SSRN Lcoe

ed
Levelized Full System Costs of Electricity
iew
Abstract
ev
Different electricity generating technologies are often compared using the Levelized Costs
of Electricity (LCOE), which summarize different ratios of fixed to variable costs into a
r
single cost metric. They have been criticized for ignoring the effects of intermittency and
non-dispatchability. This paper introduces the Levelized Full System Costs of Electricity
er
(LFSCOE), a novel cost evaluation metric that compares the costs of serving the entire
market using just one source plus storage. Like LCOE, and in contrast to alternatives such
pe
as System LCOE, LFSCOE condense the cost for each technology into one number per
market. The paper calculates LFSCOE for several technologies using data from two different
markets. It then discusses some refinements, including the LFSCOE-95 metric that require
each technology to supply only 95% of total demand.
ot
Keywords: Intermittent renewables, Levelized Costs of Electricity (LCOE), System LCOE,

Power Generation Economics, Electricity
tn
rin
ep
Pr
This preprint research paper has not been peer reviewed. Electronic copy available at: https://ssrn.com/abstract=4028640
ed
1 Introduction
The lifetime costs of an investment are key measures for decision-making. This is true for invest-
iew
ment decisions in electricity markets as well, where the most popular measure to compare different
technologies for generating electricity are the Levelized Costs of Electricity (LCOE). To calculate
the LCOE, the expected lifetime generation of an electricity generating plant and the expected
costs to generate the lifetime electricity are calculated. After dividing total costs by total genera-
ev
tion, the final number (usually in USD/MWh) is derived. Input assumptions like capacity costs,
maintenance, marginal operating costs, or average capacity factor, which is particularly relevant
for renewable sources of electricity, are crucial for the calculation and vary by study.1 For example,
r
Lazard (2019) estimates the LCOE of coal between 66 and 152 USD/MWh and onshore between
er
28 and 54 USD/MWh, whereas EIA LCOE (2020) derives LCOE for coal at of USD/MWh and
LCOE for wind 40 USD/MW. Many recent studies indicate that the LCOE are the lowest for
pe
onshore wind and utility-scale solar using photovoltaic cells (hereafter referred to as “solar PV” or
“solar”), findings frequently cited by proponents of a fast transition towards renewable electricity.
Nevertheless, if it is the cheapest source while not emitting CO2, why are countries still investing
heavily in new gas and coal power plants? Is it just because coal generation may employ more
ot
people in politically sensitive regions of the country, or are there financial reasons not reflected in
the LCOE?
tn
Critiques of LCOE are not scarce. Joskow (2011) is one of the first to point out that LCOE ignore
the costs associated with intermittency. It is easy to see the fundamental misunderstanding in
LCOE: The LCOE describe the costs of generating electricity. However, the function of supply
rin
in electricity markets is not to generate electricity but to provide a specified amount of electricity
to a specific place at a particular time. The locational aspect adds significant additional costs to
renewables that are generally less flexible about where they can be sited than fossil fuel plants.
ep
As a result, a larger grid is required to transport the electricity from, e.g., hydropower plants to
the demand in urban areas. These transmission costs are partly taken care of in some LCOE
estimates when a transmission cost adder is included in the LCOE. But the timing aspect turns
Pr
out to be even more crucial and the focus of this paper. Many renewables (like wind and solar)
1
I abbreviate the term “electricity plants that use renewable sources of energy” by “renewables”.
1
ed
are intermittent and non-dispatchable (hereafter referred to just as “intermittent” unless further
specified), and some that are not intermittent (like run-of-river-hydro) are often not fully dispatch-
able.2 As long as the share of intermittent generation is low, sufficient dispatchable generation
iew
capacity will usually be available to step in and replace missing intermittent generation output.
Economically, the fact that intermittent generation has no obligation to meet the demand can be
seen as a hidden subsidy. One can even go one step further and argue that intermittent generation
ev
is of zero value if it cannot be made available to consumers who demand a steady electricity flow.
To do that, however, supply and demand on the network must always be in balance. In effect,
the ability to schedule other generators to continuously maintain that balance is necessary to give
r
value to renewable output. The dispatchable generators thus raise the value of renewable genera-
tion, but the subsidy is “hidden” because the latter does not have to pay for it. Once the share
er
of intermittent generation increases to a certain level (and dispatchable capacity is shut down),
efforts have to be taken to maintain system reliability. But who should be responsible for these
pe
costs? How can the cost of integrating renewables into the system (which increases significantly
with their market share) be included in the evaluation of their cost?
Ueckerdt et al. (2013) address the cost of integrating renewables into a network by introducing
the “System LCOE”. The System LCOE of an intermittent source are defined as the sum of the
ot
(marginal) generation costs (the LCOE) and the (marginal) integration costs, where integration
costs can be split up into balancing costs, grid costs, and profile costs - see Section 2 for further
tn
information of System LCOE. Unlike conventional LCOE, the System LCOE of renewable sources
of electricity depend highly on their market share. If the share of wind (resp. solar) generation
rin
increases, the generation costs (i.e., the LCOE) remain constant, while the integration costs in-
crease significantly. In their calculation, the System LCOE for wind in Germany increase from 60
EUR/MWh to almost 100 EUR/MWh if the share increases from 0% to 40%.
ep
The System LCOE seem to be state of the art and reasonably accurate (see Reichenberg (2017)
for further refinement), but are apparently too complicated and “not catchy” enough to be used
2
Note that intermittency implies non-dispatchability, but not vice versa. Intermittency means that the capacity
Pr
factor is subject to external influences and varies in a short time. In contrast, a source of electricity is non-
dispatchable if the operator cannot easily control its output.
2
ed
by a non-academic audience.3 However, there is a high necessity of a cost measure that includes
the costs of intermittency and is accessible to a broader audience. Since climate change and the
accompanying transition of the electricity market became a crucial topic in public debates and
iew
politics, the LCOE have become the most popular measure to evaluate investment decisions and
market developments in electricity generation. As politicians and policymakers fail to understand
the limitations and flaws of this measure and spread the idea that solar PV and wind are the
ev
cheapest sources of electricity, there is a need for a cost measure that addresses the limitations of
LCOE yet remains accessible to a broader audience by being catchy and straightforward.
This paper introduces a novel method to evaluate the costs of electricity that is catchy and
r
includes the costs of intermittency: The Levelized Full System Costs of Electricity (LFSCOE).
The LFSCOE are defined as the costs of providing electricity by a given generation technology,
er
assuming that a particular market has to be supplied solely by this source of electricity plus
storage.4 Methodologically, the LFSCOE for intermittent or baseload technologies are the opposite
pe
extreme of the LCOE. While the latter implicitly assume that a respective source has no obligation
to balance the market and meet the demand (and thus demand patterns and intermittency can be
ignored), LFSCOE assume that this source has maximal balancing and supply obligations. This
paper shows that in both Germany and the region of the Electricity Reliability Council of Texas
ot
(ERCOT), the LFSCOE of wind and solar PV are higher than the most expensive dispatchable
technology examined in this paper.5 Simulating the effect of decreasing storage costs, we observe
tn
that although the LFSCOE for wind and solar drop significantly, even a storage cost reduction
of 90% is insufficient not make wind or solar PV competitive on an LFSCOE basis.6 Allowing
3
rin
It is worth noting that the calculation of LCOEs is by no means simple. The appendix to the Annual Outlook
of Energy requires more than 100 pages to introduce the terminology and describe the calculation process, see EIA
Model Documentation (2020).
4
While storage is necessary for intermittent sources of generation, storage can also allow conventional dispatch-
ep
able technologies to meet system load at lower costs, for example, by smoothing demand fluctuations and allowing
plants to operate at higher capacity factors.
5
Though the ERCOT market does not span all of Texas, ERCOT and Texas are used interchangeably in this
paper.
Pr
6
Note that a decrease in storage costs is to some degree equivalent to an increase in the storage factor (i.e., the
amount of MWh stored per MW installed). Some technologies (like solid-state batteries) cannot increase this factor,
while others (like flow batteries) can. This matter is briefly discussed, but a thorough analysis of an adjustable
3
ed
for losses in the charging and discharging process, it is interesting to see the small magnitude of
economic effects of even significant storage losses in such a system. Last, we extend the LFSCOE to
LFSCOE-95, which assume that only 95% of the system must be supplied by a certain technology
iew
plus storage. While the LFSCOE-95 are only slightly lower than the LFSCOE for dispatchable
technologies, they are about 50% lower for intermittent sources, which challenges the economic
sanity of 100% intermittent renewable targets.7
ev
This paper is structured as follows: Section 2 introduces the method for calculating LFSCOE
and concludes with the cost evaluations for the markets in ERCOT/Texas and Germany. Section
3 examines different changes in the model assumptions (such as storage losses), follows up with an
r
analysis of significant decreases in storage costs, and concludes with introducing the LFSCOE-95.
Section 4 discusses potential model extensions and concludes.
er
Literature To my knowledge, the cost measure and evaluation methodology introduced in this
pe
paper are new. However, studies have been conducted that address the cost of intermittent
renewables or baseload technologies when they are responsible for meeting the market demand.
Becker et al. (2014) examine the requirements of a fully renewable system in the U.S., whereas
Hartley (2017) those of a wind-only market in Texas. Denholm et al. (2012) combine renewables
ot
and nuclear with storage. For the market in Germany, Sinn (2017) discusses economic challenges
by pointing out the large curtailment and storage requirements in a wind & solar market in
tn
Germany (using existing storage in Norway), while Zerrahn et al. (2018) conclude that electrical
storage would rather not limit the transition to renewable energy. It is important to note that
the motivation of this paper is to introduce a novel methodology of calculating costs and then use
rin
this methodology to examine some relevant counterfactuals. Given the simplifying assumptions,
the numbers should not be seen as definitive.
storage factor’s economic consequences is not part of this paper.
ep
7
Decarbonization of electricity markets is often just defined as an emission reduction of at least 80%, circum-
venting the challenge to decarbonize the last few percents - see Jenkins et al. (2018).
Pr
4
ed
2 Levelized Full System Costs of Electricity
This section first introduces the concept of the Levelized Full System Costs of Electricity and
iew
compares 5 dispatchable technologies with wind, utility scale solar PV (called “solar” from now
on), and an optimal combination of wind and solar.
Assumptions A technology y has overnight capacity costs ccy , fixed operation and maintenance
ev
costs (O&M) omcy , and variable (or constant marginal) costs vcy . Dispatchable technologies
have ramping times of rampupy and rampdowny , in percentage/hour relative to their current
generation (as depict in table 1). Note that all modern technologies can technically ramp-up
r
and ramp-down within an hour, making the ramping times redundant for the analysis which
er
only considers deterministic hourly demand (see below). However, frequent fast ramp-ups and
ramp-downs are not advisable and sometimes not permitted (for example for old nuclear plants
- however, the costs in Table 1 are for new advanced plants). As a result of this, the ramping
pe
values are set to rampupy = 150% and rampdowny = 50%, meaning that the utility can change
their output by ±50% per hour. This implicitly increases the costs for dispatchable generators,
but not by much as the demand fluctuations of consecutive hours are almost always within ±50%.
ot
It is worth noting that the model only considers deterministic hourly demand and ignores any use
of certain technologies outside the wholesale generation market (e.g. a natural gas combustion
tn
turbine, which, unlike nuclear power plants, can also be used in the short-term balancing market).
In period t, intermittent and non-dispatchable generators have an hourly generation intensity of
Rent ∈ [0, 1]. This value is nothing else than the maximal capacity factor in hour t of the renewable
rin
plant - when the sun is shining or the wind is blowing, this value will be high (and close to 1), but
during the night, the intensity Rent will be equal to 0 for solar PV. The hourly demand is denoted
with Dt for t = 1, ..., H (e.g. H = 8760 if the entire year is consider). Both intensity and demand
ep
are assumed to be perfectly forecasted and are deterministic, i.e. possible demand response is
ignored. Following the calculations from EIA Assumptions (2020), the costs are averaged over
an investment period of 30 years, where the investment occurs in the first two years (thus, the
Pr
overnight capacity costs are split evenly between year 1 and year 2) and the generation goes from
5
ed
Table 1: Cost Assumptions
Overnight
Capital Costs O&M Costs Variable Costs ramp-up/down
iew
Technology
[USD/kW] [USD/kW/year] [USD/MWh] [% per hour]
Biomass 4,401 125.2 28 150%/50%

Coal (USC) 3,661 40 25 150%/50%
Natural Gas CC 1,079 14 18 150%/50%
ev
Natural Gas CT 710 7 28 150%/50%
Nuclear 6,317 121 8.4 150%/50%
Solar 1,331 15.2 0 -
Wind 1,319 26.2 0 -
r
Storage 1,383 24.7 0∗ -
er
Table 1: Main data source is EIA Costs (2020). The fixed costs include overnight capacity costs and fixed O&M. Wind fixed costs
are for on-shore wind. The variable costs include O& M and fuel costs, but do not include any carbon taxes or reserve payments for
environmental purposes.
pe
∗ Unlike for the generators, the variable costs for storage include all costs but purchasing the input fuel (which is electricity). As these
purchasing costs are just transferred to the generators, they can be ignored if we evaluate the cost of the “generator + storage” system.
year 3 through year 30.8 The cost of capital is fixed at 6.7% (and thus the annual discount factor
is β = 1/(1+0.067)) and the model implicitly assumes the same demand and hourly wind and solar
ot
capacity factor profile in each of these years.9 The variable costs for storage are zero and there are
no losses in the storage process (the no-storage-losses assumption is relaxed in section 3). Storage
tn
can store ρsf = 3 MWh per installed MW of generating capacity, which is equivalent to current
residential solid-state battery storage solutions. A moderate increase in the storage factor, which
is discussed in section 3, is equivalent to decreasing the storage costs. For simplicity, no minimal
rin
storage level or security backup is required (i.e. Ssecurity = 0 MWh), but demand has to be met at all
times. In practice, if a market is supplied solely by an intermittent source, it will be very unlikely
that no security storage is required as a backup. Furthermore, required security storage will also
ep
depend on the average hourly demand, its variability and price elasticity, installed storage as well
8
Note that these investment periods seem to be quite generous for solar PV and wind, and too short for nuclear
and coal. On the other side, the construction period is generous for nuclear and coal, and too long for solar.
Pr
9
EIA Assumption (2020) calculates weighted average costs of capital for every year. As they are almost constant,
using constant discount factor is a reasonable simplification.
6
ed
as installed generation capacity, and will vary between markets. Ignoring the security backup in
my model simplifies the analysis significantly and allows for maximal comparability.
iew
Calculating the optimal installed capacity Let x ∈ RH+1 be the independent variable for
the stored electricity in MWh (i.e. x[t] is the storage level at the beginning of hour t), and
gen ∈ RH the independent variable for the dispatchable generation in each period (i.e. gen[t] is
the generation in period t).10 In the optimization problem, denote the installed capacity by rp
ev
(for intermittent “renewable power”) and dp (for “dispatchable power”), and the installed storage
by sp (for “storage power”). Installed capacities, storage levels, and generation will be chosen to
r
minimize total system costs conditional on meeting demand in every period. For any technology
y, the net present value of all non-variable costs (i.e fixed costs) is
f cy = (
ccy ccy
2
+
2
er30
β + ∑ β u−1 omcy ).
u=3
pe
With this notation (and further explanations of each line below), intermittent sources of generation
with intensity Rent at period t solve the optimization problem
min rp ⋅ f cRen + sp ⋅ f cStorage

rp,sp,x
ot
s.t. 0 ≤ x[t + 1] ≤ x[t] + Rent ⋅ rp − Dt f or all t, (D.1)
−sp ≤ x[t + 1] − x[t] ≤ sp f or all t, (S.1)

tn
Ssecurity ≤ x[t] ≤ sp ⋅ ρsf f or all t, (S.2)
x[1] ≤ x[H + 1], (S.3)

rin
10
To enhance readability, the elements of vectors that are independent variables are denoted with brackets [t]
and the elements of vectors which are parameters are denoted with lower case t-s.
It is worth noting that electricity storage does not store electricity itself, but converts it do a different form of
ep
energy (e.g. pumped hydro plants convert electric energy into kinetic energy). However, to enhance readability,
we will use the term “storing electricity”.
Pr
7
ed
whereas conventional sources y solve
30
8760 H
min dp ⋅ f cy + sp ⋅ f cstorage + ∑ β u−1 ⋅ ∑ vcy ⋅ gen[t]
dp,sp,gen,x u=3 H t=1
iew
s.t. 0 ≤ x[t + 1] ≤ x[t] + gen[t] − Dt f or all t, (D.2)
−sp ≤ x[t + 1] − x[t] ≤ sp f or all t, (S.1)
Ssecurity ≤ x[t] ≤ sp ⋅ ρsf f or all t, (S.2)
ev
x[1] ≤ x[H + 1], (S.3)
gen[t] ≤ dp f or all t, (G.1)

gen[t + 1] − gen[t]
−rampdowny ≤ ≤ rampupy
r
f or all t, (G.2)
gen[t]
The objective function calculates the total costs to meet the demand. For intermittent renewables,
er
this is just the costs of installing the capacity, for dispatchable sources, it also includes an additional
term to account for the variable costs. Note that to calculate the net present value of these costs,
pe
dispatchable generators are implicitly assumed to have the same generation for every year. If H is
smaller than 8760, the hourly costs are extrapolated accordingly to account for an entire year.11
Constraints (D.1) and (D.2) ensure that the demand Dt is met at any period t (and allows for free
ot
curtailment/disposal of electricity if necessary), while (S.1) ensures that the storage is charged and
dispatched according to the technical maximum (which is the maximal installed power in GW).
tn
Constraint (S.2) ensures that the storage level is technically feasible and larger than the security
storage. With a storage factor of ρsf = 3, we observe that the right hand side of (S.2) ensures that
(S.1) is never binding for intermittent technologies. In these systems, bp will be so high to ensure
rin
that (S.2) is fulfilled so that (S.1) becomes redundant. Once the storage factor increases, however,
(S.1) becomes binding at one point. Constraint (S.3) ensures that technologies are compared on
11
If H = 8760, there are 17523 independent variables. To reduce the computational burden in the counterfactual
ep
analysis, the optimization problem is only solved for half the year, i.e. H = 4380. To ensure that the solution is at
least theoretically feasible for the entire year, the condition
maxcapx ⋅ dp + sp ≥ max(Dt ). (X.1)

Pr
is added. This ensures that the maximal annual demand can be met with the proposed solution. For intermittent
sources, the program can always be optimized over 8760 hours, which is particularly important as these costs are
driven by periods with low hourly capacity factor and seasonal properties.
8
ed
a level playing field by always requiring energy in storage to return to the same starting level at
the end of each simulation. Inequalities (G.1) and (G.2) restrict the generation to the maximal
capacity and the hourly change in generation to the respective ramp-up and ramp-down rate.
iew
A solution to the optimization problem above (i.e. the optimal rp (or dp), sp, gen, and x) for
technology y is denoted by (I y , Istorage
y
, GEN y , X y ).12 This is then the optimal installed capacity
y
for generation I y and storage Istorage along with the hourly generation GEN y and hourly storage
level X y . Not that the generation vector GEN y is an H × 1 dimensional vector for dispatchable
ev
technologies (and equal to 0 for intermittent sources), whereas the storage level vector X is of size
H + 1 × 1 (and potentially equal to 0 if no storage is used).
r
Definition LFSCOE After solving the optimal installed capacity, generation, and storage level
(I y , Istorage
y
I y ⋅ f cy + Istorage
y
er
, GEN y , X y ), define the LFSCOE of a dispatchable source as
30
⋅ f cstorage + ∑ β u−1 ⋅ vcy ⋅
8760 H y
∑ GENt
pe
H t=1
LF SCOEy = u=3
30 8760 H
∑ β u−1 ⋅ ⋅ ∑ Dt
u=3 H t=1
and the LFSCOE of an intermittent source as
ot
I Ren ⋅ f cRen + Istorage

Ren
⋅ f cstorage
LF SCOERen = 30 8760
.
∑ β u−1 ∑ Dt
u=3 t=1
tn
The numerator is nothing else than the total cost of the system (and thus equal to the objective
function above). The denominator averages these total costs. The crucial element of the LFSCOE
is the fact that costs are not averaged over the (discounted) lifetime generation but over the
rin
(discounted) lifetime demand that they and their associated storage support. For dispatchable
sources, the lifetime generation is equal to (or at least close to) the lifetime demand as variable
ep
generation costs that are larger than 0 penalize producing excess output and as there are no storage
losses. By contrast, the next section shows that for intermittent generators, lifetime generation
is significantly higher than lifetime demand of the system, hereby causing a large amount of
Pr
overproduction and curtailed electricity. The Levelized Costs of Electricity account for some, but
not all, of the curtailment by adjusting the capacity factor and averaging over the total generation.
12
Existence and uniqueness of this solution are discussed in Appendix A.2.
9
ed
Ueckerdt et al. (2013) solves this issue by including overproduction in their cost estimation, but
as they limit their analysis to a 40% share of renewables, the overproduction is still relatively
small (see below a comparison between the LFSCOE and the results of Ueckerdt et al. (2013)).
iew
Note that the LFSCOE are defined for one specific data set (e.g. one year for one region). If more
data on hourly demand and hourly capacity factors for a certain region is available (e.g. data for
different years, see below) and this data cannot be incorporated into a joint optimization process,
ev
the LFSCOE are calculated separately and (with slight abuse of notation) redefined as the mean
of each year (see calculation below).
r
Calculating the LFSCOE Starting with real market data for hourly demand, wind capacity
factor, and solar capacity factor from 2010/2011-2017 for Germany and 2012/2013-2019 for Texas,
er
the costs per MWh for each technology and year are calculated by solving the minimization
problem as described above.13 We then define the LFSCOE of a technology as the mean of the
pe
costs per MWh from each of the 8 years. Allowing also for a wind & solar mix, Table 2 and Figure
1 display the LFSCOE for each technology and market as well as the interval of the costs for each
year (note the different scale of the x-axis).
ot
Figure 1: LFSCOE Mean and Max/Min intervals in the German (left) and ERCOT market
tn
rin
ep
Figure 1 displays the mean and the max/min intervals for every technology (as in Table 2).
Several interesting observations can be made. First, neither wind nor solar nor the wind & solar
Pr
13
As the installed capacity of solar was very low in 2010 in Germany and 2012 in Texas, the data does not seem
trustworthy. Thus the year 2010 in Germany and 2012 in Texas are excluded in the analysis of solar and the wind
& solar mix. For all other technologies, all 8 years are considered.
10
ed
Table 2: Levelized Full System Costs of Electricity
iew
Germany Texas
Technology LFSCOE (Mean) [Min,Max] LFSCOE (Mean) [Min,Max]
Biomass 104 [100,109] 117 [112,126]

Coal 78 [76,82] 90 [86,96]
Natural Gas CC 35 [34,36] 40 [38,41]
ev
Natural Gas CT 39 [38,39] 42 [40,42]
Nuclear 106 [101,113] 122 [115,132]
Solar 1548 [1185,2058] 413 [341,579]
r
Wind 504 [438,552] 291 [229,369]
Wind&Solar 454 [319,498] 225 [178,358]
er
Table 2: Mean, Minimum, and Maximum in USD/MWh of the derived LFSCOE values. LFSCOE for Germany and Texas, using cost
assumptions as in Table 1 and a discounting factor of β = 1/(1 + 0.065).
pe
mix seem economically competitive to the dispatchable sources on a LFSCOE basis. Even the
LFSCOE value of the wind & solar mix in Texas, which is the most competitive of the renewable
technology installations considered, is almost twice as high as the LFSCOE value of the most
ot
expensive dispatchable technology.14 These differences stem from the storage requirement and
low capacity factors for wind and solar, see Appendix A.4. Second, while the LFSCOE for the
conventional sources are slightly lower in Germany than in Texas, the LFSCOE for wind and
tn
especially solar are much higher in Germany (e.g. 1548 USD/MWh for solar in Germany vs.
413 USD/MWh for solar in Texas). The higher LFSCOE for dispatchable sources in the Texas
rin
market can be explained by the seasonal variance in demand. In Texas, the peak demand in the
month with the highest demand is 67% higher than in the month with the lowest peak demand.
This percentage is only 12% in Germany, which results in a higher overall capacity factor for
ep
dispatchable technologies in Germany than in Texas (see Appendix A.1). The significantly higher
LFSCOE for wind and solar in Germany compared to Texas stem from the higher overall capacity
factor (0.35 vs. 0.20 for wind, and 0.23 vs. 0.11 for solar) and the fact that the high demand
Pr
14
This is independent of the discount factor, which rather benefits technologies with higher upfront costs - see
Appendix A.3 for LFSCOE without discounting.
11
ed
periods in Texas (during summer days) are correlated with the high capacity factors for solar,
while the slightly higher demand in winter in Germany comes along with significantly lower solar
generation. This correlation explains the fact that the effective capacity factor (i.e. the average
iew
dispatched electricity per hour) for solar is almost six times larger in Texas than in Germany -
see Appendix A.3. Third, the variation between years of LFSCOEs for dispatchable sources are
much lower than for wind and solar. This does not come as a surprise as dispatchable sources
ev
depend only on one variable vector (demand) whereas wind and solar also depend on their own
hourly capacity factor and are particularly impacted by periods with low hourly capacity factor. If
consumers are risk averse, this higher variance can come with additional costs as they are willing
r
to pay a premium for capacity that is only used in the worst case scenarios.
er
Comparison with Levelized Costs of Electricity Using similar assumptions on costs as in
Table 1, EIA LCOE (2020) derives the Levelized Costs of Electricity. As mentioned above, this
pe
calculation is by no means simple, and is described in detail in EIA Model Documentation (2020).
Table 3 summarizes their analysis and compares it with the LFSCOE derived above.
LCOE and LFSCOE are relatively similar for all intermediate-load technologies (Biomass, Coal,
and Natural Gas CC), which is a sign that the capacity factor assumed for the LCOE (and thus
ot
the lifetime generation) is similar to the average capacity factor if only one of those technology
(plus storage) is responsible for meeting the market demand. Natural Gas CT, seen as a peak
tn
load generator under LCOE assumptions, has significantly lower LFSCOE as its capacity factor
increases from 30% to almost 70% in Germany, while the capacity factor for nuclear, being the
highest under LCOE assumptions (90%), drops to just under 80%.
rin
The most striking difference can be seen for the intermittent technologies solar and wind. While the
LCOE assume no responsibility in meeting the demand and focus solely on the costs of generation,
the LFSCOE assume full responsibility of meeting the demand. This responsibility comes at a
ep
very high price, making the LFSCOE for intermittent renewables up to 40 times higher than the
LCOE.
Pr
Comparison with System Levelized Costs of Electricity Ueckerdt et al. (2013) developed
the System LCOE to address the inability of LCOE to reflect the cost of intermittency. System
12
ed
Table 3: Comparison of LCOE and LFSCOE
iew
LFSCOE
LCOE
Technology Germany Texas
[USD/MWh] [USD/MWh] [USD/MWh]
Biomass 95 104 117

Coal (USC) 76 78 90
ev
Natural Gas CC 38 35 40
Natural Gas CT 67 39 42
Nuclear 82 106 122
r
Solar PV 36 1548 413
Wind 40 504 291
er
Table 3 displays the LCOE for onshore wind, while the LCOE for offshore wind are 122 USD/MWh. The data on the hourly wind
capacity factor does not distinguish between on-shore and off-shore wind, thus the LFSCOE implicitly assume an on-shore/off-shore
mix as currently deployed in the market in the given year. The highest share of off-shore capacity in the data was in Germany in 2018,
pe
which was ∼ 5GW offshore of the total ∼ 55GW installed capacity. The LCOE also include transmission costs (ranging between 1.1
USD/MWh for nuclear to 3.6 USD/MWh for solar), which are not part of the LFSCOEs (but can be added as a mark-up to the fixed
costs).
LCOE are the sum of generation costs and integration costs for a renewable source. Integration
ot
costs are split up into overproduction costs, full-load-hour reduction costs, and backup costs (plus
grid costs and balancing costs, but they are ignored in parts of the paper and in my analysis as
tn
well).15 The System LCOE are a very accurate way of calculating the cost of renewables, but
their precision makes them less “catchy”, as they depend on the share of the renewable genera-
tion. Figure 3 shows the System LCOE for wind and solar in Germany, taken from Ueckerdt et
rin
al. (2013).
It is important to note that as the paper was published in 2013, the LCOEs (60 EUR/MWh for
wind and 120 EUR/MWh for solar) are outdated as they dropped significantly since. It is interest-
ep
15
Ueckerdt. et al (2013) define balancing costs of VRE (i.e. renewables) as follows: “Balancing costs occur
because VRE supply is uncertain. Day-ahead forecast errors and short-term variability of VRE cause intra-day
adjustments of dispatchable power plants and require operating reserves that respond within minutes to seconds.”
Pr
(page 65). The LFSCOE model assumes deterministic demand and capacity factors while the shortest time interval
is one hour. Thus, although storage is required to balance supply and demand, “balancing costs” as defined by
Ueckerdt et al. (2013) are ignored in the LFSCOE model.
13
ed
Figure 2: System LCOE for Wind (left) and Solar (right) in Germany.
iew
ev
Figure 2: Graphs are taken from Ueckerdt et al. (2013), page 72, Figure 10.
r
ing to see that both LCOE and LFSCOE can be found in a complete System LCOE study: LCOE
er
are similar to System LCOE with renewables supplying a 0% share of final electricity, whereas
LFSCOE are conceptually equivalent to System LCOE with renewables supplying a 100% share
pe
of electricity. Indeed, if the LCOE values are adjusted to current LCOE estimations (which are
40 USD/MWh for wind and 34 USD/MWh for solar) and integration costs are extrapolated, the
System LCOE ought to approach the direction of the LFSCOE as calculated in this paper. If the
integration costs are extrapolated in Figure 2 in a linear fashion, the System LCOE will end up
ot
substantially below the LFSCOE for Germany presented in Table 4 - but if an exponential growth
of integration costs is assumed, the System LFSCOE actually reach the magnitude of LFSCOE
tn
values as calculated above.

A key observation from the System LCOE analysis in Figure 2 is the increasing cost of over-
production as the share of renewables increases. The analysis of the effective capacity factors in
rin
Appendix A.3 shows that LFSCOE supports this observation: In a system with solely intermittent
generation, overproduction occurs on a large scale. However, the analysis of the effect of storage
losses in the next section shows an advantage of overproduction. It turns out that even significant
ep
losses in the storage process do not increase the LFSCOE by much, as the additional demand in
some periods (due to storage losses) is more than compensated by the overproduction in other
periods without having to invest more in expensive storage.
Pr
14
ed
Limitations of the LFSCOE As noted above, the LFSCOE can be seen as an opposite ex-
treme of the LCOE and thus share similar limitation. LFSCOE are dependent on the location
and demand patterns, and should thus be presented as intervals if derived for a global context.
iew
Furthermore, they do not internalize externalities and regulatory market conditions. Unlike LFS-
COE, they include the cost of intermittency, but are much more dependent to outliers in demand
and renewable capacity factors.16
ev
3 Counterfactuals and Model Extensions
r
Counterfactual: Impact of falling storage costs Using the initial model as described above,
LFSCOE can be determined if the costs for storage decrease significantly.
er
Figure 3: LFSCOE with decreasing capacity costs for storage.
pe
ot
tn
Figure 3: Development of LFSCOE if storage costs decrease significantly for the market in Germany (left) and Texas (right).
rin
Figure 3 supports the intuition: Technologies that require large storage facilities (like wind and
solar) benefit from a significant decrease in storage costs, whereas the effect on the LFSCOE for
dispatchable technologies (like nuclear and natural gas) is barely noticeable. A reduction in costs
ep
of storage capacity by even 95% would still not make the LFSCOE of wind, solar, or wind & solar
competitive to the dispatchable generation in Germany, but would at least in Texas move them
below the LFSCOE of nuclear and biomass.
Pr
16
It is worth noting that the the combination of demand and capacity factors during the Texas Freeze 2021 did
not constitute an outlier for the LFSCOE calculations.
15
ed
It is important to note that the dispatch constraint (B.1) is never binding for the renewables,
meaning that a drop in storage costs can also be achieved by an increase in the storage factor.
While a decrease of costs per MW installed capacity by 75% seems unrealistic any time soon (see
iew
Schmidt et al. (2019)), an increase in capacity factor by a factor of 4 (to 12 M W /M W h) can
be achieved more easily with storage technologies like flow batteries (which, however, have higher
capacity costs at the moment).
ev
Model Adjustment: Including storage losses and depreciation Next, the no-storage-
losses assumption is relaxed. To simplify, there are three fundamentally different losses that
r
occur during a storage process: Losses can occur when the storage is charged with the generated
electricity (“Charging loss”), when it is dispatched (“discharging loss”) or while storage is idle but
er
charged to a certain point (“storage depreciation”) - see Appendix A.6 for further information.
Figure 4 illustrates the effect of charging losses (left) and discharging losses (right) for the LFSCOE
pe
with German market data, assuming that there are no other storage losses (e.g., no storage
depreciation).
Figure 4: LFSCOE with charging and discharging losses.

ot
tn
rin
ep
Figure 4: Note that the LFSCOE with charging losses of 40% are lower for wind and wind & solar than those with discharging losses
of 20% (522 vs. 565 for wind and 461 vs. 501 for wind & solar). For solar, LFSCOE with charging losses of 40% are higher than for
LFSCOE with discharging losses of 20% (1752 vs. 1656), but those with charging losses of 30% are lower (1627 USD/MWh).
The most striking observation is the low cost impact of storage losses, especially charging losses,
Pr
in the wind and the wind & solar system. This stems from the substantial overproduction under
the optimal capacity installation. Without storage losses, the abundant electricity is curtailed;
16
ed
with losses, it is just lost in the storage process. Given the different scale of the x-axis, one can say
that discharging losses, though very small for wind and wind & solar, are roughly twice as costly
as charging losses. Results for Texas (displayed in Appendix A.6) support these observations.
iew
Model Adjustment: LFSCOE-95 Immediate critiques of LFSCOE address the unrealistic
assumption that an electricity market will rely on only one source of electricity (unless it is hydro,
which is basically the only source of domestic power generation for some countries with very
ev
favorable geography like Paraguay or Albania). In addition, research on a complete reliance on
intermittent renewables points out the significant cost reduction that could be achieved if the
r
system would allow dispatchable technologies to back up renewables by supplying a small share
of demand (see for example Jenkins et al. (2018)). This thought experiment is included by
er
introducing the LFSCOE-95, the Levelized Full System Costs of Electricity if only 95% of the
market has to be supplied by this respective source of electricity. To calculate the LFSCOE-95,
pe
assume that a generator is available that can generate electricity at the lowest costs available
for intermediate-load generation (i.e. mc95 = 18 U SD/M W h) but is restricted to only 5% of the
demand. After calculating the total costs associated to each considered technology (plus storage),
the LFSCOE-95 are then calculated by averaging not over the total system demand (which is
ot
jointly supplied by the respective technology and the low-cost supply) but only over the total
demand that is supplied by the respective source of electricity (which is at least 95% of the total
tn
electricity). This means that the low-cost generation is basically treated as an adjustment of the
demand curve (i.e. it reduces up to 5% of total demand at a price of mc95 = 18 U SD/M W h, but
can choose in which period it reduced the demand).
rin
Table 4 displays the mean LFSCOE-95 for the data on Germany and Texas and compares it with
the LFSCOE-100, while Figure 5 displays mean and intervals for the LFSCOE-95.
There are a few things worth mentioning: First, for dispatchable generation, the LFSCOE-95
ep
are lower than LFSCOE as the residual demand curve for the generator is flattened by the free
generation, which increases the average capacity factor. The flattened seasonal demand curve also
Pr
reduces the difference in LFSCOEs between Texas and Germany. An additional consequence of the
flattened demand curve is that the variance between years almost diminishes. For intermittent
sources, all effects observed for the dispatchable technologies are more extreme. Going from
17
ed
Table 4: Levelized Full System Costs of Electricity
iew
Germany Texas
Technology LFSCOE-100 LFSCOE-95 LFSCOE-100 LFSCOE-95
[USD/MWh] [USD/MWh] [USD/MWh] [USD/MWh]
Biomass 104 90 117 95

Coal (USC) 78 67 90 72
ev
Natural Gas CC 35 31 40 32
Natural Gas CT 39 36 42 37
Nuclear 106 90 122 96
r
Solar 1548 849 413 177
Wind 504 279 291 131
Wind & Solar 454
er
220 225 97
Table 4 - LFSCOE-95: For computational reasons, storage is not an option for dispatchable technologies anymore. Given that only at
pe
most 2% of electricity was dispatched from storage (see Table 3), this restriction does not distort the results.
LFSCOE-100 to LFSCOE-95, i.e. reducing the load responsibility of wind or solar from 100% to
95%, reduces the costs by roughly 50%. However, the LFSCOE-95 for wind and solar in Germany
are still significantly higher than the LFSCOE-100 for all dispatchable sources, but especially the
ot
wind & solar mix in Texas appears to be as competitive as all non-natural-gas thermal generation
on a LFSCOE-95 basis. In any case, the LFSCOE-95 show that having dispatchable generation
tn
to support the intermittent renewables reduces the total system costs significantly and should be
considered when planning the energy transition.
rin
4 Conclusion
Intermittency of generation makes the cost comparison between different generation technologies
ep
much more difficult. While being a good measure to evaluate the cost to generate electricity, the
most popular cost measure, the Levelized Costs of Electricity, fails to include the costs associated
Pr
with meeting the demand and providing usable electricity. On the other hand, the System Lev-
elized Costs of Electricity by Ueckerdt et al. (2013) include the cost of integration and balancing,
but do not seem to be simple enough to make it to a broader audience. Using the radical but
18
ed
Figure 5: LFSCOE95 Mean and Variance in the German (left) and ERCOT market
iew
r ev
straightforward assumption that each source of generation has to meet the demand over a given
er
year (with the help of storage), the Levelized Full System Costs of Electricity introduced in this
paper are the first cost measure to condense the cost of providing electricity to one number per
market and technology. With LFSCOE being much higher than the LCOE for wind and solar,
pe
it becomes evident that LCOE are far from being an accurate measure to include the cost of
intermittency.
Analyzing different sources of generation sources shows that the LFSCOE are much higher for
ot
wind and solar than for conventional and dispatchable fuels, which stems from the large require-
ment for storage to overcome wind and solar’s intermittency. However, even if these storage costs
tn
drop by 90%, renewables are still not competitive on an LFSCOE basis.

Introducing the LFSCOE-95, which assume that up to 5% of the annual demand can be supplied
by a very inexpensive dispatchable source of electricity, we show that reducing the responsibility
rin
of intermittent renewables to supply only 95% of the demand will cut the system costs in half.
This observation is supported by existing literature criticizing any 100% emission-free approaches
by pointing out the enormous costs of supplying the last 5%.
ep
The cost evaluation concept in this paper can be refined in different ways. First, the evaluation of
LFSCOE can be refined by including transmission costs, capacity limits, or locational differences
in capacity factors. Second, using data from other parts of the world, LFSCOE for other markets
Pr
can be evaluated. Third, LFSCOE-95 can be relaxed further to LFSCOE-90 or any reasonable
demand reduction. Any extension, however, should find a balance between increasing accuracy
19
ed
while staying simple to address a broader audience.
iew
References
Battery University. (2018). “What does Elevated Self-discharge Do?”. see
https://batteryuniversity.com/learn/article/elevating self discharge.
ev
Becker, Gary S. (1965). “A theory of the allocation of time.” The Economic Journal, 75, 493-517.
United States Census Bureau. (2019). “American Community Survey - Home Heating Fuel”.
r
see https://www.census.gov/acs/www/about/why-we-ask-each-question/heating/
er
Cleanenergywire. (2020). Data on German sources of heating, see:
https://www.cleanenergywire.org/factsheets/heating-40-million-homes-hurdles-phasing-out-fossil-
fuels-german-basements
pe
Denholm, Paul, Jeffrey C. King, Charles F. Kutcher, Paul P.H. Wilson. (2012). “Decarbonizing
the Electricity Sector: Combining Renewable and Nuclear Energy using Thermal Storage.”
Energy Policy, 44: 301-311.
ot
Hartley, Peter. (2017). “The Cost of Displacing Fossil Fuels: Some Evidence from Texas.” Rice
tn
University, Working Paper.
Ibrahim, H., A. Ilinca, and J. Perron. (2008). “Energy Storage Systems - Characteristics and
Comparison.” Renewable and Sustainable Energy Reviews, 12: 1221-1250.
rin
Jenkins, Jesse D., Max Luke, and Samuel Thernstrom. (2018). “Getting to Zero Carbon
Emissions in the Electric Power Sector.” Joule, 2: 2498-2510.
ep
Joskow, Paul. (2011). “Comparing the Costs of Intermittent and Dispatchable Electricity
Generating Technologies.” American Economic Review: Papers & Proceedings, 100:3: 238-241.
Pr
Lazard. (2019). “Lazard’s Levelized Costs Of Energy Analysis - Version 13.0”.
20
ed
Reichenberg, Lina, Frederik Hedenus, Mikael Odenberger, Filip Johnsson. (2017). “The Marginal
System LCOE of Variable Renewables - Evaluating High Penetration Levels of Wind and Solar
in Europe.” Energy, 152: 914-924.
iew
Schmidt, Oliver, Sylvain Melchior, Adam Hawkes, Lain Staffel. (2019). “Projecting the Future
Levelized Cost of Electricity Storage Technologies.” Joule, 3: 81-100.
ev
Sinn, Hans-Werner. (2017). “Buffering Volatility: A Study on the Limits of Germany’s Energy
Revolution.” European Economic Review, 99: 130-150.
Ueckerdt, Falko, Lion Hirth, Gunnar Luderer, Ottmar Edenhofer. (2013). “System LCOE:
r
What are the costs of variable renewables?”. Energy, 63: 61-75.

er
United States Census Bureau. (2019). “American Community Survey - Home Heating Fuel”.
see https://www.census.gov/acs/www/about/why-we-ask-each-question/heating/
pe
United States Energy Information Administration. (2009). “Household Energy Use in Texas”.
EIA’s 2009 Residential Energy Consumption Survey. Cited as “EIA Texas (2009)”.
United States Energy Information Administration. (2020). “Levelized Costs and Levelized
ot
Avoided Cost of New Generation Resources”. Annual Energy Outlook 2020. Cited as “EIA
LCOE (2020)”.
tn
United States Energy Information Administration. (2020). “The Electricity Market Module of
the National Energy Modeling System: Model Documentation 2020”. Annual Energy Outlook
rin
2020. Cited as “EIA Model Documentation (2020)”.
United States Energy Information Administration. (2020). “Cost and Performance Character-
istics of New Generation Technologies”. Annual Energy Outlook 2020. Cited as “EIA Costs
ep
(2020)”.
Zerrahn, Alexander, Wolf-Peter Schill, Claudia Kemfert. (2018). “On the Economics of Electri-
Pr
cal Storage for Variable Renewable Energy Sources.” European Economic Review, 108: 259-279.
21
ed
Appendix A
A.1 Comparison of demand patterns in Germany and Texas Figure 6 displays the aver-
iew
age daily peak and minimal demand for Germany and ERCOT/Texas, where years are normalized
so that each year starts with a Monday. The comparison shows that the seasonal demand varia-
tion in Germany is relatively small compared to Texas. This does not come as a surprise as air
conditioning accounts for a large share of residential electricity consumption in Texas, whereas it
ev
is rarely utilized in Germany. However, heating is more important in Germany than in Texas, but
while around 60% of the housing units in Texas use electricity for heating (see United States Cen-
sus Bureau (2019)), the main source of heating in Germany is natural gas (48.2% of households)
r
and oil (25.6% of households).17 Once heating gets electrified on a large scale, there should be a
er
larger difference between the electricity demand in summer and in winter in Germany (unless air
conditioning becomes popular in Germany as well).
pe
Figure 6: Average Minimal and Peak Demand in Germany (left) and the ERCOT market.
ot
tn
rin
Figure 6 displays the average daily minimal and maximal demand in Germany and Texas over 8 years. The days are adjusted for
weekdays, thus the first day of a year is always a Monday.
ep
17
For Germany, see Cleanenergywire (2020); for Texas, see EIA Texas (2009). In 2009, households in Texas
use 18% of their total energy for cooling, almost exclusively by electric air conditioning units - which is very high
compared to the U.S. average of 6%. While the data shows that households in Texas use 22% of the energy for
space heating (compared to 41% in the U.S.).
Pr
22
ed
A.2 Existence and uniqueness of solution Existence of a solution is easily proved as the
set of constraints is clearly convex and non-empty. In fact, any capacity installation that is large
enough can meet the constraints.
iew
Uniqueness is more ambiguous: The solutions to both the intermittent and the dispatchable
problem are not necessarily unique. In fact, without losses in the storage process, the optimization
problems for both intermittent and dispatchable sources have an infinite number of solutions as
ev
the storage can be charged at any time (unless there is no storage capacity in the optimal system,
as for natural gas CT). Adding a bonus of > 0 for keeping electricity in storage, which can be
justified as storage secures against unanticipated supply shortages, ensures the uniqueness of x for
r
every investment tuple (I y , Istorage
y
) of renewable sources. While uniqueness of the independent
variables can still not be ensured, the value of the objective function is unique, which is sufficient
for the purpose of this paper. er
pe
A.3 Storage Characteristics Table 3 displays characteristics of the storage in the system. It is
not surprising that intermittent renewables require significantly more storage than the dispatchable
technologies, out of which those with low capacity costs like natural gas CC and CT barely use
storage at all. Furthermore, as the hourly capacity factor for solar is equal to 0 in almost half of
ot
the periods (at night), it is not surprising that almost half of the electricity in the market solely
supplied by solar is dispatched from storage. The most surprising observation is probably the
tn
length of storage cycles, which are not longer than 334 hours (i.e. about 14 days) in Texas and
903 hours (i.e. about 38 days) in Germany. The main reason for the short storage cycles is the
substantial overproduction, which is apparently cheaper than investing in more storage. Relaxing
rin
some model assumption like perfect forecastability would certainly increase the length of storage
cycles, but it raises the question whether seasonal storage is required if it stays expensive (at least
in Texas).
ep
A.4 Counterfactual: No discounting Generation technologies with low capital costs but
higher variable costs benefit from a lower discounting factor (i.e. a higher cost of capital) compared
Pr
to those with higher capital costs and lower variable costs. As a thought experiment, Table 6
displays the LFSCOE (mean and min-max interval) without discounting of the future (i.e. a cost
23
ed
Table 5: Storage details on calculating the LFSCOE
iew
Germany Texas
Max Max
Technology Storage Storage storage Storage Storage storage
Generation Generation
Capacity Capacity dispatch cycle Capacity Capacity dispatch cycle
[GW] [GW] [%] [hours] [GW] [GW] [%] [hours]
Biomass 76 14 0.1% 94 53 33 1.9% 94
ev
Coal 79 5 0.05% 64 58 13 0.7% 64
Natural Gas CC 83 0.3 0% 15 69 0.3 0.0% 15
Natural Gas CT 84 0.2 0% 6 69 0 0.0% 6
r
Nuclear 73 31 0.5% 146 52 34 2.0% 146
Solar 3709 2078 48.3% 909 438 792 49.6% 332
Wind
Wind&Solar
1027
916
1030
948
9.2%
6.2% er 664
669
517
368
276
262
3.5%
1.5%
128
117
pe
Table 5 depicts the average storage capacity, generation capacity, storage dispatch (i.e. percentage of electricity coming from storage)
and the longest storing cycle. In Germany, both NGCC and NGCT use storage in only one of the 8 years. Their average dispatched
electricity is not equal to 0 (apart from NGCT in Texas), but on average lower than 0.1%. Note that the average hourly demand is 58
GWh in Germany and 40 GWh in Texas.
of capital of 0 %).
ot
A.5 Effective Capacity Factor Define the “effective” capacity factor as the average dispatched
tn
electricity per hour. This capacity factor is capped by the “technical” capacity factor, which is
the (maximal) generation of a source of electricity. The average technical capacity factor in the
data for wind (solar) is 35% (23%) in Texas and 18% (11%) in Germany. Table 8 displays the
rin
average effective capacity factor for each technology and the interval ranging from the minimal
to the maximal capacity factor for each year. As net dispatch from storage is not permitted and
there are no losses in the storage process, this can be calculated by dividing the average hourly
ep
demand by the installed capacity. Some observations can be made: First, the capacity factors for
dispatchable sources in Texas are lower than in Germany, which stems from the higher seasonal
variance of the demand. This means that capacity needs to be provided in Texas which only
Pr
becomes relevant for a few month in the summer. A direct implication of the lower capacity
factors are the higher LFSCOE values of dispatchable generation technologies in Texas compared
24
ed
Table 6: Levelized Full System Costs of Electricity without discounting
iew
Germany Texas
Technology LFSCOE [min,max] Conf.Int. LFSCOE [min,max] Conf.Int.
[USD/MWh] [USD/MWh]
Biomass 71 [69,74] 78 [75,82]

Coal 52 [50,53] 57 [55,60]
ev
NGCC 26 [26,27] 29 [28,29]
NGCT 33 [33,34] 35 [34,35]
Nuclear 60 [58,64] 68 [65,73]
r
Solar 734 [564,965] 188 [155,266]
Wind 247 [214,276] 146 [117,174]
W+S 221
er
[160,240] 111 [88,169]
Table 6: Levelized Full System Costs of Electricity without discounting (i.e. with a cost of capital of 0%).
pe
to Germany. Second, the effective capacity factors for wind and solar are significantly lower
than the technical capacity factor, implying that a large share of electricity gets curtailed. An
interesting observation is that though the technical capacity factors for solar are just twice as high
ot
in Texas than in Germany (23% vs. 11 %), the effective capacity factor is almost six times higher
in Texas than in Germany. This means that not only solar do generators generate less in Germany,
tn
they also sell significantly less electricity in a single-source market, making the investment even
less profitable.
rin
A.6 Storage losses Next, the no-storage-losses assumption is relaxed. To simplify, there are
three fundamentally different losses that occur during a storage process (see Ibrahim et al. (2008)
for a detailed description of storage technologies, and Schmidt et al. (2019) for current round-trip
ep
efficiency estimations). First, losses can occur when the storage is charged with the generated
electricity. Let α1 ∈ (0, 1] be the share of electricity that reaches the storage facility in a charging
process (and thus 1 − α1 the share that is lost in the process), meaning that α1 = 1 would be
Pr
equivalent to no losses in the charging process. For many storage technologies, this is the main
location for losses. Especially, turning water into hydrogen (so called Power-to-Gas) requires a
25
ed
Table 7: Average Annual Effective Capacity factor
iew
Germany Texas
Technology LCOE
LFSCOE [min,max] LFSCOE [min,max]
Biomass 83% 76.0% [72.0%,81.2%] 76.0% [70.9%,80.6%]

Coal 85% 73.7% [68.8%,81.2%] 68.9% [61.0%,73.5%]
NGCC 87% 69.9% [65.5%,73.8%] 58.2% [55.9%,64.8%]
ev
NGCT 30% 69.6% [65.5%,73.5%] 58.1% [55.6%,64.4%]
Nuclear 90% 79.9% [74.6%,84.8%] 76.9% [70.9%,81.8%]
Solar 29% 1.5% [1.2%,2.0%] 10.4% [5.0%,12.9%]
r
Wind 40% 5.9% [3.7%,7.5%] 7.8% [6.5%,9.6%]
Wind & Solar – 6.5% [4.5%,7.7%] 10.9% [9.0%,12.4%]
er
Table 7: Capacity factors for LCOE taken from EIA LCOE (2020), where the table displays the capacity factors for onshore wind.
The capacity factor for Wind & Solar is derived by dividing the the sum of the demand by the sum of the installed capacity.
pe
significant amount of energy that cannot be fully recaptured, which is structurally nothing else
than charging losses. Second, stored electricity can depreciate over time - this loss is denoted with
γ ∈ (0, 1]. Storage depreciation (or self-discharge) can have many different reasons: For hydrogen,
ot
it can occur due to leaks, for pumped hydro, it occurs due to evaporation (but might be offset
in part through rainfall and runoff), and for solid state batteries, it occurs due to an unwanted
chemical reaction. Third, losses can occur when the stored electricity is dispatched back to the
tn
grid - denote the share of electricity that reaches the grid by α2 ∈ (0, 1], i.e. the share of losses
is 1 − α2 . It is easy to see that “round-trip efficiency”, the most common term to describe the
efficiency of storage, is nothing else than α1 ⋅ α2 , and varies between 0.4 for Power-to-Gas and 0.88
rin
for the flywheel as depict in Table 5. Unfortunately, the technologies with the highest round-trip
efficiency sometimes come along with the highest depreciation (i.e. the lowest γ). In a context
ep
where it is desirable to store the energy for different length of time, it will be optimal to have a
diversified storage portfolio.
Pr
26
ed
Table 8: Storage losses by technology
Round-trip
iew
efficiency Self-discharge
Technology
[%] [%/day]
Pumped Hydro 78% 0%

Flywheel 88% 480%
Lithium-Ion 86% 0%
ev
Vanadium redox-flow 73% 0%
Hydrogen 40% 1%
r
Table 8 displays round-trip efficiency and self-discharge of selected storage technologies, taken from the accompanying documents of
Schmidt et al. (2019) (Table S4). The technology input parameters are from 2015 and the self-discharge is at an optimal charging
level. A self-discharge level of 480%/day could be reinterpreted to 20%/hour which can then be interpreted at γ = 0.8 if discharging
er
occurs at a fixed non-linear rate (which is, however, not the case for the flywheel).
Given the notations above, the system of equations for intermittent technologies changes to
pe
min rp ⋅ f cRen + sp ⋅ f cstorage
rp,sp
s.t. x[t + 1] − γx[t] ≤ (Rent ⋅ rp − Dt )α1 , (D.1+)

1
x[t + 1] − γx[t] ≤ (Rent ⋅ rp − Dt )
ot
, (D.1-)
α2
−sp ≤ x[t + 1] − γ ⋅ x[t] ≤ sp, (B.1)
tn
Ssecurity ≤ x[t + 1] ≤ sp ⋅ ρsf , (B.2)
x[1] ≤ x[H + 1]. (B.3)
Constraints (D.1+) and (D.1−) ensure that demand is met at any time: (D.1+) has to hold in
rin
periods where storage is charged (i.e. with an abundance of electricity), whereas (D.1−) corre-
sponds to periods where electricity is generated by discharging storage.18 Note that the system
ep
above only captures exponential storage depreciation γ. However, technologies like lithium-ion
batteries self-discharge at other than an exponential rate over time. In that case, the percentage
loss depends on both the time and the battery state/charging level, where 5% of the load de-
Pr
preciates within a day (if the battery was fully charged) and then just up to another 5% within
18
Note that by replacing the left hand side of (S.1) with max(Dt − Rent ⋅ rp) ≤ sp, which can be replaced by
α2 (Dz − Rz ) ⋅ rp ≤ sp at z = argmaxt Dt /Rent , H − 1 conditions can be removed from the optimization problem.
27
ed
the next month - see Battery University (2018). Given that storage cycles tend to be short, the
long-term depreciation can be ignored in the analysis, while the short-term storage depreciation
can be interpreted as a charging and discharging loss. As a result, γ will be fixed at 1 and the
iew
focus will lay on charging and discharging losses.
Figure 8 depicts the effect of the LFSCOE values in Texas if charging and discharging costs are
introduced. The observations of the LFSCOE in Germany hold here as well.
ev
Figure 7: LFSCOE with charging and discharging losses in Texas.
r
er
pe
Figure 7 displays the LFSCOE in Texas with charging losses and discharging losses. Note that for every technology, the LFSCOE
with charging losses of 40% are always lower than those with 20% discharging losses.
ot
tn
rin
ep
Pr
28

SSRN Lcoe

Uploaded by

Copyright:

Available Formats

SSRN Lcoe

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SSRN Lcoe

Uploaded by

Copyright:

Available Formats

ed

Levelized Full System Costs of Electricity

Keywords: Intermittent renewables, Levelized Costs of Electricity (LCOE), System LCOE,

Biomass 4,401 125.2 28 150%/50%

min rp ⋅ f cRen + sp ⋅ f cStorage

s.t. 0 ≤ x[t + 1] ≤ x[t] + Rent ⋅ rp − Dt f or all t, (D.1)

−sp ≤ x[t + 1] − x[t] ≤ sp f or all t, (S.1)

Ssecurity ≤ x[t] ≤ sp ⋅ ρsf f or all t, (S.2)

x[1] ≤ x[H + 1], (S.3)

−sp ≤ x[t + 1] − x[t] ≤ sp f or all t, (S.1)

Ssecurity ≤ x[t] ≤ sp ⋅ ρsf f or all t, (S.2)

gen[t] ≤ dp f or all t, (G.1)

maxcapx ⋅ dp + sp ≥ max(Dt ). (X.1)

I Ren ⋅ f cRen + Istorage

Biomass 104 [100,109] 117 [112,126]

Biomass 95 104 117

values as calculated above.

Figure 4: LFSCOE with charging and discharging losses.

Biomass 104 90 117 95

drop by 90%, renewables are still not competitive on an LFSCOE basis.

University, Working Paper.

 Lazard. (2019). “Lazard’s Levelized Costs Of Energy Analysis - Version 13.0”.

2020. Cited as “EIA Model Documentation (2020)”.

Biomass 71 [69,74] 78 [75,82]

Biomass 83% 76.0% [72.0%,81.2%] 76.0% [70.9%,80.6%]

Pumped Hydro 78% 0%

s.t. x[t + 1] − γx[t] ≤ (Rent ⋅ rp − Dt )α1 , (D.1+)

Ssecurity ≤ x[t + 1] ≤ sp ⋅ ρsf , (B.2)

x[1] ≤ x[H + 1]. (B.3)

You might also like

Lazard. (2019). “Lazard’s Levelized Costs Of Energy Analysis - Version 13.0”.