Policy Analysis

Carbon Emissions from U.S. Ethylene they are simultaneously produced from several processes
that rely on various feedstocks, each process produces a
Production under Climate Change unique yield of ethylene and propylene, coproducts, and
byfuels and requires varying levels of process energy. The
Policies amount of process energy use is largely dependent on the
choice of feedstock used in the steam cracker. Product yields
MATTHIAS RUTH* and process energy use can be varied by technology change
Environmental Policy Program, School of Public Affairs,
and process adjustments. For example, “high severity”
University of Maryland, 3139 Van Munching Hall, cracking, which is performed under higher temperatures and
College Park, Maryland 20742 shorter residence times, increases both the yield of ethylene
and the amount of process energy required.
ANTHONY D. AMATO Interdependencies among energy, feedstock, technology,
Environmental Policy Program, School of Public Affairs, and product mix can require complex adjustments by indus-
University of Maryland, 2100 Van Munching Hall, try to externalsmarket or policy-drivensinfluences with
College Park, Maryland 20742 results that (a) may not be obvious a priory, (b) manifest
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.
Downloaded via TEXAS A&M UNIV COLG STATION on December 20, 2024 at 02:57:18 (UTC).

themselves over years, and (c) influence the effectiveness of

BRYNHILDUR DAVIDSDOTTIR policy. Many of these interdependencies may be entirely
Department of Geography, Boston University, overlooked if one analyzes the chemicals industry as a whole,
675 Commonwealth Avenue, Boston, Massachusetts 02215 as is typically done in studies of climate change policy impacts
on industry. As a consequence of highly aggregate analysis,
policy recommendations may emerge that fall short in pro-
moting technology and environmental improvement. How-
This paper presents the results from a dynamic computer ever, disaggregate analyses of policy implications for indi-
model of U.S. ethylene production, designed to explore vidual chemicals products suffer from the need to limit
implications of alternative climate change policies for the technology choices to those that can be well described with
industry’s energy use and carbon emissions profiles. publicly available data.
The model applies to the aggregate ethylene industry but This paper presents a dynamic computer model to
distinguishes its main cracker types, fuels used as organize publicly available data and explore implications of
feedstocks and for process energy, as well as the industry’s selected climate change policies for feedstock and process
capital vintage structure and vintage-specific efficiencies. energy use of aggregate U.S. ethylene production as well as
corresponding carbon emissions. [Only a small set of policies
Results indicate that policies which increase the cost
and investment strategies are explored here, compared to
of carbon of process energyssuch as carbon taxes or the full potential of the model. To receive a copy of the model
carbon permit systemssare relatively blunt instruments for and explore your own “policy scenarios” send e-mail to
cutting carbon emissions from ethylene production. In [email protected].] The paper’s purpose is to contrast
contrast, policies directly affecting the relative efficiencies the effectiveness of policies which increase the cost of carbon
of new to old capitalssuch as R&D stimuli or accelerated of process energyssuch as carbon taxes or carbon permit
depreciation schedulessmay be more effective in systemsswith policies that directly affect the relative ef-
leveraging the industry’s potential for carbon emissions ficiencies of new to old capitalssuch as R&D stimuli or
reductions. accelerated depreciation schedules.
The paper is organized as follows. Section 2 discusses the
system boundaries applied to this study and its underlying
1. Introduction methodology. Section 3 presents an overview of data sources
The U.S. chemicals industry accounts for 25% of manufac- and econometric analyses used to specify the model. Section
turing energy use and 2.6% of the country’s carbon emissions 4 presents model results under a business as usual scenario
(1). One of the most energy-consuming steps within the and compares that scenario to results if alternative climate
chemicals industry is steam cracking of hydrocarbon feed- change policies and investment strategies are chosen. The
stocks to produce ethylene, propylene, butadiene, and paper closes with a brief summary and conclusions. Technical
aromatics (2). Since the U.S. production of ethylene alone details can be found in the Supporting Information.
accounts for approximately 28% of world capacity (3),
technology and policy choice in the U.S. are seen to have 2. System Boundaries and Methodology
far-reaching implications for the industry’s contributions to
The ethylene model traces for each individual cracker type
the global carbon cycle. However, even though steam cracking
(ethane, propane, naphtha, and butane) the flow of fuels
of feedstocks for ethylene production is highly energy
used as feedstocks, byfuels from ethylene production, and
intensive, a significantly larger share of the energy used in
purchased process energy required to produce ethylene
ethylene production is used as feedstock and ends up in the
(Figure 1). Since information on the industry’s process energy
finished product rather than directly leading to the emission
use is not directly available, we use industry cracker efficiency
of carbon into the atmosphere. The energy that is combusted
estimates along with historic production data to deduce the
during the steam cracking process is used for heating the
industry’s total energy use. From information on energy use
feedstock, compression, and separation of products.
we calculate annual carbon emission rates on the basis of
Ethylene, and to a lesser extent propylene, are the basic
historical data until the year 1998 and then project energy
components in the production of plastics in the U.S. While
use and carbon emissions profiles until the year 2020. The
* Corresponding author. Email: [email protected] Phone: model also traces indirect carbon emissions associated with
301-405-6075 Fax: 301-403-4675. the generation of purchased electricity but does not trace
10.1021/es010809r CCC: $22.00  2002 American Chemical Society VOL. 36, NO. 2, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 119
Published on Web 11/30/2001
 A key feature of the model lies in its explicit representation
of the industry’s capital vintage structure. The capital vintage
description of the model follows the perpetual inventory
methods pioneered by Jorgenson (4, 5). In that model, the
end-of period capital stock is estimated as a function of gross
new capital investments (It), existing capital stock (Kt-1), and
rate of replacement (µ), where all variables are measured in
monetary units:

Kt ) It + (1 - µ)*Kt-1

We modify Jorgenson’s approach by creating a physical

perpetual inventory. We do this by estimating the physical
size and vintage structure of the capital stock in the most
recent year available. The future size and structure of the
capital stock is then estimated as a function of gross new
capital investmentsstranslated into new physical production
capacitysand the size of the existing physical capital stock
minus the retirement of physical capital. In our adaptation
of this approach for the U.S. ethylene industry, information
on annual changes in capital stock is used to calculate
production capacity, giving an estimate of the age structure
of U.S. capacity to produce ethylene in physical units. Parallel
to the capital stock’s age structure the model traces a set of
age-specific efficiency characteristics by production process.
Changes in efficiency occur through the replacement of
old, retiring capital with new, more efficient capital. The
changing capital vintage structure gradually increases overall
efficiency of the industry and reduces the intensity of
feedstock and/or of process energy use of new capital relative
FIGURE 1. System components and boundaries. to existing capital stock in that year. Aggregate efficiency of
the ethylene industry is calculated as a weighted sum of
vintage-specific efficiencies of all processes, where the
weights are determined by production capacity of the
respective vintage class used in each process. The model
assumes that within a vintage class efficiencies remain
constant and that each capital vintage class is fully retired
after 25 years.
We explicitly trace the industry’s capital stock for two
main reasonssone theoretical, one practical. First, a growing
number of industry-specific economic analyses place em-
phasis on the role that capital vintage effects play for firms’
input choice, rate of technology change, and environmental
performance (6-9). Theoretical work dating back to Leontief
(10), Fisher (11), and Diewert (12) explores conditions under
which an aggregate capital stock can be defined and
emphasizes the relevance that relative efficiencies of capital
FIGURE 2. Model configuration. of different vintage classes do have for that definition (13).
Second, capital vintage effects have also been associated
energy used to produce final products from ethylene. Carbon with technology lock-in (14), which may impede rapid
emissions as a result of processes outside of the ethylene reductions of carbon emissions. The lock-in hypothesis posits
production process itself, such as the downstream production that incremental development of a firm’s knowledge base
of plastics from ethylene or the incineration of ethylene end over time produces standard operating procedures that can
products during waste disposal, are not accounted for by the be barriers to the adoption of new technologies. In addition,
model. since capital investments are generally financed from a firm’s
The key exogenous drivers to the model include historic own cash flow, internal investments tend to strengthen
and forecast energy prices, GDP, and fuel-specific carbon existing production processes and products. Thus, as a firm
coefficients. Each of these drivers, plus many of the functional grows older it may become less inclined to update its
relationships, can be interactively adjusted in the model to technologies and as a result closely follows a specific tech-
investigate alternative futures for the industry, making the nology trajectory. Under some conditions initial technology
model a versatile tool for investment and policy analysis of choices may gradually rigidify and lock in potentially in-
the U.S. ethylene industry. Given assumptions about exog- efficient processes. Developments at the firm level may be
enous drivers, the model endogenously calculates production accompanied, and even exacerbated, by lock-in of institutions
rates, capacity and investment requirements, efficiency that oversee or regulate an industry (15).
changes, energy and feedstock use, and carbon emissions as The role of industrial capital vintage structure in influ-
the result of interactions among the many different com- encing future investment choice and effectiveness of climate
ponents of the industry (Figure 2). Endogenous calculations change policies has led to calls for more detailed representa-
are based on econometric estimates and engineering infor- tions of capital vintage structures in models for investment
mation described in the next section and presented in more and policy analysis (16, 17). The explicit representation of
detail in the Supporting Information. the industry’s capital vintage structure in our study enables

120 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 2, 2002

us to elucidate the implications of a wider range of policies When excluding carbon from byfuel combustion, carbon
for carbon emissions, such as policies that are expressly emissions from ethylene production increase by 245%
targeted toward development and adoption of advanced between 1990 and 2020. When carbon emissions from byfuel
technologies, instead of simply dealing with policies that combustion are factored in, carbon emissions increase by
raise the cost of energy or carbon. 120% between 1990 and 2020. When we contrast these trends
with an expected increase of 50% in carbon emissions (net
3. Data Sources and Econometric Analysis of emissions from byfuels) per million pounds of product
The model components of Figure 2 and their interrelation- over the same time frame, it becomes apparent that byfuel
ships are specified on the basis of historical data and production is a crucial determinant for the industry’s carbon
engineering information. All resulting equations are solved emissions profile and that technology change occurs at faster
simultaneously with the use of a dynamic computer model rates than output increasesseven in the absence of new
specifically designed to handle large numbers of simultaneous policies.
difference equations which may contain time-lags and 4.2. Policy Scenarios. To explore potential impacts of
nonlinearities (18, 19). The model is calibrated over historic different climate change policies on energy use and carbon
time (1970-1998) and run to explore future trajectories until emissions we choose alternative policy scenarios that we
the year 2020. assume to be implemented in the year 2004. Several leverage
Parameter estimation techniques chosen for model points for policy exist to affect carbon emissions from
specification include multivariate regressions using both ethylene production. One strategy is to raise the cost of
simultaneous and lagged independent variables as well as purchased process energy on the basis of the carbon content
seemingly unrelated regressions. Model (parameter) speci- of fuels in hopes of stimulating energy savings and thus
fication is based on conventional hypothesis tests (t, F-test, emissions reductions. Another strategy is to directly stimulate
and R 2) as well as extensive econometric diagnostics such development and diffusion of more energy efficient technol-
as Lagrange multiplier tests for heteroscedasticity (20) and ogy, for example, by boosting research and development of
serial correlation (21). Estimation results as well as basic new equipment and providing tax or other incentives to
engineering information on the relative intensity of feedstocks industry to more rapidly adopt such equipment.
and process energy, feedstock shares and ethylene yields, A cost of carbon increase is explicitly modeled as a price-
purchased process energy use, and carbon contents of fuels adder for fuels, where the magnitude of the adder depends
are provided in the Supporting Information. on the fuels’ carbon content. Effects of technology-directed
policies are captured implicitly by decreasing the intensity
4. Scenario Analysis and Results of process energy of new capital relative to the process energy
4.1. Base Scenario. It is the purpose of the model to explore intensity of the existing capital stock (the so-called relative
likely trajectories for process energy use and carbon emissions process intensitysRPI). The effects of these and similar
under business as usual scenariossassuming continuation strategies can be explored in the model in isolation from
of trends embedded in the time series datasand to compare each other or in various combinations. However, each
these trajectories to those that would result under alternative scenario must assume that the structure of the industry
climate change policies. The goal of such a comparison is remains unchanged (i.e. no fundamentally new feedstocks
the identification of effective leverage points for policy and or technologies emerge to produce ethylene, capital equip-
investment decisions to influence carbon emissions from ment has at the most a 25-year lifetime, and markets for
ethylene production. While the model is based on a set of inputs and products continue to be competitive) and that
simplifying assumptions, many of which are the consequence producers do not foresee the advent of new policies. Allowing
of insufficiently disaggregated, publicly accessible industry the model to capture fundamental changes in the industry’s
data, it performs well in tracking historic trends and clearly production processes would require specification of those
suggests the extent to which alternative policies affect process processes in terms of their fuel mix and energy and feedstock
energy use and carbon emissions from the production of efficiencies. Where such descriptions exist they can be readily
ethylene. used to respecify the model. Given the capital-intensity of
The base scenario assumes that no new policies are the ethylene industry, new technologies will take time to
implemented and that new capital is 4% more efficient than diffuse. If new capital stock were needed to handle, e.g.,
the industry’s aggregate existing capital stock producing biomass feed (ethanol, methanol) or to conduct back-to-
different efficiencies for vintage classes. For example, in the polymer recycling, the associated changessin the ethylene
year 2020 the gap between ethane crackers just coming on industry aggregateswould not be felt for some time. In
line and those about to be retired is 7.2%. While this efficiency contrast, technologies such as gas-turbines placed in front
gap is not large, it is consistent with the trends in the broader of the cracker (turbine off-gases with temperature of 500 °C
chemical industry where improvements in energy efficiency are used in the burners instead of combustion air) do not
have remained relatively flat since the late 1980s (22). require that the capital stock has turned over and can be
Ethylene outputsspecified as a function of GDPsreaches added to the existing stock and thus more readily influence
almost 83 billion pounds by 2020 in the base scenario. Total efficiencies. Similarly, the use of new catalysts by ethylene
process energy usesincluding byfuels produced during producers could lower energy consumption by 20% (23)
productionsincreases by 109% between 1990 and 2020, while without requiring major turnover of the existing capital stock.
the total purchased process energy use increases by 433%. In reality, increased energy prices could make investments
The reason for this difference is that processes that are more into such technologies cost-effective.
efficient at producing ethylene are less efficient in the Changing the assumption that the maximum lifetime of
production of byfuels. Reduction in byfuel production results capital equipment is 25 years seems mute in light of his-
in an increase in the use of purchased process energy. torical trends in the industry. Allowing the model to ac-
However, we must note that purchased process energy commodate changes in market structures and producer
accounted in 1990 for only 1.6% of total process energy use expectations would require, for example, game-theoretic
in the ethylene industry. We expect process energy use per approaches that describe economic behavior under alterna-
million pounds of output to decrease by 9% between 1990 tive market conditions and behaviors of industry decision
and 2020. Over the same time frame, purchased process makers. To date, too little well-grounded information is
energy use per million pounds of output is expected to available to pursue that line of research for the ethylene
increase by 132% due to reasons detailed above. industry.

VOL. 36, NO. 2, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 121

TABLE 1. Summary of Model Results
base $75/ton RPI
scenario carbon 0.940
total production (% change from 1990 levels to 2020) 130.1 130.1 130.1
total purchased energy use (% change from 1990 levels to 2020) 432.6 428.5 348.6
total process energy use (% change from 1990 levels to 2020) 108.7 108.8 107.4
total process energy use per output (% change from 1990 levels to 2020) -9.3 -9.2 -9.8
total purchased process energy use per output (% change from 1990 levels to 2020) 131.5 129.7 95.0
years gained in efficiency improvement 0 3
gross carbon emissions (% change from 1990 levels to 2020) 120.3 120.2 118.4
net carbon emissions (% change from 1990 levels to 2020) 244.5 242.9 211.4
net carbon emissions per ton of output (% change from 1990 levels to 2020) 49.8 49.1 35.3
Financial Parameters
cumulative present value of energy expenditures for purchased process energy, 1.343 1.549 1.320
2000-2020 (billion 1994 $, 5% discount rate)
cumulative present value of cost of carbon payments, 2000-2020 0 201.1 0
(million 1994 $, 5% discount rate)
energy expenditures purchased process energy in 2020 (net of cost of 112.7 112.2 102.2
carbon, where applicable)
(million 1994 $)
cost of carbon in 2020 (million 1994 $) 0 45.632 0
energy cost per unit of output in 2020 (1994 $/short ton output) 1.343 1.337 1.217
(net of cost of carbon, where applicable)

FIGURE 3. Purchased process energy use with and without climate

FIGURE 4. Purchased process energy use per pound of output with
change policies.
and without climate change policies.
Given the prescribed structure of the industry and
available technologies, the following discussion contrasts a because of lower energy prices, the rate of improvement in
cost of carbon policy with policies that improve the relative both feedstock energy efficiency and process energy efficiency
process intensities of new to existing capital (RPI). Results declines as well. This in turn causes an increase in byfuel
indicate that reducing RPI marginally (from 0.960 to 0.959) production, triggering a decline in the use of purchased
achieves the same total carbon emissions in 2020 as a process energy.
$75/ton increase in the cost of carbon. A slightly more Carbon emissions from purchased process fuels (exclud-
ambitious improvement in RPI from 0.960 to 0.940 leads ing emissions from byfuel combustion) will continue to
the industry to achieve an approximately 20% reduction in increase but will increase less rapidly than in the base scenario
carbon emissions by 2020 over those in the base scenario (Figure 5). The same trend can be seen for carbon emissions
and the same aggregate energy efficiencies in the year 2017 from purchased process fuels (excluding emissions from
as would be achieved with an RPI of 0.96 in the year (cf. byfuel combustion) per million pounds of output (Figure 6).
Table 1).
The model indicates that if cost of carbon or RPI-directed 5. Discussion
policies are implemented, purchased process energy uses This paper presents selected results from a dynamic computer
both total and per unit outputscontinue to increase over model of U.S. ethylene production, specified on the basis of
time but at a lower rate than in the base scenario (Figures historic data, engineering information, and forecasts of GDP
3 and 4 and Table S-5 (Supporting Information)). The reason and energy prices. The purpose of the model is to highlight
for the difference in the impact of an increase in the cost of key leverage points for policy or investment decisions which
carbon on purchased process energy compared to the have the goal of reducing the industry’s carbon emissions
impacts of policies affecting RPIs is due to the different ways through process efficiency improvements. Specific attention
in which these policies would influence the relationship is given in our model to the industry’s capital vintage structure
between investment and the production of byfuels. If, for and the efficiencies of existing and new capital equipment
example, investment in energy efficient capital declines in using feedstock and process energy. An important

122 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 2, 2002

 energy efficiency by lowering fuel requirements to the
heater (24).
Our results from policies aimed at increasing the cost of
carbon for purchased process energy show limited potential
for emission reductions. However, if the cost of carbon is
increased not only for purchased process energy but also for
fuels used as feedstock, then significant reductions in
greenhouse gas emissions (GHG) may result. For example,
Greonendaal and Gielen (25) find for the European chemical
industry that increased GHG permit prices for all fuels has
the potential to significantly reduce emissions by 2030.
Emission reductions are found to be largely due to a shift
from fossil fuel-based to biomass-based feedstocks, which
have no net emissions, and to a lesser extent the recycling
of waste plastics by consumers. A comparison of model results
demonstrate that industrial emissions’ profiles are highly
sensitive to assumptions about how broadly cost of carbon
policies are applied to fuel inputs and what system boundaries
are chosen for emissions accounting.
FIGURE 5. Carbon emissions from purchased process energy with Introducing performance goals or technology standards
and without climate change policies. can produce noticeable changes in process and purchased
energy use and thus carbon emissions. Performance goals
or technology standardsscollectively reflected in our model
as impacting relative process energy intensities of new to old
capital (RPIs)smay be the product of voluntary or govern-
ment-led actions but will be most effective if they stimulate
provision of highly efficient capital to industry and lead to
an efficiency gap between new and existing capital for
industry to close. While there are clear long-term techno-
logical and thermodynamic limits on efficiency of new capital,
intensified R&D and technology adoption, combined with
disproportionally higher utilization rates of newer rather than
older capacity, can go a long way in establishing and then
exploiting the efficiency gap. Investment tax credits and
accelerated depreciation schedules are but two examples
that could help leverage the effectiveness of performance
goals and technology standards. The study results clearly
indicate that such policies are more effective than mere
increases in the cost of carbon in cutting carbon emissions.

Acknowledgments
FIGURE 6. Carbon emissions from purchased process energy per
million pounds of output with and without climate change policies. This project was made possible by support from U.S.
Environmental Protection Agency under Grant Number X
826822-01-0 and benefited from many discussions with John
distinction is made between purchased process energy and
“Skip” Laitner and Ernst Worrell. However, the paper does
process energy from byfuels.
not necessarily reflect the views of U.S. EPA or those of the
Though this and any other publicly available study of the individuals who provided input into, or commented on, the
ethylene industry are severely hampered by the lack of model on which this paper is based.
industry-specific energy and emissions data, no assessment
of alternative policies will ever have all the data it needs. But Supporting Information Available
herein lies the beauty of dynamic modelingsan ability to
build on the best available data and to explore the likely Key functional relationships, econometrically estimated
consequences of alternative assumptions. As the ethylene parameters, and engineering information used in the dy-
model above shows, those consequences are fairly distinct namic, capital vintage model of U.S. ethylene production.
if policy increases the cost of carbon or directly affects the This material is available free of charge via the Internet at
turnover of capital in favor of more efficient equipment. http://pubs.acs.org.
Our results indicate that increasing the cost of carbon for
the ethylene industry produces only small changes in carbon Literature Cited
emissions due to the structure of the industry. Almost all (1) U.S. Office of Industrial Technologies. Industry Profile; 2000;
fossil fuels purchased by the industry are incorporated into http://www.oit.doe.gov/chemicals/page11.shtml.
the product itself as a feedstock. If the cost of carbon is (2) ACS, AICE, CMA, The Council for Chemical Research, U.S.
increased only for fossil fuels that provide process energy, Department of Energy, and U.S. Environmental Protection
the industry has little incentive to change its behavior since Agency. Workshop on Carbon Reduction Technology Op-
these fuels are only a small percentage of total monetary portunities for Chemical Process Industries, Background Docu-
ments, Vol. 1 and 2, March 3-4, 1999; Baltimore, MD, 1999.
outlays. Energy efficiency will largely be determined by (3) 3.Oil Gas J. Pennwell Publishing: various years.
feedstock choice, which in turn remains based on market (4) Jorgenson, D. W. J. Political Economy 1968, 76(3), 1123-1151.
demand for ethylene and its coproducts. To induce sig- (5) Jorgenson, D. W. Investment; MIT Press: Cambridge, 1996.
nificant changes in the energy profile of the industry, changes (6) Forsund, F. R.; Hjalmarsson, L. Econometrica 1983, 51(5), 1449-
in the capital stock need to occur. For example, the inte- 1467.
gration of gas turbines can lead to further improvements in (7) Sterner, T. Energy J. 1990, 11(2), 155-167.

VOL. 36, NO. 2, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY 9 123

 (8) Jacques, C.; Lafrance, G.; Doucet, J. A. Energy Policy 2001, 29, (19) Hannon, B.; Ruth, M. Dynamic Modeling, 2nd ed.; Springer-
453-463. Verlag: New York, 2001.
(9) Ruth, M.; Amato, A. Energy Policy 2001, forthcoming. (20) Breusch, T.; Pagan, A. Econometrica 1979, 47, 1287-1294.
(10) Leontief, W. Econometrica 1947, 15, 361-373. (21) Godfrey, L. G. Econometrica 1978, 46, 1293-1302.
(11) Fisher, F. Rev. Economic Studies 1965, 32, 263-288. (22) U.S. Office of Industrial Technologies. Energy and Environmental
(12) Diewert, E. Aggregation Problems in the Measurement of Capital. Profile of the U.S. Chemical Industry, 2000
In The Measurement of Capital; Usher, D., Ed.; University of (23) Martin, N.; Worrell, E.; Ruth, M.; Price, L.; Eliot, R. N.; Shipley,
Chicago Press: Chicago, IL, 1980; pp 433-528. A. M.; Thorne, J. Emerging Energy-Efficient Industrial Technolo-
(13) Doms, M. E. Economic Inquiry 1996, 34, 78-92. gies; LBNL 46990; 2000.
(14) Arthur, B. W. Increasing Returns and Path Dependence in the
(24) Albano, J.; Olszewski, E.; Fukushima, T. Oil Gas J. 1992, 90(6),
Economy; The University of Michigan Press: Ann Arbor, MI,
55-60.
1994.
(15) Unruh, G. C. Energy Policy 2000, 28, 817 - 830. (25) Groenendaal, B. J.; Gielen, D. J. The Future of the Petrochemical
(16) Jacobi, H. D.; Wing, S. I. Energy J. Special Issue, 1999, 73-92. Industry: A MARKAL-MATTER Analysis; ECN-C-99-052.
(17) Sands, R. D.; Edmonds, J. A.; MacCracken, C. N. SGM 2000:
Model Description and Theory; Draft, Pacific Northwest National Received for review April 2, 2001. Revised manuscript re-
Laboratory: 1999. ceived October 18, 2001. Accepted October 22, 2001.
(18) Ruth, M.; Hannon, B. Modeling Dynamic Economic Systems;
Springer-Verlag: New York, 1997. ES010809R

124 9 ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 2, 2002