Effects of the Uncertainty about

Global Economic Recovery on

a
Catholic University of Louvain (UCL), CORE, Voie du Roman Pays 34,
B-1348 Louvain-la-Neuve, Belgium.
b
IFP Energies nouvelles, Economics Department, 232 Avenue Napolon Bonaparte,
92852 Rueil-Malmaison, France.
Abstract
This paper examines the impact that uncertainty over economic growth may have on global energy transition
and CO
2
prices. We use a general-equilibrium model derived from MERGE, and define several stochastic
scenarios for economic growth. Each scenario is characterized by the likelihood of a rapid global economic
recovery. More precisely, during each decade, global economy may - with a given probability - shift from the
EIA's (2010) low-economic-growth path to the EIA's (2010) high-economic-growth path. The climate policy
considered corresponds in the medium term to the commitments announced after the Copenhagen
conference, and in the long term to a reduction of 25% in global energy-related CO
2
emissions (with respect
to 2005). For the prices of CO
2
and electricity, as well as for the implementation of CCS, the branches of the
resulting stochastic trajectories appear to be heavily influenced by agents initial expectations of future
economic growth and by the economic growth actually realized. Thus, in 2040, the global price of CO
2
may
range from $21 (when an initially-anticipated economic recovery never occurs) to $128 (in case of non-
anticipated rapid economic recovery). In addition, we show that within each region, the model internalizes
the constraints limiting the expansion of each power-generation technology through the price paid by the
power utility for the acquisition of new production capacity. As a result, in China, the curves of endogenous
investment costs for onshore and offshore wind are all bubble-shaped centered on 2025, a date which
corresponds to the establishment of a global CO
2
cap-and-trade market in the model.
1. Introduction
OECD countries are already committed to reductions in greenhouse-gases emissions by 2020. In
the longer term, these countries must in any case assume a leadership role in the field of emissions
reduction, in order for less-developed economies to accept reductions in their own emissions. This
implies significant restructuring of their energy systems, through the deployment of low-carbon
emissions technologies. However, this restructuring is contingent on considerable short and medium-
term investment in infrastructures whose lifetime is in general very long. In a climate of uncertainty,
such investments must offer sufficient expected returns to attract the necessary capital.
There have been relatively few studies using dynamic general-equilibrium models under
environmental constraints to determine equilibria that take into account the influence of uncertainty

1
Corresponding author. Tel.: +32 10 474358, fax: +32 10 474301.
Email addresses: [email protected] (Olivier Durand-Lasserve), [email protected] (Axel Pierru),
[email protected] (Yves Smeers).
2
This author has been financially supported by GDF-Suez. The results in this article do not necessarily reflect the
opinions of UCL and IFP Energies nouvelles. The authors gratefully thank Thomas F. Rutherford and Geoffrey J.
Blanford for granting full access to the code of the MERGE model.
2

on agents' behavior. Manne and Richels (1995) and Manne and Olsen (1996) study the impact of the
uncertainty surrounding the sensitivity of climate damages to greenhouse-gases concentrations.
Bosetti et al. (2009) and Durand-Lasserve et al. (2010) study the impact of uncertainty relative to
regional emissions-reduction targets. Bosetti and Tavoni (2009) consider the uncertainty over the
success of R&D activities concerning a backstop technology. De Cian and Tavoni (2010) evaluate
the effect of uncertainty in future carbon prices and technology costs. However, none
3
of these
contributions investigates the impact that uncertainty over global economic growth may have on
technological choices and resulting CO
2
price.
The recent economic and financial crises, which have particularly affected certain OECD
regions, have ushered in a period of great uncertainty. Above and beyond the question of short-term
business cycle, this uncertainty affects the capacity of these economies to grow at a sufficient rate in
the longer term. Several institutional analyses (OECD (2010), IMF (2010)) emphasize in particular
the long-term impact of the rise in unemployment linked to the crisis
4
, as well as concern over the
sustainability of the national-debt burden of some industrialized countries
5
. Consequently, the recent
crisis may be followed by one or more decades of very low growth, similar to the lost decades
experienced by the Japanese economy since the beginning of the 1990s. This uncertainty over
economic growth in industrialized countries, and more generally over global economic growth, may
represent a major obstacle to the executions of the investments necessary for change in the energy
mix. Thus, in developing countries with high economic growth (where elasticity of energy demand to
GDP is relatively high) this uncertainty may slow the growth of the energy supply. In industrialized
countries with moderate economic growth (and in the longer term on a global scale), this uncertainty
may slow the adaptation of the energy mix to targeted reductions in CO
2
emissions.
In his address to the 2010 World Energy Congress, Fatih Birol, Chief Economist of the
International Energy Agency (IEA), declared that the energy industry was confronted by a "level of
uncertainty that is unprecedented", placing the uncertainty linked to "the shape and pace of economic
recovery following the worst global recession since World War Two" at the highest rank of the
crucial factors that will have an impact on investment decisions in the energy sector6. This concern
is shared by the Energy Information Administration (DOE/EIA), for which "expectations for the
future rates of economic growth are a major source of uncertainty" (EIA (2010), p.20). To reflect this
uncertainty the EIA (2010) presents three scenarios for global economic growth by 2035. A result
put forward is the extreme sensitivity of energy demand to the economic-growth scenario.

3
Research works dealing with economic-growth uncertainty (e.g., Scott et al. (1999), Webster et al. (2008)) generally use
methods relating to Monte-Carlo simulation.
4
[] there continues to be a risk that at least part of the rise in unemployment since the crisis began will prove long-
lasting. (OECD (2010), p. 40).
5
In particular, new tail-risks have arisen from the growing concerns about longer-term debt sustainability in some
countries [] (OECD (2010), p.43).
Greece accounts for less than 3 percent of the European Unions total GDP, but signs of structural problems in the
economies of Spain, Portugal, Ireland, and to a lesser extent Italy may weigh heavily on the economic recovery of OECD
Europe as a whole [] (IEO, 2010).
6
"Some economists say the recovery will take a long time. Some say we will see a second recession. And some say that
we will soon see a strong rebound. This is important for decision-makers, especially for investment in the energy
industry." in An outlook of unprecedented uncertainty, European Energy Review, 20/09/2010,
http://www.europeanenergyreview.eu/index.php?id=2353.
3

This paper studies the impact that uncertainty over economic growth may have on the choice of
the global energy mix and resulting CO
2
prices. To do so, we define stochastic scenarios for
economic growth inspired by the EIA's (2010) scenarios. The uncertainty relates to the period during
which the global economy may once again take a path of solid growth. The climate policy
considered corresponds in the medium term to the commitments announced after the 2009
Copenhagen conference, and in the long term to a reduction of 25% in global energy-related CO
2

emissions (with respect to 2005). Using stochastic dynamic programming, we solve a general-
equilibrium model similar to that described by Durand-Lasserve et al. (2010), derived from MERGE
(Manne et al. (1995)). In this model, all agents are forward looking, in the sense that each
representative household (firm) maximizes its expected sum of discounted utilities (expected present
value) under rational expectations. In other words, on every date firms base their current decisions on
consistent subsequent prices and outputs on every possible outcome. These decisions are influenced
in particular by irreversibilities resulting from investments and technological choices. At economic
equilibrium, for each variable (for instance CO
2
, oil, gas and power prices) the model derives a
stochastic scenario consistent with that assumed for economic growth.
The following section more precisely describes the stochastic economic-growth scenarios
considered. The general-equilibrium model used is then described briefly, emphasizing the different
forms of irreversibility taken into account. The CO
2
emissions targets considered for the various
regions of the world are presented. Several comments on the calibration of the model form a
particular subsection. We then present and interpret the results obtained, identifying the impact of the
uncertainty under consideration on prices and on the global energy transition. Technological
trajectories, CO2 prices and electricity prices are studied in turn. Finally, we demonstrate that the
existence of constraints limiting the expansion of each power-generation technology is equivalent to
rendering endogenous the price paid by the power utility for the acquisition of additional production
capacity. For wind technologies, the formation of this endogenous investment cost is analysed.
2. Stochastic scenario for global economic growth
We consider that the global economy may experience a period of slow growth, before returning
to a high growth rate. The uncertainty centres on the number of decades of low growth (i.e. of "lost
decades", to borrow the expression used by Hayashi and Prescott (2002) about the Japanese
economy) through which the global economy must pass before experiencing an economic recovery.
More precisely, during each decade beginning in 2015, 2025 and 2035, global economic growth
may, with the probability p, escape from the path of low growth to shift to a path of higher growth
(on which it will subsequently remain). In other words, the higher the value of p, the more likely a
rapid economic recovery.
4


Figure 1: Possible GDP growth rates in OECD and non-OECD.

Figure 2: Structure of the stochastic scenario for each variable of the model.
These two possible growth paths correspond to the EIAs (2010) high and low economic-growth
scenarios. Each path defines specific economic-growth rates for each of the regions in the model.
On a global scale, the high-growth path implies growth rates around 1% higher than those
corresponding to the low-growth path, as shown by Figure 1. However, this differential varies
according to the region considered. Up to 2015, the economic growth is deterministic and
corresponds to the EIAs (2010) reference scenario. In all regions of the world, the two growth paths
begin to converge from 2035. From 2060 on, all regional economies grow at the annual rate of 2%.
5

For each variable (price or quantity), the model will derive a stochastic scenario consistent with
that considered for economic growth. This stochastic scenario may be represented by a tree with four
branches, each branch defining one possible for path the variable under consideration. In the
remainder of the paper, these paths are named R2015 (economic recovery in 2015, with probability
p), R2025 (economic recovery in 2025, with probability p(1-p)), R2035 (economic recovery in 2035,
with probability p(1-p)
2
) and NR (no economic recovery, with probability (1-p)
3
). A new branch
appears on each possible date of economic recovery, as shown in Figure 2.
3. Presentation of the model
3.1 General overview
The general equilibrium model SMERGE used, derived from MERGE (Manne et al. (1995)), is
essentially that described by Durand-Lasserve et al. (2010). The models horizon extends from 2005
(base year) to 2100, with 5-year time periods. The world is divided into 3 OECD regions (European
Union, North America, OECD Pacific) and 8 non-OECD regions (Africa, China, India, Latin
America, Middle East, non-OECD Asia, Russia, Rest of the World). The economic-growth
stochastic scenario is common knowledge and all agents in the model are equally informed and have
perfect foresight. In each region, a representative household maximizes the expected sum of
discounted logarithmic utilities of per-capita consumption over time and states.
The good consumed by households is a composite good that includes all intermediate and
consumption goods and services, as well as all the final energy needs of the economy (firms and
households). It serves as a numeraire, measured in terms of units of purchasing power for the year
2005. This composite good can be used for final consumption (by households), investment, and
intermediate consumption (in industrial sectors). This composite good is produced in each region by
a final industrial sector using various equipment vintages. At any given period, the production
function corresponding to a new vintage of installed equipment is described by a two-level nested
Constant Elasticity of Substitution (CES) function, using four inputs: labor, capital, electricity and
non-electric energy. However, once the vintage is installed, the respective quantities of these inputs
can no longer be adjusted. The resulting coexistence of different vintages in the final sector generates
irreversibility, since this sector is somehow locked in the short run by previous technical choices. For
instance, the choice of the energy intensity of new vintages until 2015 impacts the possible
substitutions after this period between capital and energy and between electricity and non-electric
energy.
Electricity and non-electric energy are supplied to the final sector by two distinct industrial
sectors that use various Leontief technologies to transform fossil fuels and composite goods into
energy. The technologies in the non-electric sector are: oil for direct use, gas for direct use, coal for
direct use, bio-fuels, synthetic fuels, and non-electric backstop (see Durand-Lasserve et al. (2010) for
a precise description). The power generation technologies are shown in Table 1.
Our model explicitly considers sunk investment costs for each power generation technology,
which renders irreversible the cost of a capacity increase. On this point, our model differs from that
used by Durand-Lasserve et al. (2010), who, for each technology at each period, consider a total cost
equal to a levelized cost multiplied by the production during that period. In their model, the absence
6

of unrecoverable investment costs was compensated for by a constraint limiting the possible rate of
decline in the use of the technology.
Moreover, in our model, investments are subject to time-to-build: 10 years (2 periods in the
model) for nuclear technologies with CCS and non-electric backstop, 5 years (one period) for the
other technologies. Considering time-to-build increases the uncertainty over energy demand and
prices when investment decisions have to be made. Note that, for all technologies, installed capacity
decays at an annual rate of 2%.
Table 1: Power generation technologies in the model (Sources: IEA (2008a), IEA
7
(2010a,b)).
Technology
Investment cost
c

($ per kWe)
Efficiency (%)
Emission rate
(TCO
2
/MWh)
Load factor (%)
2010 2050 2010 2050 2010 2050 2010 2050
Hydropower 2410-3210 3010-3330 - - - - 23-54 30-50
Remaining nuclear - - - - - - 85 -
New nuclear 2000-2500 3670-4000 - - - - 85 85
Remaining oil - - 30-54 - 0.57-0.87 - 18-49 -
Remaining coal - - 26-39 - 0.86-1.39 - 43-75 -
Remaining coal with
CCS
a
880-1390 - 20-33 - 0.09-0.14 - 70 -
New coal 700-1750 2530-2070 42-44 47 0.76-0.89 0.71-0.78 48-80 43-88
New coal with CCS
a
1510-3120 3140-3780 34-36 39 0.05-0.06 0.05 85 85
Remaining gas - - 36-54 - 0.37-0.62 - 20-54 -
New gas 600-700 820-900 55-59 63 0.34-0.40 0.33 41-64 40-64
New gas with CCS
a
1080-1260 1430-1490 47-51 55 0.02 0.02 85 85
Biomass 2130-2480 2120-2220 - - - - 25-71 65
Wind onshore 1390-1730 1410-1520 - - - - 18-24 30-33
Wind offshore 2570-3180 1670-1740 - - - - 35-39 39-40
Solar photovoltaic 3900-4650 1910-2020 - - - - 11-17 12-33
Electric backstop
b
6140-7220 3090-3410 - - - - 40 40
a
A 10$/T CO
2
transportation-and-storage cost is considered, in accordance with Rafaj and Kypreos (2007) and the (wide)
range of estimates given by IEA (2010b).
b
Includes concentrating solar power and marine energy.
c
In addition to
investment costs, variable and fixed non-fuel O&M costs are also considered (along with a variable feedstock cost for
nuclear and biomass power plants).
The model includes constraints on the maximum expansion of each technology. These
constraints, which take the form of a maximum growth rate in technology deployment from one
period to the other, reflect real-world frictions when installing new capacity. For new technologies,
they also cover the inertia inherent in the emergence of a new industrial sector, an inertia which may
be explained by the fact that new competing technologies face infrastructural, institutional and
cultural barriers (Khler et al. (2006)). In other words, the development and spread of any
technology is in essence a gradual process, which may for example be shown by an S-shaped curve

7
Main IEA (2010b) cost assumptions available at: http://www.worldenergyoutlook.org/investments.asp (accessed 5
January 2011).
7

(e.g. Knnla and Carrillo-Hermosilla (2008)) whose take-off and acceleration phases would be
mimicked by the constraints on expansion. These constraints therefore generate a path dependence -
in the absence of learning curves in the model - linked to the organization of the sector, which makes
adaptation of the energy system even more difficult during an economic recovery, especially if it was
not properly anticipated.
In each region, a single electricity demand is considered for each period. So as to consider an
energy mix compatible with an averaged demand, the maximum market shares of nuclear, wind and
solar technologies are limited (33% of total power generation for each). Similarly, a minimum of
10% of the total installed capacity must consist of technologies based on fossil fuels, as a
complement to intermittent technologies and because of the existence of peak demands. Furthermore,
capacities in hydropower and biomass are subject to exogenous availability limitations.
The fossil fuels are supplied by a mining firm that extracts oil, coal and gas from regional
reserves. All regions are linked together by the international trade in composite good, oil and gas. An
indexation of the variables (Meeraus and Rutherford (2005)) corresponding to the structure of the
stochastic scenario for economic growth replaces the explicit specification of non-anticipativity
constraints. Additional aspects of the calibration and numerical resolution of the model are described
by Durand-Lasserve et al. (2010).
3.2 CO
2
emissions policy in the model
Table 2 describes the gradual energy-related CO
2
emissions reductions targets
8
assumed for the
various regions of the world. In the long term, the objective is to reduce global emissions by 25%
between 2005 and 2050, with all regional quotas allocated from 2050 (and up to 2100) corresponding
to the same per capita volume of emissions. Based on the demographic projections used by the EIA
(2010), this per capita volume of emissions is 2.21 tons of CO
2
per year.
To attain this objective, OECD regions, Russia, China, the Middle East and Rest of the world
will have to decrease their CO
2
emissions between 2005 and 2050. Conversely, regions where the
current per capita volume of CO
2
emissions is low may increase their emissions (+372% for Africa,
+198% for India, +136% for non-OECD Asia and +48% for Latin America). These targets of
emissions reductions for 2050 have been established within the framework of a global cap-and-trade
market for CO
2
emissions (i.e. a global market with inter-regional trading of emissions permits) from
2025 onwards. From that date, during each period, a quota of emissions permits is attributed to every
region. The series of quotas allocated to the various regions are set so as to converge linearly towards
the same per capita volume of CO
2
emissions in 2050. Unused emissions permits can be banked, to
be used during a subsequent period. By consuming fossil fuels, the sectors supplying electricity and
non-electric energy produce CO
2
emissions. In each region, in any period, the quantity de CO
2

emitted by these two sectors must therefore be smaller than the regional quota, increased by the
emission permits banked in previous periods, and by the purchase of emissions permits from other
regions.

8
Our model which does not contain any modelling of the impact of GHG emissions on climate change implicitly
considers regional CO
2
emissions permit markets.
8

In addition to this long-term objective, there are regional targets of emissions reductions for
2020. These targets (WRI (2010a,b)) follow the commitments announced after the 2009 Copenhagen
conference.
OECD Regions have committed to reduce emissions by 2020: -20% for the European Union,
-17% for North America and OECD Pacific (relative to 1990 levels of emissions). In the model, each
of these regions implements (from 2010 to 2020) a regional cap-and-trade market (not linked to other
regions) and progressively decreases the allocated quota of emissions permits (i.e., the cap) so as to
converge linearly towards its 2020 emissions target. In each region, banking of permits is allowed
from 2010 onwards, with the option to use the banked permits within the global emissions-permits
market (from 2025 on).
Table 2: Regional emissions-reduction targets
2005 2020 2050
Level
(GT)
Emission
limit
CO
2

intensity of
GDP
Base
year
Level
(GT)
Change
2005-2020
Per capita
emissions
(T)
Level
(GT)
Change
2005-
2050
North America 6.88 -15% - 1990 5.71 -17%


2.21
1.42 -79%
European Union 4.10 -20% - 1990 3.32 -19% 1.29 -69%
OECD Pacific 2.19 -15% 1990 1.82 -17% 0.41 -81%
China 5.46 - -40% 2005 10.78
c
10.27
d
+97%
c
+88%
d
3.21 -41%
India 1.23 - -20% 2005 2.63
c
2.51
d
+113%
c
+103%
d
3.68 +198%
Russia 1.57 -15% 1990 1.83 +17% 0.25 -84%
Middle East 1.32 - - 2020
a
2.10 +60% 0.80 -39%
Non-OECD
Asia
1.50 - - 2020
a
2.01 +34% 3.53 +136%
Latin America 1.00 +25% - 2005
b
1.25 +41% 1.48 +48%
Africa 0.86 - - 2020
a
1.10 +28% 4.06 +372%
Rest-of-the-
world
1.20 -20% - 1990
b
1.35 +12% 0.36 -70%
World 27.33 - - - 34.06 33.44 +25% +22% 20.49 -25%
Notes
a
Reference case of IEA (2008b) in 2020. This figure only serves to define the series of quotas from 2025 to 2050.
b
Latin America (Rest of the world) takes into account Brazils (Ukraines) commitments.
c
Figure for the R2015
scenario.
d
Figure for the non-R2015 scenarios. Sources: WRI (2010a,b), IEA (2007a, 2008b, 2010a).
China, India, Russia, Latin America and the Rest of the world are also committed to reduce
emissions by 2020. Each commitment is accounted for in the model by a constraint on the emissions
volume in 2020. The commitments made by China and India (WRI 2010a) are expressed in the form
of a decrease in CO
2
emissions per unit of GDP. These commitments therefore depend on the path
followed by economic growth. As this is uncertain, the model takes account of the two possible
objectives in 2020 for Chinese and Indian emissions.
For Africa, non-OECD Asia and Middle East (who have not made no commitment for 2020), the
2020 level of emissions in the IEA's (2008b) reference scenario is used as a starting point for
calculating regional quotas from 2025 to 2050.

9

3.3 Special issues relating to calibration on a stochastic economic-growth scenario
The efficient labor (i.e., labor productivity
9
multiplied by the population) is a key driver of
economic growth in SMERGE. As a consequence, to calibrate the model on the stochastic economic-
growth scenario described by Figure 1, we consider the same stochastic scenario for the growth rate
of efficient labor. At equilibrium, this calibration yields regional economic growth rates that are very
close - but not identical - to the targeted values. As a matter of fact, the gross domestic product of a
region is not equal to the production of composite goods, principally because of intermediate
consumption and the added value generated by the three other industrial sectors (electricity, non-
electric energy and - especially in the Middle East - mining). The economic growth rate also depends
- at least to a certain extent - on the availability of energy technologies and the environmental
constraints specified in the model. In the version of SMERGE used here, the parameters reflecting
Autonomous Energy Efficiency Improvement
10
(AEEI) have been calibrated on an energy-intensity
scenario that is consistent with EIA's (2010) projections.
More generally, let us stress that the joint maximization of all regional objective functions is
intended to portray the market equilibrium consistent with a given (stochastic) view of the future.
The choice of the corresponding parameters flows more from a descriptive than a prescriptive
approach. Thus, the different regional rates of time preference for utility are chosen so as to respect
economic-growth differentials between regions (corresponding to the calibration scenario). In fact, in
the model, the perfect mobility of the composite good is equivalent to the perfect mobility of capital.
Therefore, at equilibrium, all regional interest rates are equal. According to Ramsey rule (e.g.
Durand-Lasserve et al. (2010)), the sum of the regional rate of time preference for utility and of the
rate of growth in regional per capita consumption is therefore the same for all regions. If we consider
that growth in regional per capita consumption and regional rate of economic growth are very
close
11
, this implies that the differential in economic growth between two regions is approximately
equal to the difference between the two regional rates of time preference for utility. On this point, it
is interesting to note that, in a rather unusual way, the stochastic economic-growth scenario adopted
here leads us to consider regional rates of time preference for utility themselves to be stochastic
(since the differentials in regional economic-growth rates depend on the branch of the stochastic
scenario under consideration).

9
As emphasized by Jacoby et al. (2008), p.615, about CGE models, "it is well established that economic growth cannot
be explained only by the growth of labor and accumulation of capital. A residual productivity factor always
remains.""this phenomenon conventionally is represented in one of two ways either a change in total factor
productivity or as an increase in labor productivity".
10
All sources of non-price-induced reductions in the energy required per unit composite good are summarized in
MERGE by the AEEI parameters, which operate as scaling factors on the energy input into production. As Richels and
Blanford (2008) stress, these reductions may occur due to both technological progress (e.g. end-use efficiency) and
structural changes in the economy (e.g. shifts away from manufactured goods towards services).
11
At the steady state of a stylised growth model in which the efficient labor grows at an exogenous rate, consumption
and Gross Domestic Product both grow at this same rate (see for example Aghion and Howitt (1998) p. 22).
10

4. Results and interpretation
For a better understanding of the impact of the uncertainty surrounding economic recovery, the
equilibrium of the model is determined for each of the following five values of the probability p: 1
(deterministic scenario with R2015 certain), 0.8, 0.5, 0.2, 0 (deterministic scenario with NR certain).
4.1 Achieved economic-growth scenarios
For the various values of p under consideration, the stochastic scenario for GDP growth obtained
at equilibrium replicates relatively well the stochastic scenario of efficient labor used to calibrate the
model. This is illustrated in Figure 3 for 0.5 p = . The fact that GDP grows a little more than efficient
labor during the initial periods is explained by the regional AEEIs. Energy expenditure actually rises
less rapidly than the output of the final sector, even though the latter grows almost at the same rate as
efficient labor. For all values of p, the economic growth differential between the high and low
branches nevertheless remains at around 1%, in line with the desired stochastic scenario. In concrete
terms, all the agents in the model form their expectations by considering the stochastic scenario of
economic growth actually achieved at equilibrium.

Figure 3: Efficient labor (plain lines) and achieved GDP growth (dotted lines) at equilibrium in
OECD and non-OECD regions for 0.5 p =
4.2 CO
2
price trajectories
Figure 4 shows the CO
2
prices in the European Union - corresponding to the global-market price
from 2025 onwards - in the various scenarios. The CO
2
price appears to be very sensitive to the path
11

followed by economic growth, as well as to the agents' initial anticipations (i.e. the value of p). In
2010, the price per ton of CO
2
ranges between $26 (p=0) and $29 (p=1). According to the scenarios
and branches under consideration, the price varies between $35 (p=0.5) and $65 (p=0.2) in 2020,
between $33 (p=0.8) and $80 (p=0.2) in 2030, and between $25 (p=0.8) and $163 (p=0.2) in 2045.
Let us note that in the case of no recovery (the NR branch), the CO
2
price in 2045 is extremely
sensitive to initial anticipations: $78 if non-recovery was perfectly anticipated (p=0), but $25 if it
was very poorly anticipated (p=0.8).
An essential difference exists between the R2015 and NR branches. In the R2015 branch, in
terms of decision-making, the role of anticipations (i.e. the value of p) is limited to the 2010 and
2015 periods, since from 2020 onwards no uncertainty remains, with economic growth remaining
high. On the other hand, in the NR branch, the uncertainty surrounding a possible economic recovery
persists until 2035. Up to that date, during each period, the uncertainty consequently has two
(opposite) effects on the CO
2
price: the effect resulting from past decisions taken in anticipation of a
possible rapid recovery which in the end did not take place ("past precaution effect") and the effect
of current decisions taken in anticipation of a possible future recovery ("current precaution effect").
The past (current) precaution effect leads to a current CO
2
price that goes lower as p gets higher
(smaller).
This perspective may for example explain why, in the R2015 branch, the CO
2
prices obtained in
the various scenarios converge quite rapidly (from 2025 onwards) towards neighbouring values, with
the past precaution effect being limited to the decisions taken in 2010 and 2015. In the same way, the
figures given in the previous paragraph show that the lowest CO
2
price (still drawn from the NR
branch) is obtained overall for a probability that is higher the further away the date considered is. So,
in 2010, the only thing that matters is the current precaution effect, which leads to less and less
banking of emissions permits as the probability of a rapid recovery gets lower. The lowest price is
thus obtained for p=0. However, when we move further away in time, the past precaution effect
becomes more and more significant relative to the current precaution effect. In the NR branch in
2040, when there is no more uncertainty about the economic growth to come, only the past
precaution effect has an impact on the CO
2
price, which is therefore at its lowest ($21) when a rapid
economic recovery has initially been considered highly probable (p=0.8).
In all scenarios, emissions permits are banked right from the initial periods. This banking
connects the price of an emissions permit from one period to another, so that it then follows the
Hotelling rule. In 2020, each of the two possible branches of the scenario yields specific emissions-
permit price and interest rate (between 2015 and 2020). As a result, the price of an emissions permit
in 2015 is equal to the expected value of its price in 2020 discounted at the corresponding interest
rate.
In the R2015 branch, as one might expect, the rise in the emissions permit price is higher as the
value of p gets lower, i.e. the economic recovery has been poorly anticipated. When p=0.2, the
permits banked beforehand are used during the 2020 period to mitigate the shock on the CO
2
price.
The price of an emissions permit is therefore $65, compared to a price of $47 when economic
recovery has been anticipated perfectly (p=1). This impact is nevertheless transitory, both because
the energy system adapts and because of the "hot air" generated by non-OECD countries entering the
global cap-and-trade market in 2025 (with Chinese and Indian emissions quotas higher in the R2015
12

branch). Thus, in the R2015 branch of the p=0.2 scenario, during the 2025 period the European
Union buys on average 287 million emissions permits per year (8.8% of its total emissions).
Conversely, the absence of economic recovery, whereas such a recovery had been considered
probable, leads to a lower emissions-permit price in 2020. This patternimpact on the permit price
if the realized path (economic recovery occurring or not) was initially poorly anticipatedis
repeated in 2030 and 2040. Finally, when the absence of economic recovery was anticipated
perfectly (p=0), the CO
2
price is lower in 2025 than in 2020, which leads to all the previously banked
permits being consumed in 2020.
When the economic recovery is initially seen as fairly probable (p0.5), the European energy
system prepares itself relatively well. This enables continued banking of emissions permits in 2020,
with the permit price rising at the interest rate between 2020 and 2025. In fact, where p0.5, all
OECD regions bank emissions permits, which, by backward induction (through the Hotelling rule)
from a single global emissions-permit price in 2025, leads to a convergence of CO
2
prices in the
various regional markets from 2015 onwards.


Figure 4a
13


Figure 4b

Figure 4c
Figure 4: Trajectories of CO
2
price in the European Union in the various scenarios
4.3 Technological trajectories

Figure 5a shows the world installed capacities for each power technology in the R2015 branch
when economic recovery is either perfectly (p=1), or badly (p=0.2) anticipated. Figure 5b shows the
installed capacities in the NR branch when non-recovery is perfectly (p=0) or badly (p=0.8)
anticipated. In all cases, the development of the global electric system is fairly similar during the
initial periods. The regional emissions targets are so ambitious that, regardless of anticipations on
economic growth, the constraints limiting the expansion of some renewables are binding from the
earliest periods (so as to enable subsequent large-scale deployment).
14

From 2030, hydropower stagnates due to exogenous availability constraints. The share of
nuclear in power generation capacity increases from 9% in 2005 to between 11% and 16% in 2045.
The steep rise in gas-fired capacity between periods 2005 and 2010 is mainly explained by the
substitution of gas for coal in OECD regions so as to limit their emissions. At the same time, the rise
in coal-based production capacity is driven by non-OECD regions (which are not yet subject to
emissions constraints). Fossil-fuel-based capacities without CCS decrease from 2015, i.e. when
renewables, especially wind and biomass, have reached sufficient size to expand widely. The R2015
and NR branches diverge significantly from 2030 onwards, especially for nuclear, renewables and
CCS.

Figure 5a: Power production capacity in the R2015 branch for p=1 and p=0.2


Figure 5b: Power production capacity in the NR branch for p=0 and p=0.8

The relatively low installed CCS capacity inaccurately reflects the significant role of CCS in the
energy mix from 2025 onwards, because of its high load factor of 85% (especially relative to 18-40%
for wind) as shown in Table 1. Furthermore, as illustrated by Figure 6, the deployment of CCS is
15

highly sensitive both to the path actually followed by growth and to agents initial anticipations. In
the scenario where recovery and non- recovery in 2015 are equally probable (p=0.5), the installed
CCS capacity is 165 GW in 2025. This capacity is only 106 GW when non-recovery has been
perfectly anticipated (p=0). This sensitivity of CCS deployment to anticipations on economic growth
is in particular due to its time-to-build of two periods.


Figure 6a

Figure 6b
16


Figure 6c
Figure 6: World installed power generation capacity with CCS in the various scenarios.

Finally, the adaptation of the energy mix results not only from investment in low-CO
2

technologies, but also from adjusting the use of existing fossil-fuel-based capacities. In the first
periods, this latter mechanism is a decisive factor in the sensitivity of CO
2
prices to anticipations on
economic growth. It also largely explains the electricity prices yielded by the model.
4.4 Marginal technologies and electricity prices
Electricity prices in the European Union are shown in Figure 7 for the various stochastic
scenarios considered. Figure 7 also indicates the marginal technology in each period, i.e. the
technology whose short-run marginal cost of production is equal to the price of electricity. This
short-run marginal cost takes account of variable operating costs, including fuel and CO
2
emissions
permits. The marginal technology is characterized by partial use of available capacity.
In some cases, this short-run marginal cost is degenerate, with a left-hand short-run marginal cost
(corresponding to the technology used to produce the last kWh and whose production capacity is
fully used) and a right-hand short-run marginal cost (corresponding to the technology currently not
used, but which would be used to produce one additional kWh). The price of electricity given by the
model therefore falls between these two marginal costs, with the corresponding symbols being
represented side by side in Figure 7. Agents in the model base their decisions on this price that, at
equilibrium, ensures the clearing of the power market.
The sequence of marginal technologies demonstrates that the various types of CO
2
-emitting
technologies are successively abandoned as the emissions-permit price rises. In general, from 2010
to (in some branches) 2025, remaining coal-fired power plants are only partially used. Subsequently,
these plants are no longer used, with remaining gas-fired plants then representing the marginal
technology. After that, from 2030 (2025 in the R2015 branch of the scenario p=0.2), with remaining
gas-fired plants no longer being used, electricity-production adjustment occurs through the use of
new gas-fired capacities (even though this was installed after 2010). From 2040, or even 2035, the
new gas-fired plants are no longer in use. Renewables and nuclear, with a low short-run marginal
cost, are used at full capacity (a situation denoted by a cross and a circle side by side in Figure 7). In
17

the case of an absence of economic recovery that has been correctly anticipated (p=0, 0.2), the
abandonment of the technologies based on fossil fuels occurs earlier (2035 instead of 2040). This is
in particular due to lower investment in the final sector leading to lower demand for electricity in
2035. The achievement of a mix of power-generation technologies emitting little or no CO
2
is
therefore not contingent upon the realization or the anticipation of strong economic growth. On
the contrary, strong growth may, by driving up energy demand, cause standard fossil-fuel-based
capacities to be used for a longer time. As we can see from Figure 7, the stronger the economic
growth, the higher the electricity price.


Figure 7: Trajectories of electricity price in the European Union in the various scenarios.
Moreover, incorrect anticipations of economic growth may lead to situations of over- or under-
investment, possibly with a strong impact on the electricity price. A poorly anticipated economic
recovery in 2015 (p=0.2 and, to a lesser extent, p=0.5) leads to an abrupt rise in the price of
electricity in 2020 (up to $98 per MWh). Inadequate investment in new gas-fired plants and
renewables requires the use of some old coal-fired power plants whose short-run marginal
production cost is extremely high because of the CO
2
price. The electricity price subsequently
decreases, as investments in renewables enables the further use of the old coal-fired plants to be
avoided. In the p=0.2 scenario, even old gas-fired plants are abandoned through the combined effect
of high gas and CO
2
prices. In this way we move directly from a price in 2020 set by the short-run
marginal cost of remaining coal-fired plants to a price in 2025 set by the short-run marginal cost of
new gas-fired plants.
18

Conversely, a strongly-anticipated recovery that does not take place may give rise to an
overinvestment situation: anticipating (wrongly) an economic recovery leads to premature
investments in gas and renewables. As this recovery does not happen, the gas-fired capacities are
sufficient to allow the remaining coal-fired power plants to be abandoned, as we observe in 2020 in
the NR branch of the p=0.8 scenario. As a result of the low CO
2
and gas prices in this branch (see
Figures 4 and 8), the electricity price drops to a particularly low level.

Figure 8: Gas price in the European Union in the 0, 0.2, 0.8,1 p = scenarios.
4.5 Endogenous investment costs and net present values in the electricity sector
At the optimum of social welfare in each region, the optimality conditions of the model can be
interpreted as if a single power utility - acting as price taker and providing all electricity production
in the region - were to maximize its value while taking account of the constraints limiting the
expansion rate of each technology, the market shares of wind, nuclear and solar technologies, and the
availability of hydropower and biomass.
Let us consider a given technology. Table 1 provides an exogenous investment cost, based on
EIA (2010a,b) estimates, which, when used in the model, should be interpreted as the cost of
construction and installation for a power-generation equipment manufacturer. By charging the power
utility a price equal to this exogenous cost, an equipment manufacturer makes no profits. At a
regional scale, equipment manufacturers form a technological sector whose expansion is constrained.
The Appendix shows that the model internalizes the constraints limiting the expansion rate of this
technological sector into the investment cost paid by the power utility. The resulting endogenous cost
may be interpreted as the price actually paid (to the equipment manufacturer) by the power utility for
installing an additional unit of capacity. This price is equal to the exogenous investment cost adjusted
for the dual variables associated with the expansion constraints. This investment-cost adjustment
may be interpreted as resulting from tensions between demand for additional capacity (from the
power utility) and supply of new capacity (by the technological sector under consideration). From an
economic point of view, the expansion constraints generate an inelasticity of supply (through
bottlenecks) by the technological sector. At equilibrium, this can lead to a rise in the price paid by
the power utility for the acquisition of new production capacity.
19

Figure 9 shows the endogenous investment cost (determined using the equations in the
Appendix) of wind onshore and wind offshore technologies in the European Union, North America
and China. In each region and for each technology, the maximum annual growth rate in installed
capacity is 10%. The endogenous investment cost is higher than the exogenous investment cost until
2025 in Europe and until 2030 in North America. During this lapse of time, the growth in demand for
electricity, especially carbon-free electricity, comes up against bottlenecks in the supply of new
capacities using these two technologies. After 2030, the expansion constraints have no further
effects, because of the maturity of both technological sectors and - in a context of low growth in
electricity demand - the limitation on the market share of wind power. In the European Union, where
growth in demand for electricity is lower, the constraint on market share of wind becomes binding
sooner than in North America. The endogenous investment cost - highly sensitive to the path
followed by economic growth in North America - depends only very slightly on initial anticipations.
In China, the curves of endogenous investment costs are all bubble-shaped
12
centred on 2025, a
period which corresponds to this regions entry to the global emissions-permit market. The
bottleneck in the wind equipment sectors is very significant at that date. In a context of steeply rising
demand for electricity, the price of acquiring production capacity is relatively sensitive to the path of
economic growth followed, and, to some extent, to agents initial anticipations. Interestingly, in 2015
the endogenous investment cost for wind offshore is lower than the exogenous cost. Chinese
manufacturers therefore agree to sell below construction cost temporarily, so as to develop
manufacturing capacity and produce more equipment subsequently (when the endogenous
investment cost is considerably higher than the exogenous cost). An alternative interpretation is that,
from a social point of view, the Chinese authorities should subsidize this technological sector in
2015.

Figure 9a: European Union

12
By analogy, in North America and the European Union (regions subject to an emissions-permit market right from the
first periods), the endogenous investment costs curves may be seen as bubbles truncated on the left.
20


Figure 9b: North America

Figure 9c: China
Figure 9: Endogenous investment costs of wind onshore and wind offshore technologies in scenarios
0, 0.5,1 p = .
From the results of the model, we can determine the expected net present value of a marginal
investment by considering the endogenous investment cost, the electricity price trajectories, fixed
and variable production costs and the decay rate of the installed capacity. The cash flow achieved in
each possible state of the world is discounted at the economy's interest rate realized in that state of
the world. The expected net present value, determined as the expected sum of these discounted cash
flows, reflects the point of view of a power utility that would make the investment in an atomistic
electricity sector. This utility would effectively consider that its investment decisions had no effect
on the breakdown between technologies (i.e. maximum market shares). Table 3 shows the expected
net present values (as percentages of the corresponding endogenous investment costs) associated
21

with wind onshore and wind offshore production capacity built during the 2015 and 2020 periods in
China, North America and the European Union. In China, these expected net present values are
relatively close to zero. On the other hand, they are relatively high in European Union and North
America, as the constraints on market share for wind power are fairly soon binding, which generates
a scarcity rent at the level of both technologies. Thus, if we assume that a (maximum) market share
of 50% (instead of 33%) for wind power is technically realistic, the net present values obtained in
both deterministic scenarios are all below 8% (12%) of the corresponding endogenous investment
cost in 2015 (2020). Furthermore, Table 3 shows that anticipation of a rapid economic recovery does
not necessarily result in a markedly higher net present value.
In practice, the valuation of an investment by a private utility differs noticeably from the above
calculation. Thus, the model does not consider any fiscal transfer. Moreover, private firms generally
use higher discount rates than that used by the public authorities. Let us consider, for example, the
case of a deterministic economic recovery ( 1 p = ), taking into account a tax rate of 30% and linear
depreciation of the investment over 10 years (2 periods) for calculating the project's free cash flows.
An investment in wind onshore (wind offshore) in 2015 in the European Union would then generate
an internal rate of return of 5.9% (6.1%). This internal rate of return would be 7.3% (7.7%) for an
investment in wind onshore (wind offshore) in 2020. Without subsidies, a firm using a discount rate
greater than 8% (in real terms) would therefore not make these investments.
Table 3: Expected net present value as a percentage of endogenous investment cost for marginal
investment made in 2015 and 2020.

2015 2020
p=0 p=0.2 p=0.5 p=0.8 p=1 p=0 p=0.2 p=0.5 p=0.8 p=1
European
Union
Wind onshore 16 20 23 25 25 24 31 35 38 39
Wind offshore 18 23 26 27 28 27 34 38 41 42
North
America
Wind onshore 10 10 9 7 6 24 25 21 20 19
Wind offshore 13 13 11 9 9 27 28 24 23 22
China
Wind onshore -2 -5 -1 -6 -7 6 4 2 0 -1
Wind offshore 1 -3 1 -3 -4 8 4 4 2 1
6. Conclusion
Both economic-growth path realized and agents initial anticipations strongly influence the CO
2

price. Thus, in 2040, the global price of CO
2
may range from $21 (when an initially-anticipated
economic recovery never occurs) to $128 (in case of non-anticipated rapid economic recovery).
A rapid and poorly-anticipated economic recovery may cause an abrupt (but transient) rise in the
CO
2
price because of an insufficient number of banked permits and, to a lesser degree, unsuitable
technologies.
The case of no economic recovery whereas a rapid upturn had been considered possible is
ambiguous. The uncertainty in this case has two opposite effects on the CO
2
price: the existence of a
stock of emissions permits previously banked in anticipation of a possible rapid economic recovery
which in the end did not take place, and the need for a sufficient stock of banked permits in
anticipation of a possible future recovery. As time goes by, the effect of past banking becomes
preponderant, which leads in the long term to very low CO
2
price levels.
22

During the initial periods of the model, the differences in the quantities of emissions permits
banked observed between the different scenarios result principally from adjustments in the use of
existing fossil-fuel-based capacities. Anticipating economic recovery (and therefore higher CO
2

prices) as being probable speeds decommissioning, first of remaining coal-fired power plants, then of
remaining gas-fired power plants, and finally of more recent gas-fired power plants. Additionally, the
deployment of CCS technologies is especially sensitive to anticipations.
In the electricity sector, the model internalizes the constraints limiting the expansion rate of each
technology into the price paid by the power utility for acquiring additional production capacities. For
the wind onshore and wind offshore technologies, this endogenous investment cost which is
relatively sensitive to the realized economic growth - in general depends very little on initial
anticipations. In the medium run, assuming a technical limitation on wind-power market share
creates a rent for power utilities producing with offshore and onshore wind. From the viewpoint of a
private utility paying an income tax, a marginal investment made before 2020 nevertheless yields an
internal rate of return that is noticeably lower than 8%. A subsidy from the public authorities would
therefore be justified to attain the socially optimal level of investment. To a certain extent, a power
utility investing in wind power would today receive public subsidies and, in the future, benefit from a
scarcity rent priced by the market.
References
Aghion, P., and P. Howitt (1998). Endogenous growth theory. Cambridge: The MIT Press.
Bosetti, V. and M. Tavoni (2009). Uncertain R&D, backstop technology and GHGs stabilization.
Energy Economics 31 (Suppl.), S18S26.
Bosetti, V., Carraro C., Sgobbi, A. and M. Tavoni (2009). Delayed action and uncertain targets.
How much will climate policy cost? Climatic Change 96 (3), 299-312.
De Cian, E. and M. Tavoni (2010). Hedging against climate policy and technology uncertainty:
implications for technology mix and policy instrument choice. Proceedings of International
Energy Workshop, Stockholm, July 2010.
Durand-Lasserve, O., Pierru, A. And Y. Smeers (2010). Uncertain long-run emissions targets, CO
2
price and global energy transition: a general equilibrium approach. Energy Policy 38 (9), 5108-
5122.
EIA (2010). International Energy Outlook 2010. U.S. Energy Information Administration,
www.eia.gov/oiaf/ieo/index.html.
Hayashi, F. and E. C. Prescott (2002). The 1990s in Japan: a lost decade. Review of Economic
Dynamics 5 (1), 206-235.
IEA (2010a). World Energy Outlook 2010. International Energy Agency, Paris.
IEA (2010b). Projected Costs of generating Electricity 2010. International Energy Agency, Paris.
IEA (2008a). Energy Technology Perspectives. International Energy Agency, Paris.
IEA (2008b). World Energy Outlook 2008. International Energy Agency, Paris.
IEA (2007a). World Energy Outlook 2007. International Energy Agency, Paris.
Jacoby, H. D., Reilly J. M., McFarland J. R., and S. Paltsev (2006). Technology and Technical
change in the MIT EPPA model. Energy Economics 28 (5-6), 610-631.
23

Khler, J,. Grubb, M., Popp D., and O. Edenhofer (2006). The transition to endogenous technical
change in climate-economy models: A technical overview to the Innovation Modelling Comparison
Project. The Energy Journal. Endogenous Technological Change and the Economics of
Atmospheric Stabilisation, Special Issue, 17-56.
Knnl T., Carrillo-Hermosilla, J., and R. van der Have (2008). System transition concepts and
framework for analysing energy system research and governance. IE WP08-31. Available at
SSRN: http://ssrn.com/abstract=1646274.
Manne, A. S., and T.R. Olsen (1996). Green house gas abatement - toward pareto-optimal
decisions under uncertainty. Annals of Operation Research 68 (2), 267279.
Manne, A. S., Mendelsohn, R., and R.G. Richels (1995). MERGE: a model for evaluating regional
and global effects of GHG reduction policies. Energy Policy 23 (1), 1734.
Manne A. S. and R. G. Richels (1995). "The greenhouse debate: economic efficiency, burden sharing
and hedging strategies". Energy Journal 16(4), 137.
Meeraus, A., and T. Rutherford (2005). Mixed complementarity formulations of stochastic
equilibrium models with recourse. GOR Workshop, Bad Honnef (Germany), October 2021.
OCDE (2010), Economic outlook 2010. OECD, Paris.
Rafaj, P., and S. Kypreos (2007). Internalisation of external cost in the power generation sector:
Analysis with Global Multi-regional MARKAL model. Energy Policy 35 (2), 828843
Richels, R. G., and G. J. Blanford (2008). The value of technological advance in decarbonizing the
U.S. economy. Energy Economics 30 (6), 2930-2946.
Scott, M.J., Sands, R. D., Edmonds J., Liebetrau, A. M. and D.W. Engel (1999). Uncertainty in
integrated assessment models: modeling with MiniCAM 1.0. Energy Policy 27 (14), 855-879.
Webster, M. D., Paltsev, S., Parsons, J., Reilly, J. and H. Jacoby (2008). Uncertainty in Greenhouse
Emissions and Costs of Atmospheric Stabilization. MIT Joint Program Report No 165, MIT Joint
Program on the Science and Policy of Global Climate Change, Cambridge, MA.
WRI (2010a). Summary of GHG reduction pledges put forward by developing countries. World
Resources Institute, June 2010.
WRI (2010b). Comparability of annex I emission reduction pledges. World Resources Institute,
February 2010.








24

Appendix: Constraints on expansion rate and endogenous investment costs
Let us consider a given power-generation technology in a given region. Like renewables in the
model, this technology is supposed to have a low short-run marginal cost, with the capacity installed
in the region therefore being entirely used during each period. At the optimum of the regional social
welfare, the optimality conditions of the model can be interpreted as if a price-taking regional power
utility were maximizing the expected net present value generated by its policy of investment in the
technology considered. In a consistent manner, we here directly consider the (decentralized) power
utility's value-maximization problem.
To alleviate the notations, only one probability node is here considered. An uncertainty is thus
resolved between periods 1 t and t, with two possible states from period t onwards (for example,
economic recovery and absence of economic recovery from 2020). Each state s has a probability of
occurrence denoted by
s
p . Where necessary, each variable or coefficient is indexed as a function of
time (subscript) and state (superscript). The power utility therefore maximizes the program below,
with the dual variable associated with each constraint appearing in parentheses to the right of the
constraint:

( ) ( ) ( )
( ) ( )
( ) ( )
( ) ( )
( ) ( )
( ) ( )
1 2
1 1 1
1 1 1
1
1 1
1 1
1 1
1
1 0 1
1 0 1
1 0 1, 2
. .
1 0 1, 2
1 0
t T
s s s s s s s
j j j j j t t t t t j j j j j
j s j t
j j j
j j j j
s s
t t t
s s s
t t t t
s s s
j j j
Max z a k c i p z a k c i z a k c i
g k k j t
i k d k j t
g k k s
s t
i k d k s
g k k t
o

= = = +

+ +
+ +

| | | |
+ +
| |
|
\ . \ .
+ > s s
+ = s s
+ > =
+ = =
+ >

( ) ( )
1 1
2 , 1, 2
1 0 2 , 1, 2
s s s s
j j j j
j T s
i k d k t j T s

+ s s =
+ = + s s =


The coefficient
j
a represents the operating cash flow generated in period j by the use of one kW
of installed power. This coefficient takes account of the regional electricity price and operating costs.
The coefficient
j
c represents the cost of construction and installation - by the technological sector
considered - of one kW of power-generation capacity in period j. It corresponds to the exogenous
investment cost given in Table 1. The capacity built in period j (assumed to be positive) is decided in
period 1 j and is denoted by
1 j
i

. The total installed capacity owned by the power utility in period j
is denoted by
j
k , with
0
k being fixed. Because of the expansion constraints, the growth rate in this
installed capacity from one period to another cannot exceed g. In addition, the installed capacity
decays at the periodic rate d. The present value of a cash flow generated in period j is obtained by
applying the discount factor
j
z (derived from the economy's interest rate in the model) to this cash
flow.
By differentiating the Lagrangian of this problem with respect to
s
T
k and
1
s
T
i

, we obtain:
0
s s s s
s T T T T
p z a o =
25

0
s s
T s T T
p z c =
By combining these two equations, we have:
s
s T
T T s s
T
a c
p z
o
= + (1)
At optimum, in period T (2100 in the model), a marginal investment generates a zero net present
value: the present value of the expected operating cash flows (here summed up as
s
T
a ) is equal to the
investment cost. According to (1), the power utility therefore considers that, in state s, installing at
period T one additional kW of capacity costs
s
T
T s s
T
c
p z
o
+ .
Let us now differenciate the Lagrangian with respect to
s
j
k and
1
s
j
i

, with 1 t j T + s s :
( ) ( )
1 1
1 1 0
s s s s s s s
j j j j j j
p z a g d o o
+ +
+ + + = (2)

s s s
j j j
p z c = (3)
Combining (2) and (3), we obtain:
( )
( )
1 1
1
1
1
s s s
j j j s
j j j s s s
j j
g z
a c d c
p z z
o o
+ +
+
+
= + (4)
Let us define
s
j
c such that:
s s
j j j
c c e = + (5)
with:
( )
( )
1 1
1
1
1
s s s
j j j s s
j j s s s
j j
g z
e d e
p z z
o o
+ +
+
+
= + ,
s
s T
T s s
T
e
p z
o
= .
Combining (5) and (4), we obtain:
( )
1
1
1
s
j s s s
j j j s
j
z
a c d c
z
+
+
= (6)
Equation (6) enables us to interpret
s
j
c as an investment cost derived endogenously from the
model. Anticipating the installation of 1 d kW of capacity by one period (i.e., installing 1 kW in
period j instead of 1 d kW in period j+1) enables, through production during period j, generation of
the additional operating cash flow
s
j
a . However, advancing this installation from period j+1 to period
j generates an additional investment cost equal to
( )
1
1
1
s
j s s
j j s
j
z
c d c
z
+
+
. Equation (6) simply indicates
that, at optimum, anticipating (or delaying) this marginal investment by one period does not modify
the value of the objective function. We may also note that, formally, equation (6) corresponds to
equation (4) written without the term containing the dual variables associated with expansion
constraints (since the endogenous investment cost internalizes these constraints).
By recurrence we obtain:
( ) ( )
1
1
1
1
T
k j
s s s
j j k s s
k j
j
e g d d
p z
o o

= +
| |
= +
|
\ .

In state s, the endogenous investment cost is therefore (with 1 t j T + s s ):
26

( ) ( )
1
1
1
1
T
k j
s s s
j j j k s s
k j
j
c c g d d
p z
o o

= +
| |
= + +
|
\ .

(7)
By differentiating the Lagrangian with respect to
t
k ,
1 t
i

and
s
t
i , we have:
( ) ( ) ( )
2
1 1
1
1 1 0
s s s s s
t t t t t t
s
p z a d g o o
+ +
=
+ + + + =

2
1
s
t t s t
s
c p z
=
=


1 1 1
s s s
t t t
P z c
+ + +
=
Combining these equations, we obtain:
( ) ( )
2 2 2 2
1 1 1
1 1 1 1
1 1
s s s s s s s
t t t s t t t t t
s s s s
p z a c p z g d c p z o o
+ + +
= = = =
= + +

(8)
As the decision to invest was taken one period before, the endogenous investment cost
t
c is
considered to be independant of the state realized in period t (since it was set in period 1 t when the
order for capacity construction was issued). At optimum, the expected gain obtained by advancing by
one period (i.e. in t rather than 1 t + ) the installation of 1 d kW of capacity should be equal to the
expected additional investment cost resulting from this decision. The endogenous investment cost
t
c
must therefore satisfy the following condition:
( )
2 2 2
1 1
1 1 1
1
s s s s s s s s
t t t t t t
s s s
p z a c p z d p z c
+ +
= = =
=

(9)
Combining (8) and (9), we obtain:
( ) ( )
2 2
1 1 1
1 1
2
1
1 1
s s s s
t t t t
s s
t t t t
s s
t
s
g d p z e
c c e c
p z
o o
+ + +
= =
=
+ +
= + = +

(10)
Following the same line of reasoning as before, for 1 1 j t s s we have:
j j j
c c e = +
where:
( ) ( )
1 1 1
1 1
j j j j
j
j
g d z e
e
z
o o
+ + +
+ +
= .
Using the values of the dual variables at the model's equilibrium, the endogenous investment cost
can be backwardly computed with Equations (1), (5) and (10).