تعادل عمومی، فن آوری تولید برق و هزینه کاهش کربن: تجزیه و تحلیل حساسیت ساختاری

Electricity generation is a major contributor to carbon dioxide emissions, and abatement in this sector is a key determinant of economy-wide regulation costs. The complexity of an integrated representation of economic and electricity systems makes simplifying assumptions appealing, but there is no evidence in the literature on how important the pitfalls may be. The aim of this paper is to provide such evidence, drawing on numerical simulations from a suite of partial and general equilibrium models that share common technological features and are calibrated to the same benchmark data. We report two basic findings. First, general equilibrium inter-sectoral effects of an economy-wide carbon policy are large. It follows that assessing abatement potentials and price changes in the electricity sector with a partial equilibrium Marshallian demand can only provide a crude approximation of the complex demand-side interactions. Second, we provide evidence that widely used top-down representations of electricity technologies produce fuel substitution patterns that are inconsistent with bottom-up cost data. This supports the view that the parametrization of substitution possibilities with highly aggregated production functions is difficult to validate empirically. The overall picture that emerges is one of large quantitative and even qualitative differences, highlighting the role of key structural assumptions in the interpretation of climate policy projections.

Electricity generation is a significant contributor to carbon dioxide (CO2) emissions, and potentially has an important role in abatement efforts. The current research paradigm for ex-ante carbon policy assessment mainly involves two classes of models (Hourcade et al., 2006). On the one hand, technology-rich ‘bottom-up’ models provide a detailed representation of generation technologies and the overall electricity system. By construction, these models are partial equilibrium, and typically include no or very limited interactions with the macroeconomic system. On the other hand, economy-wide ‘top-down’ models represent sectoral economic activities and electricity generation technologies through aggregate production functions. While these models are designed to incorporate general equilibrium effects, the use of smooth functions is not well suited to capture the temporal and discrete nature of technology choice.1 The integration of bottom-up technology representation and economy-wide interactions into ‘hybrid’ models is the subject of a large literature. For example, reference is often made to ‘soft-linked’ models, where the combination of the two models either fail to achieve overall consistency (Drouet et al., 2005, Hofman and Jorgenson, 1977, Hogan and Weyant, 1982 and Jacoby and Schäfer, 2006), or complement one type of model with a ‘reduced-form’ representation of the other, thereby lacking structural explicitness (Bosetti et al., 2006, Manne et al., 2006, Messner and Schrattenholzer, 2000 and Strachan and Kannan, 2008). An alternative and more recent approach, referenced to as ‘hard-linked’, is to directly embed a set of discrete generation technologies into a top-down model (Böhringer, 1998, Böhringer and Rutherford, 2008 and Sue Wing, 2006). Under this approach, however, the representation of technological detail significantly increases the dimensionality of the model, thus severely constraining large-scale applications. Finally, a decomposition algorithm by Böhringer and Rutherford (2009) employs an iterative solution procedure to solve top-down and bottom-up model components consistently. This approach is essentially a soft-linked approach, but overcomes issues of dimensionality and consistency, and has been employed in the context of U.S. climate policy in Tuladhar et al. (2009). Despite the large literature documenting efforts to reconcile top-down and bottom-up modeling paradigms, and a tendency towards ever more detailed models, there is no quantitative evidence on the pitfalls of different simplifying assumptions. The objective of this paper is to explore the implications of different structural assumptions concerning electricity supply and demand for the assessment of economy-wide carbon policies, thereby going beyond the usual parametric sensitivity analysis. As it is impossible to derive general qualitative propositions for such an issue, we employ a suite of numerical partial equilibrium (PE) and general equilibrium (GE) models that share common technological features and are calibrated to the same benchmark equilibrium. Our benchmark model consistently integrates a bottom-up technology representation of the electricity sector within a general equilibrium setting based on the decomposition method by Böhringer and Rutherford (2009). The economy-wide component is based on a static version of the MIT U.S. Regional Energy Policy (USREP) model, a multi-sector multi-region numerical general equilibrium model designed to analyze climate and energy policy in the U.S. (Rausch et al., 2010a and Rausch et al., 2010b). The electricity sector is represented by a multi-region model based on a comprehensive database of electric generators from the Energy Information Administration (EIA, 2007a), and features detailed plant-level information on the generation costs and capacity, fuel switching capabilities, and season-specific load profiles.2 We assume imperfect factor mobility in the economy and fixed capacity of electricity generation technologies, so that the response to a policy shock is of short- to mid-term horizon.3 Our results are as follows. First, we find that general equilibrium income and substitution effects induced by an economy-wide carbon policy are of first-order importance to evaluate the response of the electricity sector. Changes in electricity prices and abatement potentials are largely driven by both the slope and the location of the demand schedule. Following the suggestion in an early and influential article by Hogan and Manne (1977), we explore whether price elasticities of electricity demand simulated from a GE model can approximate general equilibrium effects in a partial equilibrium setting. We report evidence that such a modeling strategy is not sufficient to capture the underlying economy-wide changes, as represented in an integrated model. For example, we calculate that general equilibrium effects mitigate electricity price increases by up to 20% in the case of even moderate carbon prices of around $25 per metric ton of CO2. Our second set of results relates to the representation of electricity generation technologies in general equilibrium top-down models by means of aggregate substitution elasticities. We implement two top-down technology specifications based on nested constant elasticity of substitution (CES) functions (Bovenberg and Goulder, 1996 and Paltsev et al., 2009) which are widely adopted for ex-ante climate policy assessment. Our analysis suggests that these representations produce fuel substitution patterns that are inconsistent with bottom-up cost data, mainly because top-down representation of electricity markets implies that the price of electricity reflects the total carbon content of generation. This contrasts with real markets (and the bottom-up approach), where the carbon price is reflected in the electricity price through the carbon content of the marginal producer at a given point in time (Stavins, 2008). In our setup, structural assumptions about the technology representation translate into country-wide welfare costs that differ by as much as 60% for an emissions reduction target of 20%. We further observe large heterogeneity in regional discrepancies, mostly driven by the benchmark shares of carbon-intensive technologies. On a more general level, our findings demonstrate the significance of structural assumptions embedded in top-down and bottom-up modeling approaches for the assessment of carbon and energy policies. While both approaches rely on the assumption of fully rational behavior, the structural setting makes empirical validation of the behavioral response in each modeling approach is difficult. Any analysis inevitably involves simplifications from a more complex reality, but we usually do not know how misleading assumptions might be when drawing policy conclusions from quantitative analysis. By providing evidence on the magnitude of structural assumptions, albeit in the context of models also using a set of restrictive assumptions, we believe that our investigation contributes to an improved understanding of the theoretical and methodological basis for carbon policy assessment with large-scale simulation models. The remainder of this paper proceeds as follows. Section 2 provides an overview of the economy-wide model, describes the top-down and bottom-up representations of the electric power sector, and presents the integrated model. Section 3 investigates the importance of general equilibrium factors and the implications of top-down versus bottom-up technology representation for carbon policy assessment. Section 4 concludes.

Large-scale numerical models have become a popular and widespread tool to assess the economic implications of climate and energy policies. While the virtue of top-down models is their representation of general equilibrium effects, a major source of critiques is their reliance on smooth aggregate production functions to describe the technology choice in the electric power sector. In contrast, bottom-up models have a rich technological underpinning but typically do not account for general equilibrium effects. By developing an integrated benchmark model that embeds a bottom-up technology representation of the electricity sector within a multi-sector general equilibrium framework, we generate numerical evidence on (1) the importance of general equilibrium effects for partial equilibrium bottom-up models of the electricity sector, and (2) the implications of top-down versus bottom-up representations of electric generation technologies for assessing the cost and environmental effects of CO2 control policies. In the context of U.S. climate policy, our numerical analysis suggests that the general equilibrium effects and the mode of representation of electricity technologies are of crucial importance for projecting electricity prices and demand, carbon abatement potentials, and welfare costs. Moreover, the elasticity parameters needed for a reduced-form model response are difficult to estimate from empirical observations, for two reasons. First, general equilibrium effects associated with carbon policies are complex and difficult to identify from historic data. Second, while nested CES function can accommodate any substitution patterns, the empirical validation of these structures to represent substitution among electricity generation technologies is difficult. In our framework, bottom-up and top-down models represent a structural representation of the electricity supply and demand respectively, and our comparison exercise generates quantitative insights on the implication of these assumptions for policy assessment. As a final note, we emphasize that our quantitative results are model-specific and abstract from a number of features. First, we do not represent the dynamic response to a policy shock. Specifically, the response of the electricity sector is constrained by the existing generation capacity, and most of the actions take place in the substitution between coal and gas. While our analysis is informative in this respect, carbon-free technologies such as wind, solar, hydro and nuclear are used at their effective capacity in the benchmark, and thus cannot expand as they get more competitive under a carbon price. The expansion of renewable technologies under a carbon price is obviously an important research question, but in our view the structural assumption about electricity sector will be even more important in a capacity-expansion model. Thus we expect discrepancies between modeling frameworks to be even more important in a forward-looking framework. Second, our analysis abstracts from physical constraints on the transmission network which are likely to hamper the flexibility in the substitution among technologies and might increase the welfare costs of carbon policy. We note that such constraints will be difficult to represent accurately in a highly aggregated top-down representation of electric power technologies. Finally, while the assumption of marginal cost pricing makes the comparison across different modeling paradigms more transparent, carbon abatement policies is likely to be affected by the extent of state regulation and imperfect competition in the U.S. electricity markets. Quantifying the effects of market structure on the cost of carbon regulation is, however, beyond the scope of the present paper.