Learning Curves for Photovoltaics

Gregory Nemet
nemet@wisc.edu
University of Wisconsin – Madison

June 2007
International Energy Agency

Gregory Nemet Learning Curves for PV 1

Learning curves for photovoltaics

Session 1 items: 2 points:

• latest rates 1 If experience curves. . .
• prices vs costs reliability of predictions
• structural changes 2 If tech4 more generally. . .
expectations
• stability of l-rates
• prediction
• geographic scope
• niche markets

Gregory Nemet Learning Curves for PV 2

Cost of electricity from PV

Gregory Nemet Learning Curves for PV 3

1. Reliability of experience curve projections

Experience curve for PV modules

Gregory Nemet Learning Curves for PV 4

1. Reliability of experience curve projections

Frequency distribution of learning rates calculated in 156

learning curve studies.

Data: Dutton and Thomas (1984); McDonald and Schrattenholzer (2001);

Nemet (2007)
Gregory Nemet Learning Curves for PV 5
1. Reliability of experience curve projections

Experience curve for PV modules

Gregory Nemet Learning Curves for PV 6

1. Reliability of experience curve projections

Learning rates calculated for varying periods

0.25
Learning rates

0.2

0.15

0.1
1985 1990 1995 2000 2005
End year of learning interval

Gregory Nemet Learning Curves for PV 7

1. Reliability of experience curve projections

Calculated learning rates

30

25
Frequency (n=231)

20

15

10

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Learning rate

5th to 95th percentile = 0.17 to 0.24

Gregory Nemet Learning Curves for PV 8

1. Reliability of experience curve projections

Year at which price of PV equals that of competing technology

30
Frequency (n=231)

20

10

0
2010 2020 2030 2040 2050 2060 2070 2080
Breakeven year

5th to 95th percentile = 2029 to 2040

Gregory Nemet Learning Curves for PV 9

1. Reliability of experience curve projections

Cost to subsidize PV through breakeven

Frequency (n=231)

30

20

10

0
0 100 200 300 400 500 600 700
Billions of 2005 $s

5th to 95th percentile = $50b to $166b

Gregory Nemet Learning Curves for PV 10

2. Decomposition of cost-reductions

• Which factors were most important in reducing cost?

• What were the primary drivers of change in those factors?

Gregory Nemet Learning Curves for PV 11

2. Decomposition of cost-reductions

Questions:
• Which factors were most important in reducing cost?
• What were the primary drivers of change in those factors?

Approach:
1 Identify technical and economic factors
2 Time-series data on each factor
3 Model impact of factors
4 Assess influence of experience, etc.

Gregory Nemet Learning Curves for PV 12

2. Change in each factor 1975-2001

Gregory Nemet Learning Curves for PV 13

2. Model results: contribution of each factor

Portion of cost reduction accounted for by each factor,

1980-2001
$25.30
$25

$20
2002 $/W

$15

$10 43%

30%
$5 $3.68
12% 3% 3%
2% 2% 5%
$0
1979 plant efficiency Si wafer Si yield poly- un- 2001
price size price size used x-stal explained price

Gregory Nemet Learning Curves for PV 14

2. Drivers of change in cost-reducing factors
Learning-by-doing (lbd) only weakly explains change in the
most important factors. . . other drivers play a role

Factor Cost impact Drivers of change in each factor

Plant size 43% Expected future demand
Efficiency 30% R&D, some lbd for lab-to-market
Silicon cost 12% Spillover benefit from IT industry
Wafer size 3% Strong lbd
Si use 3% Lbd, but spillover for wire-saws
Yield 2% Strong lbd
Poly share 2% New process, lbd possible
Other factors 5% Not examined

Nemet, G. F. (2006). “Beyond the learning curve: factors influencing cost

reductions in photovoltaics.” Energy Policy 34(17): 3218-3232.

Gregory Nemet Learning Curves for PV 15

2. Did experience enable increase in plant size?
• Several firms make rapid expansions with little experience.
• Biggest plants built by firms with access to capital.

15
Output per plant (MW/year)

10

0
1975 1980 1985 1990 1995 2000

Managing risk of large, uncertain investments w/ lengthy

payoffs more important than overcoming technical problems.
Gregory Nemet Learning Curves for PV 16
Summary

If we are going to rely on experience curves. . .

1 Need to be explicit about reliability of predictions
2 Policy decisions should reflect variation in L-rates

Gregory Nemet Learning Curves for PV 17

Summary

If we are going to rely on experience curves. . .

1 Need to be explicit about reliability of predictions
2 Policy decisions should reflect variation in L-rates

If we consider the drivers of technological change itself. . .

• Need to create incentives for investments in
cost-reducing activities
• Investments drive cost reductions
• Expectations about future demand drive investment
• Government activities affect these expectations
• . . . and perceptions of risk

Gregory Nemet Learning Curves for PV 17

Appendix

Extra slides. . .

Gregory Nemet Learning Curves for PV 18

Experience curves to predict change in costs

(Wright 1936, Arrow 1962, Conley 1970)

 −b
qt
Costt = Cost0
q0

q = cumulative capacity, b = learning coefficient

Advantages Limitations
• Data availability • Sensitive to timing
• Model consistent with • Discontinuities
narratives • Technical constraints
• High goodness-of-fit • Highly aggregated
• Dynamic predictions • Other causal factors
• Single parameter • 1-dimensional quality

Gregory Nemet Learning Curves for PV 19

Prices vs. costs: change in competition

1975-1979: from 2 firms to dozens.

Gregory Nemet Learning Curves for PV 20