accounting for productivity and spillover effects in emerging energy technologies

23
Accounting for productivity and spillover effects in emerging energy technologies: the case of wind power Richard Bowers Proposal for Capstone Project ECPA Economics Department UMBC Advisor: Virginia McConnell June 6, 2014 Abstract The need to bring renewable energy sources into the market to replace more traditional sources in order to reduce greenhouse gas (GHG) emissions are great, and will only increase over time. Many renewable energy sources such as wind and solar photovoltaic (PV) currently have high costs relative to traditional energy sources. The costs for new products tend to decline over time as the productivity improves. This paper will examine the evidence for cost changes in the wind power industry as it has emerged in its major market in California between 1985 and 1995. Productivity improvements over time are hypothesized to occur for a number of reasons including technology improvements, scale economies, learning within firms (intra firm learning), and learning across firms (inter firm learning) and the industry as a whole. A Reason for policy intervention due to market failures such as learning spillovers, and R&D are identified. An externality or spillovers can also exist on the demand side of the market and these will be discussed.

Upload: richard-bowers

Post on 14-Jan-2017

18 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for productivity and spillover effects in emerging energy technologies: the case of wind power

Richard Bowers

Proposal for Capstone Project ECPA

Economics Department UMBC

Advisor: Virginia McConnell

June 6, 2014

Abstract The need to bring renewable energy sources into the market to replace more traditional sources in

order to reduce greenhouse gas (GHG) emissions are great, and will only increase over time. Many

renewable energy sources such as wind and solar photovoltaic (PV) currently have high costs relative to

traditional energy sources. The costs for new products tend to decline over time as the productivity

improves. This paper will examine the evidence for cost changes in the wind power industry as it has

emerged in its major market in California between 1985 and 1995. Productivity improvements over time

are hypothesized to occur for a number of reasons including technology improvements, scale economies,

learning within firms (intra firm learning), and learning across firms (inter firm learning) and the industry

as a whole. A Reason for policy intervention due to market failures such as learning spillovers, and R&D

are identified. An externality or spillovers can also exist on the demand side of the market and these will

be discussed.

Page 2: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

2

I. Introduction

There is a good deal of focus on emerging technologies for electric power generation around the

world because of the high rate of greenhouse gas (GHG) emissions from carbon based sources such as

coal, oil and natural gas. Many of these emerging technologies, such as wind and solar photovoltaic

(PV), are at early stages of development and have relatively high costs compared to more traditional

sources. While these high costs are prohibitive There is a great deal of pressure to subsidize these

emerging technologies with the hope of driving costs down, and allowing them to become mature

competitive sources in the future.

It is common for costs of new products to decline over time as the technology and manufacturing

processes mature. This paper will examine the reasons costs tend to decline, including factors within a

firm, and spillovers between firms. The arguments for declining costs will first be discussed, then

evidence from the literature on which of the various factors have been shown to be most important in

identifying learning will be presented.

In addition this paper will seek to establish when subsidies would be most appropriate. Economic

theory suggests that subsidies are appropriate as a correction to spillovers created in certain situations; 1)

research and development (R&D) and 2) learning by doing. Externalities on the demand side of the

market may also exist and will provide rationale on different types of subsidies. The focus of the paper is

then on the costs of wind power provision in California from 1985 to 1995. We attempt to identify the

existence and possible effect of operational learning by doing, or learning occurring from the day to day

use of a technology or process, by examining the increases in quarterly electricity output of individual

wind plants given wind speed changes and capital depreciation to determine if operational learning by

doing occurs in the wind energy production industry.

II. Arguments for Declining Costs for Emerging Industries

The issue of technology change and learning by doing has been a focus in economics as far back as

Arrow (1962) and the early development of knowledge accumulation and shifts in production (Arrow,

Page 3: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

3

1962). Declining costs for emerging industries over time can occur for a number of reasons. We can

summarize them as:

A. Costs reductions within the firm: including R&D, Learning by Doing, and scale economies

B. Cost reductions due to spillovers from other firms within the same industry: including Learning

by Doing across the industry and R&D by other firms within an industry

C. Demand spillovers caused by increased usage of a given technology.

A. Cost Reductions within the Firm.      

Research and development (R&D) is one method by which firms can attempt to reduce costs. While

internal firm learning involves optimizing the given inputs to minimize costs, R&D focuses on the

generation of new technologies, or process, to attempt to reduce costs. R&D, however, is often expensive

and resource consuming (Griliches, 1992). In deciding whether or not to engage in R&D, or deciding on

the level at which to engage in R&D, a firm will attempt to balance the general cost of R&D against

expected future profits that the effort might yield, adjusting for the risk of such an investment (Jaffe,

1996). In theory, firms would engage in R&D when its own expected benefits outweigh the costs.

However, due to R&D spillovers firms tend will tend not to realize benefits from R&D in excess of R&D

investments (Griliches, 1992). This can vary a greatly by industry.

Another way for firms to reduce costs is to improve at production methods. As a firm repeats the

same processes over and over again, it is generally well accepted economic fact that the costs associated

with such processes should decrease over time as the firm realizes efficiency improvements. This

behavior of knowledge accumulation (learning by doing) at the individual firm level has been

documented in numerous industries including farming, semi-conductor manufacturing, shipbuilding, and

aeronautics (Griliches, 1992; Wright, 1936; Thornton and Thompson, 2001; Irwin and Klenow, 1994).

Learning by doing is generally measured by a progress ratio or learning curve which represents the

decrease in production costs associated with a doubling of cumulative production at the individual firm

level (Epple, et al, 1991; McDonald & Schrattenholzer, 2001). Since learning is the process of

Page 4: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

4

experience accumulation it only takes place through the attempt to solve a problem, which requires

activity (Arrow, 1962). No learning occurs from a firm sitting idle.

Learning can also take various forms. In a seminal paper, Bahk and Gort (1993) identify three

principle elements of knowledge accumulation by the firm, or learning by doing as:

- Organizational Learning: the matching of employees and tasks based on the knowledge and

experience of each employee or the accumulation of independent knowledge of each

employee which is not transferable from employee to employee; or managerial learning

reflected in improved scheduling and coordination among departments and in the selection of

external suppliers of services or products.

- Capital Learning: associated with the increase in knowledge about characteristics of psychical

capital (tolerance to which parts are designed, special tools or devices, and plant layout).

- Manual Task Learning: learning associated with repeating execution of manual and semi-

manual tasks (Bahk and Gort, 1993).  

Finally, economies of scale is another reason why costs can fall within a firm. Economies of scale

are reached when a firm reaches a production volume over a given period of time which yields a

reduction in inputs due either to optimization of labor capital ratios, and/or a reduction of the price of

inputs brought on by higher volumes of production. The cost reductions derived from economies of scale

are reductions in average cost brought on by higher production volumes over a given period while

learning by doing reduces costs over time as total cumulative production increase (Arrow, 1962; Bahk

and Gort, 1993; Nemet, 2011).

B. Cost Reductions due to Industry Spillovers  

In addition to firm-level reductions in costs, there can be cost reductions that can spillover over

from one firm to another. Innovation and development by one firm may affect design, production, and

the associated costs of these activities to other firms. This can be through R&D by one firm that allows

for improvements that other firms in the same industry can adopt. The key is that these are spillover

externalities – a single firm does not capture all of the returns to its own investment in R&D, or to

Page 5: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

5

innovation in production strategies because some of the returns from innovation spillover to other firms

that produce similar products. Since a firm does not capture all the benefits of R&D itself, due to

spillovers into other firms, a firm is likely to invest into R&D at a level that is less than socially optimal,

preferring to free-ride on the R&D of other firms in the industry (Griliches, 1992) (Nemet, 2011). Such a

situation has occurred in the semiconductor industry in which many producers in the industry have

benefitted from the R&D advancements of an individual firm since the products produced are relatively

homogeneous (Irwin & Klenow, 1994).

Generally spillovers will refer to benefits accrued by a firm or industry which incurred minimal

or no costs obtaining those benefits. Spillovers differ slightly from free-ridership since spillovers are

generally derived from an action in which the costs are incurred by one firm, with the benefits initially

intended for that firm inclusively. Free-ridership normally occurs when a firm (or individual) obtains a

benefit without incurring any costs and when the benefits were intended for the larger society, at a cost.

Any time that a firm gains a benefit at the cost of another firm or a firm does not capture the full value of

an investment into new technologies it will be assumed that a spillover is present in the market.

C. Demand Spillovers  

Increased production comes from an increased demand for the product. Thus, diffusion of the

new technologies on the demand side can also be important for reducing costs. (Rao, Keepo, & Riahi,

2006). Technology diffusion refers to the adoption of a technology and then the subsequent adoption of

that technology by additional users - in short it refers to the ability of a technology to ‘catch-on’ by users

resulting from the technology being visible in the market place and used by early adopters. This diffusion

of technology shifts the demand curve for the given technology outward overtime and allows production

to increase and costs to fall.

D. Spillovers and Wind Energy  

Spillover effects are likely to be important for new technologies in the wind energy market, where

there is the need and potential for innovation, where there will likely to trial and error, and where there

are a relatively small number of players who can learn from each other. In addition to innovation and

Page 6: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

6

spillovers in product development, installation siting, a major determinant to electricity output for wind

technologies, can also create spillovers. This is another case in which the actions of one firm can benefit

other firms even if not intended. Yet, another example for wind energy is the development a new

methodology for resource measurement, say wind speed measurement and predictions. If one firm begins

to place wind turbines in areas that are were once proven unprofitable, other firms will likely take notice

and can then benefit from that firm’s investments.

While technological change that lowers costs is brought on through learning, innovation and

R&D, costs are also reduced by increasing production.

III. Role of Subsidies for Emerging Energy Technologies

Subsidies are very common today for renewable energy firms. Here we focus on the economic

rationale for subsidies. The first argument for subsidies is the existence of un-priced externalities

associated with carbon dioxide (CO2) emissions from carbon-based fuels. In this case a better policy than

a subsidy might be a carbon tax, which would disadvantage carbon-based fuels relative to renewables.

But there are several economic rationales for subsidies based on the above discussion. There are two

possible spillovers or externalities that may result in renewables being under-supplied. One is the

spillover from R&D, as discussed above.

The second is the potential spillover in learning by doing, when those spillovers are not fully

accounted for by individual firms. Subsidization to account for these spillovers if it is equal to the

magnitude of the spillovers will increase social welfare. Developing some empirical estimates of those

spillovers is what this paper will address.

One important issue with spillovers from learning by doing is that both learning within firm and

learning between firms may be subject to diminishing returns. Information does not travel perfectly

between firms, due to information asymmetries between those who initially acquire the knowledge and

the adoption of that knowledge by others. Since information travels in an imperfect manner learning, both

between firms and within a firm, is subject to diminishing returns whenever the knowledge is

accumulated second-hand.

Page 7: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

7

Another key issue is that knowledge gained from experience may depreciate over time, and

sometimes rapidly as learning occurs and knowledge accumulates certain learning is likely to be forgotten

as new knowledge is accumulated if not utilized often. This can occur for a variety of reasons including

employee turnover, change in corporate practice, and non-utilization of hyper-specialized knowledge

involved with a rare occurring event in project operation (Bahk and Gort, 1993; Nemet, 2011).

Also, since there are large spillovers involved with R&D, the level of R&D in which firms

engage in might not be the socially optimum level (Griliches, 1992). The inability of an individual firm to

capture all of the returns to R&D suggests the need for either public R&D or industry R&D subsidization.

Since the price paid by consumers is lower than the full price needed to capture the R&D value the

consumers receive a positive benefit from R&D spillovers. Non-optimal spending on R&D alone is not

enough to warrant government intervention in the form of subsidies, otherwise the government would

need to subsidize every industry. Even if government subsidization of R&D did occur, the generally

inelastic supply of scientists would result in higher salaries than more scientists, the latter of which would

be necessary to promote more R&D projects (Griliches, 1992).

IV. Application to Wind Power Generation

In order to determine the presence of learning in the wind energy sector in California between

1985 and 1995, this paper examines quarterly electricity output of wind projects and notes increases in

quarterly electricity output given wind speed and fixed but depreciating capital. These increases are

designed to reflect operational learning. Learning in this manner can be caused by wind project

owner/operators better adapting to changing wind conditions by controlling the pitch and yaw of the

turbine and scheduling maintenance during forecasted low wind periods. Accumulated knowledge

depreciates over time. Therefore we expect to see increases in electricity output at a given wind farm to

occur at diminishing rate, assuming that capital depreciation occurs in the manner which we will model it

(see below).

Early research on learning by doing utilized production theory to represent the occurrence of

learning. Rapping (1965) looked at shipbuilding output during World War II, and utilized a basic

Page 8: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

8

production function to determine how labor inputs engaged in learning during this period to increase

output in the absence of capital gains or technology improvements (Rapping, 1965). The basic production

function is as given:

𝑋!" = 𝐴𝑀!"!!𝐾!"

!!𝑉!", (1)

where X is the annual rate of ship output, M is the annual rate of physical labor inputs, K is the annual rate

of capital inputs, V is a random disturbance term, and i and t are production facility and time indices

respectively (Rapping, 1965). Rapping (1965) found an increase in X over time despite no increases in

capital or changes in technology, implying that labor learning occurred allowing for a higher output

(Rapping, 1965). The learning by doing literature expands on this basic production function approach.

Bahk and Gort (1993) use a standard production function and include a term that captures the

stock of accumulated knowledge:

𝑌! = 𝐹 𝐿! ,𝐾! ,𝐸! (2)

where, Y is output, L is labor, K is capital, and E refers to a stock of knowledge for a given relevant time

period t. The stock of knowledge for a given organization is a measure of the cumulated gross output

since the organization’s birth, or:

𝐸! = ℎ 𝑆! (3)

where, h’ > 0, 𝑆! = 𝑦!!!! , which is the cumulated gross output from the birth of the organization up

through the previous time period t.

Other research has extended this basic model to look at the wind energy production industry in

particular. This paper seeks to build upon the previous work of Gregory Nemet (2011) in looking at

knowledge accumulation spillovers in wind power for the state of California (Nemet, 2011). In

determining the presence and scale of knowledge accumulation spillovers Nemet utilized a modified

production function employed by Bahk and Gort (1993

In modifying the production function used by Bahk and Gort (1993) for general industries, Nemet

(2011), attempts to represent learning by doing in the wind energy production industry. His analysis

Page 9: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

9

includes the addition of a W component presenting the wind energy available at the location of a wind

project and a Q, quality, component to account for exogenous improvements in the quality of equipment

that is not captured in higher purchase prices of capital (Nemet, 2011). An assumption is made of zero

returns to increasing labor inputs, so L is also dropped (Nemet, 2011). The assumption of zero returns to

increasing labor inputs is justified since wind turbine installation is a capital machine intensive process

and the ratio of labor to capital is unlikely to change unless there are intensive changes in the

methodology of installation.

The analysis here will focus not on installation but operation of turbines. To capture the effect of

learning E, we consider only operational learning at a given site, as discusses below. Given an individual

project, E represents the operational learning occurring over time which can take a variety of forms

including better maintenance scheduling and capturing more of the available wind energy from pitch and

yaw control of the turbine. When looking at operational learning, E, among different projects owned by

the same operator spillovers within a firm can be demonstrated. The resulting production function is:

𝑌! = 𝐹 𝐾! ,𝐸! ,𝑊! ,𝑄! , (4)

The estimated production function above provides the electrical output for each quarter t. Observing

changes in the electrical output over time at the same plant accounting for other changes over time will

support the hypothesis that gains from learning are occurring. In order to determine spillovers the

electrical output between projects operated by different owner/operators across quarter will support the

hypothesis that spillovers between firms exist since operating experience should be equal across all firms.

Output, intuitively is a product of inputs (labor, capital, materials), but the level of these inputs

can be changed due to other factors. The implementation of a policy can affect other things that may

make production easier. Implications of policy can be derived from the inclusion of a policy variable into

the production function:

𝑌! = 𝐹 𝐾! ,𝐸! ,𝑊! ,𝑄! ,𝑃! ,     (5)

Page 10: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

10

where Pt is the indicator of an incentive policy in in effect during quarter t. Several policies exist for

rationale of increase output, but no single policy expands enough of the study period to be considered to

have an measurable effect on production.

This production function acts as the base model for determining the presence of operational

learning by estimating the effect of operational learning, E, on quarterly electricity output, Y. Spillovers

will be measured by comparing operational learning, E, for 1) different projects under the same

owner/operator to measure spillovers within a firm, and 2) different wind projects under different

owner/operators to measure spillovers between firms.

V. Data

This analysis will utilize a data set of wind turbine projects installed in California from 1985-

1995 of (n=144 projects) obtained from the Wind Performance Reporting System (WPRS) at University

of California Davis (eWPRS, 1985-1995). The WPRS dataset provides quarterly electricity output as well

as detailed turbine data for each project (eWPRS, 1985-1996). By providing information on every project

installed in California since the beginning of the industry locally in the state the survival bias is avoided.1

Figure 2, Figure 3, and Figure 4 depict the locations of turbines by capacity in the three wind

energy areas of interest for this analysis; Altamont, Tehachapi, and San Gorgino Pass. While not all

turbines depicted are included in the analysis due to the USGS map including turbines installed after

2003, the three figures gives a broad overview of the distribution of turbines across the roughly 40 km2 in

each figure.

1 The survival bias occurs when data points which are not present throughout the every year of the analysis are dropped out of the data set prior to it being available which then biases results to those data points which lasted (survived) the entire time period of the analysis.

Page 11: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

11

Figure 1: Location of Three Wind Areas

Figure 2: Altamont Wind Turbine Locations

Page 12: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

12

Figure 3: Tehachapi Wind Turbine Locations

Figure 4: San Gorgino Pass Wind Turbine Locations

Page 13: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

13

VII. Estimation

To estimate the effects of learning and scale economies, we estimate equations above, using a

panel data set of quarterly data from wind power plants in California between 1985 and 1995. We rewrite

equation (6) including the subscript i as an indicator for individual wind projects. Estimating a

production function for wind we use equation (6) above. Here we explain in more detail the components

of equation (6) and the hypothesized coefficients of model.

𝑌!" = 𝐹 𝐾!" ,𝐸!" ,𝐸!" ,𝑊!" (6)

Table 1 contains list of the variables used the analysis of the wind industry in California. While

most of the variables used in the analysis are measured fairly straightforward, a few of the defined

variables are detailed below.

Table 1: Variable Notation

We assume plants are installed at a point in time and then the capital depreciates over time. We

then define learning as an increase in quarterly electricity production given wind speed and depreciated

Symbol Description Units

t Calendar time; t = 1 in Q1-1985 Quarters

i Project identifier Category

Yit Electrical output by project i in quarter t kWh/qtr

υiτ Turbines installed in quarter τ by project i Turbines

λ Knowledge depreciation; remaining after 1 quarter %

δ Quarterly rate of knowledge depreciation; =  − ln(λ) %

Eit Depreciated operating experience kWh

EIt Total depreciated operating experience for the industry kWh

Vit Average wind speed in quarter t at project i m/s

Wit Wind energy available in quarter t at project i kWh/m2*qtr

Ut Dummy for windy season; 1= Q2 and Q3 Binary

h Number of hours per quarter Hours

Gi Generation capacity of each turbine at project i kW

Tit Number of turbines in quarter t at project i Turbines

ct Cost of wind turbine capacity in quarter t (2008$/kW)

γ Quarterly rate of capital depreciation %

Kit Depreciated capital stock in quarter t at project i (2008$)

Pt Value of policy dummies in quarter t Binary

Page 14: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

14

capital, and the other variables in equation (6). Since knowledge accumulation depreciates due to a

variety of reasons including, but not limited to; employee attrition; knowledge becoming less relevant due

to changes in demand, technology, or structure; or even ability to comprehend and adapt to indicators

from operation. Depreciated operating experience in the production function is modeled as: (Nemet, 2011)

𝐸!" = 𝑌!"𝑒!! !!!!!!! , (7)

where Yiτ is cumulative electricity for project i from the beginning of the data set until period t, where the

time period range is from τ = 1 until τ = t. The depreciation rate is denoted by δ and is estimated from

experience-derived knowledge depreciation values from other studies (we use 0.42).

The regression analysis includes both an Eit and an Eit2 term to account for the possibility of

learning being non-linear over time as discussed above. The derivative of output Yit with respect to Eit,

!!!"!!!"

, will yield the output change to the firm from operational experience gains. Therefore we hypothesize

the following:

HYPOTHESIS 1a. Operational experience at a firm is positively associated with output, but at a

diminishing rate over time.

HYPOTHESIS 1b. Operational experience across an industry is positively associates with output, but at

a diminishing rate over time.

Spillovers, as discussed above, occur when firms within an industry can readily adopt operational

practices that other firms in the industry have established. In order to measure the industry operational

learning variable will take the following form:

𝐸!! = 𝐸!"!!!! (8)

where, EIt is the summation of all depreciated operational experience for all projects in a given quarter. A

positive coefficient should suggest that spillovers are occurring within the industry, between firms. Due to

the nature of the wind power industry as suggested by the literature above, we hypothesize:

HYPOTHESIS 2: Industry learning will be positively associated with electrical output as firms should

learn from one another indirectly by observing behaviors and practices of projects.

Page 15: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

15

Wind energy directly impacts the output and performance of a wind turbine project. The wind Wit

variable of the production function denoted available wind resource at project i in quarter t. Since

available wind is being measured by quarter the diurnal and smaller variation in wind speeds is averaged

away. Available wind for project i in quarter t is given as:

𝑊!" = 𝐹!"! 𝑅 𝑉!" ∙ ℎ,2 (9)

where R is a Rayleigh distribution with mean, Vit, h is the number of hours in a quarter, 2192, and 𝐹!"!is

the power curve relationship between wind speed and electricity output for the given turbine used at

project i. There exists a physical cubic relation between wind speeds and electricity output, meaning that

electricity output increases by a power of 3 relative to the change in wind speed. This relationship only

holds between the cut-in speed (minimum necessary wind speed) and the rated power of the turbine

(given by the manufacturer). This relationship can be seen in Figure 5 Part C. This relationship

demonstrates the maximum rated power output of a turbine at a given wind speed, not actual achieved

power. The power curve of a wind turbine is based on the modeled output by the manufacturer and only

measures the potential output of the turbine.

The term capacity factor is often used to identify the relationship between actual power generated

at a plant and the potential power. Mathematically the capacity factor is the ratio of the actual output

achieved to the maximum potential output of the turbine determined by the manufacturer.

Increases in electricity output will demonstrate operational learning in the face of a depreciated

capital stock, with the electricity output gain likely derived from learning how to better position the yaw

and pitch of the turbine to maximize captured wind speeds. An increase in capacity factor for a given

project from quarter to quarter suggests operational learning of a different type.

Capacity factor can increase from operational learning in pitch and yaw control of the turbine, but

can also increase in learning from operational and maintenance (O&M) scheduling. Utilizing the

2 This equation captures the cubic relationship between wind speed and power output in the power curve for the General Electric 2.5 MW turbine.

Page 16: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

16

WindPower Program software, the power curve of each turbine model can be determined (output of the

software appearing in Figure 5). (WindPower, 2012).

Figure 5: WindPower Program Output for GE 2.5 MW turbine.A)Turbine name, rotor diameter, and cut-in/cut-out wind speeds, B)Output (in kW) for given wind speed, C) graphically represented power curve, D) mean power output given a mean hourly wind speed, E) wind speed probability distribution function, F) turbine output given increasing wind speeds.

The measurement of available wind maintains the cubic relationship between wind speed and

electricity output while not overestimating output due to high winds which cause cut-out of the turbine

(Nemet, 2011). Sensitivity analysis will be conducted on the available wind by utilizing different power

curves of turbines not at the technology frontier by utilizing the power curves for each turbine model at a

given project i provided by the WindPower Program. Since wind speed is the most impactful input in

electricity output we can hypothesize the following:

HYPOTHESIS 3. Available wind defines maximum potential electricity output and is therefore positively

associated with output.

Capital, like experience, depreciates over time. Following Nemet (2011) the capital stock

depreciation is measured by:

𝐾!" = 𝑐!𝑇!"𝐺!"𝑒!! !!!!!!! , (10)

A

B

C D

E

F

Page 17: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

17

where 𝑐! is the average cost of wind turbines sold in California each year, multiplied by the number of

turbines installed at project i in quarter τ, multiplied by the generation capacity of each turbine at project i

in quarter τ, depreciated by rate γ. Since the capital stock used for this analysis is a depreciated one, the

expected contribution of capital to electricity output from one quarter to another is expected to decrease.

As capital depreciates, other things the same we would expect output to fall over time, therefore we

hypothesize the following:

HYPOTHESIS 4. The greater the capital stock the greater the electricity output, therefore capital stock

will be positively associated with output, but at a decreasing rate. .

The analysis utilizes a basic linear form with project fixed effects, and four models that expand

upon each other:

MODEL 1: 𝑌!" = 𝛼 + 𝛽!𝐾!" + 𝛽!𝑊!" + 𝛽!𝐸!" + 𝛽!𝐸!" + 𝑢!" (11)

MODEL 2: 𝑌!" = 𝛼 + 𝛽!𝐾!" + 𝛽!𝑊!! + 𝛽!𝐸!" + 𝛽!𝐸!"! + 𝛽!𝐸!" + 𝛽!𝐸!"! + 𝑢!" (12)

MODEL 3: 𝑌!" = 𝛼 + 𝛽!𝐾!" + 𝛽!𝑊!" + 𝛽!𝐸!" + 𝛽!"𝐸!" + 𝛽! 𝐾!"𝐸!" + 𝛽! 𝐾!"𝐸!" + 𝑢!" (13)

MODEL 4: 𝑌!" = 𝛼 + 𝛽!𝐾!" + 𝛽!𝑊!" + 𝛽!𝐸!" + 𝛽!𝐸!"! + 𝛽!𝐸!" + 𝛽!𝐸!"! + 𝛽! 𝐾!"𝐸!" + 𝛽! 𝐾!"𝐸!" + 𝑢!" (14)

Under this form a positive coefficient of a variable will suggest an increase in electricity output of

a given wind project in a given quarter, other things the same. The coefficient on Wit should be positive

for the reasons stated above. Since capital depreciation, K, does occur we should see a larger capital

stock associated with larger electrical output with diminishing returns to this output as the capital stock

depreciates over time. Learning will be observed by the operational learning within the firm, Ei, and the

diminishing of maintained learning, Ei2.

The important spillover variable is industry experience, EIt. Industry experience should, in theory

show whether or not spillovers occur between firms. The inclusion of the variable EIt, which is the

summation of output of all projects in the region, should have a positive coefficient, as discussed above if

there are learning spillovers between firms within the industry. Two interaction terms are also included

between capital and operational experience within firms and capital and industry experience to deal with

Page 18: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

18

the possible effect of project size on learning from experience. Small projects and large projects might

learn at differing rates. Smaller projects are likely to have higher rates of learning since the projects tend

to be less complex and easier to manage, although the output gains are not expected to be as large as

larger projects.

Table 3: Definition of Variables Used

VIII. Results

Equation (11) was estimated with project fixed effects in order to determine operational learning

between quarters, which is noted by increase electricity output, relative to wind speeds and capital

depreciation. In order to deal with large numbers all variables were converted from kilowatt-hour units to

megawatt-hour units and the variables were given the suffix “mw” to denote the difference. Spillovers are

measured by observing the coefficient of the industry learning variable. A positive coefficient suggests

spillovers are occurring between firms within the industry. Summary statistics for relevant variables are

provided in Table 4.

Symbol Variable Description Units

Yit outputmw Electrical output by project i in quarter t MWh/qtr Kit capitalmw Depreciated capital stock in quarter t of project i (2008$) Wit avalmw Wind energy available in quarter t at project i MWh/m2*qtr Eit expermw Depreciated operating experience in quarter t of project i MWh Eit

2 expermw_squ Depreciated operating experience squared in quarter t of project i MWh EIt ind_expermw Experience of the industry MWh

KitEit capital_exper Interaction between capital and experience (2008$)MWh i Project identifier Category I Industry identifier Category

uit Error term

Page 19: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

19

Table 4: Summary Statistics

Correlation coefficients for relevant variables are outlined in Table 5.

Table 5: Correlation Coefficients

We then ran the panel regression using project fixed effects and a linear form utilizing the

megawatt-hour unit versions of the variables. The output of the fixed effect linear model is provided in

Table 6.

Symbol Variable Description Units

Yit outputmw Electrical output by project i in quarter t MWh/qtr Kit capitalmw Depreciated capital stock in quarter t of project i (2008$) Wit avalmw Wind energy available in quarter t at project i MWh/m2*qtr Eit expermw Depreciated operating experience in quarter t of project i MWh Eit

2 expermw_squ Depreciated operating experience squared in quarter t of project i MWh EIt ind_expermw Experience of the industry MWh EIt

2 ind_exper_squ Experience of the industry squared MWh KitEit capital_exper Interaction between capital and experience (2008$)MWh KitEIt capital_ind Interaction between capital and industry experience (2008$)MWh

i Project identifier Category I Industry identifier Category

uit Error term

plantsizemw 0.6425 0.8606 0.2513 0.8776 0.1415 1.0000

ind_expmw 0.1025 -0.0498 -0.0262 0.0781 1.0000

avalmw 0.6784 0.7406 0.1782 1.0000

expermw 0.4738 0.4757 1.0000

capitalmw 0.6066 1.0000

outputmw 1.0000

outputmw capita~w expermw avalmw ind_ex~w plants~w

Page 20: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

20

Table 6: Linear Panel Regression with Project Fixed Effects

The regression equations has a fairly high degree of explanatory power, with all R2 being about

0.50, but model 4 provides the greatest explanatory power. Capital, which was hypothesized to have a

positive coefficient only holds up to that hypothesis in model 2 and model 4, both where it is significant.

In model 4 a $1000 increase in the size of the capital of the project is associated with gains of around 46

kWh increase in output. Since each megawatt of rated capacity costs $4500, as a project increases in size

by 10 megawatts, production should increase by 2070 MWh.

Experience with in a firm shows that learning is occurring when the coefficient is positive, which

is seen in model 1 and model 4. Positive experience within a firm suggests that from quarter to quarter the

individual project is learning from itself and increasing output. Although experience within a firm is

significant across all models it is not robust, and thus we only look at output results for model 4. As

cumulative output of a give plant increases by 10 MWh, on average a project, all else the same, increases

productivity by roughly 6 MWh from the quarter previous. The experience squared variable which was

hypothesized to capture diminishing returns to experience is positive and significant, rather than negative,

but only affects the output of a project after large gains to cumulative production.

capitalmw -­‐0.026 *** 0.031 *** -­‐0.0060863 0.0460995 ***0.01 0.01 0.01 0.01

expermw 0.502 *** -­‐0.353 *** -­‐0.6553248 *** 0.6068446 ***0.02 0.04 0.09 0.10

avalmw 0.028 *** 0.027 *** 0.0273408 *** 0.0260875 ***0.00 0 0 0

ind_expmw 0.0000044 0.0002725 * -­‐0.0001179 ** 0.0004135 ***0.00 0.00 0.00 0.00

expermw_squ 0.0000035 *** 0.00000551 ***0 0

ind_exper_squ -­‐2.54E-­‐11 ** -­‐2.7E-­‐11 **0 0

capital_exper 0.00000319 *** -­‐0.00000405 ***0 0

capital_ind 6.89E-­‐10 5.96E-­‐09 ***0 0

Constant -­‐1718.241 *** -­‐2334.112 *** -­‐850.7861 * -­‐4169.513 ***331.85 346.54 364.75 388.14

R-­‐sqr 0.5781 0.591 0.5724 0.6196

Model  4b/se

*  p<0.05,  **p<0.01,  ***p<0.001

Model  1b/se

Model  2b/se

Model  3b/se  

Page 21: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

21

Spillovers between firms can be derived by looking at the coefficient of the industry experience

variable. Based on the regression, a 100 megawatt-hour increase in industry experience output is only

associated with a 4.135 kilowatt-hour increase in output of a given project. While this number seems

small relative to increases from project experience it still suggests spillovers between projects. Industry

experience across the models shows that industry experience is not robust and only positive and

significant in models 2 and 4, with model 4 providing the best fit overall. While the existence of

spillovers between firms is illustrated by the positive coefficient of industry experience, the rationale for

policy intervention is not as clear cut and will be discussed later.

The inclusion of two interaction terms, one between capital and experience within a project and

capital and industry experience provide some interesting, but possibly inconclusive results. The first point

about the interaction terms is that the coefficients are incredibly small, although both are significant in

model 4. The negative coefficient on the interaction between capital and project experience suggests that

as capital gets larger there are loses to experience gains, perhaps due to complexities of the project which

prevent larger projects from learning faster than smaller projects. The interaction between capital and

industry experience is positive however, which suggests that larger projects learn more from other firms

than smaller firms. This could be due to the ability of larger projects which might have more resources

obtaining market reports or perhaps being a larger member of the industry and having the ability to better

monitor other projects. There could also be some rationale not observed by the data that provides this

result.

IX. Conclusions

The goal of this project was assess whether and why there have been gains in productivity in the

wind industry. In particular we were interested in learning on the part of wind plants in California as

industry was getting started, and whether such learning spills over across firms. The data was drawn from

California plants during the period of 1985 to 1995. the conclusions drawn from the analysis suggest a

few things; 1) spillovers between firms do exists but they are rather small, 2) own firm learning is

important but these gains from own firm experience may not diminish over time as has been found for

Page 22: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

22

other industries, and 3) while larger plants produce more there is not a strong relationship between

experience gains and project size.

One important question for new renewable energy industries is whether or not subsidies are

appropriate to promote learning by doing. The evidence here suggests that spillover effects have been

relatively small, and that large subsidies based on such spillover externalities would not be warranted.

This does not mean that subsidies are not appropriate at all, because they could be justified on grounds of

environmental externalities, for example, that wind energy has lower greenhouse gas emissions than

alternative sources of power. While the idea of a subsidy for the wind electricity production industry is

not ruled out by this analysis providing a subsidy, on the ground of learning by doing and encouragement

of spillover, does not seem strong enough given the analysis results. This analysis only utilized an

available data set from California during 1985-1995 due to the proprietary nature of the data of the

industry. In expanding this analysis, future considerations would be put into obtaining the national level

dataset from the National Renewable Energy Laboratory (NREL) provides resources available. In

continuing my education I intend to pursue this issue further while working on my PhD in Marine Policy

at the University of Delaware if able to obtain the national level dataset to determine how learning has

occurred on land based wind energy generation in the United States and how this learning can help reduce

the learning time for offshore wind energy generation.

Page 23: Accounting for Productivity and Spillover Effects in Emerging Energy Technologies

Accounting for Cost Reductions and the Role for Subsidies in Emerging Energy Technologies

23

Works Cited Alsema, E. (1998). Energy Requirements and CO2 Mitigation Potential of PV Systems. Proceedings of

Photovolatics and the Environmnent. Keystone, CO. Arrow, K. J. (1962). The Economic Implications of Learning by Doing. The Review of Economic Studies,

29(3), 155-173. Bahk, B.-H., & Gort, M. (1993). Decomposing Learning by Doing in New Plants. Journal of Politcal

Economy, 561-583. Benthem, A., Gilligham, K., & Sweeney, J. (2007). Learning-by-Doing and the Optimal Solar Policy in

California. Griliches, Z. (1992). The Search for R&D Spillovers. The Scandinavian Journal of Economics, S29-S247. Irwin, D. A., & Klenow, P. J. (1994). Learning-by-Doing Spillovers in the Semiconductor Industry.

Journal of Politcal Economy, 1200-1227. Jaffe, A. B. (1996). Economic Analysis of Research Spillovers Implocations for the Advanced

Technology Program. Jaffe, A. B., Newell, R. G., & Stavins , R. N. (2004). A Tale of Two Market Failures: Technology and

Environmental Policy. Resources for the Future, Washington, DC. Kato, K., Murata, A., & Sakuta, K. (1997). Energy Payback Time and Life-Cycle CO2 Emissionsof

Residential PV Power Systems with Silicon PV Module. Utrecht University. Margolis, S. E. (n.d.). Network Externalities (Effects). S.J. Liebowitz Management School. McDonald, A., & Schrattenholzer, L. (2001). Learning Rates for Energy Technologies. Energy Policy, 29,

255-261. Nemet, G. (2011). Spillovers from Learning by Doing in Wind Power. Palz, W., & Zibetta, H. (1991). Energy Payback Time of Photvoltaic Modues. International Journal of

Solar Energy, 211-216. Rao, S., Keepo, I., & Riahi, K. (2006). The Importance of Technology Change and Spillover in Long-

Term Climate Policy. The Energy Journal, 123-140. Wright, T. P. (1936). Factors Affecting the Costs of Airplanes. Journal of Aeronautical Sciences, 122-

128.