Experience Curves and Photovoltaic
Technology Policy
Robert M. Margolis
Human Dimensions of Global Change Seminar
Carnegie Mellon University
October 16, 2002
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 1
Outline
•
•
•
•
•
History/Origins of Experience Curves
Background on PV
Application to Solar PV
Thinking Prospectively Using Experience Curves
Concluding Thoughts
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 2
Origins of the Learning Curve
• The “learning curve” describes how marginal
labor cost declines with cumulative production
(for a given manufactured good and firm).
– Wright’s 1936 study of airplane manufacturing found
that the number of hours required to produce an
airframe (an airplane body with out engines) was a
decreasing function of cumulative airframes, of a
particular type, produced.
– Learning curves reflect a process of learning-by-doing
or learning-by-producing within a factory setting.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 3
Origins of the Experience Curve
• The “experience curve” generalizes the labor
productivity learning curve to include all the cost
necessary to research, develop, produce and
market a given product.
– Boston Consulting Group’s 1968 study across a range
of industries.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 4
Boston Consulting Group’s 1968 Study
– Empirically BCG found that, “costs appear to go down
on value added at about 20 to 30% every time total
product experience doubles for the industry as a whole,
as well as for individual producers.”
– In BCG analysis cost included, “all costs of every kind
required to deliver the product to the ultimate user,
including the cost of intangibles which affect perceived
value. There is no question that R&D, sales expense,
advertising, overhead, and everything else are included
” (p. 12).
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 5
The General Form of the Experience
Curve is the Power Curve
• P(t) = P(0) x [q(t)/q(0)]^-b
Where:
P(t) = average price of a product at time t
q(t) = cumulative production at time t
b = learning coefficient
• PR = 2^-b
Where:
PR = progress ratio. For each doubling of cumulative
production the MC decreases by (1-PR) percent.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 6
Illustrative Learning for Three Progress
Ratios
$100
$100
PR = 90%
Average Price (per unit)
Average Price (per unit)
$90
$80
$70
$60
PR = 90%
$50
$40
$30
PR = 80%
$20
PR = 80%
PR = 70%
$10
PR = 70%
$10
$0
$1
0
50
100
150
200
250
Cumulative Units Produced
300
350
1
10
100
Cumulative Units Produced
Where: P(0) = 100
q(0) = 1
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 7
1000
Average Revenue per Unit (constant $)
Example from BCG Study:
Silicon Transistors, 1954-1969
100.
PR = 90%
10.
PR = 80%
1.
PR = 70%
0.1
0.01
0.01
0.1
1.
10.
100.
1,000.
Key Milestones:
1947: 1st Demonstration
(Bell Labs)
1954: 1st Commercial
Production (TI)
1960s: Expansion into
Televisions (Sony)
10,000.
Cumulative Units Produced (million)
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 8
Why Might Marginal Cost of
Production Decline?
• Changes in production
– process innovations, learning effects and economies of
scale.
• Changes in the product itself
– product innovations, product redesign, and product
standardization.
• Changes in input prices
 Experience curves typically aggregate all of these
factors.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 9
Distribution of Progress Ratios
22 Field Studies (Dutton and Thomas 1984)
14
12
Frequency
10
Note: These progress
ratios are firm level (not
industry wide) studies.
8
6
4
2
0
50
60
70
80
90
100
Progress Ratio
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 10
Distribution of Energy Progress Ratios
(McDonald and Schrattenholzer 2001)
7
6
Frequency
5
4
3
2
1
0
60-65
70-75
80-85
90-95
100-105
110-115
Progress Ratio
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 11
Background on PV
• PV technology overview
– Types of cells
• PV market overview
–
–
–
–
Historical Production
Shifting Market Segmentation
PV Module Historical Experience
An Industry Projection
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 12
Two Main Classes of PV Cells
• Mono/Polycrystalline Crystalline Silicon Cells
– Industry standard
– Multiple cells assembled into a module
– Accounted for > 80% production during the 1990s
• Thin-film Solar Cells
– Main contenders include: Amorphous Silicon,Cadmium
Telluride (CdTe), and Copper Indium Diselenide (CIS)
– Monolithic design
– Emerging technologies
* A Single Learning Curve?
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 13
Global PV Cell/Module Production
80%
5 Year
Average Annual
Growht Rate
(1996-2001)
= 35%
360
320
280
70%
60%
240
50%
200
40%
160
30%
120
20%
80
2000
1998
1996
1994
1992
1990
1988
0%
1986
0
1984
10%
1982
40
1980
PV Cell/Module Production (MW)
400
Year
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 14
ROW
Europe
Japan
U.S.
U.S. Share
Note: ROW =
Rest of World
Shifting PV Market Segments
• Industrial (off-grid)
– Telcom, Instrumentation,
Warning Signals, etc.
• Rural (off-grid)
– Rural Solar Home Systems,
Water Pumping, etc.
• Consumer
Market
Segment
Industrial
Rural
Consumer
Grid-connected
1994
1997
38%
28%
14%
20%
29%
27%
8%
36%
Source: Strategies Unlimited 1998, p. 14.
– Watches, Calculators, etc.
• Grid-connected
– Rooftop, Building
Integrated, Central
Generation.
Notes:
Production grew between 1994 and 1997
from 58 to 110 MW.
In 2000 grid-connected systems accounted for
approximately 50% of the 300MW market.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 15
Example Rural PV Applications
Solar Home
System
(China)
PV Water Pumping
(Tanzania)
Solar Home
System
(Honduras)
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 16
Example Grid-Connected Applications
Sanyo Building
with PV
Façade (Japan)
BP Service
Station (Spain)
House with
PV Integrated
Roof (Japan)
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 17
Historical Experience for PV Modules, 1976-1998
(Data from Various Sources)
PV Module Price (1996$/Wp)
100
EIA/DOE
(US)
Maycock
(PV News)
Harmon
(Global)
Strategies
Unlimited
Tsuchiya
(Japan)
10
1
0.1
1
10
100
Cummulative Production (MWp)
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 18
1000
An Industry Projection, 2000-2005
900
1 GW
800
Non-subsidized Residential Rooftops
Commercial Building Facades (BIPV)
PV Integrate Products
PV Industry MW
700
600
500
400
Grid-tied
Rooftops (Japan, Germany)
Remote Habitation
Telecommunications
Corporate Image (BP)
300
Large Scale Power
Off-Grid-Rural
200
100
Off-Grid Industrial
0
1999
2000
2001
2002
2003
2004
2005
Source: Strategies Unlimited, April 2000
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 19
Application to Solar PV
• Why should government have a role in
encouraging learning?
• Summary tables from other studies
• A typical learning based projection for PV
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 20
A Government Role?
• Benefits of learning have a tendency to spillover
to other firms.
– A firm can hire employees from a competitor, use
reverse engineering, rely on informal contacts with
employees at rival firms, or pursue industrial espionage.
– As a result firms tend to under-supply a technology that
exhibits strong learning effects.
• For a technology that exhibits strong learning
effects and provides public benefits, the case for
government sponsored demand-pull efforts is
particularly strong.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 21
PV Progress Ratios
Study
Progress
Ratio
Maycock and Wakefield (1975)
78%
Williams and Terzian (1993)
81.6%
# of
obs
16
17
Years
Scope
Cost/Price Measure
1959-1974
1976-1992
US
Global
Cody & Tiedje (1997)
78.0%
13
1976-1988
Global
Williams (1998)
Maycock (1998)
82.0%
68.0%
19
18
1976-1994
1979-1996
Global
Global
PV Module Sale Price
Factory Module Price, based
on Strategies Unlimited Data
(from 1993)
Factory Module Price, based
on Strategies Unlimited Data
(from 1989)
PV Module Price
PV Module Price, from text
Tsuchiya (2000)
83.8%
20
1979-1998
Japan
Harmon (2000)
79.8%
21
1968-1998
Global
IEA (2000)
65%
11
1985-1995
EU
IEA (2000)
84%
53%
79%
9
4
10
1976-1984
1984-1987
1987-1996
EU
PV Module Government
Purchasing Price, vs. Japanese
cum. Production (sign.
fluctuation over time)
PV Module Price, based on
mix of sources (including
Maycock, Ayres, NREL,
Thomas, and Watanabe)
PV Electricity Costs
(ECU/kWh), vs. cum. kWh
produced using a PV system
(includes BOS and shift from
SHS to BIPV)
PV Module Price, based on
EU-Atlas project data, vs
global production.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 22
PV “Buy-down” Cost Estimates
Study
Time
Period
Target
Level
Estimated BuyDown Cost
Notes
Neij (1997)
1995various
5
cents/kWh
$100 billion
(0.8 PR)
$20 billion
(0.7 PR)
Subsidize all future purchases
above target cost. Total system
cost.
Wene (1999)
1997-not
specified
$0.5/Wp
$60 billion
Subsidize all future purchases
above target cost. PV module
only.
Wene (1999)
19982007
$3/Wp
$1.2 billion
Japanese government learning
investment. Total system cost.
(Global subsidies are estimated to
be $18 billion.)
Williams (1998)
19972006
$1000/kW
$50 billion (x-Si)
$120 million (a-Si
targeted program)
Subsidize all future purchases
above target cost. PV module
only.
Williams and
Terzian (1993)
19952020
$1100/kW
$5.4 billion (Net
benefits accounting
for env. ext.)
Subsidize only additional
purchases relative to baseline.
Total system cost.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 23
A Typical Learning Based
Projection for PV
PV Module Price (1996$/Wp)
100.
Maycock
(PV News)
80% PR
10.
70% PR
90% PR
1.
13,000 MW
51,000 MW
0.1
0
1
10
100
1,000
10,000
Cummulative Production (MWp)
100,000 1,000,000
Note: Total US Electricity Capacity
in 1998 was 775,000 MW
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 24
Thinking Prospectively
•
•
•
•
•
What is the right target level?
Module vs. System Costs
Need to calculate impacts relative to a baseline
Single progress ratio is an over simplification
What are realistic cost reductions over time?
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 25
What’s the Right Target Level?
• Depends on targeted application
 Rooftop/BIPV: Retail Electricity Rate
– Large-Scale Power: Wholesale Rate
– Telecom: Currently competitive in many remote
locations
– Solar Home Systems: Economically viable when
remote from the grid
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 26
Average Residential Electricity Prices for
Selected Countries, 1997
Country
USA
Japan
Germany
Switzerland
Netherlands
Average Price
(cent per kWh)
8.5
20.7
16.1
13.6
13.0
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 27
PV System vs. Electricity Costs
44
Cost of Generated Electricity (cents/kWh)
40
Capacity Factor = 0.2
(U.S. Average)
36
32
28
Japanese Retail Rate
24
20
German Retail Rate
Capactiy Factor = 0.25
(South West U.S)
16
Pennsylvania Retail Rate
12
8
4
0
$9.00
Additional Assumptions:
System Lifetime = 20 years
Real Interest Rate = 6%
O&M = 0.1 cent per kWh
$8.00
$7.00
California Retail Rate
$6.00
$5.00
$4.00
$3.00
$2.00
Installed PV System Cost ($/Wp)
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 28
$1.00
$0.00
Module vs Systems Costs
• Really a compound learning curve
– PV module
– Balance of System components
• Rooftop/BIPV offers many opportunities for cost reduction
– Elimination of Storage
– Substitute structurally
– Elimination of frame
– Installation
• Different components may have different learning
rates.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 29
Impacts Relative to a Baseline
• PV has niche markets that are likely to grow
• Can target subsidies (as in Japan and Germany)
• A simple illustration:
– Baseline annual growth: 10% or 20% (?)
– Subsidy increases annual growth to 30%
– PR = 0.8, System Cost in 1998 = $7/Wp
 Achieve $3/Wp: 2020 (10% growth); 2012 (20%
growth); 2009 (30% growth).
 Subsidy required:
All Production = $11 billion;
Extra Prod. = $6 billion (increase growth from 10% to 30%);
Extra Prod. = $4 billion (increase growth from 20% to 30%).
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 30
Projected PV Annual Production Under
Three Growth Scenarios
Annual Global PV Module Production (MW)
3,000
2,500
2,000
30% Growth
20% Growth
10% Growth
1,500
1,000
500
0
1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2020
Year
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 31
Projected PV System Price
Three Growth Scenarios (PR = 0.8)
8
Average PV System Price ($/MW)
7
6
5
30% Growth
20% Growth
10% Growth
4
3
2
1
0
1998
2000
2002
2004
2006
2008
2010
2012
2014
2016
2018
Year
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 32
2020
Using a Single Progress Ratio?
• There is considerable uncertainty in historical
progress ratios
– What is the relationship between R&D and progress
ratios?
– The potential for breakthroughs is difficult to quantify
• Results are highly sensitive to progress ratio
 Need to include sensitivity analysis.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 33
Sensitiveity of Global PV System Subsidy Cost to PR
$140
Subsidy Cost (billion $)
$120
Assumes:
System Cost in 1998 = $7/Wp
Buy down all systems to $3/Wp
$100
$80
$60
$40
$20
$0
70%
75%
80%
85%
Progress Ratio
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 34
90%
Japanese Rooftop Program Experience, 1994-2000
18
120
PV System Price (1998$/Wp)
100
14
12
80
10
60
8
6
40
4
20
2
0
1993
1994
1995
1996
1997
1998
1999
Year
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 35
2000
0
2001
PV Rooftop Systems Installed (MW)
16
PV Budget Comparison (Million US$)
Program Area
R&D
U.S
Japan
Germany
(FY2002) (FY2002) (FY2001)
72
56
27
Demonstration
--
41
6
Market
Stimulation
Total
*
473
30
72
570
62
Source: Maycock (2002); EIA (2001)
* There is considerable activity at the state level in the U.S.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 36
Illustrating a Breakthrough in PV Technology
x-Si Curve
Alternative PV Curve
1.
Transition
Period
PV Module Price (1996$/Wp)
10.
0.1
100
1,000
10,000
100,000
Cummulative Production (MWp)
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 37
1,000,000
What are Realistic Cost Reductions?
• Material Costs
– Crystalline Silicon: $0.43-0.68/Wp (silicon, glass,
EVA, other), for 13% efficient cell (Maycock 1998)
– Amorphous Silicon: ~ $0.31/Wp (Carlson 1993)
– CdTe: ~ 0.48/Wp (Zweibel 1999)
• Average Module Production Cost in 2000 for U.S.
PVMaT participating companies was $2.94/Wp
• Average Module Price in 1998 was $3.82/Wp
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 38
PVMaT Cost-Capacity Historical Data (1992-2000)
and Projection (2001-2006)
Average PV Module Production COST
(1996$/Wp)
5
1992
4
3
Historical
Projected
2000
2
2006
1
0
0
100
200
300
400
500
600
700
Cummulative Production CAPACITY (MWp)
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 39
800
900
Concluding Thoughts
• Process of innovation is inherently uncertain
– prospects for learning with existing technologies,
– breakthroughs (i.e., through R&D investments),
– market developments (i.e., how rapidly will the gridconnected and rural home markets grow).
 Need to be cautious!
– Simplistic use of industry-wide experience curves can
easily mask the underlying dynamics of the process of
innovation.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 40
Concluding Thoughts (cont.)
• With respect to PV technology we are in what
Cowan (2000) calls the “narrow windows” and
“blind giants” stage of technology development.*
– There is a wide range of emerging PV technologies.
– It is currently unclear which PV technology will
dominate the market in the long-run.
 Government should strive to encourage the
development and diffusion of a diverse set of PV
technologies (to avoid lock-in to an inferior PV
technology).
* That is, effective policy-making is only possible during the early stages of competition
between technologies, yet that is when analysts and policy-makers know the least about
what to do.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 41
Policy Recommendations
• Pursue (demand-pull and supply-push) policies that will
encourage the range of emerging PV technologies to
continue to advance from the laboratory into production:
– Demand-pull: net metering, setting interconnection standards,
instituting a federal subsidy (tax credit or loan subsidy) for rooftop
and building-integrated PV systems, and instituting a renewable
portfolio standard (RPS) that would include distributed PV
systems.
– Supply-push: Expand R&D programs like the Thin-Film PV
Partnership project and the PVMaT project. These projects have
fostered innovation in a wide array of PV technologies and should
continue to pursue a multi-technology strategy.
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 42
Directions for Research
• Need to improve our understanding of the
mechanisms underlying the process of
technological change
– How do demand-pull and supply-push policies interact?
– How much can we speed up the process of learning?
• Can we quantify the value of R&D?
R. M. Margolis, HDGC Seminar, Oct. 16, 2002, page 43