bottom-Up Analysis B. Welfare 1.4 1.4 1.2 1.2 % Change in Welfare from Baseline % Change in Welfare from Baseline 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0 0.0 -0.2 -0.2 2015 2020 2025 2030 2035 2015 2020 2025 2030 2035 Abatement by Rest of Sector (kTCO2) 1% 2% 3% 5% 7% 10% -50 Bottom-Up Analysis 1% 2% 3% 5% 7% 10% Bottom-Up Analysis C. Sectoral CO2 Abatement and Leakage 0 400 350 -100 Abatement by Rest of Sector (kTCO2) Abatement by Rest of Sector (kTCO2) -50 300 250 -100 -150 200 150 -150 -200 100 -200 50 0 -250 -250 -50 0 100 200 300 400 500 0 600 100 700 200 0 300 200 400 400600 500 1000 600 800 1200 700 1600 1400 Abatement by Discrete Technologies (kTCO2) Abatement by Discrete Technologies Abatement (kTCO by Discrete 2) Technologies (kTCO2) 1% 2% 3% 5% 7% 10% Bottom-Up Analysis 1% 2% 3% 5% 7% 10% Bottom-Up Analysis Annual rate of increase in µ: 1% 2% 3% 5% 7% 10% Bottom-Up Analysis 20 Both countries exhibit similar potential for reducing emission without penalizing eco- nomic growth, but differ in the elasticity of their GDP and abatement responses to increas- ing technology-specific capital markups. Even with an aggressive 5% annual rate of increase in the technology markup, Armenia’s GDP remains above its baseline trajectory by 2035. By contrast, for Georgia’s GDP to exceed its baseline trajectory by 2035, markups must be much smaller—equivalent to a growth rate of efficiency penetration of 2% per annum— otherwise the opportunity costs of technology investment quickly outweigh the productivity benefits. Allied with this, the maximal increase in GDP above baseline experienced by Arme- nia (about 1%) is over four times as large as that in Georgia. This difference has its roots in the uniformly smaller magnitudes of the capital cost elasticity of efficiency improvement for equivalent classes of technologies in Georgia as compared to Armenia, which simply results in Georgia experiencing smaller productivity and abatement benefits for the same opportunity cost of abatement capital. The results also highlight the need for caution when attempting to use MAC curves to assess the prospects for low-carbon economic growth. The dashed line shows the results of simulating the bottom-up technology penetration scenario from Section 3.2 in which resi- dential space conditioning and appliances, as well as residential and commercial lighting, are intensively targeted. Under this policy stimulus the locus of the economy’s response has a similar shape, but in both regions the peak GDP increase is sharply curtailed, as is the total reduction in emissions. The explanation is that mandated increases in capital costs are concentrated in technologies whose elasticity of energy efficiency improvement is relatively small, thus dissipating the policy’s gross environmental, and net economic, benefits.15 We consider this result in more detail below. Figure 3.B indicates that an improvement in aggregate consumption and gross welfare that for the most part parallels the increase in GDP. For annual rates of increase in the technology capital markup of up to 5% these benefits continue to increase over the simula- tion horizon, but with more rapid penetration diminishing returns become evident, as the marginal benefits of improved productivity fail to keep pace with the marginal opportunity cost of ever more expensive technology-specific capital. Additional longer-run simulations indicate that this phenomenon is pervasive, in the sense that even with a comparatively slow pace of cumulative technology deployment, the point of diminishing returns will eventually be reached—albeit well after the 30-year horizon of interest here. These results are also a reminder of how misleading GDP can be as measure of aggregate well-being. Section 3.2’s bottom-up technology penetration scenario, even though associated with comparatively small increase in GDP that quickly peters out, nonetheless generates a sustained improvement in welfare due to the effects of the policy on commodity prices and household consumption. Figure 3.C illustrates the policy’s impacts on abatement by the technology and non- technology components of technology-rich sectors. The two economies exhibit markedly dif- ferent patterns of intra-sectoral emission leakage. In Armenia, the direct effect of increased energy efficiency is to reduce CO2 by up to 500 kT annually, but expansion of fuel use by the non-technology component of technology-rich sectors dissipates one half to one third of this 15 A salutary example is commercial lighting in Georgia, which has one of the lowest capital-energy elas- ticities and is among the most capital intensive technologies, yet is subject to one of the most stringent penetration mandates—by 2035 (1 + µ) > 6! 21 abatement, with the rate at which such “internal” leakage occurs declining with increasing technology deployment. By contrast, in Georgia the technology and non-technology compo- nents’ demands for fuel appear to be largely complementary, with a much larger quantity of primary emission reductions (1.2 GT) coinciding with negative leakage that enhances abate- ment by 10%-30% over most of its range. Abatement and leakage responses in the bottom-up scenario are qualitatively similar but much smaller in magnitude. We also test the extent to which the bottom-up analysis’ assumptions of options’ techni- cal potential might be biased by recalculating the discounted value of CO2 directly emitted by our slate of technologies over the 2015-2034 period. For the bottom-up scenario, total discounted abatement is 689 kT in Armenia, slightly underestimating Figure 1.A. However, in Georgia we find less than 1.2 MT of discounted abatement as compared to 7 MT predicted by engineering assessments, suggesting that the latter are wildly optimistic. Our results also capture the joint consequences for emissions of general equilibrium effects such as rebound and increasing capital prices’ drag on broader economic activity by repeating these calcula- tions for total economy-wide abatement. Armenia’s discounted abatement shrinks by almost 40% to 421 kT, while Georgia’s rises slightly to 1.3 MT. These results can be understood by examining the responses of sectors under a 2% annual energy efficiency improvement scenario, shown in Figure 4. Panel A indicates that a key contributor to Armenia’s increase in welfare (and, by implication, household consumption) is a rapid increase in output of the private dwellings sector (1% per annum on average), which is in turn propelled by productivity benefits of the concentration of elastic residential technologies. Over the simulation horizon, the chemicals, rubber & plastics sector is a a distant second, expanding by only 8% relative to the baseline, while the majority of industries exhibit increases, but of less than half this magnitude. By contrast, Georgia’s sectors are evenly split between small output increases and outright losses from their baseline levels, which accounts for its smaller improvement in GDP and welfare. At the the root of this outcome are sectors’ relatively inelastic energy efficiency and productivity responses to the technology mandate. Sectoral dynamics also help explain the two countries’ similar abatement performance. By 2035 Armenia experiences an 8% drop in electricity output and associated emissions, while Georgia’s electricity and coal sectors shrink by 4% and 10%, respectively. Figure 4.B illustrates the abatement and leakage consequences of changes in the outputs of technology rich sectors. In Armenia, several sectors exhibit “internal” leakage rates greater than 100%, with relative prices changes inducing increases fuel use in their non-technology components that swamp the reduction in fuel use in their technology components. Chief among these are sectors which undergo intermediate expansion, such as chemicals, rubber & plastics and other industry. Crucially, emission leakage by the private dwellings sector is small relative to the large reduction in CO2 that it directly generates. This compensates for the leakage from other parts of the economy, and also accounts for the sharp divergence between the emission reductions attributable to technology options and the overall quantity of abatement identified earlier. Georgia’s experience is very different, however. Technology-rich sectors’ abatement responses tend to be larger and more evenly distributed. Intra-sectoral emission leakage is far smaller, with sectors such as chemicals, rubber & plastics, services and other industry exhibiting negative leakage that drives the responses in Figure 3.C. Figure 5 elucidates the microeconomic underpinnings of our sectoral results by interpolat- 22 Figure 4: 2% Annual Energy Efficiency Improvement: Sectoral Impacts, 2015-2035 Armenia Georgia A. Output 25 25 20 20 % Change in Output from Baseline % Change in Output from Baseline 15 15 10 10 5 5 0 0 25 -5 -5 20 % Change in Output from Baseline -10 -10 2015 2020 2025 2030 2035 2015 2020 2025 2030 2035 Agriculture Food Products Chemicals, Rubber & Plastics Agriculture 15 B. CO2 Abatement and Other Intra-Sectoral Food Products Chemicals, Rubber & Plastics Leakage (Technology-Rich Sectors Only) Non-metallic Minerals Iron & Steel Other Industry Non-metallic Minerals Iron & Steel Industry Services Dwellings Rest of Economy Services 5 Dwellings Rest of Economy 5 Gas Petroleum Electricity Gas 10 Petroleum Electricity Coal Abatement by Rest of Sector (kTCO2) Abatement by Rest of Sector (kTCO2) 0 0 -5 -5 -10 5 -10 -15 -15 -20 0 -20 -25 -25 -30 -30 -5 -35 -35 -40 -40 -20 0 20 -1040 60 80 100 120 140 0 20 40 60 80 100 120 140 2015 Abatement by 2020 Discrete Technologies (kTCO2) 2025 2030 by Discrete Technologies Abatement 2035 (kTCO2) Food Products Chemicals, Rubber & Plastics Food Products Chemicals, Rubber & Plastics Agriculture Food Products Chemicals, Rubber & Plastics Non-metallic Materials Iron & Steel Non-metallic Materials Iron & Steel Other Industry Non-metallic Minerals Services Iron & Steel Other Industry Services Other Industry Dwellings Services Dwellings Dwellings Rest of Economy Gas Petroleum Electricity Coal 23 ing across scenarios to trace out the locus of technologies’ responses to the policy stimulus.16 In general, fewer technologies show declining trajectories in Armenia as compared to Geor- gia, reflecting the latter’s smaller efficiency improvement elasticity values. The impact of the markup and concomitant productivity boost on the change in the technology-specific capital price and level of output relative to the baseline solution is shown in Figure 5.A, where technology substitution dynamics lead to three types of trajectories. First, a handful of options are economically unattractive candidates for efficiency improvement, leading their output to decline relative to baseline levels (e.g., commercial lighting in both countries; resi- dential appliances and mechanical devices in Georgia’s food products sector). Symmetrically, other options are competitive, and continue to expand their output over the entire period of the simulation (e.g., building shell insulation in both countries, and residential lighting and commercial space conditioning in Armenia). Most options fall between these two extremes, exhibiting increased output relative to the baseline for one or more periods before declining, and then eventually falling below baseline levels. The steepest drop-offs occur in residen- tial appliances and space conditioning in Armenia, and industrial process technologies in Georgia’s iron & steel and non-metallic materials sectors. The fact that these technologies’ efficiency responses are relatively elastic demonstrates the critical importance of the satura- tion in efficiency’s direct productivity-enhancing effect (i.e., the 1/ηθ factor in the first term of (27)). The consequent non-monotonic response highlights the shifting balance between the diminishing marginal productivity benefit of efficiency improvement and the increasing marginal opportunity cost of the mandate as ηθ increases. This is the key driver of aggregate output response. We close by examining the consequences for technologies’ direct and embodied CO2 emis- sions in Figure 5.B. Abatement is pervasive: there is only one technology for which output expansion increases emissions above its baseline level. Notwithstanding this, technologies that experience large expansions in output generally abate less.17 As a group, abatement by electricity-using technologies is more vigorous than that by non-electric technologies in Armenia, but the reverse is true for Georgia. This can be traced to the larger (smaller) values of ξθ for Armenia’s (non-) electric technologies compared to Georgia’s. Increasing activity reflects declining unit production costs as the productivity benefit of energy efficiency out- strips the rising cost of capital acquisition. Technologies for which this is the case experience a decoupling of output from fuel use that ends up causing their emissions to decline relative to the baseline. Crucially, however, the same emission trajectory can result if productivity falls short of opportunity costs, production costs and output prices rise, and activity and the demand for fuel both decline. In both cases the abatement benefit is the same—the key difference lies in the economic consequences. Examples are space conditioning and lighting in Armenia’s service sector, and Armenia’s residential building shell technologies juxtaposed with Georgia’s residential appliances. 16 The loci are nonparametric lowess regression fits to the pooled datapoints from all simulations. 17 Building shell technologies in Armenia’s residential and Georgia’s commercial sectors are exceptions. 24 Figure 5: Technology Dynamics Armenia Georgia A. Output (i) Electricity-Using Technologies % Change in technology output 40 % Change in technology output 40 20 20 0 0 −20 −20 −40 −40 0 100 200 300 400 0 100 200 300 400 % Change in technology−specific capital price % Change in technology−specific capital price (ii) Non-Electric Technologies Appliances Lighting Appliances Lighting Technologies Technologies % Change in technology output 40 % Change in technology output Bldg. Shell HVAC 40 Bldg. Shell Mechanical Dwellings Chemicals Food Prod. Non−Metallic Mat. 20 Sectors 20 Sectors Services Dwellings Iron & Steel Services 0 0 −20 −20 −40 −40 0 100 200 300 400 0 100 200 300 400 % Change in technology−specific capital price % Change in technology−specific capital price Bldg. Shell Bldg. Shell Technologies B. CO2 Emissions Technologies High−Temp. Proc. (i) Electricity-Using Technologies Chemicals Iron & Steel Other Ind. Chemicals Food Prod. Non−Metallic Mat. Sectors Sectors 0 Food Prod. Non−Metallic Mat. 0 Dwellings Iron & Steel % Change in CO2 emissions % Change in CO2 emissions −25 −25 −50 −50 −75 −75 % Change in technology output 40 −100 −100 0 100 200 300 400 0 100 200 300 400 % Change in technology−specific capital price % Change in technology−specific capital price 20 (ii) Non-Electric Technologies % Change in technology output Appliances Lighting 40 Appliances Lighting Technologies Technologies 0 Bldg. Shell HVAC 0 Bldg. Shell Mechanical % Change in CO2 emissions % Change in CO2 emissions 0 20 Dwellings Chemicals Food Prod. Non−Metallic Mat. −25 Sectors −25 Sectors Services Dwellings Iron & Steel Services −20 0 −50 −50 −40 −20 −75 −75 0 100 200 300 400 % Change in technology−specific capital price −100 −40 −100 Chemicals Food Prod. Non−Metallic Mat. Services Sectors0 100 200 300 400 0 0 100 100 200 200 300 300 400 400 % Change inIron Dwellings & Steel Other Ind. technology−specific capital price % Change % Change in technology−specific in technology−specific capital capital priceprice Bldg. Shell Bldg. Shell Appliances Technologies High−Temp. Proc. Mechanical Chemicals Food Prod. Technologies Non−Metallic Mat. Services Technologies Sectors High−Temp. Proc. Bldg. Shell Lighting HVAC Dwellings Iron & Steel Other Ind. Chemicals Iron & Steel Other Ind. Chemicals Food Prod. Non−Metallic Mat. Sectors Sectors Appliances High−Temp. Proc. Mechanical Food Prod. Non−Metallic Mat. Technologies Dwellings Iron & Steel 25 Bldg. Shell Lighting HVAC 5 Discussion and Conclusions A recent paper by Fowlie et al (2014) notes that “Widely publicized engineering estimates [. . .] suggest that consumers are systematically bypassing opportunities to invest in cost- effective energy efficiency improvements that lower their energy expenditures and reduce externalities associated with energy production. As a result, many policymakers are cham- pioning programs designed to encourage energy efficiency as a cost-effective strategy to con- front climate change.” Like these authors, we find that bottom-up estimates tend to over- state the benefits of energy efficiency mandates in terms of energy and emissions savings. But our analysis comes at this issue from the opposite direction, examining the aggregate- level environmental and economic effects of technology deployment strategies for low-carbon economic economic growth. Building on the fundamentals of engineering cost assessments, we develop a stylized bottom-up/top-down general equilibrium model which accounts for the several influences: the opportunity costs of increasing the installed base of energy effi- cient technology-specific capital, the direct productivity benefits to economic sectors of the resulting increases in energy efficiency, and the broader impact of rebound effects. We then implement this model as a CGE simulation which integrates energy technology data into the system of open-economy input-output accounts for Armenia and Georgia. Our main finding is that mandated increases in the penetration of bottom-up energy effi- ciency technology options can mitigate CO2 up to around 4 % of baseline emissions without adversely impacting real GDP or welfare. Even so, as capacity in these more efficient but rel- atively expensive technologies expands, the productivity benefits to technology-rich sectors are progressively diminished by increasing opportunity costs. The good news for transition economies such as Armenia and Georgia is that even modest increases in energy efficiency (e.g., our 1% or 2% per annum deployment rates) appear to generate both welfare improve- ments and reductions in the rate of growth of emissions—albeit small in magnitude. The bad news is that, depending on the structure of the economy, the direct abatement realized by higher energy efficiency tends to be undermined by substantial emission leakage—due to in- creased fossil fuel use by less efficient conventional technologies in both target and non-target sectors. Moreover, despite the usefulness of our methodology for using abatement costs and potentials to identify technology options whose energy productivity response is elastic rel- ative to opportunity costs, these options’ potential contributions to abatement or national income growth are unknown ex ante and remain an emergent property of the economy-wide simulation. Overall outcomes depend critically on the structure of the economy (particularly the fuel- and capital-intensities of sectors targeted for deployment) as well as the abatement technology portfolio, which together shape the general equilibrium effects that determine technologies’ macroeconomic costs, output demand, and leakage. All these considerations suggest that our findings, rather than being viewed as the definitive conclusion, should be seen as marking the start of a research program to rigorously assess the potential of a range of template low-carbon economic growth strategies in different country settings. 26 Appendix A. Further Details of the Modeling Approach Our focus is on elucidating the tradeoff between the abatement and productivity benefits of improved energy efficiency and the opportunity costs of acquiring expensive relatively expensive technology-specific capital. To that end, we strive to keep the analysis as trans- parent as possible by choosing to represent the remainder of the economy in a simple and straightforward fashion, following the canonical model in Sue Wing and Balistreri (2014). In sectors where discrete technology options are not represented, the production function is specified according to the nested CES structure of the non-technology component identified by the dotted area in Figure 2.B. In each region the representative agent has CES preferences denominated over the 14 commodities, and a consumption elasticity of substitution of 0.5. Production of the government good has a similarly parameterized CES structure. Distor- tions are specified in terms of ad-valorem equivalents and are maintained at their benchmark levels given by the GTAP database. We do not impose government budgetary balance, or constrain revenue and expenditure in the counterfactual scenario to match their baseline trajectories. Similarly, we simulate the economy’s full-employment equilibrium, eschewing labor market closures such as labor-leisure choice that capture the potential impacts of pro- ductivity improvements on, and introduce the additional general equilibrium feedbacks of, elastic labor supply. To be sure, results such as the employment consequences of energy pro- ductivity gains will be of interest to policy makers, but they are secondary to our purposes here, and as such we leave their elaboration to future research. 27 References Allan, G., N. Hanley, P. McGregor, K. Swales, K. Turner (2007). The impact of increased efficiency in the industrial use of energy: a computable general equilibrium analysis for the United Kingdom, Energy Economics 29: 779-798. Allan, G., M. Gilmartin, P. McGregor, J. Kim Swales, K. Turner (2009). Modelling the economy-wide rebound effect, in H. Herring and S. Sorrell (eds.), Energy Efficiency and Sustainable Consumption, 67-98. Allcott, H. and M. Greenstone Is there an Energy Efficiency Gap? Journal of Economic Perspectives 26: 3-28. Aldy, J.E., S. Barrett and R.N. Stavins (2003). Thirteen Plus One: A Comparison of Global Climate Policy Architectures, Climate Policy 3: 373-397. Anderson, S.T. and R.G. Newell (2004). Information programs for technology adoption: the case of energy-efficiency audits, Resource and Energy Economics 26: 27-50. Broberg, T., C. Berg and E. Samakovlis (2015). The economy-wide rebound effect from improved energy efficiency in Swedish industriesA general equilibrium analysis, Energy Policy 83: 26-37. Brooke, A., D. Kendrick, A. Meeraus, and R. Raman (1998). GAMS: A Users Guide, Wash- ington D.C.: GAMS Development Corp. Dirkse, S.P., and M.C. Ferris (1995). The PATH Solver: A Non-Monotone Stabilization Scheme for Mixed Complementarity Problems, Optimization Methods and Software 5: 123-156. Ferris, M.C., and T.S. Munson (2000). Complementarity problems in GAMS and the PATH solver, Journal of Economic Dynamics and Control 24: 165-188. Fowlie, M., M. Greenstone and C. Wolfram (2014). Do Energy Effciency Investments Deliver? Evidence from the Weatherization Assistance Program, Working Paper. Hallegatte, S., G. Heal, M. Fay and D. Treguer (2011). From growth to green growth—a framework, World Bank Policy Research Working Paper No. WPS5872. International Monetary Fund (2014). IMF World Economic Outlook (WEO): Legacies, Clouds, Uncertainties, Washington DC: IMF. Jacoby, H.D., R. Schmalensee and D.M. Reiner (1997). What does stabilizing greenhouse gas concentrations mean? in B. Flannery et al. (eds.), Critical Issues in the Economics of Climate Change, London: International Petroleum Industry Environmental Conservation Association. Jacoby, H.D., R. Schmalensee and I. Sue Wing (1998). Toward a Useful Architecture for Cli- mate Change Negotiations, Proceedings of the OECD Workshop on the Economic Mod- eling of Climate Change, Paris, 17-18 September. 28 Kiuila, O. and T.F. Rutherford (2013a). Piecewise smooth approximation of bottom-up abatement cost curves, Energy Economics 40: 734-742. Kiuila, O. and T.F. Rutherford (2013b). The cost of reducing CO2 emissions: integrating abatement technologies into economic modeling, Ecological Economics 87: 62-71. McKinsey & Co. (2010). Impact of the financial crisis on carbon economics: Version 2.1 of the global greenhouse gas abatement cost curve, http://www.mckinsey.com/~/ media/McKinsey/dotcom/client_service/Sustainability/cost\%20curve\%20PDFs/ ImpactFinancialCrisisCarbonEconomicsGHGcostcurveV21.ashx Narayanan, G.B., A. Aguiar and R. McDougall, eds. (2012). Global Trade, Assistance, and Production: The GTAP 8 Data Base, Center for Global Trade Analysis, Purdue University. Popp, D. (2012). The Role of Technological Change in Green Growth, NBER WOrking paper No. 18506. Rozenberg, J., A. Vogt-Schilb and S. Hallegatte (2013). How capital-based instruments fa- cilitate the transition toward a low-carbon economy: a tradeoff between optimality and acceptability, World Bank Policy Research Working Paper No. WPS6609. Rutherford, T.F. (1999). Applied General Equilibrium Modeling with MPSGE as a GAMS Subsystem: An Overview of the Modeling Framework and Syntax, Computational Eco- nomics 14: 1-46. Sikharulidze, A., G. Timilsina and E. Karapoghosyan (2015). How do We Prioritize the GHG Mitigation Options in Georgia? Development of Marginal Abatement Cost Curves for the Building Sector, draft report, World Bank. Stewart, R.B., and J.B. Wiener (2001). Reconstructing Climate Policy: The Paths Ahead, Policy Matters No. 01-23, Washington D.C.: AEI-Brookings Joint Center for Regulatory Studies. Sue Wing, I. and E.J. Balistreri (2014). Computable General Equilibrium Models for Economic Policy Evaluation and Economic Consequence Analysis, in S.-H. Chen, M. Kaboudan and Y.-R. Du (eds.), Oxford University Press Handbook on Computational Economics and Finance, forthcoming. Timilsina, G., E. Karapoghosyan and S. Suren (2015). Economic Analysis of Energy Effi- ciency Measures in the Building sector in Armenia, draft report, World Bank. Timilsina, G. and F. Landis (2014). Economics of Transiting to Renewable Energy in Mo- rocco, World Bank Policy Research Working Paper No. 6940. U.S. Dept. of Energy/Energy Information Administration (2014). International Energy Out- look, Washington DC: DOE/EIA-0484(2014). van den Bergh, J.C.J.M. (2011). Energy Conservation More Effective With Rebound Policy, Environmental and Resource Economics 48: 43-58. 29 Vogt-Schilb, A., S. Hallegatte and C. de Gouvello (2014). Long-term mitigation strategies and marginal abatement cost curves: a case study on Brazil, World Bank Policy Research Working Paper No. WPS6808. World Bank (2015). Health Nutrition and Population Statistics: Population estimates and projections, Washington DC: World Bank. 30