WPS5803 Policy Research Working Paper 5803 When Starting with the Most Expensive Option Makes Sense Use and Misuse of Marginal Abatement Cost Curves Adrien Vogt-Schilb Stéphane Hallegatte The World Bank Sustainable Development Network Office of the Chief Economist September 2011 Policy Research Working Paper 5803 Abstract This article investigates the use of expert-based or there is a longer-term target. The best way to achieve Marginal Abatement Cost Curves (MACC) to design Europe’s goal of 20 percent reduction in emissions by abatement strategies. It shows that introducing inertia, 2020 is different if this objective is the ultimate objective in the form of the “cost in timeâ€? of available options, or if it is only a milestone in a trajectory toward a 75 changes significantly the message from MACCs. With percent reduction in 2050. The cheapest options may an abatement objective in cumulative emissions (e.g., be sufficient to reach the 2020 target but could create a emitting less than 200 GtCO2 in the 2000–2050 period), carbon-intensive lock-in and preclude deeper emission it makes sense to implement some of the more expensive reductions by 2050. These results show that in a world options before the potential of the cheapest ones has been without perfect foresight and perfect credibility of the exhausted. With abatement targets expressed in terms of long-term carbon-price signal, a unique carbon price in emissions at one point in time (e.g., reducing emissions all sectors is not the most efficient approach. Sectoral by 20 percent in 2020), it can even be preferable to start objectives, such as Europe’s 20 percent renewable energy with the implementation of the most expensive options if target in Europe, fuel-economy standards in the auto their potential is high and their inertia significant. Also, industry, or changes in urban planning, building norms the best strategy to reach a short-term target is different and infrastructure design are a critical part of an efficient depending on whether this target is the ultimate objective mitigation policy. This paper is a product of the Office of the Chief Economist, Sustainable Development Network. 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 vogt@centre-cired.fr and hallegatte@centre-cired.fr. 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 When Starting with the Most Expensive Option Makes Sense: Use and Misuse of Marginal Abatement Cost Curves a∗ b,c Adrien Vogt-Schilb , Stephane Hallegatte September 20, 2011 ∗ Corresponding author; e-mail: vogt@centre-cired.fr. a CIRED, 45bis avenue de la belle Gabrielle, F-94736 Nogent-sur-Marne, France b 1 The World Bank , Sustainable Development Network, Washington D.C., USA c Ecole Nationale de la Meteorologie, Meteo-France, Toulouse, France Abstract This article investigates the use of expert-based Marginal Abatement Cost Curves (MACC) to design abatement strategies. It shows that introducing inertia, in the form of the cost in time of available options, changes signicantly the message from MACCs. With an abatement objective in cumulative emissions (e.g., emitting less than 200 GtCO2 in the 2000-2050 period), it makes sense to implement some of the more expensive options before the potential of the cheapest ones has been exhausted. With abatement targets expressed in terms of emissions at one point in time (e.g., reducing emissions by 20% in 2020), it can even be preferable to start with the implementation of the most expensive options if their potential is high and their inertia signicant. Also, the best strategy to reach a short-term target is dierent depending on whether this target is the ultimate objective or there is a longer-term target. The best way to achieve Europe's goal of 20% reduction in emissions by 2020 is dierent if this objective is the ultimate objective or if it is only a milestone in a trajectory toward a 75% reduction in 2050. The cheapest options may be sucient to reach the 2020 target but could create a carbon-intensive lock-in and preclude deeper emission reductions by 2050. These results show that in a world without perfect foresight and perfect credibility of the long-term carbon-price signal, a unique carbon price in all sectors is not the most ecient approach. Sectoral objectives, such as Europe's 20% renewable energy target in Europe, fuel-economy standards in the auto industry, or changes in urban planning, building norms and infrastructure design are a critical part of an ecient mitigation policy. Keywords: marginal abatement cost curves, MACC, inertia, when-exibility, how-exibility, opti- mal abatement strategy, merit-order, timing, dynamic eciency. 1 The views expressed in this paper are the sole responsibility of the author. They do not necessarily reect the views of the World Bank, its executive directors, or the countries they represent. 1 1 Introduction To design the best policies to cope with climate change, decision-makers need information about the various options to reduce greenhouse gas (GHG) emissions. Such information has been provided to the public in many dierent ways, including in the form of Marginal Abatement Cost (MAC) curves. Kesicki (2010) calls expert-based MAC curves (here, we simply refer to them as MAC curves or MACCs ) the curves that represent information on abatement costs and potentials for a set of mitigation activities. 2 MAC curves are usually constructed for a specic country or region, and for a specic time horizon. They report abatement potentials that can be achieved as a function of the abatement cost (Fig. 1). They are constructed following a bottom-up approach: they assess a set of available mitigation options, each taken independently. Where they have been developed, they have proved useful to communicate about abatement options and potentials. MACCs rank potential mitigation options from the least to the most expensive one. By doing so, they look like merit-order curves, which are used to describe various production options that are instantly available at a given point in time to satisfy a demand at a given price. In the power generation sector, for instance, merit-order curves are used as supply curves (Fig. 2): crossing them with demand curves actually gives the optimal dispatch (Ongsakul, 1999). This similarity in the presentation could suggest that MAC curves can be used as conventional supply curves too  this is done for instance in an early attempt by Jackson (1991). If MACCs were conventional supply curves, an abatement activity should be implemented when the carbon price is larger or equal to its marginal abatement cost. In this case, a cost-minimizing strategy would exhaust the cheapest available options, before progressively turning to more expensive options until the committed abatement level is reached. This paper shows that MACCs are dierent from merit-order curves because they include ac- tivities that could take decades to implement. Because of this inertia (Grubb et al, 1995), the implementation schedule of the various options  and even the choice of mitigation options  suggested by a naive interpretation of MACCs is suboptimal. The objective of this paper is to investigate the optimal timing of GHG emissions abatement (choice across time) along with the optimal dispatch of the reduction burden (choice across abatement activities). To do so, we introduce inertia  in the form of a cost in time of each activity  in a MAC curve and use an inter-temporal optimization model to investigate the when-exibility on the implementation of various abatement options. The cost in time makes it possible to distinguish 2 In this paper, we call these abatement options  activities , following the terminology from Rodrik (2004). Activities include changing technologies, notably in the car and power sectors, but also non-technological options such as modal shift in the transportation, waste recycling, reforestation or building retrotting sectors. While the terms MAC curve can also refer to dierent curves, as those named model-derived MAC curves by Kesicki (2010) and studied for instance by Klepper and Peterson (2006), we focus in this paper on expert-based MAC curves. 2 Cost $/tCO2 N Y N-1 5 c(i) 4 3 X 1 2 amax (i) Abatement potential MtCO2/yr Figure 1: An expert-based MAC curve exhibits abatements options 1..N characterized by their maximum potential amax and their marginal cost c, ranked from the least to the most expensive one. The classical in- terpretation, challenged in this paper, is that the optimal strategy for achieving abatement X is to implement all the activities cheaper than Y. available abatement activities not only using their costs and potentials, but also the time it takes to implement them. For instance, it allows taking into account the fact that urban planning may be cheaper and have a higher potential to reduce emissions than technological change in the car industry, but is also much slower and requires much more anticipation to be eective (Gusdorf et al, 2008). Conrming previous results from Lecocq et al (1998) and Jaccard and Rivers (2007), introducing inertia changes the order in which various options should be implemented, and  under some conditions  can make it optimal to start with the introduction of the most expensive options. Moreover, with inertia, a uniform carbon price across sectors cannot trigger the optimal dispatch of activities, unless its long-term evolution is perfectly credible and economic actors have perfect foresight. In a realistic setting  one with inertia, public-policy long-term credibility issues, and imperfect foresight  it can make sense to use sector-specic complementary policies such as Europe's 20% renewable energy target in Europe or fuel economy standards in the auto industry (An et al, 2007). The article starts with a review of the literature on two parallel research questions: the MAC curves methodology and limits (Section 2.1) and the timing of mitigation strategies (Section 2.2). Section 3 then presents a model that computes the least-cost abatement strategy to reach a climate objective, taking into account technological ceilings, costs and inertia. In Section 4, this model is used to design an optimal abatement strategy to reach an objective expressed in terms of cumu- lative emissions (for instance emitting less than 200 GtCO2 over the 2000-2050 period). We show that, because of the interplay of inertia and discounting, this optimal abatement strategy will not 3 Marginal cost $/MW 8 D S 7 6 p 5 4 3 2 1 q Capacity MW Figure 2: Merit order in the power generation sector. The graph exhibits a supply curve (S) made by aggregating 8 available power plants and a demand curve (D). Each power plant is characterized by available capacities (in MW) and a marginal cost (in $/MW). The optimal response to the demand is to call power plants 1 to 6 and to let 7 and 8 idle. necessarily exhaust all cheap abatement options before starting to implement more expensive ones. In Section 5, we examine optimal strategies to achieve emission targets at one point in time (e.g., Europe's goal of reducing emissions by 20% by 2020). In this case, we demonstrate it can make sense to start with the most expensive options rather than with the cheapest. We also show why MAC curves should not be used as supply curves to nd the optimal activity mix to achieve a particular abatement at a given time horizon. Instead, this mix strongly depends on longer-run objectives, so that the optimal activity mix to reach a given target in 2020 is strongly dependent on the objective for later years (e.g. 2050). Our conclusion (Section 6) discusses how to translate a general climate objective (whether expressed in cumulative emissions, carbon budget or concentration) into sectoral objectives in terms of emissions at one point in time, and shows that MAC curves could be usefully extended to account for additional information on inertia. 2 Literature review 2.1 Expert-based MAC curves usage and their limitations Expert-based MAC curves are built in a bottom-up, engineering, approach. They investigate avail- able technical options to mitigate GHG emissions, and present them in terms of their abatement potential and marginal cost. Kesicki (2010) proposes a denition of expert-based MAC curves and a review of their use in the literature. They dier from model-based MACC curves (Klepper and Peterson, 2006) which are outside of the scope of this paper and therefore not discussed here. McKinsey and Company have built expert-based MAC curves assessing potentials in the USA 4 in 2030 (McKinsey and Company, 2007) and Ireland in 2030 (Motherway and Walker, 2009). The World Bank has assessed Poland's and Mexico's reduction potentials in 2030 in the form of MACCs (Poswiata and Bogdan, 2009; Johnson et al, 2009). And Sweeney and Weyant (2008) have proposed a MAC curve for California in 2020. MAC curves are designed to provide synthetic and easily usable information to policymakers. Their use has however been criticized. For instance, Ekins et al (2011) mention that: • MAC curves generally neglect non-climate benets such as air pollution reduction or increase in energy security. They also neglect co-costs, like negative distributional impacts. • They do not account for synergies between abatement activities (e.g., MACCs do not report that promoting electric vehicles and green electricity together would allow to save more GHG than the sum of the two isolated abatement activities). • They also neglect possible rebound eects (Greening et al, 2000) and the need for behavioral or institutional changes. • They assess project or technological costs only, excluding institutional barriers and transac- tion costs. This is the classical explanation for why MAC curves often report negative-costs abatement options  so-called no-regret options (Bréchet and Jouvet, 2009). These weaknesses are due to the fact that MAC curves are developed in a bottom-up framework. However, this approach has advantages compared with integrated assessments, which often appear as black-boxes, and has proved useful for communication and debate. In this paper, the main criticism we address to the MAC curves is not related to the way they are built, but rather to the interpretation made of them. As we will show, one major issue with the MAC curves is that they do not take into account inter-temporal dynamics and inertia, which play a key role in determining the optimal timing of abatement options. 2.2 The timing of mitigation strategies The optimal timing of greenhouse gases abatement to limit climate change is widely discussed in the literature. In many cases, these discussions are developed with models that represent the economy as one single aggregated sector. For instance, the topic has emerged from arguments in favor of postponing emission reductions, proposed by Nordhaus (1992), Hammitt et al (1992) and Wigley et al (1996). Those arguments include the risk of sunk investments if climate change reveals less threatening than expected, and the role of discounting. Others have studied the eect of technical change (Goulder and Mathai, 2000; Manne and Richels, 2004; Sue Wing, 2006), the risk of early impact of climate change and the interplay between uncertainty in the carbon and climate cycles and irreversible investments (Ha-Duong et al, 1997; Ambrosi et al, 2003). A more complete literature review can be found in Kverndokk and Rosendahl (2007). 5 A dierent approach investigate the link between when- and how-exibility using models with multiple sectors. Lecocq et al (1998), Gilotte (2004) and Jaccard and Rivers (2007) focus on the role that long-lived capital plays in the design of mitigation policies. They argue that the dierences in inertia across dierent economic sectors matter, and that the design of optimal abatement strategies cannot therefore be investigated using aggregated models. Lecocq et al (1998) conclude that in absence of perfect foresight, classical policy instruments (such as a carbon price) might not be enough to decarbonize economies at low costs, and they call for specic policies directed toward green infrastructure and long-lived capital. Jaccard and Rivers (2007) nd that early action is preferable in long-lived capital sectors, even if marginal costs are higher there. A third branch of the literature studies the interplay between technology and environmental policies (Jae et al, 2005). They often focus on the eect of the learning by doing (LBD) when there are several available technologies and imperfect foresight (Kverndokk and Rosendahl, 2007; del Rio Gonzalez, 2008). In this framework, carbon prices cannot trigger investments in the most promising solutions if initial costs are too high; targeted policies are therefore needed to encourage early action and initiate costs reductions. Our contribution is threefold: rst, we to develop a new  independent from LBD  argument for early action in some expensive activities, based on the interplay of inertia and high abatement potentials; second we show that factoring in inertia is not only necessary for sectors like infrastructure and urbanism, but that information on the implementation inertia of all abatement activities is critical to design optimal strategies; third, we show that MAC curves can easily be extended to account for this additional information. 3 Model description Our model is called MACinert (Marginal abatement cost curves and inertia). It is consistent with the MAC curves framework (see Section 2.1), i.e. abatements are achieved by applying changes to a xed baseline. We do not incorporate more realistic but complex dynamics, such as capital accumulation, learning-by-doing, sectoral interactions, or crowding-out eect on investment. Instead, a social planner controls GHG abatements from an emission baseline, by spending time and money in abatement activities, characterized by a MAC curve and a cost in time  the latter takes inertia into account. We then use this model to carry out simple numerical experiments. Parameters corresponds roughly to the European Union situation, but these simulations have an illustrative purpose only. They demonstrate what MACCs can and cannot do, and provide insight on how to use them. 6 3.1 GHG emissions There are N abatement options, indexed by i. The model is run on a period that goes from 2000 to 2050 with a time step, ∆t, of one year. At each time step t, emissions are computed from the baseline emissions Ebase (t) and the abatement achieved with each activity i. N E(t) = Ebase (t) − a(i, t) (1) i=1 We assume constant baseline emissions, that is Ebase (t) = 5 GtCO2 /yr. The cumulative emissions M (t) are then computed as the sum of emissions: M (t) = E(t) · ∆t + M (t − 1) (2) M (0) = 0 (3) 3.2 Potentials, costs, and inertia Abatements eorts in each sector are subject to two restrictions. First, each activity i has a technical ceiling, i.e. a maximum abating potential amax (i), expressed in avoided annual emissions, i.e. in MtCO2 /yr. This potential is commonly represented by the rectangles width in MAC curves (see Fig. 1). For instance, switching to more ecient thermal engines for personal vehicles could save a fraction of GHG emissions associated with private mobility, but not more. a(i, t) ≤ amax (i) (4) In the MAC curve, each activity i is qualied with a constant marginal cost of abatement c(i) (see Fig. 1). Here, we assume that marginal abatement costs are independent of cumulative abatements and of time. In particular, we do not model technical change and learning-by-doing. Therefore, abatements a(i, t) achieved thanks to activity i at time t have a cost I(i, t) which reads: I(i, t) = a(i, t) · c(i) (5) The innovation here is the introduction of inertia in the MACCs. Inertia is dened by a  cost in time : a given amount of abatement requires to spend a positive amount of time for its implementa- tion. This cost could be expressed in years per MtCO2 , and is assumed independent of the nancial cost of the option. In other words, there is a maximum amount of supplementary abatement that is achievable with each activity during a given time interval, which is expressed in MtCO2 per year. 3 2 Let α(i), in MtCO2 /yr , be the maximal incremental amount of GHG emissions abatement achievable with activity i over a year. Note that α(i) is the algebraic inverse of the cost in time. For 3 This modeling diers from the time-to-build à la Kydland and Prescott (1982). Time-to-build would reect the idea that there is an incompressible lag between investment decisions and actual abatements. With time-to-build, an arbitrary large amount of abatements would require as much time to be implemented as a small abatement (if achieved through the same activity). With our cost in time, in contrast, the time expenditures are proportional to the amount of abatement, in the same way as nancial expenditures. 7 the sake of simplicity, we assume α(i) is constant and does not depend on the previously achieved abatements nor current time step t. In other terms, there is no learning-by-doing leading to a reduction in the cost in time of an activity. This α, by introducing inertia in the modeling framework, also introduces path dependency: achievable abatements at time t+1 directly depend on already achieved abatements at time t. a(i, t) ≤ a(i, t − 1) + α(i) · ∆t (6) These costs in time may come from several factors, such as (i) availability of skilled workers, (ii) availability of productive capacities, (iii) emissions being embedded in capital, and (iv) incompressible institutional requirements. Points (i) and (ii) could be overcome by training workers or redirecting unemployed workers and unused capital; but training and redirecting are activities per se and cannot be done overnight either. The third point is related to capital vintages and turnover: if one sees emissions as embedded in capital (Davis et al, 2010; Guivarch and Hallegatte, 2011), decarbonization cannot be faster than capital turnover, except by wasting valuable productive capital through early scrappage (see Section 2.2). In the case of urban planning, transforming a city cannot be done in less than decades, and an aggressive action can lead to strongly negative distributional impacts within a city (Gusdorf et al, 2008). The issue of institutional or organizational delays is well documented (World Bank, 2010). Reducing them is an activity per se, and takes time. We thus see the maximal amount α as an exogenous constraint  independent of the cost  that the social planner must take into account when searching for optimal abatement strategies. 3.3 Social planner objectives The objective of the social planner is to achieve a climate-related target while minimizing abatement costs. The social planner minimizes C, the total present cost of abatements, discounted at rate Ï? over the period: T N I(i, t) C= (7) t=0 i=1 (1 + Ï?)t Theoretically, the social planner could control GHG emissions in order to equalize the marginal costs of mitigation and adaptation, in a cost-benet approach as in Nordhaus (1992). Because of uncertainty surrounding both climate response to a change in GHG emissions and adaptation costs, and because decision-making is done at national scale (and not at the global scale as would be required for a global public goods issue like climate change), it is common to adopt a cost-eectiveness approach (Ambrosi et al, 2003). In our model, this can be done by constraining cumulative emissions M to remain below a given objective Mobj . M (t) ≤ Mobj (8) 8 Cumulative emissions can be used as proxies for climate change (Matthews et al, 2009; Mein- shausen et al, 2009; Allen et al, 2009); in practice, however, governments and other public agencies can hardly implement them. In contrast, they may provide emission objectives for given points in time. To do so, they mostly rely on emissions targets. 4 For instance, the European Commission has the objective of cutting its emissions by 20 % of 1990 levels by 2020. Other countries have dierent objectives, such as the UK objective of cutting emissions by 75 % by 2050. In our model, this type of objectives can be implemented by dening a set of milestones indexed by m, and constraining emissions at each milestone: obj E(tm ) = Em (9) 3.4 Numerical values For illustrative purpose, we assumed a MAC containing only two contrasted activities (N = 2), labeled  cheap  and  deep . Cheap has a lower abatement cost than deep, but the latter has a greater abatement potential (see Tab. 1 and Fig. 3). Cheap could represent for instance the activity of switching energy sources in the buildings, and deep could stand for retrotting those buildings. In the auto industry, cheap could stand for energy eciency gains in the internal combustion engines and deep for switching to other energy sources, such as electricity or biofuels. 2 The cost in time is arbitrarily assumed to be equal to 50 MtCO2 /yr in both activities. As shown by the ratios amax /α, this means that deep presents more inertia than cheap, because implementing the full potential of deep takes 60 years, while it takes only 30 years for cheap. Abatement cost Abatement potential Maximal increment 2 c ( $/(tCO2 )) amax ( MtCO2 /yr) α ( MtCO2 /yr ) Cheap 30 1 500 50 Deep 60 3 500 50 Table 1: Numerical assumptions These values are not meant to represent accurately concrete sectors of the economy. We use them to carry out illustrative experiments, which help draw more general conclusions. We solve this simple model using a linear programing algorithm provided by GAMS (Brook et al, 1988). The source code is available upon request. The main results are presented in the next two sections. 4 In the auto industry, it is common to adopt intensity objectives, expressed in gCO2 /km (An et al, 2007). In the framework of a MAC curve, nonetheless, intensity targets can be treated as emission targets. The MACCs are built assuming a given baseline; in particular, they assume that the aggregate driven distance is given (Sweeney and Weyant, 2008). The multiplication of this distance by the intensity objectives provides an emission target. 9 Cost $/tCO2 60 30 0.75 1.5 5 Abatement potential GtCO2/yr Figure 3: The MAC curve used in our experiments. According to the supply-curve interpretation, an abatement of annual emission by 750 MtCO2 /yr should be achieved through the implementation of cheap only. 4 Optimal timing with cumulative-emissions objectives In this section, we investigate the optimal abatement pathway when using a constraint in terms of cumulative emission, i.e. with full when-exibility on how to reduce emissions. Cumulative emissions can be seen as good proxies for climate change (Matthews et al, 2009; Meinshausen et al, 2009; Allen et al, 2009). This is implemented in our model by excluding Eq. 9, and using Eq. 8. We then test several values of the objective (Mobj ), and assess the consequence on the optimal reduction pathway. 4.1 Using expensive options before the potential of cheap ones is ex- hausted Figure 4: Optimal abatement strategy to limit cumulative emissions below 175 GtCO2 . Because of inertia and discounting, the deep option enters before the potential of cheap has been exhausted. Figure 4 shows the optimal strategy for maintaining cumulative emission below 175 GtCO2 . 5 5 Cumulative emissions in the baseline amount to 5 Gt/yr during 51 years, with a total of 255 Gt. 10 This particular value is used for illustrative purpose, and will allow us to make some comparisons with subsequent simulations with emissions targets (see Section 5). The abatement paths have triangular or trapezoidal shapes; this shows that one of the inertia (Eq. 6) or ceiling (Eq. 4) constraint is always binding. In this case, the intuitive ranking of abatement activities is respected: the social planner starts implementing cheap before deep. But the social planner does not use the full potential of cheap before starting using deep, which enters in 2017 while cheap does not reach its ceiling before 2030. Moreover, a more stringent objective would force deep to start even earlier (see below). The optimal implementation strategy does not strictly follow an hypothetical merit-order in which all the potential of the cheapest solutions is used before more expensive solutions are introduced. This example illustrates why the MAC curves should not be interpreted as abatements supply curves. A more systematic analysis using dierent cumulative emission objectives (Fig. 5) conrms that  for any objective  it is never preferable to implement the expensive deep before cheap. But it shows that if the objective is stringent enough (about 210 GtCO2 ), deep has to begin before all the potential of cheap has been exploited; the implementation is not sequential. If it is even more stringent (about 135 GtCO2 ), deep is forced to start in 2000, at the same time as cheap. Figure 5: Entry date et ceiling date of each abatement activity as a function of the cumulative emission objective. The entry date is the date when an activity begins to be implemented; the ceiling date is the date when the full potential of an activity is achieved. 4.2 Expensive options may be useful even when cheaper ones appear sucient Let us rst analyze a case in which the climate objective is not very restrictive, say 210 GtCO2 . This translates into cumulative abatements of 45 GtCO2 over the period. 6 Cheap has a cumulative abate- 6 Cumulative emissions in the baseline amount to 5 Gt/yr during 51 years, with a total of 255 Gt. 11 ment potential of 54 GtCO2 . 7 It is then possible to achieve the abatement objective by implementing cheap only, without implementing deep. A naive strategy could then focus on implementing cheap and not implementing deep, because cheap has the lowest marginal cost. But our simulations show that this is not the optimal strategy, because there is a trade-o between (i) implementing only the cheapest solutions, but starting early to give them enough time to reach the objective; (ii) delaying abatements in order to save present value (thanks to the discounting), but undertaking both cheap and deep to be more aggressive and reach the objective in spite of the delayed action. In our simulations (Fig. 5), the optimal strategy to reach the (lax) 210 GtCO2 objective calls deep from year 2038, which makes it possible not to call cheap before 2009 (for a strategy starting in 2000). The additional cost of using deep is more than compensated by the delay in the implementation of cheap (again, thanks to the discounting). In other words, the optimal strategy uses an expensive activity even when a cheaper activity appears sucient to attain the objective, in order to take advantage of the discounting. Again, this result challenges the interpretation of MAC curves as abatement supply curves providing a merit-order ranking. Furthermore, it rules out one of the most common interpretation of the MACCs, namely that only the cheapest options should be implemented in order to reach a given amount of abatements (see Fig. 1). These experiments illustrate the importance of taking into account the interplay of discounting with dierentiated inertias when designing a mitigation strategy. Interpreting a MAC curve as a merit-order curve could lead to postpone expensive abatements after the cheap options are exhausted, which would lead to suboptimal mitigation pathways. With discounting and inertia, it makes sense to implement expensive abatement activities in parallel to cheap ones, to be more aggressive and be able to postpone action. Of course, this conclusion does not take into account policy and behavior inertia, and the incentive for policy-makers to delay action beyond their term of oce, which justify the use of emission targets (at one point in time) instead of cumulative-emission objectives. Policymakers should be informed of abatement potentials and costs, and MAC curves provide this information. But policymakers also need to be informed on the duration of the implementation process of these activities. MAC curves would be more operational if they had a third dimension, namely our cost in time for each abatement option. 5 Optimal abatement pathways with emission targets in 2050 Commitments in terms of cumulative emissions are dicult to introduce and enforce. Indeed, with cumulative emissions, there is an incentive for decision-makers to delay investments and eorts beyond their mandate. Alternatives policies include the denition of emission targets at one or several points in time. They can be enforced with tradable emissions permits, as the ETS system in Europe. 7 Its annual abatement potential is 1.5 Gt/yr and takes 30 years to implement in full (see Tab. 1); adding the cumulated potential during the take o phase (30 yr × 1.5 Gt/yr)/2 and the potential when annual abatements have reached their ceiling 21 yr×1.5 Gt/yr gives a total of 54 Gt. 12 When these abatement commitments are well designed, the gain in realism and enforceability could compensate for the loss of when-exibility. In the next two sections, we assume that commitments are made in terms of abatement levels, at dierent points in time. We thus exclude the cumulative-emissions constraint (Eq. 8) from our model, and make use of obj the emission constraint with m ∈ {1}, t1 = 2050 and varying E1 (Eq. 9). In absence of inertia  i.e., with an innite α in Eq. 6  the optimal response to an emission objective would be to remain in the baseline emissions pathway from 2001 to 2049, and to implement abatement options in 2050 only. 8 With inertia  i.e., with a nite α in Eq. 6  the optimal mitigation strategy depends on the emission target. 5.1 Implementing expensive options before cheap ones Figure 6: Optimal abatement pathways to achieve, in 2050, ambitious abatements (3.25 GtCO2 /yr). The expensive option with large abatement potential is implemented before the cheaper option. Figure 6 shows the optimal abatement pathway for achieving an ambitious reduction in emissions in 2050, namely a reduction by 75%. It leads to a cumulative emissions of about 175 GtCO2 , which makes this simulation comparable with the one proposed in Section 4.1. In this case, the optimal strategy implements the expensive deep before implementing cheap. Indeed, the emission objective translates into abatements by 4.75 GtCO2 /yr in 2050, which cannot be achieved by implementing cheap alone. The cheapest way to achieve this objective in 2050 is to use cheap to abate as much GHG emissions as possible, i.e. 1.5 GtCO2 /yr. Because cheap cannot 2 penetrate faster than 50 MtCO2 /yr , it has to enter in 2021. Then 3.25 GtCO2 /yr remain to be abated by deep in 2050. To do so, deep has to enter as soon as 2006, 15 years before cheap. 8 One could say that this would be done by starting with the cheapest activity and continuing with the more expensive one until the emission objective is achieved. But in this context, the terms starting and continuing would not have a chronological meaning, as the abatement activities would both be implemented instantly in 2050. Instead, those words would denote the fact that the social planner, while designing the optimal strategy, would rst consider to implement cheap and then to implement deep. 13 This examples highlights the fact than when establishing a merit-order to design an optimal abatement strategy, time may play a more important role than cost. The fourfold reduction in emissions leads to cumulative emissions of 175 GtCO2 , and is thus com- parable to the simulation proposed in Section 4.1. Compared to the cumulative-emission-constrained simulation (CC), this emission-constrained simulation (EC) leads to start cheap later and deep sooner. Short-term abatements are lower  for instance, they amount to 750 MtCO2 /yr in 2020 in EC, against 1.2 GtCO2 /yr in CC  but long-term abatements are higher. The loss of when-exibility eventually raises the present cost of abatements, which amounts to 390 G$ in the CC case and to 630G$ in the EC simulation for the same nal atmospheric concentration. 9 This illustrates the fact that cumulative emission objectives, with full when-exibility, allow the social planner to nd lower-cost pathways to reach equivalent climate targets than emission objectives. A more systematic analysis is presented in Fig. 7. It gives the optimal entry dates of both activities (cheap and deep ), as a function of the 2050 emission target. It shows that when the emission target is lower than 2 GtCO2 /yr (i.e. when the abatement objective is higher than 3 GtCO2 /yr), the optimal strategy starts to implement the expensive activity before the cheap one. Figure 7: Entry date of each activity as a function of emission objective for 2050. For ambitious emission targets (below 2 GtCO2 /yr), the expensive option with large abatement potential is implemented before the cheaper option. The fact that, with emission targets, expensive options may have to be implemented before cheap ones is problematic. Considering the diculty for the government in creating a credible long-term signal for the price of carbon  and in government ability to commit in general (Kydland and Prescott, 1977; Dixit and Lambertini, 2003), economic actors cannot rely on long-term prices. If actors consider only the current carbon price, then a carbon price of 60 $/(tCO2 ) would be necessary to trigger the entry of deep (see Tab.1). Fig. 6 shows that this activity should be implemented as early as 2005 to reach the stringent objective (emissions of 750 MtCO2 /yr in 2050) at the lowest 9 Note that 390 G$ is the lowest possible cost to reach the carbon budget constraint, while 630G$ is the lowest cost for reaching the same carbon budget through one aggregated emission target in 2050. 14 possible cost. But this high carbon price would also trigger the implementation of cheap (because its marginal cost, 30 $/(tCO2 ), is lower than the signal) in 2005, i.e. too soon, leading to a suboptimal abatement pathway. In this case, achieving the optimal pathway through dierentiated carbon prices across sectors makes sense. 10 This nding is even stronger when taking into account a shorter-term target, such as the EU target to abate by 20 or 30% in 2020. Short-term targets are a priori relevant because there is visibility over the short term on technology availability, macroeconomics trends and institutional frameworks. But focusing on aggregated short-term objectives has also its drawbacks, as the next section will show. 5.2 The inuence of long-term objectives on short-term strategies Here, we compare two simulations with the same short-term aggregated target, but dierent long- term targets. The rst simulation, labeled SO (Short-term Only), has a short-term constraint for 2020, but no long-term constraint: E(2020) = 4.25 GtCO2 /yr (10) The second simulation, labeled SL (Short-term and Long-term objectives), provides the optimal abatement strategy to reach the same short-term target for 2020, but with an additional long-term constraint, namely a fourfold reduction in GHG emissions in 2050. 11 In this simulation, there are thus two emission milestones (see Eq. 9) with the following emissions objectives: E(2050) = 1.25 GtCO2 /yr (11) E(2020) = 4.25 GtCO2 /yr (12) Our objective is to assess the dierence  over the short-term  between a strategy aiming at a short-term target and a strategy aiming at both short-term and long-term targets. Our purpose here is to explore the impact of long-term emission objectives on the short-term strategy. Figure 8 compares the optimal abatement strategies from 2000 to 2020 in the two cases. With both the 2020 and the 2050 objective (simulation SL, panel a.), the social planner starts implementing deep in 2006, and does not implement cheap before 2020 (as in Section 5.1). In contrast, when the 2050 milestone is disregarded (simulation SO, panel b.), the climate objective seems easier to 10 More technically, this reminds us that in a dynamic framework, the equimarginal principle does not necessarily translate in equal instantaneous carbon prices across economic sectors (Gilotte, 2004; Kverndokk and Rosendahl, 2007), as it does in a static framework (del Rio Gonzalez, 2008). Our model does return equal marginal values of investment in both activities at each time step; but these marginal values cannot be interpreted directly as a static carbon shadow price at those time steps, because they arise from an intertemporal optimization and encompass information on inertia. 11 In this case, with perfect credibility of the 2050 target, the 2020 milestone does not add anything, since a simulation with only the 2050 target already leads to emissions of 4.25 GtCO2 /yr in 2020 (this simulation is thus identical to the simulation presented in Fig. 6, panel b.). 15 a. Taking into account both 2020 and 2050 objective b. Taking into account 2020 objective only Figure 8: Comparison of optimal abatement strategies to reach the same target for 2020, taking into account or disregarding the longer-term 2050 objective. With an ambitious long-term target, the short-term strategy is based on the more expensive option with higher potential, not on the cheapest option. achieve. As a result, the social planner starts abating latter (in 2010 vs 2006) and uses cheaper and lower-potential options, namely cheap and deep instead of deep only. The discounted expenditures in abatement measures amounts to 28 G$, against 112 G$ when the 2050 objective is taken into account: the optimal short-term nancial eort is much higher if the long-term target is taken into account (even though the abatment in MtCO2 is the same). If the 2050 target is not taken into account in the 2000-2020 period, then the 2050 target may appear in 2020 extremely costly, or even impossible, to achieve. In this illustrative example, the fourfold reduction in emissions becomes indeed impossible to achieve in 2050 if the long-term objective is not taken into account from the beginning, in 2000. In other words, despite aggregate emissions being abated at the same pace in SO as in SL, the SO pathway produces a lock-in in a carbon intensive pathway that cannot be reversed in the second period from 2020 to 2050. In plain language, the optimal strategy to reach the 2020 target is dierent (and more expensive) if the 2050 objective is included in the optimization. With an ambitious long- term objective, the short-term target needs to be achieved through the implementation of the options with the largest potentials and the largest inertia, not with the cheapest solutions, as the MAC curve could suggest (see Fig 3). 6 Conclusion This article investigates the use of expert-based MACCs to design abatement strategies. It shows that introducing inertia, in the form of a maximum amount of abatement that can be achieved over a given period of time, changes signicantly the message from MACCs. This cost in time, which complements the nancial cost, has a large inuence on the rank order of various strategies. In 16 particular, the dynamic aspect makes MACCs radically dierent from merit-order curves: at one point in time, the best approach is not to set an instantaneous carbon price and introduce all the abatement options with a marginal cost below it. With an objective in terms of cumulative emissions over a long period of time  a good proxy for climate change , it is preferable to start by implementing the cheapest strategies, but it makes sense to implement the more expensive ones at the same time, or at least before all the potential of the cheapest options has been exhausted. Reaching ambitious objectives requires the implementation of abatement options that are slow to act, such as urban planning. This means that the implementation date is also dened by the time constraint, not only by the cost of options. We also tested objectives expressed in terms of aggregated abatements at one point in time, closer to the actual practices, e.g., the -75% in 2050 of the United Kingdom. In that case, the order can even be reversed, it can be preferable to start with the implementation of the most expensive options, if their potential is higher and their inertia is large. This optimal schedule cannot be enforced with a carbon price in a world without perfect foresight and perfect credibility of the long-term carbon price signal. Moreover, we show that short-term emission target can dangerously mask longer-term targets. In the European Union, the best way of reducing emissions by 20% by 2020 is dierent, depending on whether this is the ultimate objective or only one milestone in a trajectory toward a 75% reduction in 2050. With an ambitious long-term objective, the short term target needs to be achieved through the implementation of the options with the largest potentials and the largest inertia, not with the cheapest solutions. These results conrm the need to account for inertia in various economic sectors in the design of climate policies. Transforming climate objectives into emissions pathways cannot be done with ag- gregated models if perfect foresight and long-term policy credibility are not assumed. Without these assumptions, emissions pathway need to be multi-sectoral, distinguishing in particular heterogeneous capital turnovers (Lecocq et al, 1998; Jaccard and Rivers, 2007), and promising activities subject to learning by doing (del Rio Gonzalez, 2008). Policymakers could for instance use the information provided by MAC curves and additional information on inertia and learning-by-doing potential to identify a decarbonization path in each sector (or activity). In that case, emissions target can be made credible and enforceable, and mimic the abatement pathways obtained in models with more appropriate targets that allow for when-exibility. There is a balance to maintain, however: emission objectives (or sectoral policies) should be targeted enough to distinguish dierences in inertia, but broad enough to let economic agents select the best options and technologies to reach them. Because of information asymmetry and the risk from rent-seeking behavior, micro-managing mitigation by dening too targeted objectives can be counter-productive (Laont, 1999). Also, objectives need to be updated when new information is available; for instance if one activity turns out to be less promising than expected. In a nutshell, 17 the denition of mitigation policies should take into account recent propositions on industrial policy (Rodrik, 2008). In the European Union, the current mitigation objective has been set for 2020 only. It could therefore lead to economic actors to focus on cheap and rapid solutions, and to neglect high-potential but high-inertia options which will however be required to meet an ambitious objective in 2050. One solution could be to count on the 2020 target (and the corresponding carbon price created by the EU ETS system) to take care of the fast-to-implement solutions, and to add policies targeting other solutions with high inertia and potential. Examples are the existing objective of 20% of renewable energies in 2020, the fuel economy standards in the auto industry, and proposed changes in land-use planning, building norms and infrastructure design. None of those policies would necessarily imply equal carbon prices across sectors at one point in time. Acknowledgments The authors wish to thank Patrice Dumas, Sebastien Duquesnoy, Etienne Espagne, Céline Guivarch, Marianne Fay, Baptiste Perrissin-Fabert, Philippe Quirion, Julie Rozenberg, and Vincent Viguié for useful comments and remarks. The remaining errors are entirely the authors'. The views expressed in this paper are the sole responsibility of the authors. They do not necessarily reect the views of the World Bank, its executive directors, or the countries they represent. References Allen MR, Frame DJ, Huntingford C, Jones CD, Lowe JA, Meinshausen M, Meinshausen N (2009) Warming caused by cumulative carbon emissions towards the trillionth tonne. Nature 458(7242):11631166 Ambrosi P, Hourcade J, Hallegatte S, Lecocq F, Dumas P, Ha Duong M (2003) Optimal control mod- els and elicitation of attitudes towards climate damages. Environmental Modeling and Assessment 8(3):133147 An F, Gordon D, He H, Kodjak D, Rutherford D (2007) Passenger vehicle greenhouse gas and fuel economy standards: A global update. Tech. rep., The international council on clean transportation, Washington DC Bréchet T, Jouvet P (2009) Why environmental management may yield no-regret pollution abate- ment options. Ecological Economics 68(6):17701777 Brook A, Kendrick D, Meeraus A (1988) GAMS, a user's guide. SIGNUM Newsl 23(3-4):1011, ACM ID: 58863 18 Davis SJ, Caldeira K, Matthews HD (2010) Future CO2 emissions and climate change from existing energy infrastructure. Science 329(5997):1330 1333 Dixit A, Lambertini L (2003) Interactions of commitment and discretion in monetary and scal policies. The American Economic Review 93(5):pp. 15221542 Ekins P, Kesicki F, Smith AZ (2011) Marginal abatement cost curves: A call for caution. Tech. rep., UCL Energy Institute Gilotte L (2004) Incertitude, inertie et choix optimal. PhD thesis, Cired Goulder LH, Mathai K (2000) Optimal CO2 abatement in the presence of induced technological change. Journal of Environmental Economics and Management 39(1):138 Greening LA, Greene DL, Diglio C (2000) Energy eciency and consumption  the rebound eect  a survey. Energy Policy 28(6-7):389401 Grubb M, Chapuis T, Duong MH (1995) The economics of changing course : Implications of adapt- ability and inertia for optimal climate policy. Energy Policy 23(4-5):pp. 417431 Guivarch C, Hallegatte S (2011) Existing infrastructure and the 2C target. Climatic Change Letters accepted Gusdorf F, Hallegatte S, Lahellec A (2008) Time and space matter: How urban transitions create inequality. Global Environmental Change 18(4):708719 Ha-Duong M, Grubb M, Hourcade J (1997) Inuence of socioeconomic inertia and uncertainty on optimal CO2-emission abatement. Nature 390(6657):270273 Hammitt JK, Lempert RJ, Schlesinger ME (1992) A sequential-decision strategy for abating climate change. Nature 357(6376):315318 Jaccard M, Rivers N (2007) Heterogeneous capital stocks and the optimal timing for CO2 abatement. Resource and Energy Economics 29(1):116 Jackson T (1991) Least-cost greenhouse planning supply curves for global warming abatement. En- ergy Policy 19(1):3546 Jae AB, Newell RG, Stavins RN (2005) A tale of two market failures: Technology and environmental policy. Ecological Economics 54(2-3):164174 Johnson TM, Alatorre C, Romo Z, Liu F (2009) Low-Carbon development for mexico. Low-Carbon Development for Mexico 1:1156 Kesicki F (2010) Marginal abatement cost curves for policy making: Expert-Based vs. Model-Derived curves. In: IAEE's 2010 International Conference. Rio de Janeiro 19 Klepper G, Peterson S (2006) Marginal abatement cost curves in general equilibrium: The inuence of world energy prices. Resource and Energy Economics 28(1):123 Kverndokk S, Rosendahl KE (2007) Climate policies and learning by doing: Impacts and timing of technology subsidies. Resource and Energy Economics 29(1):5882 Kydland FE, Prescott EC (1977) Rules rather than discretion: The inconsistency of optimal plans. Journal of Political Economy 85(3):pp. 473492 Kydland FE, Prescott EC (1982) Time to build and aggregate uctuations. Econometrica 50(6):1345 1370 Laont JJ (1999) Political economy, information, and incentives. European Economic Review 43:649 669 Lecocq F, Hourcade J, Ha Duong M (1998) Decision making under uncertainty and inertia con- straints: sectoral implications of the when exibility. Energy Economics 20(5-6):539555 Manne A, Richels R (2004) The impact of learning-by-doing on the timing and costs of CO2 abate- ment. Energy Economics 26(4):603619 Matthews HD, Gillett NP, Stott PA, Zickfeld K (2009) The proportionality of global warming to cumulative carbon emissions. Nature 459(7248):829832 McKinsey, Company (2007) Reducing US greenhouse gas emissions: How much at what cost? Tech. rep., McKinsey & Co Meinshausen M, Meinshausen N, Hare W, Raper SCB, Frieler K, Knutti R, Frame DJ, Allen MR (2009) Greenhouse-gas emission targets for limiting global warming to 2C. Nature 458(7242):1158 1162 Motherway B, Walker N (2009) Ireland's Low-Carbon opportunity. Tech. rep., Sustainable Energy Ireland Nordhaus WD (1992) An optimal transition path for controlling greenhouse gases. Science 258(5086):13151319 Ongsakul W (1999) Real-time economic dispatch using merit order loading for linear decreasing and staircase incremental cost functions. Electric Power Systems Research 51(3):167173 Poswiata J, Bogdan W (2009) Assessment of greenhouse gas emissions abatement potential in poland by 2030. Tech. rep., McKinsey & Co, Warsaw del Rio Gonzalez P (2008) Policy implications of potential conicts between short-term and long-term eciency in CO2 emissions abatement. Ecological Economics 65(2):292303 20 Rodrik D (2004) Industrial policy for the Twenty-First century. CEPR Discussion Papers 4767, C.E.P.R. Discussion Papers Rodrik D (2008) Normalizing industrial policy. Commission for Growth and Development. Working Paper 3, World Bank Sue Wing I (2006) Representing induced technological change in models for climate policy analysis. Energy Economics 28(5-6):539562 Sweeney J, Weyant J (2008) Analysis of measures to meet the requirements of california's assembly bill 32. Tech. rep., Stanford University Wigley TML, Richels R, Edmonds JA (1996) Economic and environmental choices in the stabilization of atmospheric CO2 concentrations. Nature 379(6562):240243 World Bank (2010) Doing Business 2011. World Bank, Washington DC 21