Michael Grubb
Energy and Environmental Programme, Royal Institute of International
Affairs,
London, England
Jae Edmonds
Technical Leader, Economic Programs, the Pacific Northwest Laboratory,
Washington, DC, USA
Patrick ten Brink and Michael Morrison
Caminus Energy Limited, Cambridge, England
KEY WORDS:energy, greenhouse, climate change, greenhouse emissions reduction
cost
TABLE OF CONTENTS
MODELLING AND COSTING DEFINITIONS AND PARADIGMS
ABATEMENT TECHNOLOGIES AND TECHNOLOGY COST CURVES
A REVIEW OF SYSTEM-WIDE ABATEMENT COST ESTIMATES
KEY DETERMINANTS: TECHNOLOGICAL VS ECONOMIC PERSPECTIVE
KEY DETERMINANTS: ABATEMENT STRATEGIES AND SCOPE OF ANALYSIS
In the late 1980s, interest nourished in the issue of global climate change. Many studies focussed on the options for limiting anthropogenic emissions of greenhouse- related gases and managing the consequences of global warming and climate change. Making appropriate policy choices requires information on both the costs and benefits, as they occur over time, of policy interventions, and an increasing number of studies have sought to quantify the costs especially of limiting CO2 emissions, as the dominant anthropogenic source. Such analyses now form an important part of overall policy assessments and influence international negotiations on policy responses. However, these studies are not well understood. In this paper we seek to analyze the literature on the costs of CO2 abatement.
The majority of work in estimating the costs of reducing greenhouse gas emissions has occurred since 1988, but interest in the issue of costing emissions reductions began more than a decade earlier with the work of Nordhaus (1, 2). Nordhaus's early work focussed on the issue of reducing fossil- fuel CO2 emissions, as did that of Edmonds & Reilly (3, 4), Kosobud et al (5), Seidel & Keyes (6), Rose et al (7), Lovins et al (8), Williams et al (9), Manne ( 10), Perry et al (11), Nordhaus & Yohe ( 12), and Mintzer (13), among others. Only Seidel & Keyes, Perry et al, and Mintzer examined non- CO2 emissions, and these studies treated them separately and in an ad hoc manner; none of the studies took land- use change into account explicitly.
While not the primary focus of their analysis, some of the studies conducted prior to 1988 analyzed the cost of emissions reductions. The results of these studies foreshadow the current debate. Edmonds & Reilly (14) noted in their 1985 literature assessment:
The economic costs of CO2 abatement policies have only been partially analyzed at this time. Edmonds and Reilly, Kosobud et al, and Nordhaus, each using a different model, indicate that the reduction in aggregate GNP associated with even stringent punitive strategies is not large, usually only a few percentage points. Lovins et al argued that the costs might actually be negative.
The purpose of this paper is fourfold. First, we seek to give a broad and
accessible guide to the main studies reported over the past five years.3
Second, we seek to clarify the issues involved in estimating abatement costs
through a systematic study and classification of the relevant concepts. Third,
through critical analysis of reported results, we suggest ranges of plausible
estimates. Finally, we highlight the most important areas of uncertainty or
confusion and suggest areas on which future research needs to concentrate.
To this end, we start (Section 2) by noting differing uses of the term "costs"
and the way in which scope and definition of analysis affects results.
2Note that the list of relevant emitted gases differs importantly
from the list
of greenhouse gases. The list of relevant greenhouse gases, that is, those
gases that are effectively transparent to incoming sunlight but that absorb
in
the infrared spectrum, includes CO2, CH4, N2O,
O3, H2O, CFCs, and CFC
substitutes. Greenhouse-
related
emitted gases are linked to greenhouse gases through natural processes such
as
atmospheric chemistry and albedo.
1In finalizing this review, we have sought to reference the most
accessible,
relevant, and general sources, rather than obscure or superseded ones. In
panicular, the series of papers by Cline, and by Manne &Richels, have each
been brought together in books; various studies for the European Commission
have been brought together in a two-
volume
edition of the European Economy; and many of the reports by the
Organization for Economic Cooperation and Development (OECD) Economics
Department have been reproduced in a special issue of OECD Economic Sludies.
All these volumes were published dunng 1992, and to the extent possible
we
reference the books rather than the many separate research papers.
We clarify the way in which we use the term in this paper so that results are
to the extent possible comparable.
Sections 3 and 4 then review abatement cost estimates. Section 3 summarizes
estimates derived directly from studies of the technologies available for
limiting emissions, and ways of interpreting them. Section 4 summarizes the
results of studies that have sought to model the impact of CO2 abatement
on whole energy systems.
Sections 5 and 6 then explore the modelling and assumption differences that
affect cost estimates. Section 5 explains and classifies the different kinds
of
models that have been applied, and Section 6 reviews the impact of variations
in key numerical parameters. Sections 2-6 draw heavily on the review of literature
performed for Phase I of the United Nations Environment Programme (UNEP) Greenhouse
Gas Abatement Costing Studies (16).
The paper then draws together the material in sections 2-6, to examine critically
the nature and relative importance of these various sources of cost difference,
and the implications that follow from this. Section 7 analyzes the economic
and engineering perspectives, the differences between which are a major source
of cost differences; the discussion includes the role
of energy-efficiency and of low-carbon supply technologies, as well as resolution
of these perspectives. Section 8 then examines issues relating to the strategy
of abatement and scope of analysis. Finally, Section 9 draws general conclusions
from the study, and suggests some implications for future research.
MODELLING AND COSTING DEFINITIONS AND PARADIGMS
The cost of emissions reductions is always computed as a difference in a given
measure of performance between a reference scenario and a scenario that
involves lower emissions. By far the most commonly used measures of performance
are the net direct financial costs to the energy sector assessed at a specified
discount rate; and the estimated impact on gross national product (GNP), or
its
close cousin GDP. GNP is the monetary value of new final goods and services
produced in a given year, and it provides a measure of the scale of human
activities that pass through markets, plus imputed values of some nonmarket
activities. It is generally assumed that financial costs in the energy sector
can be closely related to impacts on GNP, though as noted below this is not
always the case.
Neither direct financial costs nor GNP provide direct measures of human
welfare. One factor is that human welfare does not necessarily increase
linearly with the degree of consumption; a given loss of income will likely
matter far more to poor people, or poor countries, than to richer ones, for
example. Some studies attempt to capture this through "equivalent welfare"
measures, but these still rely centrally on a marketed-products
basis. A broader limitation is revealed by the fact that there are many
examples where GNP moves in the opposite direction to human well-being.
For example, a disease that increases the sale of medicine may boost GNP but
make individuals worse off; environmental disasters can stimulate economic
activity, but the environment (and human enjoyment of it) is diminished.
This reflects the fact that GNP does not incorporate many nonmarket factors
that affect welfare. Some studies have sought to examine explicitly the impact
of abatement on various external costs, and concluded that these can be very
significant (Section 8.4). However, in general, studies focus on financial
costs or GNP impact. In the broader literature, other welfare indices have been
attempted (such as the United Nations Development Programme's (UNDP) Human Development
Index), but data are rarely adequate to quantify impacts in such terms in abatement-costing
studies. At present, for quantifying results there is little practical alternative
to working with monetary cost and GNP impacts, but the caveats about these as
measures of welfare impacts need to be borne in mind.
Nor is GNP necessarily a good measure of consumption. For example, some forms
of carbon taxation can move resources from consumption to investment, which
can boost GNP but for many years may lower consumption. It is unclear whether
welfare has improved or declined. Alternatively, tax revenues might be returned
to households, which could raise household consumption but depress long-term
GNP.
This also raises the issue of comparing costs in different periods. Results
concerning abatement costs are sensitive to the assumed discount rate. This
is
particularly important with respect to evaluating the importance of the
potential impacts from climate change, where the appropriate discount rate is
both crucially important (because of the long timescales) and very uncertain
(because of the timescales and because it is an attempt to make an explicit
valuation of long-term public welfare); for a discussion see Cline (17). For
assessing abatement
costs, the timescales are less and the discount rate has to be related to the
actual rates revealed or set by government for the sector in which the
abatement investments are being made, so this is a less central (though still
significant) issue. In this study we simply report results as estimated by the
studies concerned, given the discount rates they assume (which, for the major
energy investments considered in this study, are typically about 5-8% real discount
rate).
Almost since the beginning of costing studies, a clear division has existed
between those that fundamentally use an economic approach, which relies on
observed market behavior and which generally assume that markets operate equally
efficiently in the reference and abatement case; and those that
use a technology-engineering approach, which emphasizes a technically optimal
abatement scenario (which may be contrasted with a reference case that is by
implication not optimal). The choice of "cost paradigm" in this sense is a fundamental
determinant of results--including often the sign of abatement costs--and these
differences form an important theme of this paper.
Economic studies use "top-down" models, which analyze aggregated behavior based
on economic indices of prices and elasticities, and focus implicitly or explicitly
on the use of carbon taxes to limit emissions. These studies have mostly concluded
that relatively large carbon taxes (e.g. that could much more than double the
minemouth cost of coal) would be required to achieve goals such as the stabilization
of fossil-fuel carbon emissions .
Technology-oriented studies use "bottom-up"
engineering models, which focus on the integration of technology cost and
performance data. Many such studies have concluded conversely that emission
reductions could be achieved with net cost savings.
The division between the "economic paradigm" and the "engineering paradigm"
is closely related--but not identical--to the division between "top-down" and
"bottomup" models, as it has emerged in the literature. These differences,
as a major source of differing estimates, form a strong theme in this paper:
the formal modelling differences are clarified in Section 5.1, and the
underlying "paradigmatic" issues are explored in Section 7.
There are also many modelling differences within each category. Most notably,
since 1990 an important general division has become evident between the
application of top-down models that have been developed for long-run
"equilibrium" analysis of energy and abatement costs (reflecting an idealized
economy with optimal allocation of resources), and conventional macro-
economic models designed for shorter-run analysis of the dynamic responses of
economies (which reflect many existing imperfections). Longrun equilibrium models
generally estimate the costs of reducing emissions to be positive and high by
the standards of most environmental measures implemented to date. Macro-economic
models indicate a far more complex pattern of responses and cost indicators,
which may move in different directions and vary over time. The distinctions
are explored further in Section 5.2.
Rather than initiate our survey with a detailed analysis of models, however,
we start by summarizing the results that have been presented, with a review
and interpretation of technology cost curves (Section 3) and then a summary
of general results from system modelling studies (Section 4).
ABATEMENT TECHNOLOGIES AND TECHNOLOGY COST CURVES
Technology Cost-Curve Results
Clearly a major determinant of CO2 abatement costs will be the costs
and adequacy of technologies that can reduce emissions. Many studies of the
technologies that could help to limit greenhouse gas emissions have been conducted;
major reviews are given in IPCC (18), Fulkerson et al (19), IEA/OECD (20), Grubb
(21), and Goldemberg et al (22). In addition, many international databases with
information on energy technologies have now been established; the UNEP study
(16, Appendix I) lists no less than 13 technology databases now available.
Technology cost curves provide a useful way of summarizing the technical
potential for limiting emissions as identified in such studies. The simplest
approach is to stack up different technologies in order of the cost of emission
reductions or energy displaced, though cost curves can be used to represent
the
output for almost any degree of sophistication in modelling. There are various
ways of generating cost curves of successively greater sophistication and
consistency, as discussed in UNEP (16).
Discrete technology cost curves for various developed countries are presented
by Lovins & Lovins (23), Mills et al (24), Jackson (25), and Krause et al
(26). An EPRI (27) study examined potential savings in the US electricity
sector and concluded that "if by the year 2000 the entire stock of electrical
end-use stock were to be replaced with the most efficient end-use
technologies (nearly all of them estimated to be cheaper than equivalent
supply), the maximum savings could range from 24% to 44% of electricity
supply." Lovins & Lovins (23, 27) suggest much higher potential savings
still. The cost curves associated with these two analyses are shown in Figure
1, which demonstrates that a good database does not necessarily ensure
comparable results; the potential estimated by EPRI (the upper curve) is
clearly very much smaller than that estimated by Lovins (the lower curve).
Compared against typical US electricity prices of at least 6-7 cents per kWh,
however, both illustrate that substantial emission reductions appear to be
available at net cost savings.
Technology cost curves have by no means been confined to developed countries.
A number are presented in studies of the Asian Energy Institute network summarized
in (16); Figure
2 shows a discrete technology cost curve estimated for CO2 savings
available by 2000 in Brazil. Figure 3 shows a continuous
version of a technology cost curve for Poland.
A consistent set of national abatement cost curves, derived by aggregating
technology studies, is presented by COHERENCE for the Commission of
the European Communities (29). Figure
4 summarizes the results, which indicate considerabic variation between
counilics and highlight the dependence on the baseline projections; the technologies
identified offer stabilization of emissions relative to a base year (19RS) at
little or no cost for the northern European countries, in which basclinc emissions
growth is expected to be slow, but at much greater cost for the southern European
countries, which expect rapid growth in baseline emissions.
Finally we note that the scope and basis of cost curves can vary greatly. The
Tata Energy Research Institute has developed a detailed appraisal that for India
focusses on investment requirements, rather than net cost/savings.
Most curves have been confined to the energy sector. or parts thereof, but
others (such as the Indian study) include reforestation and some addresse
other gases as well. Nordhaus (30) presents an approximate composite curve that
seeks to include chlorofluorocarbons (CFCs) as well in a global greenhouse-gas
abatement curve. The Nordhaus analysis also reinforces the fact that cost curves
are a way of presenting data, not of generating results; it seeks to summarize
a wide range of results from economic models, and has no "negative cost" section.
Limitations and Interpretation
Abatement cost curves reflect the weaknesses and strenghts of the
procedures used to produce them. The simplest technology curves usefully
summarize technical data, but may have substantial limitations as
guides to actual abatement costs. In part this is because, unless they are developed
iteratively using quite sophisticated system models, they may
neglect interactions among abatement options and thus "double count"
some emissions savings--the CO2 savings from reducing electricity
demand may for example be much reduced if nonfossil sources are introduced later
to displace coal power generation. They may also neglect interactions between
various end-uses, for example that between heat and lighting in widely diverse
building enviromnents. Frequently also they do not reflect adequately the timescales
involved in bringing the technologies into place and the underlying growth in
demand that may occur in the interim.
Even after such issues are carefully incorporated, technology assessments
and cost curves (particularly for end-use
technologies) still demonstrate a large potential for emission savings
apparently at "negative cost"--technologies that would both reduce emissions
and yield net financial savings. Typically, these suggest a potential to reduce
emissions by well over 20% at net cost savings. the uncertainties, however,
are very large, and, to be meaningful, numbers have to be defined very
carefully in terms of scope and timescale. Based on an extensive review of
technologies and related cost-curve studies, Grubb (21, Chapter 2) concludes
that "substituting identified and cost-effective technologies in OECD countries
could in principic increase the efficiency of electricity use by up to 50%,
and of other applications by 15-40%, over the next two decades. Fully optimizing
energy systems would yield larger savings, but it is far from clear how much
of this potential can be tapped."
Thus, technology studies and cost curves show that a large
"energy-efficiency gap" exists between the apparent technical potential for
cost-effective improvements, and what is currently taken up in energy
markets. In well-functioning markets, cost-effective options should be exploited
anyway, since someone should profit by doing so.
Some of the cost-effective potential may be taken up over time. But if such
technologies are not being exploited, this may indicate that other important
factors are not captured in technology analysis. For example, there may be hidden
costs, or people may be unaware of the options, or there may be other obstacles
to uptake. The acceptability of different options may also vary, for least cost
is by no means the only criterion that matters to people. This
illustrates the fact that the apparent technical potential in fact comprises
a
number of different components. As illustrated in Figure
5, realizable gains consist of:
1. those that are economically attractive in their own right and that will
be installed without policy changes;
2. those that would not be obtained unless institutional constraints
and barriers are removed, and/or other micro-economic policies are implemented
to increase the take-up of cost-effcctive options;
3. those that are justified on the basis of nongreenhouse external benefits
(e.g. reduced other environmental impacts, increased energy security).
In addition, some apparently cost-effective savings cannot be realized. The
real economic potential differs from apparent potential due to:
1. "take back" or ''robound'' Or savings (improved) efficiency
reduces the cost of the associated energy service and therefore stimulates
increased demand for the energy service);
2. unavoidable hidden costs (there may be costs associated with the use of
a technology or policies to stimulate its uptake that are not revealed in a
simplified analysis);
3. consumer preference (the technologies may not be a perfect substitute in
the provision of the energy service, which may either increase or decrease
consumer readiness to take up more efficient technologies, depending on their
characteristics). For example, it would be technically possible and highly "cost-effective"
to mandate a doubling of car efficiency in most countries; but it would probably
make the vehicles available smaller and less powerful, which people may consider
unacceptable (often this can also be considered as an aspect of hidden costs).
These different components need to be understood before drawing conclusions
about the scope for "cost-free" reductions. We underline that such cost curves
identify a technical potential but there is no expectation that all of this
can be realized. Part of the key is to consider specific policies for introducing
hotter technologies. set against explicit baseline scenarios that incorporate
some "business as usual" uptake. Johansson et al (31) present a cost curve
(described partly in terms of particular policy changes to introduce more efficient
energy technologies in the United States, drawing in part on much more detailed
implementation studies. Drawing on such studies, estimates of the practical
potential based on such cost curves are
discussed in Section 7.3 below.
To conclude, cost curves can be extremely useful and flexible ways of
displaying data, and those that summarize technical potentials consistently
indicate a large potential for technologies that can reduce CO2 emissions
with net cost reductions. But they can only reflect the strengths and weaknesses
of the models used to generate the data, and the extent to which these models
reflect interdependencies, hidden costs, and issues of implementation and timescales.
What matters is not the cost curve itself, but the underlying methods and models
used to produce results. In this context, modelling studies that seek to encompass
whole systems and economies have attracted particular attention, and it is to
these that we now turn.
A REVIEW OF SYSTEM-WIDE ABATEMENT COST ESTIMATES
A wide variety of abatement costing studies at the global and national
levels have been carried out. !4 In this section,
we summarize the results from cost studies first at the global level, and then
from various national cost
estimates across a range of modelling methods and countries, focussing first
on the extensive range of US studies, then other OECD studies, and finally studies
of non-OECD regions. The subsequent sections examine the reasons for differing
cost results.
Global Studies
In this section we summarize results from six major models used to estimate
the global costs of limiting CO2 emissions, listed in Table
1. As summarized immediately below (modelling terms and classification are
discussed in Section 5), four of these use fundamentally different global energy/
economic models, one uses an economic growth modelling framework oriented towards
technology development, and one is a global bottom-up study. Other global models
have been developed, notably the models of Rutherford (CRTM) and Peck & Teisberg
(CETA). the IEA (32), McKibben &
Wilcoxen (33), and the ICF "global macro-energy model" used as part of the "atmospheric
stabilization framework (ASF)" for US Environmental Protection Agency (EPA)
studies (34). These and others (35) are not included here because they are either
based on models already covered or do not generate results in a relevant comparabic
form.5
The six major global models/studies reviewed here are:
1. Whalley & Wigle (36, 37) use a comparative static, economic
general equilibrium model, incorporating trade but with only two fuel types
(carbon and noncarbon) and no representation or backstop or other
technologies. Their analysis focused on trade and the implications of
different ways of applying taxes and distributing emission constraints.
2. The Global 2100 model of Manne & Richels (38-40)
is a top-down general equilibrium model with a small selection of supply-side
energy technologies, including carbon-free "backstop" technologies, which are
available in unlimited quantities when the price becomes high enough. The model
is fully optimizing within each of five regions, but there is limited trade
between regions, so the result is not a global least-cost abatement. (The constraints
used involved relatively high losses for the Soviet Union and China in particular.)
Derivatives of this model include Rutherford's Carbon Rights Trade Model (CRTM)
(41), which relaxes the no-trade constraint, and the Carbon Emissions Trajectory
Assessment (CETA) (42),
which aggregates the regions and incorporates a climate damage function.
3. The Edmunds-Reilly-Barns model (ERB) (3, 4) is an energy-greenhouse
gas simulation model with detailed representation of energy supply technologies,
including cost curves, with energy trade between each of nine regions. The model
has been widely distributed and used by different authors. The model contains
a highly simplified macro-economic
linkage intended to reflect feedback effects, not for GNP evaluation; emissions
reduction costs are much better inferred from the incremental costs incurred,
which is the measure reported in this paper (see Section 5.3).
4. The OECD's GREEN model (43, 44) is a 12-region
general equilibrium model, which in its more recent versions is a multiperiod
model with capital stock modelling that encompasses both trade and backstop
technologies
5. Anderson & Bird (45) employ a simple economic growth model to illustrate
bounds on the economic impact of abatement strategies that are based on expansion
of renewable energy. The key feature of their analysis is inclusion of a relationship
between investment and cost reduction in alternative supply technologies.
6. The Goldemberg et al (46) study Energy for a Sustainable
World is the only bottom-up global study. While not an abatement costing
study as such, its detailed disaggregation of global energy use and available
technologies concludes that global energy demand could increase by only 10%
from current levels by the year 2020 through full exploitation of cost-effective
technologies for improving energy-efficiency.
In Figure
6a we summarize the major published results from these modelling
studies, plotting the degree of abatement against the cost measured directly
(for energy sector models) or as GNPloss, relative to the projected GNP. Figure
6a shows the results in terms of reductions from the baseline projection generated
by the model; simple calculation shows that the average rate of abtement in
almost all these studies, despite their apparent diversity, is 1.3-2.0% per
year below the baseline projection. Figure
6b illustrates the same results, but in terms of the level relative to the
base year (usually 1990). Contrasting the two curves shows that emission changes
from a base year show less of a pattern, due primarily to the wide variation
in the baseline emission projections, discussed in Sections 6 and 7.2. Because
of such variation, in this study we concentrate primarily on reductions relative
to the reference projection without CO2 constraints (the baseline).
Thus, in the presentation of results from studies we seek to abstract from the
scale of the economy by normalizing results to both the reference GNP and the
reference fossil-fuel CO2 emission-trajectory. To the extent that
system scale is sharply a linear multiplier of results, this makes results comparable
relative to the baseline. The essential linearity of costs with scale has been
shown to characterize thc ERB model (81). The scale (e.g. of baseline emissions
growth), however, has powerful implications for the ease or diffictilty of achieving
a specific target emission
reduction.
Additional information on these global studies is provided by Figure
7, which shows the relationship between the relative CO2 reduction
and required "carbon tax" (marginal abatement cost as reported by the model
in the target year), and by Figure
7b, which shows the relationship between the reported carbon tax and the
average GNP loss in the target year. Interestingly, whereas for a fixed analysis
the marginal cost must always be greater than the average cost, this plot shows
no such relationship. For some models this may reflect anomalies in the way
that marginal cost impacts are translated into GNP impacts (see Section 5.3);
another source of such behavior is that in some models the CO2 constraints
are introduced so as to impose a much higher
marginal cost over the first few decades than in the very long run (see Section
8.3). All these global studies impose a fixed path of emission constraint;
only the simpler CETA(42) and DICE (35) models optimize the path of abatement
to reach a given concentration level.
Because of such anomalies, there is no uniquely preferable single measure
for cost comparisons. We have to draw on the models and results as generated
to date, and for all subsequent analysis in this paper we choose the target
date GNP loss (or total additional energy sector cost) relative to projected
baseline GNP as the most appropriate single cost indicator.
Even when normalized relative to the differing baseline projections,
however, the results still show great variation. The Whalley-Wigle
results define the high end of the cost spectrum. The relatively high costs
probably reflect the limitations imposed by only having one generic carbon
fuel, the lack of technology representation, and an extraordinarily high
baseline projection, which scales up the global energy system 10-fold
over the century, because there is no allowance for autonomous efficiency
improvements (see Sections 6 and 7).6 The Goldembarg
et al bottom-up results define a lower bound, with emissions reductions perhaps
up to 50% estimated at no cost.7 However,
some of the sensitivity runs of the global top-down models, when given more
optimistic estimates of energy-efficiency improvements and supply technology
development, can also produce very low costs (e.g. see discussions in Section
6).
Not including the Whalley-Wigle outlier for the reasons above, the resulting
spread of results roughly indicate that the costs of a long-run
50% reduction in global CO2 emissions could range from negligible
to a loss of about 3% in global GNP. Reductions of 80-90% degrees GNP by 2-6%;
at the other end of the curve, the global economic models (as well as the engineering
models) suggest that emission reductions of at least 10-15% can he obtained
at very low cost.
These global models are highly aggregated and capture technical issues,
macro-economic issues, regional differences, and trade effects with
varying degrees of detail. They inevitably sacrifice local and technological
detail in order to represent important regional differences and (in some) incorporate
trade. To start to narrow down the range requires a fuller examination, which
we present below, of the factors affecting model results.
Abatement Costing Studies for the United States
A wide range of national studies has been carried out: of these. the US
ones have received the most attention. Table 2 and
Figure 8 show the range of
estimates of the impact on GNP of CO2 emissions reductions relative
to the projected baseline emissions.
The survey of US studies shows predominantly top-down economic studies. Many
bottom-up/engineering studies have also been carried out, but most have been
confined to cost-curve or subsectoral studies. In addition to the earlier examples
cited in Section 3, important recent studies include those of the National Academy
of Sciences (NAS) (47), and the Office of Technology Assessment or the US Congress
(OTA) (48).8 These. like most engineering-based
studies maintain that major efficiency improvements (and hence CO2
reductions) can be obtained at little or no cost. These estimates have been
included in Figure 8 as indicative bottom-up
cost estimates.
As with the global studies, a wide range of cost estimates is observable.
For example, for the same loss of GNP of about 2%, CO2 emission reductions
can range from 20% (US Congressional Budget Office-CBO-78) to 80% (Manne & Richels-40)
below baseline. One major reason for such differences is that the reductions
are sought on different timescales; the CBO study seeks a 20% reduction by 2000,
while the 80% reduction in the Manne & Richels study is achieved at the end
of the next century.
Excepting the very rapid reductions imposed in the CBO studies for
the year 2000, the early Goulder (74) studies form a high-cost outlier.9 Bottom-up studies, and those that examine the recycling
of tax revenues (discussed below) have produced some very low and negative abatement
cost estimates. These studies are discussed in Sections 7 and 8 below.
Excepting these, the spread of results is pretty consistent with the
global results, with 50% emission reductions from baseline yielding losses up
to a little more than 2% of GNP. The coincidence between US and global results
may reflect partially the importance of US emissions and costs, but also the
dominance of US-based modelling approaches in global studies.
Non-US OECD Studies
Table 3 lists a number of emissions reduction studies
of non-US
OECO countries. One striking feature is that nearly all these studies have a
much shorter term focus than the global studies or most of the US studies: most
in fact are focussed either on emissions stabilization by 2000 or on the
"Toronto target" of 20% reduction by 2005.
The results are displayed in Figure 9. Immense variation
is once again
apparent, even in terms of reductions from the baseline projection. No clear
pattern emerges, other than the fact that the bottom-lip
studies again give much lower costs (see also the detailed bottom-up
comparative studies of EC member countries discussed in the previous section).
We have not included these or many other bottom-up
studies that repeat the message; nor have we included studies by Data Resources
Inc. for the US Department of Commerce (49) for different reasons.10
In general, the cost range is broader than in the global studies, perhaps
reflecting the impact of transitional costs arising from the relatively rapid
reductions required in some of these studies, captured by the short-run
models, an issue discussed further below. The detailed EC studies of the macro-economic
impacts of carbon taxation, using short-run
macro-economic models, produce a wide variety of results; these results depend
heavily on how the tax revenues are used, as discussed in Section 8.1 below
where we argue that the extremes of the short-run cost estimates (high and low)
are not useful as a guide to real CO2 abatement
costs because they reflect rather the use of a tax to ship resources from one
kind of economic activity to another. The Finnish study by Christensen (141),
with very high losses for modest reductions by 2010, is a similar outlier on
the high side, and two of the Japanese studies also yield exceptionally high
costs (for the Yamaji study (118) at least this is because the carbon tax
revenues are removed from the economy). The variations make it hard to discern
any pattern, but even with these outliers excluded, the relative costs for these
short-run, national reductions are mostly somewhat larger than for equivalent
long-run reductions in the US and global studies, with several exceeding 3%
GNP losses.
The Transitional Economies
Studies of abatement costs for the former centrally planned economies of
Eastern Europe, listed in Table 4, have also used
both top-down
and hottom-up approaches, but data are insufficient to summarize usefully as
a scatter diagram. An energy technology cost curve estimated for Poland (50,
51) has been shown above (Figure 3). Figure 10 shows
a more general curve of the cost of energy savings, based on a number of Soviet
technology studies, set against the estimated cost of supply, for the former
Soviet Union (52). All of these bottom-up studies indicate a large potential
for reducing CO2 emissions with net economic savings; in Figure 10,
the marginal cost of savings only rises above that of new supply for savings
well in excess of 10 EJ, which is more than 20% of Soviet primary energy demand
in 1989.
This potential arises from the history of highly subsidized energy prices
in these regions, and other cumulative inefficiencies in the structure of
incentives. Note also from Figure 10 that the economic savings potential is
around 10% greater if the Soviet Union can access Western technologies.
Unterwurzacher & Wirl (53) estimated in 1991 that increasing prices to
world market levels in Poland, Hungary, and Czechoslovakia would reduce
emissions by 30%. Of course, the realizable potential may be a very different
matter and depends in part on the progress of economic restructuring. In fact
CO2 emissions in the former East Germany have collapsed by at least
30% as uncompetitive heavy industry has more or less shut down in the process
of unification. In Poland, provisional trend/ technology results also indicate
significant CO2 reductions as a by-product
of the economic restructuring process (.54).
Manne & Kichels include the Soviet Unionas an independent region
in their Global 2100 model. and calculate much higher GNP losses [5% in
(39), reduced to 3% in (40)] there than for the rest of the world in the
first half of the next century. This striking contrast with bottom-up
studies reflects the difficulties top-down studies have with economies undergoing
restructure. They rely on the existence of a market mechanism that in many instances
is barely functioning, and such studies will likely miss many of the important
features of these economies, such as current structural inefficiencies. It may
take some time before a market -based
modelling approach bccomes appropriate. For the present, bottom-up
engineering assessments appear much more relevant.
Developing Countries including China
Some of the above issues also apply to developing countries. Although
concerns are frequently expressed about the limited data available concerning
developing countries, considerable data exist concerning the situation in mast
major developing countries especially with respect to commercial energy supply.
Data on detailed end-uses, agriculture, and noncommercial energy sectors is
sparser and less reliable, though usable estimates exist.
The past few years have also seen a number of studies of the potential for
abating greenhouse gases in developing countries. Some of these are reported
in Table 5. For example, the country reports to the
IPCC Energy and Industries
subgroup (55) did include a number of studies from developing countries, some
drawing on more extensive internal work. In addition, Sathaye (56) reported
initial results from a series of nine developing-country
studies coordinated by the US Lawrence Berkeley Laboratory. However, these
studies focussed on scenarios and did not attempt to estimate abatement
costs. Some of the detailed studies for the Asian Energy Institute network
reported in the UNEP study (16, Chapter 5) attempt bottom-up
estimates of short-term abatement costs or investment requirements.
Some of the global models do separate developing-country and oil-exporting
regions. The global studies have also highlighted the growing importance of
China in contributing to greenhouse gas emissions during the next century. The
early Manne & Richels results (39) suggested that Chinese GNP in the year 2050
could be depressed by up to 8% below the (greatly increased) reference level,
if emissions are restricted to no more than double current levels, partly because
of the apparent lack of alternatives to coal and modelling inability to import
energy [though this was reduced to a 2% GNP loss, rising to 5% by 2100, in revised
analysis (40)]. The GREEN model suggests much lower abatement costs for China
(44).
In estimates derived from global studies that model trade (e.g. 36, 37),
GNP impacts for some of the smaller developing countries especially can be
dominated by trade and price effects arising from the action of other
countries. These studies emphasize the large potential GNP loss for
energy-exporting developing countries: the loss could arise from abatement
efforts elsewhere that depress the market for traded fuels. Conversely, energy-importing
countries (which include the poorest countries) would gain from such effects.
Concerning domestic abatement efforts, a number of cost curves have been
estimated (see e.g. Figure 2 for Brazil), which indicate substantial technical
potential for savings with net economic benefits. However, the only integrated
system-wide cost estimates that the authors could find, excepting those from
global models for China, are those of Blitzer et al (57, 58) for Egypt and an
unpublished study of Zimbabwe by the UK consultants Touche Ross, reported in
UNEP (16).
The Blitzer et al studies estimate a very large potential GNP impact from
stabilizing CO2 emissions in Egypt, with losses in some cases of
more than 10% of GNP. This is based on a short-run macro-economic
model of the Egyptian economy that is modeled with very limited capita!
mobility between sectors, and with oil and gas as the only future energy supply
options. It recognizes none of the technical inefficiencies in the economy (i.e.
assumes zero scope for cost-free energy-efficiency
improvements). Excepting the modest abatement available from switching from
oil to gas, emissions savings can only be achieved by reducing energy consumption,
which, given the constraints on capital mobility, can only be achieved by reducing
economic activity or changing its structure. Consequently, the costs reported
are clearly excessive. However, separate runs of the model that placed emission
constraints on each sector of the economy individually did emphasize that GNP
losses would be greater still if these additional restrictions were imposed.
The Touche Ross study of Zimbabwe reached precisely opposite conclusions.
Using an engineering approach, widespread cost-effective options were identified
that could both limit emissions growth and improve overall economic performance.
However, these assessments neglect a variety of
hidden costs and fundamental institutional obstacles; they also include some
elements that are expected to be achieved anyway as part of current structural
adjustments in the Zimbabwean economy. These and other limitations, which suggest
that abatement costs in this case may be substantially underestimated, are summarized
in the Zimbabwean case study of the UNEP report (16).
The limited range and appropriateness of studies for the transitional and
developing economies make the use of scatter diagrams, as were used for
presenting OECD results, not in our judgment very meaningful in this case. We
do, however, note one striking observation; the gap between top-down
and bottom-up approaches is larger even than observed for OECD countries. Top-down
models mostly report restricting developing-country emissions, even relative
to projected increases, to be more expensive than equivalent relative constraints
in OECD countries [e.g. Rutherford (41)11].
It
is not clear why this should be the case. Bottom-up studies, conversely, identify
a potential for improving energy-efficiency in these regions at a net economic
benefit that is even larger than that identified for OECD countries.
Despite this, it seems possible to draw two firm observations from the
existing developing-country emissions abatement studies: many cost-effective
technology options exist for improving energy-efficiency;
but such potential will be swamped by the pressure for emissions growth in such
rapidly expanding economies, so that actually stabilizing developing-
country emissions at current levels is nevertheless likely to be very costly.
More sophisticated and quantitative system-wide analysis of abatement impacts
is, however, only just beginning, and as outlined in UNEP ( 16), the complexities
are such that it may take many years to mature towards
consensus even on very rough cost estimates and understanding of the key
issues.
The review of cost estimates in the previous two sections shows the
enormous disparity in modelling results. It is very difficult to disentangle
the various reasons for these differences, due to the nonlinearity of the
relationships involved, the diversity of the tools used to develop emissions
reductions cost estimates, the many and varied assumptions employed, and
the enormity of the task required to obtain all of the models, establish a
protocol for analysis, and systematically unravel the relative contributions.
First steps in that direction have been taken by teams at the Energy Modelling
Forum (59-62) and OECD (63, 64). In both cases, participating modellers were
asked to adopt standard assumptions to the degree possible and to provide standardized
model results. Both sets of comparisons have focussed on top-down models (the
Energy Modelling Forum recently embarked on a similar exercise with "bottom-up"
models). It is clear from these activities that great variation in results can
be generated through the use of different models. This variation is greatly
reduced through the use of standardized assumptions.
In this section we identify the different types of models used, and discuss
some of the implications that might be expected to flow from the selection of
a particular model type. The following section then discusses data differences.
Then, in Sections 7 and 8 we analyze the factors that appear to have the most
significant impact on abatement cost estimates.
We begin by noting that all models share certain unavoidable
limitations:
1. First, a model is necessarily a simplified representation of reality, in
terms of what the concerned researchers feel are important aspects that should
be captured. A given model may not capture all the important economic relationships.
2. Second, despite these simplifications, all such models are still rather
complex and must necessarily rely on a large quantity of data and numerous
parameter estimates. Robust estimation of these is itself a major research
undertaking, and serious doubts may arise about the validity of many of the
actual numerical values employed. Studies of model sensitivities, and
structured uncertainty studies in which the values of key parameters are varied
over plausible ranges, are required to examine how much these uncertainties
may affect model results. Such studies have often not been adequately performed.
3. Third, the timescales involved require assumptions to be made
about changes in technology and life-styles. Conjectures about such changes
are inevitably uncertain and cannot be formally validated.
These limitations may be exacerbated by the fact that most studies employ models
that were not initially designed to shell light on the cost of emissions reductions
[exceptions are the OECD's GREEN model (43. 64) the ERB
(3, 4), and the forthcoming Second Generation Model (SGM) (65)].
Consequently, all modelling results need to be treated with some caution,
depending in part on the timescale of application, the care with which the
model has been developed, the extent to which it is appropriate to the
application used, and the care with which inputs have been formulated. The
ultimate argument for such modelling efforts is not that they give precise and
certain answers, but rather that they are the only consistent way of estimating
abatement costs at all and of identifying the important factors that affect
them.
Energy-economy models can be classified in various ways. In this section we
draw distinctions along six dimensions of classification.
"Top-Down" and "Bottom-up" Models
We have noted the major distinction between "top-down" economic models and "bottom-up
engineering/technology-based models. We noted that high positive abatement costs
are frequently associated with top-down/economic
approaches and low and negative costs are frequently associated with bottom-
up/engineering approaches.
As outlined in Section 2, the underlying theoretical distinction lies
between the economic and engineering paradigms. This can be discussed in terms
of a relatively simple illustration (Figure 11). In
economics. technology is featured as the set of techniques by which inputs,
such as capital. labor and
energy, can be transformed into useful outputs. The figure shows a graph of
energy versus other (e.g. capital) inputs. Each cross represents an individual
technique or technology. The "best" technique define the "production
frontier," as illustrated. In principle, efficient markets should result in
investment only in the technically efficient techniques on this frontier
(after allowing for lags associated with old stock), because such investments
can reduce all costs compared with other technologies.
Economic models all assume that markets work efficiently in the sense that
all new investments (after allowing for hidden costs) define the "production
frontier." This is assumed to be consistent with cost-minimizing
(or utility-maximizing) behavior in response to the observed price signals:
various models can encompass other inefficiencies, such as externalities and
fiscal imperfections arising from taxes and subsidies, but still share this
assumption.
Observed behavior (historical data) combined with the optimizing assumption
defines an observed production frontier. The models assume that no investments
are available that lie beyond this frontier (though future technical change
may move it). To the extent that real-world
inefficiencies exist, they are implicitly incorporated in the inferred
frontier. Relative price changes move investments along this frontier (e.g.
substitute labor for energy) as defined by the estimated elasticities; a
purely economic model has no explicit technologies, which are simply implicit
in the elasticities used.
Studies using engineering models have often focussed on identifying
potential least-cost abatement opportunities by assessing directly the costs
of all the technological options. Such assessment is independent of observed
market behavior. This also defines a "production frontier." If markets are
technically efficient, the "frontier" revealed by market behavior should
correspond to that calculated by engineering studies. As illustrated
previously, this is not the case. Engineering studies reveal widespread
potential for investments beyond the limit of the "production frontier"
suggested by market behavior and built into economic models. The explanation
can he considered in terms of the contrasting limitations of the economic and
engineering paradigms. Economic models are slave to the assumption of cost-minimizing
behavior noted above. Limitations of purely engineering models include:
1. The cost concept is based on an idealized evaluation of technologies and
options. The existence of hidden costs is typically ignored.
Top-down and bottom-up are very imprecise terms. Although models generally known
as "top-down" all determine energy demand through aggregate, economically driven
indices (GNP and/or productivity growth, and price elasticities), they can vary
greatly in the modelling of energy supply. Some of the "top-down" models are
purely economic, with supply changes being driven only by substitution elasticities.
Others are primarily economic, but incorporate a "backstop technology"--a technology
that can come in, in unlimited quantities, once a certain price threshold is
reached. Yet in other "top-down" models (such as the ERB), supply is driven
largely on an engineering basis of supply technology costs, chosen from a database
of supply technology cost curves.
Nearly all "bottom-up" models contain extensive representation of supply technologies,
but the key practical distinction as it has emerged in the CO2 costing
literature centers on the modelling of energy end-
use and the introduction of end-use technologies. Bottom-up end-use
studies indicate a large potential for reducing both emissions and costs
relative to a traditional top-down extrapolation of energy demand. In other
words, they show that the top-down projections are not optimal in terms of the
technologies available; and the major savings come by contrasting this with
a scenario that is an engineering optimum.
Why does this create such a large difference between bottom-up
and topdown studies? The primary reason is to be found in Figure 5. which was
discussed in Section 3. Neglecting the segment referring to externalities,
which may or may not be reflected in either top-down or bottom-up
studies, top-down projections of energy demand incorporate only efficiency improvements
corresponding to the bottom segment of the column--the "business as usual" takeup.
Bottom-up models, on the other hand, include all the available technologies,
without distinction as to under which category in the column they fall. Consequently,
we can conclude with some confidence that, neglecting externalities:
1. top-down modelling studies tend to underestimate the potential for lowcost
efficiency improvements (and overestimate abatement costs) because they ignore
a whole category of gains that could be tapped by nonprice policy changes; whereas
2. bottom-up end-use modelling studies overestimate the potential (and underestimate
abatement costs) because they neglect various "hidden" costs and constraints
that limit the uptake of apparently cost-effective
technologies .
Which is more "realistic" depends on the relative size of different
segments in Figure 5--something that cannot be determined
without separate
study of specific implementation policies and costs, discussed in Section 7.2.
But we can say with some confidence that the real near-term potential for limiting
CO2 emissions at low or negative costs lies somewhere between the
optimism of such bottom-up studies, and the relative pessimism of many top-down
studies. Finally, we note that although these issues are most important relative
to demand, they also can apply to energy supply, particularly concerning apparently
"cost-effective"
decentralized renewable energy options that are nevertheless not being
exploited.
We emphasize again that the systematically differing results are largely a
reflection of the non-optimality of the baseline implicit in such bottom-
up studies--and the questions it raises about the assumptions built into top-
down model baseline and abatement projections. This is perhaps the key difference
between the modelling approaches. If the baseline in bottom-up
studies used optimal technologies, the baseline emission projections would be
much lower. This was illustrated clearly in a study by Morris et al (66), which
included end-use technologies in the MARKAL engineering model and found that
obtaining basecase emissions anything near as high as official or macro-economic
forecasts proved almost impossible: the model chose more energy-efficient end-use
technologies, and more renewable energy technologies, irrespective of CO2
constraints. Further reductions were, however, relatively expensive, relying
more on supply substitution, as the stock of more efficient end-use technologies
was largely selected already in the optimal baseline.
Thus there is no inherent reason why "top-down" studies should yield positive
costs or "bottom-up" models should yield negative costs. The sign of the cost
hinges critically on the approach applied to computing costs, in particular,
assumptions regarding optimality of the baseline. For example, Bradley et al
(67) and Edmonds, Barns,Wise, and Ton (68) recognize a non-
optimal baseline and illustrate negative abatement costs within a "topdown"
energy-economy approach; whereas the Morris et al (66) study uses a wholly engineering
model and obtains positive costs for any reductions beyond those captured in
the (optimal, and much lower) baseline.
Some attempts have been made to integrate top-down and bottom-up
models explicitly. Most notably, the Global 2100 model has been linked with
the MARKAL engineering model (by replacing the energy technology submodel in
Global 2100 that formed the energy component previously) in a
bid to combine the best features of both into a single computational framework
(69). However, this still does not resolve the dilemma about whether projections,
for baseline or abatement scenarios, adopt the engineering optimum or econometric
extrapolation of energy demand, and this linked model has been criticized on
the grounds that one still dominates the other (70).
Time Horizons and Adjustment Processes: Short term Transitional versus
Long-term Equilibrium
Different models are designed for application over different timescales.
There are no standard definitions, but in many relevant branches of economic
analysis the short term is taken to be less than 5 years, the medium term is
between 3 and 15 years, and the long term is more than 10 years. The timescale
is a distinction of major importance, particularly for economic models, because
different economic processes are important on different timescales, and thus
the timescale for which models are designed fundamentally affects their structures
and objectives. Models for relatively long-run analysis may to a reasonable
approximation assume an economic equilibrium in which resources are fully allocated.
Short-run models focus on "transitional" and disequilibrium effects such as
transitional market responses, capital constraints, unemployment, and inflation.
This
distinction parallels the structural distinction drawn by Boero et al (71)
between resource allocation models and macro-economic
models.
MEDIUM TO LONG TERM: EQUILIBRIUM/RESOURCE ALLOCATION MODELS These models focus
on the allocation of available resources, within the energy sector or the broader
economy. This category includes both optimizing bottom-up models (which seek
to optimize resource allocation within the bounds set by available technology).
and all the main long-run and global energy/C02 models. The latter
are generally termed equilibrium models .
At one extreme of the long-term modelling dimension are models that can only
consider the energy/inveshnent mix for a "snapshot" year and compare this to
another, without any information on the transition between them; these are comparative
static models, such as the Whalley & Wigle model (36, 37) and an early form
of the GREEN model (72). Such models can enable detail in representing the
system, but at the expense of modelling developments over time. In contrast,
dynamic models cover medium-and long-term phenomena, extending across several
time periods.
At the opposite of this extreme within the equilibrium models, some are
designed to run in annual steps over a period of a few decades. These
can include considerable detail on different sectors, whose use of different
resources in response to price changes is estimated econometrically from data
over previous years. The main examples are the Jorgensen/Wilcoxen (73) and
Goulder (74) models for the United States.
Equilibrium models such as Global 2100 and GREEN [in its more recent
versions (43, 64)] and the ERB model lie between these extremes. They are
designed to operate in steps of 5-15 years, to look at the changing allocation
of resources under different constraints over periods of many decades, and the
way this may change under CO2 constraints.
The treatment of capital stocks in these models can have important
implications for costs. "Putty-putty" models represent capital stocks as perfectly
interchangeable between sectors and over time. Nuclear power plants can be transformed
into solar photovoltaic arrays instantaneously and without cost. "Putty-clay"
models, on the other hand, allow no transfer of capital between applications.
Once an investment has occurred, the technology cannot be altered (sec Section
8.3). Resource allocation/equilibrium models cannot, however, model other
aspects of transitional costs arising from disequilibria. In this respect,
and
in their assumption of optimal investments subject to constraints, they have
been criticized for underestimating likely abatement costs (though the satire
caveats apply to the reference projection as well, which is similarly optimal
within constraints and free from disequilibria).
SHORT TO MEDIUM RUN: MARKET SIMULATION/MACROECONOMIC MODELS Short-run models
by contrast focus primarily or exclusively on the dynamics of transition, rather
than the long-term equilibrium allocation of resources.
One class of short-run models are sectoral market simulation models, such
as detailed models of electricity or oil markets and pricing, or of industrial
sector energy demand. The diversity of such models reflects the range of specific
markets that they have attempted to model. Few such models as yet appear to
have been applied to assessing abatement costs; one notable exception is the
application by Ingham, Maw, and Ulph (75) of separate market models for the
industrial, domestic, and transport sectors in the United Kingdom. However,
such models may come to assume much greater importance as governments move closer
to considering detailed policy measures tailored to specific sectors. Frequently
(as in 75) they focus on sector responses to carbon taxes rather than costs.
Of more general interest for costing is the recent application of models
known usually simply as macro-economic models. This is the name usually
(if imprecisely) given to the models developed over many years for studying
the short-run dynamic behavior of national economies. Typically these models
contain explicit representation of investment and consumption in different sectors,
and markets do not necessarily clear; there can be
unemployment, idle production capacity, or capital shortage. Such models
generally contain a strong Keynesian component, though many other aspects of
macro-economic theory have also been brought to bear in them. Recent applications
to CO2 abatement are discussed below.
Such models can generate a wide range of macro-economic indices such as GNP,
inflation, employments, etc. For this reason they are of particular interest
for assessing the short-term macro-economic impact of CO2 abatement,
However, such models may contain very limited representation of the energy sector,
and some may not even model energy as a separate good within the economy. Few
contain a representative set of energy-technology options. Such short-run macro-economic
models are thus highly country- and model-specific, and vary greatly in the
extent to which they can be applied to assessing CO2 abatement. The
models are best at representing transitional costs; results may become unstable
and questionable when the models are run too far ahead, because
the economic feedbacks that keep economies from straying too far frolic
economic equilibrium are generally not well represented.
No study has focussed primarily on a comparison of short-run
macrocconomic with general equilibrium (GE) models, but such a comparison is
implicit in the studies of Shackleton et al (76). This took two macroeconomic
models (DRI and LINK) and two general equilibrium models (DGEM and Goulder),
all of which the model authors considered appropriate to run over a period of
2-3 decades. Each was subject to the same carbon tax ($40 per ton C). Figure 12 shows that the models behaved in very different
ways. Concerning the impact on CO2, the macro models suggest a reduction
of 0-8% depending on the way in which carbon tax revenues are used (see Section
8. 1; even a C02 increase is observed from a tax recycling that boosts
economic growth); the equilibrium models suggest 20-30% reductions depending
only on which model is used. Even more striking, the macro models show a wide
variety of GNP responses, varying greatly over time, and including substantial
increases from some tax recycling options; the GE models show a much smoother
response, with more modest impacts.
All this corresponds to the theoretical differences. The short-run
macro models reflect the resistance of the economy to change, but also the
cumulative impact of greater investment unconstrained by equilibrium
requirements. The GE models allow capital to move easily across the economy
to respond to the price changes, giving a much stronger CO2 response.
It seems reasonable to suggest that the longer-
term results from the short-run macro-economic models are questionable (and
highly sensitive to various assumed macro-economic
investment and other responses), as are the short term GE results; the most
realistic outcome may be to assume a progression over time from the macro towards
the GE results, but even this is speculative. Beyond this we cannot generalize,
but we emphasize the importance of resolving such great variation based purely
on model structure. McKibben & Wilcoxen
(33) are the only researchers to have yet applied a model that can
simultaneously address unemployment and GE class problems.
Finally we note that between the short-and long-run models, some progress has
been made in developing medium-run models that combine elements from both short-run
macro-economic and resource allocation/ equilibrium models [see discussion in
Boero et al (71)]. These may be seen as short-run
models that are expanded to include adjustment to long-term equilibria. That
is, medium-run models are mainly demand determined and allow for market disequilibria;
however, a central part of the models aims to describe the adjustment process
from short-term market disequilibria to longterm market equilibria. Finally,
we note that there are some top-down modelling approaches that do not fit into
these categories at all, such as the growth models employed by Nordhaus (35)
and Anderson & Bird (45).
Sectoral Coverage: Energy versus General Economy
Another important distinction is that between models that address only a
limited part of the economy (in this case, the energy sector) and those that
encompass the whole economy, which usually have a much more simplified
representation of any particular sector. Among equilibrium models, the
distinction is known as that between partial equilibriumand
general equilibrium models; the parallel distinction for short-run
models is that between energy market and macro-economic
models; a more general terminology would be that of [energy] sector general
economy model.
Sectoral models focus heavily or exclusively on particular economic sectors
(in this case, usually just the energy sector or parts thereof): insofar as
the
rest of the economy is represented, it is in a highly simplified way. They
address the problem of describing behavior in a single area of the economy,
for exampic energy, but ignore or treat summarily all other economic functions.
An energy market model would have no description of the labor or capital markets.
Energy sector models come in many varieties. Bottom-up technology models are
all sectoral; so are many equilibrium models. such as the ERB (3, 4). Investment
planning models, such as the electricity expansion models widely used for assessing
power sector investments, are more focussed sectoral models. Models of the international
oil market--a very important but
little analyzed issue in assessing abatement costs--are sectoral but global
models.
Sectoral models yield an estimate of costs in a particular sector, but
cannot take account of the macro-economic linkages of that sector with the rest
of the economy. They cannot, for example, estimate how the labor or investment
requirements in that sector may affect the resources available to other sectors.
By contrast, general economy models encompass all major economic sectors
simultaneously. They recognize feedbacks and interrelationship between sectors.
In principle, this enables them to estimate the full long-run GNP impacts of
restrictions (such as CO2 constraints). In practice, this is obtained
at the expense of considerable simplification of the energy sector.
General economy models are the only kind that can reflect important
features in the rest of the economy. This may include, for example, economic
distortions outside the energy sector. In particular, existing taxes impose
varying burdens on economies. If the revenues from a carbon tax are used to
reduce such taxes, the gains may in principle offset the losses, completely
altering assumptions about the net impact on the GNP of such taxes. This seems
to have attracted attention only recently [Dower & Zimmerman (77)];
relatively early discussions of this are given by the CBO (78) and Grubb (79),
and subsequent modelling studies are reviewed in Section 7.
The past few years have seen the development of linked models that seek to
integrate the detail of energy sector models with the economic consistency of
general economy models. In Europe, the macro-economic HERMES model has been
linked with the MIDAS energy supply model and applied to
analyzing abatement strategies in the four largest EC countries, including
interactions between these economies (80), to generate the results
illustrated in Figure 9. The Manne & Richels GLOBAL
2100 model (39) contains considerably more energy supply detail than other general
equilibrium models by integrating an energy technology model (ETA) into a simple
general equilibrium model (MACRO, which clears markets
for all goods and services treated as a single aggregate). The SGM (65),
which is under development, extends this approach. It is a general equilibrium
model that was designed to address greenhouse-related
issues, and thus desegregates economic activities on the basis of importance
to
the greenhouse issue.
The ERB model is an energy market rather than general economy model. but it
contains a "GNP feedback" parameter, which was incorporated to ensure that the
impact of large changes in energy sector costs on the scale of economic activity
would be reflected in terms of its impact back on energy demand. The parameter
is not intended to provide consistent estimates of the costs of emissions reductions,
though many studies have used it as such.l2
As noted above, costs must be developed using consumer plus producer surplus
techniques (81, 82). Other sectoral models, such as Fossil2 (83), require similar
approaches to make consistent cost estimates.
Optimization and Simulation Techniques
Largely independent of the above differences. models can adopt different
approaches to optimization. Some models optimize energy investments over time
by minimizing explicitly the total discounted costs (or per capita
consumption), using linear or nonlinear techniques. Several engineering models
use linear programming, most notably the EFOM model (29) used extensively within
the EC, and MARKAL (84), promoted internationally by the ETSAP program of the
International Energy Agency. The top-down
Global 2100 model uses nonlinear dynamic optimization, as does its derivative
CETA (42). Such optimization approaches in effect assume perfect investment
(within the confines of the model), with perfect foresight. It is debatable
whether this is a drawback or advantage.13
Modelling partial foresight is, however, very difficult, and the main
alternative approach is to simulate investment decisions on the basis of
"static expectations," i.e. static projection of conditions at the time of
investment. This "myopic" assumption is used in ERB. GREEN. and the CRTM trading
derivative of Global 2100 according to Dean & Hoeller (64), no
software yet exists for solving such large dynamic general equilibrium models
under the assumption of perfect foresight. The SGM model is being developed
to incorporate a variety of options for determining investments on the basis
of future expectations (including a fomiulation of partial foresight).
Thus the mechanism for selecting investments is fundamentally different
between investment simulation and optimization models, and this might be
expected to have a major impact on results. In fact, this does not appear to
be
the case. A recent comparison by the Energy Modelling Forum (unpublished) of
results from global 2100 (dynamic optimization) with those from the ERB model
(investment simulation) shows that standardizing for key assumptions leads to
remarkably similar energy and emission results. Assuming a competitive economy,
variations in key input assumptions (such as those discussed below) thus appear
far more important than the approach used for selecting investments in the model.
Short-run macro-economic models are concerned primarily with the simulation
of aggregated investment responses in terms of labor, capital, etc but do not
seek to optimize divestments; they do not represent specific technologies at
all. Some other models are purely for simulation of system operation, without
automated invested modelling, and they report on the implications of an investment
strategy that is specified externally [e.g. the Danish BRUS model (85)]. This
can enable much greater detail in representing the system. and avoids the limitations
of linear optimization especially, though there are inevitably drawbacks from
having to specify investments manually and check their consistency [UNEP (16),
Chapter 3)].
Level of Aggregation
Models differ greatly in their degree of disaggregation. To some
extent this is the obverse of the model scope. Models that, for example, focus
on household electricity demand can represent this and the options for
improving household electricity efhciency, in great detail. Global, economy-wide
models have to be highly aggregated.
At one end of the spectrum are models such as LEAP (86) and the BRUS model (85)
used in the Danish Energy 2000 study (87). Their demand sectors are generally
disaggregated with respect to specific industrial subsectors and
processes. residential and service categories, transport modes, etc. with the
aim of achieving homogeneous entities whose long-term behavior can be defined
through consistent scenario projections. Similarly, energy conversion and supply
technologies are represented at the plant type and device levels. This allows
detailed modelling of the alternatives for technical innovation, fuel switching,
etc.
With regard to emissions of pollutants and CO2, this type of disaggregation
into specific technologies makes it possible to take account of the different
characteristics of energy technologies. A very detailed analysis of abatement
options can thus be carried out. This will include energy savings at the end-use
level, changes in the conversion system, and fuel substitution. At the other
end of the spectrum are models such as GREEN and Global 2100, which treat energy
and the world in a highly aggregated manner.
In general, the level of aggregation is closely related to the other aspects;
for example, a multiperiod, global, general-economy model by necessity will
have a highly aggregated representation of energy demand and supply. with little
if any technological detail. Great detail in representing energy supply, conversion,
and end-use markets and technologies is only possible in models that are specific
to the energy sector, and focus on simulation rather than full system optimization.
The benefits need to be weighed against these limitations.
Geographic Coverage, Trade, and "Leakage"
Another important division of scope is geographical. Global models
describe the world economy divided into "regions" such as North America,
Europe, the Organization of Petroleum-Exporting Countries (OPEC), Southern Asia,
etc. and many can represent interactions via
trade and monetary transactions between different parts of the world economy.
Global models have been developed and applied primarily to examine aggregate
questions such as the likely rate and pattern of emissions growth, the relative
gains and losses from differing distributions of international CO2-reduction
and international interactions of abatement efforts, including trade issues.
A limited number of global models have been applied, as summarized in Section
4.1.
National models focus on specific aspects of single countries and can give
more detailed descriptions of the economic interactions within the country.
World market conditions are normally taken as exogenous. It is possible,
however, to link national models; trade and monetary transactions between
countries may be endogenized in order to analyze the effects that national
policies in one country may have on the economy of other countries.
Allowing trade in goods and emission targets lowers overall abatement
costs. Cline (17, Chapter 4) criticizes the Global 2100 model for its lack of
trade in either production or energy resources, which leads to each region
effectively optimizing its economy using only local resources, and thus increasing
the global cost of reducing CO2 emissions. Trade in emissions rights
can lead to significantly lower costs in achieving overall emissions reductions
(see Section 8.3). However, while overall costs may be reduced through trade,
individual nations may either gain or lose, depending on the allocation of emissions
rights [Edmonds et al (68)].
Global abatement efforts will affect internationally traded fuel prices,
and thus have particular impacts on countries that depend heavily on energy
exports or imports. Any study that does not take the global perspective into
account may underestimate the economic impact of measures to reduce CO2
on energy-exporting countries and overestimate that on energy-importing countries.
Marks et al (88) address this issue by looking at the effect of a fall in the
world price of coal; Perroni & Rutherford (89), tile OECD studies (notably Ref.
90), and Whalley & Wigle (36, 37) also address the
issue of trade. Global models are also required to examine the issue of
"leakage," by which abatement in one region may be offset by international
price and trade reactions (discussed in Section 8).
The cost estimates of studies that take these trade issues into account do
not therefore simply reflect the impact of the domestic policies on the
economy. A distinction needs to be made between the costs that stem from the
domestic policy options and the economic impact of trade effects and capital
transfers.
Modelling Classifications: A Resume
An understanding of different models is required because different
models have different strengths and weaknesses. Models for studying CO2
abatement costs have been developed in different ways, often by adapting existing
models. They are able to handle some issues (or sectors) better than others:
different models are thus suited to different purposes. Models cannot--or at
least should not--be interpreted as giving complete and accurate answers, but
rather used for the insights they offer when the results are combined with an
understanding of the model structure and limitations.
There is no universal or accepted way of classifying models. In this
section we have noted at least six dimensions. Of these the division between
"top-down" economic and "bottom-up" engineering is of great practical importance,
despite its occasional ambiguity. Within top-down
models, the distinction between long-run equilibrium and shorter-run
macro-economic models is central, as is that between partial (sectoral) and
general economy models. Neither of these latter distinctions is relevant to
bottom-up models, for which important distinctions are those between partial
models (generating a cost curve of savings related to a topdown or unspecified
baseline), and full system representation, and within the latter, the choice
between optimization and simulation. Table 6 classifies
the major models discussed in this paper according to this schemes which has
a
pragmatic focus on the factors of greatest importance to abatement costing.
Other classification approaches by Boero et al (71), and Beaver (60, 61) have
similar elements but differ he detail.
NUMERICAL ASSUMPTIONS AND SENSITIVITIES
Assumptions drive model results. Critical parameters can be usefully
(though not exclusively) divided into those that govern the overall scale of
the system and reference missions, and those that directly affect the relative
cost of emissions reductions. GNP and population growth rates, income
elasticities and the rate of "autonomous end-use energy-intensity
improvement" ("AEEI") primarily affect the baseline scale; background fossil-fuel
prices strongly affect both the baseline emissions and abatement costs; the
cost of low-carbon technologies and price elasticities largely drive abatement
costs though they also affect baseline emissions.
Population and GNP Growth Rates
The demand for energy is driven by population and per capita energy
demand, and all economic models at least assume that the latter is driven by
per capita GNP. It is consequently much more difficult to restrict the growth
in emissions for a developing country with a high population and economic
growth rate, such as India, than for a more slowly growing developed economy,
for example, Germany, which has a static or declining population.
The uncertainties in future global population are reflected in abatement
studies; estimates for the year 2025 for example include 9.5 billion for the
NAS (47) and 8.2 billion for Edmonds & Barns (81). However, the latter study
found that there was little impact from a reduction in population growth for
approximately 15 years, that is until labor force and therefore GNP was affected.
Projections suggest that the increase in global GNP will be much greater anal
more uncertain than population growth. and so
differences in GNP projections account for a greater part of variation in
baseline emissions.
Different baseline GNP and energy demand assumptions across studies
complicate comparisons of the GNP loss associated with a target CO2
reduction. In models that derive GNP from labor productivity, this is correspondingly
critical. For example, in the United States in 2020, GNP baseline estimates
range from $7.5 to $11.5 trillion [CSIS (91) and Edmonds & Barns (82), respectively];
the low costs of CO2 stabilization in the Jorgenson & Wilcoxen (92)
study have been attributed partially to their lower GNP projection.
The difference that baseline GNP makes to energy demand is more complex, as
growth in any economy will inevitably vary across sectors over time; greater
GNP growth in reality would not necessarily imply that the growth within all
sectors of the economy is increased proportionately.
Energy/GNP Relations/tips and the "AEEI"
While GNP is a major determinant of energy demand, many factors can
affect the relationship between them. A few models incorporate explicitly a
non-unitary energy-income elasticity, which implies a changing energy/GNP ratio
as GNP grows. Most models, however, express such a change, if any, in terms
of an exogenous parameter that defines the rate at which the energy/GNP ratio
would change in the absence of price changes. This rate of exogenous (or "autonomous")
end-use energy-intensity improvement (AEEI) then becomes a major determinant
of baseline energy demand for long-term
projection; the higher the rate of energy-intensity improvement, the lower will
be the baseline CO2 emissions, and the lower the costs of reducing
relative to a given base year. The parameter has been widely described as a
measure of technical progress, but as we emphasize below (Section 7.2), it compounds
many different elements.
A range of AEEls has been adopted. The main studies with the GREEN (44) and
ERB (3, 4, 68, 81, 82) models assume an AEEI of 1% per year for all of the
regions of the world. Manne & Richels (39) assumed a more pessimistic set
of values averaging 0.4% for the world, while Mintzer (13) adopts a much more
optimistic 1.5%. The difference in values between the Manne & Richels and the
ERB studies account for a very large difference in long-term
projected baselines (respectively, 40 and 23 Mt C per year by 2100) . The
sensitivity study by Edmonds & Barns (81) confirms the importance of this
parameter. In their later sensitivity analysis, Manne & Richels (40, 94)
used AEEls ranging from 0% to 1.5%. With an AEEI at 1.5%, the energy
requirements at the end of the next century would be one-fifth of the demand
had an AEEI of 0% been used.
Future Energy Prices, Resource Modelling, and Supply Elasticities
High background fossil-fuel prices lower energy demand and CO2 emissions,
and reduce the relative costs of
moving to lower carbon fuels. Limited resources (e.g. of oil) have the same
effect, implicitly or explicitly raising the prices as resources are depleted.
Resources are, however, uncertain, and the course of fuel prices is even more
uncertain.
Various approaches may be taken towards estimating the future cost and
availability of different fuels. National studies may define national
production costs but define exogenous global prices with great variation; for
example, Chandler & Nicholls (95) assume that prices of natural gas and oil
will rise by 2.5% per year and those for electricity and coal by 1% per year,
while the OTA (48) projects prices for these fuels to rise at around 4% per
year and nearly 2% per year, respectively.
Global models reflect resource/supply cost issues explicitly, through
direct estimation of the resource base and supply elasticities. High supply
elasticities mean lower fuel price rises as supply increases. GREEN assumes
zero supply elasticity for oil outside OPEC (i.e. volume set by fixed
production constraints), but price is determined by OPEC supply elasticities
varying from 1 to 3; supply elasticities are higher for gas and coal and much
lower for nonfossil sources until backstop technologies become relevant.14
2. The cost of implementation measures (e.g. information campaigns,
standard setting, and compliance processes) is not included.
3. Market imperfections and other economic barriers mean that the technical
potential can never be fully realized.
4. Macro-economic relationships (multiplier effects, structural effects, price
effects) and indicators (GNP, employment, etc) are not included in the models
Price/Substitution Elasticity of Demand for Energy
The impact of price changes on energy demand is determined by the price elasticity of energy demand, or in general economy models, the substitution elasticity between energy and other factors of production. The lower the relevant elasticity, the less energy demand is curtailed by higher energy prices, and the greater the tax that is required to reduce energy demand and consequently CO2 emissions. The long-run elasticities assumed in the EC bottom-up study [COHERENCE (29)] vary among countries and range from --0.4 for France and the United Kingdom to -1.0 for Belgium; the short-run elasticities range from --0.1 to --0.25 for the same countries. These differences account for part of the national differences; evidence for elasticities especially in EC countries is discussed in Mors (96) and Pearson & Smith (97).
In the global models, assumed long-run elasticities range from (--)0.3-0.4 for the Manne & Richels US and Global 2100 studies, to (--)0.6-1.0 for the OECD's GREEN model;15 short-run elasticities are about one-tenth this value. The OECD values were chosen after an extensive literature search (43, 64). For the United States, Jorgenson & Wilcoxen (73) estimate -0.15; Barrett (98) and Capros et al (99) discuss elasticities in the European context.
The notion of elasticities assumes a symmetric response to price changes; if prices rise and then fall back to previous levels, the energy-intensity will (after allowing for lags) return to former levels. As noted below (Section 7.2), this basic assumption, and the values assumed for modelling, are disputed and have been particularly called into question by recent trends and studies.
Technology Developments and Costs
Assumptions concerning the cost and rate of implementation of more efficient or lower carbon technologies affect both baseline emissions and relative abatement costs. The initial Manne & Richels results for the United States (38) were strongly criticized by Williams (100) as being based on unreasonably pessimistic assumptions for efficiency improvements and the costs of alternative supply technologies. In response, Manne & Richels (94) examine three different background scenarios, which they termed technology optimistic, technology intermediate, and technology pessimistic. The last of these corresponds to their initial famous estimate that CO2 abatement could cost the US $3.6 trillion over the next century, and yields some of the higher cost points on Figures 6 and 8 above. Assumptions for the first, "technology optimistic," reduced these total costs by a factor of 20 for the (fixed) abatement target set, because of the combined impact on baseline emissions (primarily from the higher AEEI) and the halved costs for "backstop" low-carbon supply technologies. As a result, Manne & Riehels noted that "the direct economic losses are quite sensitive to assumptions about both demand and supply... for the losses [from carbon constraints to approach zero, however, the most optimistic combination of supply and demand assumptions must be adopted".
This confirms that results are very sensitive to the assumptions concerning technology costs, again a result noted in the sensitivity study by Edmonds & Barns (81). Almost all studies fix the costs of supply technologies as exogenous data; Anderson & Bird (45) is the only study in which technology costs decline with increasing investment.
Energy Sector Impact on GNP
The impact of changes in energy demand and energy sector costs on GNP is complex. General economy models capture the relationship consistently, but the resulting elasticity can still vary considerably. For example, the Global 2100 modelling approach (and by implication, the CRTM and CETA models also) contains a "nested CES" (constant elasticity of substitution) production function16 to relate energy input to economic output. Cline (17, Chapter 4) criticizes the parameters chosen, claiming that they yield an excessive impact of energy sector changes on GNP--a claim disputed by Manne & Richels.17
For sectoral or partial equilibrium models, the impact of energy sector costs or carbon taxes on GNP may be estimated (if at all) by a direct elasticity ("GNP feedback") parameter, as available in the ERB model. Such studies have been criticized for overestimating the feedback and consequently overestimating the cost of reducing CO2; as noted in Section 5.3 above, Edmonds & Barns (81) recognize this limitation and suggest the use of consumer plus producer surplus changes as the best method of computing cost for the ERB model.
Interfuel Substitution Elasticities
Substantial CO2 reduction can he achieved by switching towards less carbon-intensive fuels. In models with a purely engineering approach to supply this is captured by supply technology costs; for models with econo-metric supply modelling, it is governed by the interfuel substitution elasticity. GREEN assumes a long-run interfuel elasticity in production of 2.0 and a short-run value (reflecting existing supply infrastructure) of 0.5. Halving these values lowered global baseline emissions by 13% in 2050; the impact on abatement costs may be expected to be much larger unless the bulk of substitution is governed by backstop technologies.
KEY DETERMINANTS: TECHNOLOGICAL VS ECONOMIC PERSPECTIVES
Introduction
We have explored the range of reported results on the costs of limiting CO2, and the technical modelling and data issues that affect such estimates. We now turn to a deeper analysis of the issues that determine abatement costs and the more credible results. This section examines the gulf that may be characterized (not always accurately) as that between "technology-engineering" and "economic" perspectives, in terms of assumptions concerning energy demand and policy, structural, and technological assumptions. The subsequent section examines how the abatement strategy and scope of analysis affect the reported results.
It would be facile to suppose that the differences between "economic" and "engineering" views are confined to a few modelling parameters. They reflect very different perspectives, almost paradigms, about driving forces in the energy economy. A report from a UNEP workshop that sought to bridge the divide (101) observed that:
To economists, energy is a factor of production: it is an input into economic growth, and one which can substitute for labor or capital, depending on relative prices. While energy-efficiency may improve due to technical development, so does that of other factor inputs; and efficiency improvements lower the relative price or energy, increasing the extent to which it may be applied. Also, fossil fuels dominate because of demonstrable convenience and how cost. Thus whilst recognizing the potential importance of technical improvements, and even market imperfections which prevent optimal energy use, to most economists there is every reason to expect energy consumption to grow with expanding (GNP, and no particular reason to expect technical developments to reduce CO2 emissions relative to the business-as-usual case without incurring substantial costs It is a compelling case, with much long-term historical weight behind it.
To scientists and engineers, on the other hand, energy is not an abstract input but a physical means to particular ends. The applications which consume much energy are those of heating, heavy construction, metals, etc--basic infrastructure and comfort--and travel. In developed economies, most infrastructure needs have been met, travel may be approaching limits of congestion and time budgets, and much new economic activity is in areas which consume trivial amounts of energy, such as information technology, general entertainments, etc. Thus to most scientists and engineers economic growth is becoming less and less relevant to energy needs [in developed economies]. In addition, very large technical improvements in efficiency, which need not incur much extra costs, are readily demonstrable; technology is powerful and adaptable to changing conditions (such as requirement for nonfossil sources): and it is hard to believe that human society is incapable of finding ways of putting such options into effect (this applies primarily to developed economies, but may also be of great relevance to developing ones if they can move directly to advanced technologies). This too appears to be a strong case. with at least partial support from recent trends, but it leads to a very different outlook from the "economist's" outlook sketched above.
Baseline Energy Debated and the AEEI
Section 6 notes that energy demand in the absence of abatement measures depends on GNP (population times per capita GNP) growth rates, energy prices, and the response of demand to these. and the parameter usually translated as the rate of "autonomous end-use energy-intensity improvement" (AEEI). The impact of GNP and energy price changes is recognized by all analysts, as are the uncertainties. The major contrast in views arises from the differing assumptions about the AEEI. The enormous impact of this parameter has been noted in Section 6.2. Dean & Hoeller (64) state that "unfortunately there is relatively little backing in the economic literature for specific values of the AEEI ... the inability to tie it down to a much narrower range ... is a severe handicap, an uncertainty which needs to be recognized".
Among the major global and US studies, Whalley & Wigle (36, 37) and Manne & Richels (38, 39) adopt the lowest values for AEEI (0 and around 0.5 respectively). Williams (100) strongly criticized such values as too low; Manne & Richels (40) defend their value of AEEI on the grounds that it appears consistent with observed trends in the United States. However, it is difficult to separate the various factors in their analysis (e.g. price, income distribution, and time sampling effects), and Wilson & Swisher (70) strongly dispute their interpretation, concluding that "one can produce an experiment that justifies whatever AEEI one likes within very broad ranges".
We cannot suggest a definite value for this parameter, but it is important to understand it. The parameter has been badly misnamed: it is a measure of all nonprice-induced changes in gross energy-intensity--which may be neither autonomous, nor concern energy-efficiency alone. It is not simply a measure of technical progress, for it conflates at least three different factors. One indeed is technical developmentsthat increase energy productivity. But another is structural changei.e. shifts in the mix of economic activities (which may require widely different amounts of energy per unit value added). The third is policy-driven uptake of more efficient technologies,due to regulatory (as opposed to price) changes, greater than would occur without those changes.
Technical change is indeed hard to predict. Studies by Meyers & Schipper (102, 103) suggest that in manufacturing alone, technical change has increased energy productivity in OECD countries by about 2% per year for at least the past two decades; this includes price effects, but in fact there is no clear change in the trend that correlates with the energy price shocks (perhaps because of the lags in manufacturing equipment). Technical change appears more closely related to the price shocks in other sectors.
Structural change encompasses the phenomena of saturation in energy-intensive activities (such as home heating and primary heavy industries), and shifts towards less energy-intensive activities. A range of studies have noted that structural change, both between sectors and within manufacturing industry, has played an important part in restraining energy demand in the OECD in the past 20 years (103). Williams et al (104) provide considerable evidence for expecting the observed trend in manufacturing to continue. On the basis of this and other relevant literature, and various saturation effects, Grubb (79, Chapter 6) argues that an autonomous structural trend towards lower energy-intensities (i.e. rising AEEI) is to be expected as countries develop and as economic growth moves towards increasingly refined products and services.
We tentatively suggest that the lower values of AEEI in long-term studies (significantly below 1.0 especially for OECD countries) are dubious because of saturation and structural change effects. If correct, this points to baseline emissions towards the lower end of the range of long-run model predictions, making any fixed target much easier to reach and also reducing the scale and hence relative costs of reductions. This is,however, tentative; there are substantial uncertainties and a clear need for greater understanding of technical and structural trends, and integration of such studies in abatement cost modelling.
The AEEI is not the only controversial issue surrounding the relationship between CNP and energy demand. As noted in Section 6.4, considerable uncertainty surrounds estimates of the price elasticity of energy demand. Most studies have sought to estimate elasticities from the response of demand to the energy price rises of the 1970s and early 1980s. More recent trends, if anything, increase the uncertainties. Although energy demand has started to rise again in OECD countries following the price fails since the mid-1980s, the response has not been nearly as great as predicted by the simple reversible statistics of elasticity. Engineers have long maintained that the efficiency gains would not be lost, because they are embodied in better knowledge and techniques that will not be abandoned even if energy prices fall. The recent trends at least partially support this view, and econometric studies (105) have now started to question the basis of constant elasticities that assume symmetric responses to price increases and fails. Weisacker (106) also emphasizes the importance of recent analysis of price responses, and shows how it could have substantial implications for longterm abatement costs and strategies.
Regulatory Instruments and the Energy-Efficiency Gap
Most top-down models assume the optimal operation of markets in response to observes price signals (Section 5.1). in which case economic theory suggests taxes to be the optimal means of abatement. The observation of the large efficiency gap demonstrated by engineering studies (Section 3) calls this into question. Many economic discussions reject the relevance of this data; Nordhaus (30) for example states in his review that believing in such free lunches requires an act of faith that is not warranted by economic evidence.
As discussed in Section 3.2, the "efficiency gap" comprises many different components. If it can he wholly explained in terms of lags in take-up, unavoidable hidden costs, etc. then it is a phenomenon of little direct relevance to the actual costs of limiting CO2 emissions. However, many barriers to the uptake of more efficient technologies have been identified; Reddy (107) gives an extensive and clear analysis of the different kinds of barriers, and Grubb (79, Chapter 4) notes at least eight different categories. The reality of unexploited opportunities is beyond doubt.
It is accepted that some things can be done to improve the uptake of efficient techniques, for example, with government campaigns to improve information and the awareness of consumers of energy-efficiency and conservation. Most top-down studies in effect assume such measures to be enacted irrespective of CO2 abatement, and ignore or dispute the scope for other more direct cost-effective actions. However, experience and modelling studies of regulatory policies demonstrate that such measures can and have been effective. Examples include studies of the US National Appliance Standard and Car Efficiency (CAFE) standards [Rolin & Beyea (108)]; of US building standards [Norberg-Bohm (109), Bradley et al (67)]; and a variety of measures internationally reviewed in Johansson et al (31).
Regulatory changes to encourage utility demand-side management programs have been widely advanced as a way of capturing the "free lunch" by getting utilities to invest in end-use efficiency. The "hidden costs" in such policies are debated. Joskow & Marron (110) conducted a survey of experience in US programs and concluded that "reported costs exceed those of the technology potential analysis because program costs are higher and energy savings are lower than these studies assume ... although many of the programmes still appear cost effective".
The important fact remains that modeiling studies of specific regulatory options frequently yield lower, rather than higher, estimates of abatement costs than those derived from carbon taxation designed to achieve similar abatement. Such studies of regulatory measures thus contrast with the common economic assumption that regulatory options are more expensive than using economic incentives. This is because such policies address areas of significant market "failure".
In terms of economic model parameters, this may also be understood in terms of the AEEI, for policy-driven changes is its third component. The critical question is not then just its value, but the extent to which it in fact is a variable that may be affected by policy.
So how large is the potential "free lunch" of zero-cost energy-efficiency improvements captured by regulatory change? The end-use technology studies discussed in Section 3 frequently suggest a long-run technical potential to reduce energy demand without extra costs by 20-50%. Schipper et al (103) discuss the implementation and potential of energy-efficiency programs across the OECD in detail. They too conclude that there is large potential, but emphasize that the achievable potential will always be substantially smaller than the apparent technical potential, and that exploiting it will depend on more aggressive and sophisticated policies. Taking account of the various implementation issues points at best to the lower end of the 20-50% range being available--though we note that further technological development would also be expected.
This, combined with the estimates of Schipper (103) and a range of other more sector-specific studies (such as the regulatory studies noted above) suggest that over a couple of decades, targeted energy-efficiency programs might reduce energy demand by up to 20% of the level projected in the absence of any such policies, at costs lower than that of the displaced supply. Cost-effective savings of 20% is also the figure adopted by Cline from his review of engineering studies (17, Chapter 5). Furthermore, even many topdown studies indicate that reductions of 10-20% may be available at very low cost. While acknowledging the many uncertainties, we consider 20% reductions to be a reasonable estimate of the credible size of the "free lunch."
It is still debatable quite how this may be interpreted in relation to top-down studies. The US National Academy of Sciences study (47) compares bottom-up and top-down results, and highlights the large differences between the results up to moderate emission reductions, but also states there is a broad overlap at higher emissions; the "negative cost" of bottom-up studies corresponds to "low cost" in top-down studies, after which both approaches converge on rising marginal costs for further abatement.
A different view is given by other comparisons of top-down and bottom-up results. A critique and comparison of data from economic modelling as summarized by Nordhaus (30), and their relationship to bottom-up results, is presented by Wilson & Swisher (70), who indicate that bottom-up studies suggest cheaper abatement right across the range. up to abatement of 70% or more. Their data suggest that the whole cost curve from the top-down studies reviewed by Nordhaus has to be shfted by 20-40% to reflect the technical opportunities identified by bottom-up studies, which would greatly alter the pattern shown in the graphs of Section 4.18
In reality, the data from studies are too scattered to reach a definitive conclusion; the actual situation is likely to lie between these extremes, implying a need for some reduction in abatement costs from top-down models across the range to allow for the potential of regulatory-driven improvements in energy-efficiency. A further factor is that, with expanded markets for more energy-efficient technologies, these technologies might develop faster, thus permanently raising the AEEI. The impact of such a change has already been noted.
Technology Assumptions and Modelling
Ultimately, the costs of limiting CO2 emissions will depend heavily on the technologies availabie, not only technologies for more efficient use or energy, but also for the production, conversion, and utilization of lower-carbon energy sources. The importance of technology, as well as assumptions concerning its developments and costs, has been widely acknowledged: it forms the central element of Williams's (100) critique of the Manne & Richels (38) conclusions, and sensitivity studies by Manne & Richels (40) and Edmonds & Barns (81) have demonstrated that estimates of abatement costs depend crucially on technology assumptions.
Yet the care with which such assumptions have been developed varies widely, and the models employed to date have limited representation of technology issues. Models that do incorporate some explicit representation of technology include the Global 2100 model and its derivatives (CRTM and CETA), more recent variants of the OECD GREEN model, and the ERB model, which has a fuller representation of technology in compensation for its weaker macroeconomic linkages. Sensitivity studies with all these models illustrate the crucial importance of technological assumptions.
To establish reasonable assumptions, it is pertinent to start with current data, and visible trends and options. With respect to supply-side options, data such as that collated and summarized in many publications [e.g. Refs. 18- 22] for review of sources see UNEP (16)] illustrates the immense range of possibilities. They span technologies that are proven and largely mature (such as combined-cycle gas turbines CCGTs), proven but still developing (such as wind energy and higher-efficiency clean coal conversion), confidently predicted [such as much cheaper photovoltaics (PV) systems and integrated biomass gasification], to a wide variety of lesser or more speculative options. Most recently, an immense study of the prospects for modernized renewable energy technologies (111) argued that these could meet about half global energy demand by the middle of the next century at little if any additional cost; the studies of wind energy in this volume estimated that the costs of modern wind turbines were already almost competitive against coal power stations for large-scale exploitation in countries such as the United States with extensive wind energy resources. None of the CO2 abatement models contain explicit representation of wind energy technology, and for large countries such as the United States and the former Soviet Union, this alone might substantially lower abatement cost estimates.
No models can capture the full range of options available; by inevitably excluding some, there can be an in-built tendency to overestimate abatement costs (unless this is offset by using over-optimistic assumptions concerning those that are included). Some abatement studies use data already being rendered obsolete as options already identified for cost reductions are exploited. Williams (100) provides one of the most detailed critiques of the technological assumptions used in the Manne & Richels base case assumptions and their derivatives, and argues that many of the assumed costs are higher than can already be predicted with confidence.
This leads naturally to the issue of technology development and cost reductions. This is uncertain terrain. but not a complete black box. Technologies do not arise, improve, and penetrate markets at random, especially for large and complex technologies such as those involved in energy provision. Technology development follows market demand, with the associated public and private R&D investment and learning processes as technologies develop towards market maturity. Yet despite this well-attested and understood fact, almost all the abatement costing studies to date model technology development as "exogenous"--the costs of abatement technologies are defined as input data and do not vary explicitly with the level of investment, incentives, and market penetration in the model. That alone must be considered as a severe limitation.
Anderson & Bird (45) provide one of the few abatement cost analyses to date that explicitly includes production scale economies. They apply this to renewable technologies within a simple investment/growth model of global economic expansion. Their analysis produces lower cost estimates than most of the more complex studies that model the economy and energy sector in more detail, but wholly neglect the issue of technology development.
A very different approach to the issue of endogenous technological development is that by Hogan & Jorgenson (93) . This econometric study related changes in productivity trends (which are equated with technical progress), in different sectors of the US economy, to price changes in the different inputs. Although energy productivity did improve when energy prices rose, this was more than offset by reduced productivity growth in other factor inputs at the same thile. They found that overall. "technology change has been negatively correlated with energy prices . . . if energy prices increase, the rate of productivity growth will decrease." However, such results may be very sensitive to the model specification, and as argued in Grubb (79, Chapter 3), the transition from "has been" to "will" in this excerpt conceals the importance of innumerable extraneous factors in the years analyzed, most notably the macro-economic impact of the sudden and externally imposed oil price shocks. It is highly debatable whether the data reveal anything useful about the economy-wide and long-term technological impact of smoother price changes arising from domestic policies such as carbon taxes, and other abatement policies .
The potential for substantial cost reductions associated with larger-scale deployment of low CO2 technologies, combined with the observations above about possible irreversibilities in the impact of price changes, points to the possibility that there may be various choices of technological trajectories differing little in cost. One is to continue along a carbon-intensive path. Another is to invest enough to alter the course of new investments over the next decade or two towards more efficient, and low-carbon, technologies. As investment patterns and institutions and infrastructures adapt to these new technologies. their costs will fall, perhaps until they become the naturally preferred options. The world would be on a different technological trajectory. Although the transition may be costly, especially if it is forced rapidly, given the nature of technology development and economies of scale it cannot be assumed that this would be a much more costly long-term path (147).
This is an example of the issue of "bifurcation," identified especially by Hourcade (112) as a concept that "encompasses many network industries where market forces tend, beyond a bifurcation point, to reinforce the first choice ... in a self-fulfilling process." Hourcade highlights that this is not only a matter of scale economies; once investment is made in transport infrastructure or town planning, for example, it attracts a major network of other investments that reinforce the original choice and make later changes much more costly than if development occured along a different trajectory. Hourcade develops a technology-oriented economic model and projects various scenarios that differ by up to 50% in long-run CO2 emissions for the same estimated costs.
The prospects for technology development, production scale economies, and exploitation of bifurcations to lower emissions suggest that the cost estimates in many economic studies are implausibly high. Indeed, some use data that appear to indicate costs higher than those of some currently identified technologies, and make little or no allowance for future technical improvements especially in nonfossil sources.
We do, however, note that there are considerable constraints on the rate at which such technologies could be developed and deployed; as modelled most clearly by Anderson & Bird (45), the deployment of major new supply technologies will take many decades.
On these grounds we consider the higher long-term (beyond c.2025) costs illustrated on the graphs in Section 4 to be relatively implausible.
KEY DETERMINANTS: ABATEMENT STRATEGIES AND SCOPE OF ANALYSIS
Even when issues relating to technology costs and deployment are put on a comparable basis, there are many other important sources of difference among economic modelling studies arising from the form of abatement strategy and the scope of analysis. In this section we examine the most important of these.
Subsidies, Tax Forms, and the lJse of Tar Revenues
Various forms of taxes (and subsidies) can be used to limit emissions. Different types of taxes lead to different reductions in C02 and to different impacts on the economy [Scheraga & Leary (113)]. Most of the economic models considered assume abatement to be achieved by a carbon tax, imposed on the carbon content of primary fuels. However, taxes could be applied to subsets of fuels, downstream on derived fuel products, or otherwise not in proportion to carbon content of fuels, e. g. on gasoline only or on the energy content of fuels . This generally results in greater economic costs (lower tax efficiency). Thus, a gasoline tax is less efficient than a carbon tax at reducing carbon omissions (95); and taxes on electricity production are much less efficient than a tax on input fuels; the latter do not encourage fuel switching, only reducing CO2 by depressing demand for electricity (114).
The distributional effects between countries vary greatly according to whether the tax is imposed by producers or by the consuming countries [Whalley & Wigle (37)]. Carbon/energy taxes also have substantial distributional effects within countries, as they frequently have greater impact on the poor and always have greater impacts on energy-producing sectors. Distributional impacts can often be offset by accompanying measures that redistribute some of the revenue back to those adversely affected.
Of greater relevance to the assessment of total abatement costs is that energy production in most countries is subject to a complex set of taxes and subsidies, and abatement costs inevitably depend on the existing tax structure. Where heavy taxes are already imposed (as with oil products in many OECD countries), the macro-economic impact of additional taxation is likely to be greater than in the absence of existing taxation; conversely, where energy is subsidized, removal of subsidies (or equivalent taxation) will often yield macro-economic benefits. Many models ignore initial subsidies and taxes.
The OECD (115, 116) and the World Bank (117) identify a range of energy subsidies, widespread outside the OECD but also widely applied to coal in OECD countries. These subsidies amount to an estimated $235 billion, equating to a carbon subsidy of $92 per ton outside the OECD and indicating a significant potential for limiting emissions at net economic benefits by removing subsidies. These figures are dominated by the former Soviet Union ($160 billion), which is anyway undergoing radical price reform, but the potential impact in countries such as India and China is clearly important. however, in such countries the distributional impact of removing subsidies is especially severe because of the poverty and lack of any social security protection; social impacts and political constraints are central in practical considerations. We also note that structures of subsidies and taxes are not arbitrary; oil taxes in OECD countries, for example, reflect perceived external costs associated with dependence on foreign oil.
Where net taxes are applied, the impact on GNP depends on how the revenues generated are used. Yamaji (118) assumed that the carbon tax revenue left the Japanese economy, likening an imposition of a carbon tax to the oil price shocks of the 1970s. The resulting estimates of the impact on GNP (a 5% loss) are much larger than if most of the revenue were kept within the economy. The US Congressional Budget Office (78) compared the effect of a revenue-raising tax (which removes money from the economy to reduce the federal budget deficit) with a fiscally neutral carbon tax, and found the impact on GNP to be much lower in the latter case
The short-term CBO studies (78) have since been complemented with more broad-ranging and longer-term studies of tax-recycling issues, notably Bradley et al (67), Brinner et al ( 120), Scheraga & Leary (113), the European Commission ( 121), and Shackleton et al (76), which show evidence that while initially depressed, GNP could be raised in the long term by some recycling strategies These papers explore the implications of a variety of tax recycling options Table 7 illustrates key results, qualitatively because of the uncertainties in these (mostly shorten macro-economic) models, and the extent to which the impacts vary over time, as discussed in Section 5.2 Figure 13 shows the results of two alternative tax recycling options (lumpsum rebate and investment tax credit) for four different models as compared by Shackleton et al (76)
If the tax revenues are taken out of the economy (e.g. unaccounted for or spent abroad), all impacts on the national economy are negative. For other uses of the tax revenues, different macro-economic indices frequently move in different directions. For the carbon tax levels considered (up to about $80 per ton C), nearly all these studies find some ways in which the net effect is to boost GNP, as compared to a projection in which existing tax structures are unchanged.
These results reflect at least two different factors. First, a carbon tax raises money largely from consumption; this may be transferred to qualitatively different economic activities. If the reverse are used to stimulate investment directly, this reduces consumption but soon increases GNP. If the revenues are used to reduce budget defieits, consumption and GNP are initially depressed but they may slowly recover as the lower interest rates etc improve the investment climate; Brinner et al (120) suggest that the GNP impact in the United States becomes positive within 15 years, although EC studies suggest a slower recovery. Conversely, if the revenues are used to boost consumption, investment is diminished and shortly thereafter GNP growth. It is doubtful the extent to which the gains or losses should really be credited to CO2 reduction, since they partly reflect a channeling of re
