Learning-by-Doing vs. Barriers to Cost Reduction:
David vs. Goliath or Bambi vs. Godzilla?
A Skeptical Reaction to Richard Duke’s “Learning By Doing and the Photovoltaics Case”
RFF Workshop on Learning-by-Doing in Energy Technologies
June 17-18 2003, Washington D.C.
Ian Sue Wing
The paper by Dr. Richard Duke surveys and synthesizes the literature on the economics of solar
photovoltaics (hereafter PV), analyzes the extent to which learning-by-doing justifies their
subsidization, and draws conclusions for policies to support the penetration of PV into electricity
markets. Unfortunately, however, the paper’s conceptual premise and the analysis that follows
from it suffer from problems that obscure both the mechanisms through which learning-by-doing
(LBD) can contribute to reducing the cost of PV electricity, the nature of the barriers to it doing
so, and, consequently, the magnitude of the uphill battle that PV faces in becoming competitive
with other sources of electricity supply. While Duke appears to be a true believer in the potential
for PV to provide a significant share of future electricity generation, I am quite skeptical for the
key reasons of cost and technical potential, on which I shall elaborate below.
Accounting Fallacies: Watts-Peak vs. Kilowatt-Hours
PV systems have little intrinsic value to end-users of energy, whose main concern is keeping the
lights on at low cost. The degree to which PV modules are demanded will depend on the
availability and cost of the electricity they produce relative to that generated by other sources.
The paper—and the PV industry—use cents per watts-peak (¢/Wp) as a measure of cost, which
is the cost of conversion efficiency (cents per watts of electricity generated from a standard
wattage of incident radiation per unit area), not the cost of the energy thus generated. The
question to which we seek an answer is not whether LBD reduces the cost of PV modules in
¢/Wp, it is whether it reduces the cost of PV electricity in ¢/kWh, to the point where PV is costcompetitive with other sources of supply, that may themselves enjoy cost-reductions due to
learning or other factors! Thus, LBD is only consequential if it permits reductions in the cost of
PV output that are larger than those experienced by other electricity supply technologies for the
same incremental investment in capacity. This is a fundamental point that I think deserves
explicit recognition in the paper, about which there seems to be general confusion in the
literature on LBD in energy technologies.
Cost, Cost, Cost (of Electricity, That Is!)
PV electricity is more expensive than that generated by other sources, and our purpose here is to
understand why that is and what can be done about it. A systematic way to proceed is to (1)
estimate how much more costly PV generation is than the competitive market price of electricity,
(2) account for the key components of PV’s cost, (3) analyze the characteristics of these
components to get a sense of the technological, economic and political forces to which cost
responds, and (4) develop strategies to reduce the cost of PV electricity to competitive levels by
exploiting the margins of adjustment uncovered in the step (3). This is what I shall attempt to do
here.
1
To begin with Step 1, it is natural for an economist to assume that, when faced with a choice
between expensive electricity from PV or cheap electricity from competitor technologies, a
consumer would choose the latter, less expensive option. On p. 17 Duke argues that PV
technology is a good candidate for subsidization because there are few substitutes for it, in effect
suggesting that the choice I have just outlined will be made based on things other than cost. But
to my mind this can only happen in specific niche markets. It is hard to envision that consumers
will not notice that at some point on the world electricity demand curve PV electricity can be
more cheaply generated by other sources, and will not thereupon make the appropriate
substitutions. Moreover, Table 1 suggests that they will not have to go very far down the demand
curve for this to happen. Comparing the estimates of current and future cost of PV in columns (5)
and (6) with data on other technologies reveals that PV currently is, and is projected to continue
to be, a relatively costly method of generating electricity. This conclusion is corroborated by
McVeigh et al (1999), who find that at over 20 ¢/kWh, PV is more than four times as expensive
as other renewable sources of electricity.
Low Demand for Output = Low Demand for Inputs
Such a large relative cost difference is a difficult hurdle to surmount. PV systems produce a
small quantity of high-priced electricity, which translates into low demand for the inputs to PV
generation: the most important of which are PV modules. Therefore, absent more vigorous
attempts by government to stimulate demand for PV modules over the near- to medium-term, it
is unlikely that the volume of solar arrays demanded will be sufficient for LBD to generate
reductions in their cost necessary to reduce the cost of PV electricity to competitive levels, in
spite of high progress ratios in solar cell manufacturing.
This vicious cycle is the main barrier to cost reduction, and highlights to what I see as a serious
flaw the analysis on pp. 11-13 of Duke’s paper. Duke derives a demand curve for PV modules in
the U.S., which he then extrapolates (using what appear to be tenuous assumptions) to create a
global demand curve for PV systems that shifts out over time. But the question is, how can there
by such vigorous growth in demand for the input to PV generation when the output of generation
is so expensive? Given the high current cost of PV electricity and the consequent low demand for
arrays, the growth in demand that Duke seems to assume will occur on its own has to be the
result of deliberate policies by governments to stimulate demand for PV systems through
guaranteed purchasing of PV electricity output. This means that Duke’s no-subsidy scenario is
actually premised on the continued existence of policies to subsidize PV output over at least a
number of decades. Given the large cost-differential between PV electricity and that generated
by other supply sources, such policies will likely be very costly. Duke’s analysis treats these
costs as an off-budget item, making his conclusions about the welfare impact of subsidizing PVs
highly optimistic.
Insolation: A Forgotten (and Forlorn) Input
Steps 2 and 3 reveals more bad news. Duke does not deal with the fact that there is another input
to the production of PV electricity that is an determinant of its cost: the geographically
heterogeneous solar resource and the area of the Earth’s surface on which it is incident. My
concern is that this input is both costly and limited in supply and in ways that are difficult for
LBD (or for that matter anything else) to mitigate, and that this in turn will serve to keep the cost
2
of PV electricity high, inhibiting PV output, demand for new PV modules, and learning,
perpetuating the vicious cycle identified earlier.
First and foremost, insolation determines PV’s potential contribution to electricity supply.
Hearteningly, the technical potential of the solar energy resource is large compared to that of
other renewables, as Table 2 shows. However, the economic potential of solar may be far
smaller, primarily because of its competition with other uses for land surface area (e.g., WEC
(1994) assessed its potential in 2020 to be 5-230 exajoules per year). Geographic variations in
land use constraints or subsidies for PV are often the result of economic and political factors that
are spatially uncorrelated with the distribution of the solar resource. For instance, Figures 1-4
show that in the U.S. many of the incentives for solar PV at the state level are concentrated
outside the southwest, where the highest insolation occurs (Figure 5). This is an even bigger
problem internationally. Developing countries possess abundant solar resources (Figure 6), but
their low incomes and technological capabilities diminish the attractiveness of deploying costly
and scientifically advanced technologies, and their sizeable share of GDP in agriculture implies
significant opportunity costs—both private (e.g. agricultural output) and social (e.g. food
security)—of using land for PV arrays. It therefore comes as no surprise that developing
countries produce only 15 percent of the world’s PV electricity (IEA, 2002). Moreover, it seems
implausible to me that OECD nations will make large income transfers to developing countries
in order to increase global demand for PV modules as a way of kick-starting LBD. The upshot is
a mix of technical, economic and political barriers to exploitation of the highest grade of the
resource input to PV generation, which means that its cost will remain higher, and its quantity
lower, than if a social planner could completely cover the sunniest locations around the globe
with solar collectors. It is an open question how consequential this cost premium is, but in light
of the limits to solar PV’s potential, it is worth asking whether we should not reconsider this
technology to be a minor player in the global energy picture, instead of one in which investment
will yield a large contribution to future supply.
A second key factor is the fundamentally intermittent and spatially distributed character of
electricity generation that makes use of insolation. Duke seems to suggest in several places in the
paper that grid-connected distributed generation systems are the emerging growth market for PV
modules. However, with this type of generation each array will have to be able to push electrons
back into the grid, which requires additional equipment in the form of inverters and control
electronics that must be replicated at every generating site. To me, this suggests that it will be
difficult to achieve economies of scale in distributed PV electricity production, which raises the
question of exactly how the cost of this technology can be reduced. Similarly, today’s electricity
grid is designed to support sources of supply that are centralized, fairly stable, and ultimately
controllable—criteria that grid-connected PV systems do not satisfy. The multiplicity of access
points to the grid required for distributed PV, coupled with the stochastic fluctuations in its
output due to variations in insolation, imply that growth in its share of total generation will likely
increase problems of grid control and reliability. I can see this having two effects. On the one
hand, it will necessitate additional investment in power system control equipment, further adding
to PV’s cost. One the other hand it is likely to feed political opposition to widespread
deployment of distributed generation by electric utilities, who have legitimate concerns about the
interconnection of generation equipment due to their responsibility to maintain the safety and
3
reliability of the grid, but who also have oligopolistic incentives to discourage self-generation by
electricity consumers.
Implications for PV and LBD
I do not have much to say about Step 4. To me, the foregoing caveats suggest that the bulk of
new PV capacity in the foreseeable future will still be of the central station type, at least until the
technical issues of PV’s intermittency and the associated costs of grid control can be jointly
addressed. Prior to innovations that mitigate these problems, it is optimistic to think that PV can
potentially supply anywhere near 10 percent of the 55 exajoule (and growing) annual global
demand for electricity (IEA, 2002) that is envisioned by various analyses (e.g. DOE/EPRI 1997).
This state of affairs may seem pretty dismal, but it does highlight two important general
principles. First, barriers to reducing production costs may arise from bottlenecks in markets for
both the outputs of the production process and non-capital-related inputs to the production
process, and may make it difficult to exploit economies of scale through LBD. Second, there is a
potential complementarity between technical problem solving (i.e., R&D) to remove these
bottlenecks and learning (e.g., as found by Lieberman (1984) and others). Investment in “debottlenecking” (which may not target the production process at all) can be conceptualized as a
process by which R&D expands a stock of productive knowledge that creates the potential for
cost reduction to be achieved through LBD.
R&D and Learning: Two faces of Cost Reduction? An Application to PV
To see how the intuition of the foregoing section may be applied to the case of PV electricity in a
formal way, consider the following dynamic model of an industry that generates PV electricity.
In each time period t the industry generates electric output y(t) by combining inputs of existing
generating capacity k(t) and insolation-surface area s(t) according to the production function F(·)
which is almost Leontief:
(1)
y (t ) = F ( k (t ), s (t )) .
The limited economic potential of the insolation resource as an input implies that there is an
upper bound s to its supply:
(2)
s(t ) ≤ s .
The sector make investments i(t) to expand capacity and offset depreciation, which is assumed to
occur at the rate :
(3)
k (t ) = i (t ) − δk (t ) .
It also undertakes R&D r(t), which augments a stock of cost-saving knowledge about the
production process h(t), which accumulates according to the knowledge production function
M(·):
(4)
h (t ) = M ( r (t ), h (t )) .
4
The net social benefits of PV generation NSB(t) depend on five factors:
• The private revenues from the sale of electricity y(t) at the market price p (assumed for
simplicity to be constant)
• The net social benefits of PV output (e.g., avoided costs of mortality and morbidity as a
result of reduced air pollution from the displacement of electricity generated using fossil
fuels) that are represented by a function B(y) and are assumed to be positive (By > 0, Byy < 0)
• The private cost of making and installing new PV modules, which rises with investment in
each period, falls with increasing experience x(t) in module manufacture/installation and with
advances in knowledge, and is represented by a function V(i, x, h) (Vi, Vii, Vxx, Vhh > 0; Vx, Vh,
Vix, Vih, Vxh < 0)
• The social opportunity cost of using land-surface area for PV generation, represented by the
function Z(s) (Zs, Zss > 0), and
• Expenditures on R&D.
For simplicity the variable costs of PV generation are assumed to be zero, which is realistic
because once manufactured and installed solar cells just sit out in the sun and generate
electricity, without requiring much further intervention:
(5)
NSB(t ) = py (t ) + B ( y (t )) − V (i (t ), x(t ), h (t )) − Z ( s (t )) − r (t )
Finally, experience is a stock that accumulates according to the history of investment:
(6)
x ( t ) = i (t ) .
The objective of a social planner is choose paths of investment in new PV capacity, land usage
and R&D to maximize welfare W, given by the present value of the net social benefits provided
by the technology, discounted at the rate :
∞
max W = NSB(t )e − ρt dt
i ( t ),r ( t ), s ( t )
0
subject to equations (1)-(6). A key feature of the model is that the net social benefits of PV
generation are negative in the initial period, i.e.
V (i (0), x (0), h (0)) + Z ( s(0)) + r (0) >> py (0) + B ( y (0)) ,
which raises the question of how strong the individual effects Vx and Vh, and the
complementarity between them Vxh, have to be in order to support a positive trajectory of
investment.
I have not yet analyzed this model, but my intuition is that something like it (in either analytical
or computational simulation form) can yield insights into the contribution of the learning process
to reducing the cost of PV generation, and the potential for PV to expand its share of the future
electricity market. It would probably be a useful component of a research agenda on the subject.
5
References
1. Department of Energy and EPRI (1997). Renewable Energy Technology Characterizations,
U.S. Department of Energy Topical Report No. 109496.
2. Energy Information Administration (2000). Energy Consumption and Renewable Energy
Development Potential on Indian Lands, Washington DC.
3. International Energy Agency (2002). IEA Energy Balances 2002, Paris: OECD/IEA.
4. Lieberman, M. (1984). The Learning Curve and Pricing in the Chemical Processing
Industries, Rand Journal of Economics15: 213-228.
5. McVeigh, J., D. Burtraw, J. Darmstadter and K. Palmer (1999). Winner, Loser or Innocent
Victim: Has Renewable Energy Performed as Expected? Renewable Energy Policy Project
Research Report No. 7, Washington DC.
6. United Nations Development Program (2000). World Energy Assessment: Energy and the
Challenge of Sustainability, New York: UNDP.
7. World Energy Council (1994). New Renewable Energy Resources: A Guide to the Future,
London: Kogan Page.
6
Table 1. Solar Photovoltaics Compared to Other Renewable Energy Sources: Key Characteristics
Technology
Biomass energy
Electricity
Heat
Wind electricity
Solar photovoltaic
electricity
Solar thermal
electricity
Low-temperature
solar heat
Hydroelectricity
Large
Small
Geothermal energy
Electricity
Heat
Marine energy
Tidal
Wave
Current
OTEC
(1)
Increase
in
installed
capacity
1993-98
(percent
per year)
(3)
Capacity
factor
(percent)
(2)
Operating
capacity,
end 1998
(GW)
(4)
Energy
production
1998
(TWh)
(5)
Current
energy
cost of
new
systems
(¢/kWh)
(6)
Potential
future
energy
cost
(¢/kWh)
3
3
30
30
40
> 200
10
0.5
25–80
25–80
20–30
8–20
160
> 700
18
0.5
5–15
1–5
5–13
25–125
4–10
1–5
3–10
5–25
5
0.4
20–35
1
12–18
4–10
8
18
8–20
14
3–20
2–10
2
3
640
23
35–60
20–70
2510
90
2-8
4–10
2-8
3–10
4
6
8
11
45–90
20–70
46
40
2–10
0.5–5
1–8
0.5–5
0
—
—
—
0.3
—
—
—
20–30
20–35
25–35
70–80
0.6
—
—
—
8–15
8–20
8–15
—
8–15
—
5–7
—
Source: UNDP (2000), Table 7.25.
7
Table 2. Solar Photovoltaics Compared to Other Renewable Energy Sources: Technical Potential
Resource
Hydropower
Biomass energy
Solar energy
Wind energy
Geothermal energy
Ocean energy
Current Use
9
50
0.1
0.12
0.6
—
Source: UNDP (2000), Table 5.26.
8
Technical
Potential
(EJ)
50
> 276
>1,575
640
5,000
—
Theoretical
Potential
(EJ)
147
2,900
3,900,000
6,000
140,000,000
7,400
Figure 1. State Net Metering Programs
D.C.
State-wide net metering rules
Net metering applies to select utilities only
Arizona, Arkansas, California, Colorado, Connecticut, Delaware, District of Columbia, Florida,
Georgia, Hawaii, Idaho, Illinois, Indiana, Iowa, Kentucky, Maine, Maryland, Massachusetts,
Minnesota, Montana, Nevada, New Hampshire, New Jersey, New Mexico, New York, North
Dakota, Ohio, Oklahoma, Oregon, Pennsylvania, Rhode Island, Texas, Utah, Vermont, Virginia,
Washington, Wisconsin, Wyoming
Source: Database of State Incentives for Renewable Energy, http://www.dsireusa.org
9
Figure 2. State Income Tax Incentive Programs for Renewable Energy
Personal and Corporate Income Tax Incentives
Personal Income Tax Incentives Only
Corporate Income Tax Incentives Only
Personal and Corporate Tax Incentives: California, Colorado, Georgia, Hawaii, Idaho, Kansas,
Louisiana, Maryland, Massachusetts, Montana, New York, North Carolina, North Dakota, Ohio,
Oklahoma, Oregon, Utah, West Virginia
Personal Income Tax Incentives: Alabama, Arizona, Rhode Island, Puerto Rico
Corporate Tax Incentives: Iowa, Missouri, Nebraska, New Mexico, South Dakota, Texas,
Virginia, Wyoming
Source: Database of State Incentives for Renewable Energy, http://www.dsireusa.org
10
Figure 3. Rebate Programs for Renewable Energy
State and utility programs
Utility programs
State programs
Local programs
State and utility programs: California, Florida, Illinois, New York, Washington, Wisconsin
Utility programs: Arizona, Hawaii, Nevada, Oregon, Texas
State programs: Delaware, Maryland, Massachusetts, Minnesota, Montana, New Jersey, Rhode
Island
Local programs: Colorado, Pennsylvania
Source: Database of State Incentives for Renewable Energy, http://www.dsireusa.org
11
Figure 4. States With At Least One Grant Program for Renewable Energy
Alabama, California, Connecticut, Florida, Idaho, Illinois, Indiana, Iowa, Kansas, Massachusetts,
Michigan, Montana, New York, Oregon, Pennsylvania, Rhode Island, Utah, Washington,
Wisconsin
Source: Database of State Incentives for Renewable Energy, http://www.dsireusa.org
12
Figure 5. Solar Photovoltaic Resource Potential
Source: EIA (2000)
13
Figure 6. Global Potential Solar Resources
14