Title

Analytical models for wind power investment

Advisor(s)

Wu, FF; Zhong, J

Author(s)

Cheng, Mang-kong.; .

Citation

Issued Date

URL

Rights

2011

http://hdl.handle.net/10722/174453

The author retains all proprietary rights, (such as patent rights)

and the right to use in future works.

Analytical Models for Wind Power Investment

by

Henry Mang-kong Cheng

B.Eng. HKU; M.Econ. HKU

A thesis submitted in partial fulfillment of the requirements for

the Degree of Doctor of Philosophy
at the University of Hong Kong

September 2011

Abstract of thesis entitled

Analytical Models for Wind Power Investment

Submitted by
Henry Mang-kong Cheng
for the degree of Doctor of Philosophy
at the University of Hong Kong
in September 2011

Wind power generation has experienced an explosive growth worldwide. It is a

promising renewable energy source to countries that are short of fossil fuels, e.g. China.
While wind power is a distinctive direction to go for, it is still necessary to examine the
rationale behind such investing mania, and this thesis analyzes the issue by collectively
investment modeling.
For investment analysis, it is necessary to first identify the relevant market
background before inferring to any analytical model. Chapter 2 identifies a number of
wind power investment scenarios in accordance to modern electricity market regime,
primarily American and European structures. Among them, two main scenarios are
investigated and modeled subsequently: fixed tariff wind power project invested by
independent power producer and wind power project undertaken by utility. It has to be
emphasized that different market scenarios would lead to different modeling
methodologies for best representing the reality.
Wind power is intermittent and uncertain. One way to describe the probabilistic
energy production is by statistical characterization of wind power in a period of time.
Chapter 3 presents a standalone analytical model of the wind power probability
distribution and its higher order statistics. Large-scale deployment of wind power would
influence power system in unprecedented ways. High penetration wind power poses a
need of refinement to existing methodologies on production costing and reliability

evaluation. The applications of the probabilistic wind power model to these topics are
outlined in this chapter.
In Chapter 4, investment of fixed tariff wind power project is analyzed. Operation
of wind farm is very passive and as long as wind keeps blowing, such wind power
investment has minimal risk in annual revenue. The low-risk profile facilitates debt
financing. This leads to the attempt to manipulate the project capital structure to
maximize the project levered value. Yet the default probability is raised and associated
with a subjective value of default probability there is a value-at-risk debt level. I therefore
propose an optimization formulation to maximize the wind power project valuation with
debt as decision variable subject to the value-at-risk debt constraint.
Apart from independent wind power producers, many policy and market factors
driving wind power development are actually put on the utility side, e.g. Renewable
Portfolio Standard (Renewable Energy Target) in U.S. (Europe) and Green Power
Programs. It implies that utility has to have wind power (or other renewable) capacity
ready by a certain date. In practice, utility may take action earlier if conditions are
favorable or optimal. The conditions considered here are fossil fuel prices or in more
general setting, electricity contract prices. Define the total fuel cost saving from
conventional units as the benefit of wind power. If fuel prices are high enough,
substituting load demand by wind energy is profitable, vice versa. The investment
decision is analogous to premature exercising of an American option, in which the wind
power project is modeled as real option. Chapter 5 offers detailed formulation of this idea.
(485 words)

DECLARATION
I declare that this thesis represents my own work, except where due
acknowledgement is made, and that it has not been previously included in a thesis,
dissertation or report submitted to this University or to any other institution for a degree,
diploma or other qualification.

Signed.
Henry Mang-kong Cheng

ACKNOWLEDGMENT
I would like to express my heartfelt thank and appreciation to my supervisor, Prof.
Felix Wu, for his continual guidance and insightful advice on my study. Prof. Wu is a
role model in many ways and I learn far more than research from him.

I would also like to thank Dr Jin Zhong and Dr Yunhe Hou, for their opinion,
encouragement, and support, in particular when Prof. Wu is busy.

My gratitude is also delivered to other CEES members, Peter and Clara, for being
friendly and helpful to me.

Special gratitude goes to CLP Power, for awarding me the CLP Fellowship in
Electrical Engineering as a generous financial support.

LIST OF ORIGINAL IDEAS

The following highlights in particular are original contribution of this thesis:

A new probabilistic wind power generation model furnished with analytical

formulae of any higher order moment/cumulant. It can be used in conjunction with
reliability evaluation and production costing in power system literature.

A stochastic optimisation framework of levered firm valuation for the investment

modelling of wind power project under feed-in tariff, subject to value-at-risk debt
constraint. It points to an optimal debt level for maximizing the firm valuation.

Application of a bivariate real option model to determine the optimal investment

timing and value of a wind power project undertaken by utility for meeting the
requirement of renewable energy target. The financial model successfully
incorporates probabilistic production costing result as a power system consideration.

Any error and omission are my own responsibility.

Table of Contents
DECLARATION ................................................................................................................ 3
ACKNOWLEDGMENT..................................................................................................... 4
LIST OF ORIGINAL IDEAS............................................................................................. 5
List of Tables .................................................................................................................... 10
List of Figures ................................................................................................................... 11
List of Notations ............................................................................................................... 13
List of Abbreviations ........................................................................................................ 15
1

Overview of Generation Planning and Investment ................................................... 16

1.1

Introduction...................................................................................... 16

1.2

Conventional Generation Planning .................................................. 17

1.3

Distributed Generation Planning...................................................... 19

1.4

Generation Investment ..................................................................... 20

1.4.1

Financial risk management of generator profit.............................................. 20

1.4.2

Valuation of generator in spot market............................................................ 21

1.5

Research Motivations....................................................................... 22

1.6

Objective and Expected Contribution.............................................. 23

1.7

Thesis Outline .................................................................................. 24

1.8

References........................................................................................ 25

Market Scenarios for Wind Power Investment ......................................................... 29

2.1

Background and Scope .................................................................... 29

2.2

Feed-in Tariff ................................................................................... 30

2.2.1

German wind tariffs........................................................................................ 31

2.2.2

German photovoltaic tariffs ........................................................................... 32

2.2.3

Concluding remark for investment modeling ................................................. 33

2.3

Obligation American Experiences ................................................ 33

2.3.1

Renewable Portfolio Standard ....................................................................... 34

2.3.2

Integrated Resource Planning ........................................................................ 34

2.3.3

Green Power Programs.................................................................................. 35

2.3.4

Tax Credits and Production Incentives .......................................................... 36

2.3.5

Concluding remark for investment modeling ................................................. 37

2.4

Obligation European Experiences................................................. 37

2.4.1

EU Energy and Climate Package................................................................... 38

2.4.2

EU Emission Trading System ......................................................................... 38

2.4.3

Nordic Energy Perspectives ........................................................................... 39

2.4.4

NEP Modelling Methodology ......................................................................... 40

2.4.5

Concluding remark for investment modeling ................................................. 44

2.5

Wind Power in Spot Market ............................................................ 45

2.6

Auction and Tendering .................................................................... 46

2.7

Summary .......................................................................................... 46

2.8

References........................................................................................ 47

Probabilistic Wind Power Generation Model ........................................................... 49

3.1

Introduction...................................................................................... 49

3.2

Wind Speed Distribution.................................................................. 50

3.3

Wind Turbine ................................................................................... 51

3.3.1

Ideal Power Curve.......................................................................................... 51

3.3.2

Aerodynamic principle ................................................................................... 52

3.3.3

Wind turbine generator type........................................................................... 53

3.3.4

Power regulation............................................................................................ 55

3.3.5

Empirical power curve ................................................................................... 55

3.4

Wind Power Distribution ................................................................. 56

3.4.1

Analytical Formulae of Wind Power Statistics............................................... 58

3.5

Wake Effect and Wind Direction..................................................... 59

3.6

Evaluating Production Cost and Reliability with Wind Power ....... 60

3.7

Data Source ...................................................................................... 61

3.7.1

Royal Netherlands Meteorological Institute................................................... 62

3.7.2

Vermont Small-scale Wind Energy Demonstration Program......................... 62

3.8

Data Pre-processing ......................................................................... 62

3.8.1

Wind speed measuring height......................................................................... 62

3.8.2

Wind speed partitions and the parameter lambda.......................................... 63

3.8.3

Empirical power curve ................................................................................... 63

3.8.4

Wake effect ..................................................................................................... 65

3.9

Simulated and Empirical Results ..................................................... 66

3.9.1

Historical wind speed analysis ....................................................................... 67

3.9.2

Mean and standard deviation of annual average wind power ....................... 74

3.9.3

Monte Carlo simulation for wind power statistics ......................................... 76

3.9.4

Comparison between analytical and empirical wind power PDF.................. 77

3.9.5

Regional wind power distribution .................................................................. 79

3.10

Remarks ........................................................................................... 83

3.11

References........................................................................................ 83

Fixed Tariff Wind Power Investment Model............................................................ 88

4.1

Introduction...................................................................................... 88
4.1.1

4.2

Accounting Preliminaries................................................................. 91
4.2.1

Definition of cash flow ................................................................................... 92

4.2.2

Net present value ............................................................................................ 93

4.2.3

Other discount rates ....................................................................................... 94

4.2.4

Capital structure............................................................................................. 96

4.3

Model Formulation for FIT Wind Power Investment...................... 98

4.4

Trial Data for the Model ................................................................ 102

4.4.1

4.5

Scope .............................................................................................................. 90

Base case financial parameters.................................................................... 103

Numerical Example ....................................................................... 104

4.5.1

Base case ...................................................................................................... 104

4.5.2

Sensitivity analysis ....................................................................................... 106

4.6

Summary ........................................................................................ 108

4.7

References...................................................................................... 108

Real Option Wind Power Investment Model.......................................................... 110

5.1

Introduction.................................................................................... 110

5.2

Literature Review and Comparison ............................................... 112

5.2.1

Review of selected real option applications in energy sector....................... 112

5.2.2

Comparison with existing works................................................................... 114

5.2.3

Preliminaries of option pricing theory ......................................................... 115

5.3

Contingent Claim and Real Option................................................ 117

5.3.1

Contingent claim and justification for delta hedging ................................... 118

5.3.2

Solution of contingent claim as project valuation ........................................ 121

5.3.3

Real option accounting delay of investment ................................................. 122

5.4

Bivariate Binomial Lattice for two Fuel Prices ............................. 125

5.4.1

Univariate binomial model........................................................................... 126

5.4.2

Bivariate binomial model ............................................................................. 128

5.4.3

Risk neutral valuation .................................................................................. 128

5.4.4

Extension to multi-fuel displacements .......................................................... 130

5.5

Categorization of Parameters ......................................................... 131

5.5.1

Annual average wind energy production...................................................... 131

5.5.2

Fuel displacement......................................................................................... 132

5.5.3

Wind turbine capital and maintenance costs................................................ 134

5.5.4

Choice of discount rate................................................................................. 134

5.5.5

Carbon price and renewable credit.............................................................. 136

5.6

Parameters Estimation ................................................................... 139

5.6.1

Fossil fuel price drift and volatility .............................................................. 139

5.6.2

Fossil fuel price correlation ......................................................................... 141

5.6.3

Risk-free rate ................................................................................................ 141

5.6.4

Risk-adjusted discount rate by CAPM.......................................................... 142

5.6.5

Wind turbine costs ........................................................................................ 143

5.6.6

Fuel consumption by PPC ............................................................................ 144

5.7

Numerical Example ....................................................................... 147

5.7.1

Base case results .......................................................................................... 147

5.7.2

Sensitivity analysis ....................................................................................... 148

5.8

Summary and future works ............................................................ 153

5.9

References...................................................................................... 155

Conclusion .............................................................................................................. 160

Appendices.............................................................................................................. 162

I.

Wind Power Probability Distribution ..................................................................... 162

II.

M&M Propositions I and II with Corporate Taxes................................................. 168

III.

Solution of an Ordinary Second Order Non-homogenous Differential

Equation

170

IV.

Solving Ordinary Second Order Homogenous Differential Equation

with Boundary Conditions .............................................................................................. 172

V.

Matching Mean and Variance of a Bivariate Binomial Lattice with Geometric

Brownian Motions .......................................................................................................... 173

VI.
8

Moment/Cumulant and Gram-Charlier series................................ 176

Publications............................................................................................................. 179

List of Tables
Table 2.1 Summary of German feed-in tariffs for land and sea wind power ... 32
Table 2.2 Summary of feed-in tariffs for various photovoltaic installations.... 32
Table 2.3 EUA prices for various scenarios in Nordic Energy Perspectives.... 43
Table 2.4 Summary of attributes of groups of NEP models ............................. 44
Table 3.1 Summary of distribution parameters of selected wind speed data.... 68
Table 3.2 Annual figures of the 3.2M wind turbine placed at Station 225
IJmuiden.................................................................................................... 70
Table 3.3 Comparison between simulated and empirical average powers of the
3.2MW wind turbine placed at various locations. .................................... 71
Table 3.4 Average wind power and its standard deviation............................... 74
Table 3.5 Statistical properties of residuals ...................................................... 75
Table 3.6 Statistical properties of residuals (monthly basis) ............................ 76
Table 3.7 Wind power statistics: analytic vs simulation................................... 76
Table 3.8 Wind turbine breakdown by capacities in Denmark 2009................ 80
Table 4.1 Financial parameters for fixed tariff wind power project ............... 103
Table 5.1 Drifts and volatilities derived from fossil fuel prices ..................... 140
Table 5.2 Correlations between three pairs of fossil fuel ............................... 141
Table 5.3 U.S. Treasury bond yields (Dec 2010) ........................................... 142
Table 5.4 Beta for wind power project ........................................................... 143
Table 5.5 Assumed cost data for wind turbines .............................................. 143
Table 5.6 One-area generator data .................................................................. 144
Table 5.7 Generator outage cumulants ........................................................... 145
Table 5.8 Wind power under-capacity cumulants .......................................... 145
Table 5.9 Expected energy productions before and after wind capacity addition
................................................................................................................. 147
Table 5.10 Annual fuel reductions to an IEEE-RTS96 area by 28.5MW wind
capacity ................................................................................................... 147
Table 5.11 Base case valuation of a 28.5MW wind power project ................ 148
Table V.1 Discretization outcomes of two correlated GBMs......................... 173

10

List of Figures
Fig. 3.1 An ideal wind turbine power curve ..................................................... 52
Fig. 3.2 Wind turbine characteristics for maximum power extraction (Courtesy
of [59]) ...................................................................................................... 54
Fig. 3.3 Comparison between power curves of fixed speed and variable speed
wind turbine generators (Courtesy of [65]) .............................................. 54
Fig. 3.4 Comparison between power curves of pitch control and stall control
wind turbine generators (Courtesy of [64]) .............................................. 55
Fig. 3.5 Visual comparison between three-segment and four-segment power
curve.......................................................................................................... 57
Fig. 3.6 Empirical power curve determined by regressing real data ................ 64
Fig. 3.7 Waked wind speed density function (b) compared with its original
Rayleigh source (a) ................................................................................... 65
Fig. 3.8 Effective average power of waked wind turbine ................................. 66
Fig. 3.9 Average powers along specific months of all years ............................ 72
Fig. 3.10 Annual average wind power, Station I.D. 210, Valkenburg.............. 73
Fig. 3.11 Annual average wind power, various Dutch locations. ..................... 73
Fig. 3.12 Normality test for residuals (differences between global mean and
annual averages)........................................................................................ 75
Fig. 3.13 Wind power PDF synthesized from simple power curve .................. 77
Fig. 3.14 Wind power PDF synthesized from improved power curve ............. 77
Fig. 3.15 Modeling empirical wind power by analytical PDF.......................... 78
Fig. 3.16 Successive convolution of individual wind turbine outputs.............. 81
Fig. 3.17 Standardized PDF of correlated cumulant method of 7 variables .... 82
Fig. 3.18 Standardized PDF of correlated cumulant method of 31 variables .. 82
Fig. 4.1 Project NPV and levered NPV of one MW onshore wind capacity
investment ............................................................................................... 105
Fig. 4.2 The VaR debt level of one MW onshore wind capacity investment . 105
Fig. 4.3 Sensitivity analysis of VaR debt to debt interest rate and default
probability for onshore wind farm .......................................................... 106

11

Fig. 4.4 Sensitivity analysis of VaR debt to debt interest rate and default
probability for offshore wind farm ......................................................... 107
Fig. 4.5 Sensitivity analysis of maximum levered NPV to debt interest rate and
return on unlevered equity for onshore wind farm ................................. 107
Fig. 4.6 Sensitivity analysis of maximum levered firm value to debt interest rate
and return on unlevered equity for offshore wind farm .......................... 108
Fig. 5.1 A time step of a binomial model ....................................................... 115
Fig. 5.2 Bivariate binomial lattice and iteration of its option value ............... 130
Fig. 5.3 Fuel prices for electric power use in U.S........................................... 141
Fig. 5.4 S&P 500 and Dow Jones Utility Average since 1980 ....................... 143
Fig. 5.5 Sensitivity analysis of land wind project NPV over fuel prices ........ 149
Fig. 5.6 Sensitivity analysis of land wind project IRR over carbon price ...... 150
Fig. 5.7 Sensitivity analysis of land wind project NPV over carbon price and
emission policy arrival rate ..................................................................... 150
Fig. 5.8 Sensitivity analysis of sea wind project NPV over fuel prices.......... 151
Fig. 5.9 Sensitivity analysis of sea wind project IRR over carbon price ........ 151
Fig. 5.10 Sensitivity analysis of sea wind project NPV over carbon price and
emission policy arrival rate ..................................................................... 152
Fig. 5.11 Synthetic trading values of wind power investment real options.... 153

12

List of Notations
A
a
B
c
C
C co2

Area swept by wind turbine rotor; constant in contingent claim analysis

Probability of default
Constant in real option solution
Call option; central moment
Wind turbine operating cost
Carbon dioxide content of fossil fuel (lbs/MBtu)

CP

Power coefficient

CT

Thrust coefficient

D
d
Dp

Debt
Debt instalment
Depreciation

dq
dz
E
EW

Poisson process
Wiener process
Equity
Annual wind energy production

f(.)
F(.)
F(S,t), F
g

In general means a function or a PDF

In general means a CDF
Wind power investment (real) option
Dummy for growth rate, e.g. g C means growth rate of wind turbine operating cost

gm

Rated power of wind turbine net of electrical loss

Fuel heat (MBtu), e.g. H o means heat content of oil

h
iD

Wind turbine hub height

Debt interest

K
k
LK i

Option strike price

Shape parameter of Weibull distribution; wake decay constant
System load cumulant to the order i

m
O CF

Moment
Operating cash flow

Oi

Operating income

OK i

Generator outage cumulant to the order i

p
P air

Probability of price evolution in binomial model, have subscript u or d

Power in free wind speed

P wt

Power extracted by wind turbine

q
R
r, r f

Risk-neutral probability
Radius of wind turbine blade
Risk-free rate

RA

Cost/return of asset

13

List of Notations (contd)

R co2

Carbon saving/revenue

RD

Cost/return of debt

r DSC

Debt service coverage ratio

RE

Cost/return of equity

ri

Return of an individual stock i

rm

Return of market portfolio

RU
S

Unlevered cost of capital

S co2

Price in GBM in general, e.g. S c means coal price ($/MBtu)

Carbon price (/ton)

S RC
T
tC

Renewable credit price

TC

Corporate tax

tW
v
V
V(S,t),V
V*
VL

Wind energy tariff

VU
w
WK i

Unlevered firm value

yD
z

Debt coupon rate

(S,t)

Terminal time period

Corporate tax rate

The drift of additive Brownian motion

Dummy for firm value
Contingent claim of wind power project
Optimal project value
Levered firm value
Dummy variable of wind speed
Wind power cumulant to the order i
Roughness length
Drift rate of GBM
Beta coefficient in CAPM; Constant in real option solution
The delta of hedging
Convenience/dividend yield, in general means

Infinitesimal time period

Cumulant
Parameter of exponential distribution; scale parameter of Weibull distribution; tip
speed ratio
Risk-adjusted discount rate by CAPM
Net cost saving (profit function) of wind turbine
Correlation between two GBM; air density
Volatility rate of GBM; standard deviation
Angular speed of wind turbine

14

List of Abbreviations
BS
BSDE
CAPM
CDF
CDM
CER
CPUC
DFIG
DG
Dp
DSC
EBIT
EENS
EIA
ETS
EUA
FIT
GBM
GHG
IPP
IRP
IRR
LDC
LOLP
M&M
NEP
NPV
NREL
NWC
OCF
PDF
PPC
PTC
PURPA
PV
REC
RET
RPS
Tc
VaR
WACC

Black Scholes
Black Scholes differential equation
Capital Asset Pricing Model
cumulative distribution function
clean development mechanism
certified emission reductions
California Public Utilities Commission
doubly-fed induction generator
distributed generation
depreciation
debt service coverage
earning before interest and tax
expected energy not served
Energy Information Administration
Emissions Trading Scheme
European Union allowance
feed-in tariff
geometric Brownain motion
greenhouse gas
independent power producer
Integrated Resources Planning
internal required rate of return
load duration curve
loss-of-load probability
Modigliani and Miller
Nordic Energy Perspectives
net present value
National Renewable Energy Laboratory
net working capital
operating cash flow
probability density function
probabilistic production costing
production tax credit
Public Utility Regulatory Policies Act
photovoltaic
renewable energy certificate
Renewable Energy Target
Renewable Portfolio Standard
corporate tax
value-at-risk
weighted average cost of capital

15

Chapter 1

1 Overview of Generation Planning and Investment

Abstract
This thesis works on investment modeling of wind power, and theoretically
photovoltaics as well. Under the broad umbrella of generation planning and investment,
three topics, namely, conventional generation expansion planning, distribution generation
planning and the more contemporary generation investment are first identified and
discussed. Historical developments of modeling approaches to these topics are reviewed.
Then, in my opinion, the ways of constructing wind power investment model should
consider two aspects. On one hand, renewable energy investment may be recognized as
part of the overall generation investment; coherency with existing modeling works has to
be strived for. On the other hand, renewable generation may have unique characteristics
that could only be catered by new modeling techniques; in this case consistency with its
own technical specifics is more desirable. It is this special orientation of renewable
generation that requires careful justification of the choice of investment modeling
methodology.

1.1

Introduction
In power system literature, wind power investment analysis is relatively scarce. It

seldom exists as standalone, comprehensive investment model. Rather, wind, together

with other renewable, appears only as component in generation planning model. It can
also be grouped under distributed generation planning. However, both planning cases are
not readily transformable into the open market scenario if investment modeling is
required. Although there is a breakthrough of evaluating profit of conventional generator

16

in spot market by real option methodology, wind power or other renewable investment
does not readily utilize this approach. The dilemma of positioning wind power investment
analysis is further explained in Section 1.5 Research Motivations of this chapter. Here I
intentionally start the thesis by reviewing conventional generation planning and
investment, so as to give a preliminary for wind power investment to come in place. Then
I briefly discuss the approach of modeling and expected contribution in Section 1.6
Objective and Expected Contribution.

1.2

Conventional Generation Planning

Conventional generation planning or generation expansion planning is the electric

utilitys decision on generating capacity additions to meet future load demand. The task
composes of a series of questions of when, where, what type and capacity of generators to
be built in long run. In the past, electric utility was vertically integrated with generation,
transmission and distribution together, essentially monopolistic in its own geographical
area. Therefore electricity tariff necessitated a cap or regulation. The business model or
objective of electric utility is to minimize total cost without jeopardizing reliable supply
to customers. This translates generation planning into a constrained optimization problem:
to minimize total costs subject to some constraints.
There are many applications of optimization in power systems; economic dispatch
is probably the most common one. It is a non-linear optimization (programming) problem
as the input-output characteristics of condensing generators, hence the objective cost
function, is nonlinear [1]. Power balance is the equality constraint. Economic dispatch is
run every moment, e.g. a couple of minutes, throughout system operation. During such
period of time, system load may be regarded as constant or deterministic, therefore
economic dispatch tells the optimal generator outputs corresponding to the load of that
moment. Later we will see situations that the load cannot be treated as deterministic but
has a few random scenarios, e.g. long-term load forecast, so that when optimization is
applied the problem becomes stochastic optimization.

17

Economic dispatch yields optimal solution for a particular moment. On the other
hand, unit commitment calculates multi-period generator outputs that are as a whole
optimal for the time period concerned. As a generator has on and off states, binary
variables 1 and 0 respectively are needed in the objective cost function. Hence the
optimization becomes a multi-period, mixed-integer, and nonlinear programming
problem. The optimal sequence of generator states is chronological in nature in order to
minimize the total operating cost.
Multi-period optimization technique has been extended to the context of
generation planning, which takes the following features. First, the basic objective
function to be minimized is the present value of total cost. Total cost comprises of
investment costs of all types of generator at any capacity incurred throughout the whole
timeline and the corresponding operating costs, primarily fuel costs. It is sufficient to
approximate the generator input-output curves by linear segments because the level of
details of non-linear objective function is not necessary for long-term planning. Therefore
generation planning can be as simple as a linear programming problem with annual
capacity additions and energy outputs as decision variables. In terms of constraints, the
most important one is reliability. The common reliability criterion is the loss-of-load
probability (LOLP). Since all generators have outage probability, the total cost is
minimized subject to a pre-defined value of LOLP as constraint. Furthermore, concepts
of load duration curve and probabilistic production costing capture economic dispatch in
generation planning by loading units according to their incremental costs. An example of
generation planning in simple linear programming setting is called sequential linear
programming [32].
The formulation of [32] has catered unit forced outage and loss-of-load
probability by probabilistic production costing and reliability evaluation respectively, but
still it has limited capability to handle other broader planning uncertainties, such as load
growth rate and fuel cost growth rates, as they are only represented as deterministic
parameters. The set of parameters could be describing a particular scenario, or average
value of a few scenarios. Essentially, the resultant expansion plan is optimal only to a
particular set of deterministic equivalent parameters. Reference [3] made quite a precise
description on the limitations of deterministic linear programming applied to generation

18

planning, and also on the solution approach by decision analysis and min-max strategy
for multi-objective function. Subsequently multi-period mixed integer stochastic
optimization was proposed [3]. The resultant objective function is divided into two subproblems and solved by Benders decomposition technique. Its numerical technique is not
explored further here. Apart from linear or non-linear programming, generation planning
can also be done by dynamic programming, either deterministically [23] or stochastically
[5]. For completeness, emerging techniques on generation planning other than
conventional mathematical programming are also reported [17]. Up to here a gentle
review on generation planning and its optimization techniques is completed.

1.3

Distributed Generation Planning

Optimization techniques have also been applied in distribution planning. Very

generic mixed-integer programming models for distribution planning, in terms of

substation capacity and feeder capacity, are presented in [30]. A major subset of
distribution planning is distributed generation (DG) planning, including renewable energy.
DG faces a number of technical constraints in the distribution network, such as short
circuit level and voltage level; optimal allocation (location and rating) of DG can be
formulated as an optimization problem. Linear programming to determine the maximum
DG capacity with respect to the network constraints is reported in [2]. Mathematical
programming is a good formulation to cover as many constraints as possible, but it limits
the problem nature as a planning model. What may be more contemporary, as mentioned
in [31], is DG investment, which describes distribution utility considering DG as an
alternative to meet future load demand. In competitive electricity market where
distribution utility acts as a buyer, it bids in the spot market or purchases electricity
directly from other generators through bilateral contracts. But in principle, both
distribution utility and end customers can own DG [31]. The optimal DG investment
decision, again in terms of location and capacity, is determined from a proposed heuristic
approach to minimize total cost, which consists of DG investment and operating costs,
network upgrade cost, electricity purchase (spot or contract) and unserved load.

19

In course of literature review of distributed generation, one widely mentioned

attribute is the so-called economic benefit. The benefits recognized are network upgrade
deferral, reduced network loss and avoidance of electricity purchase. The authors of [15]
believe that efficient economic systems require that those who create a benefit to
someone else be economically compensated. If DG is found good to a system, it is
necessary to quantify its benefits and make them visible to regulators for the development
of incentives or commercial mechanisms that can allocate those benefits back to DG
owners and improve the profitability of their investment. In turn this encourages
implementation of DG that is valuable to the system and society as a whole. Specifically,
the same authors have offered a quantification of the deferral value (benefits of deferring
substation and feeder investments) of a hypothetical DG in a testing system [16].

1.4

Generation Investment
The unbundling of generation assets from the grid has fundamentally abolished

the traditional concept of generation planning. There is no such entity as the vertically
integrated utility that could look for a least-cost generation expansion plan anymore.
Instead, generation companies sell electricity in competitive wholesale or spot market,
and through bilateral contracts. Their common objective is to maximize individual profits.
Generation investment models are very often evaluated in the context of bidding in dayahead spot market. Two main research areas that are under the hierarchy of generation
investment are identified: financial risk management of generator profit and valuation of
such generator.

1.4.1 Financial risk management of generator profit

Risk management and assessment of generator profit in contemporary electricity
markets is very broad, comprehensive survey papers [19] and [25] serve as good
introduction. Selected topic, for example, is the optimization of portfolio of contacts hold
by generation company. A portfolio of contracts comprises of revenue contracts and fuel

20

contracts. A utility function, defined in terms of expected return of the portfolio, is

maximized subject to a stated level of risk denoted in standard deviation of the expected
return. A very general setting of portfolio comprising revenue contracts, including futures
contracts, and fuel contracts is constructed and then the tradeoff between return and risk
is illustrated by efficient frontier [26]. A slight modification to incorporate multi-market
conditions, both spot market and bilateral contract, is also found [20]. Other topics
include hedging generator operations with forward contracts [21] and futures contracts
[10]. Nevertheless, there is argument from the economists point of view that hedging by
electricity futures is not the same as other commodities [24]. A spot transaction in the
future can be hedged by the next-to-expire futures (futures with expiration date right after
the transaction). The difference between the prospective spot price and the futures price,
called basis risk, would be normally small and stable as it is affected by delivery force
only. This is true for most commodities. However, electricity is non-storable, so
electricity price can fluctuate greatly with no guarantee that the price now would be
similar to the next hours due to, e.g. sudden forced outage. Hence basis risk would still
be large. More recently, with the prerequisite that the probability distribution of generator
profit in spot market is available [33], analytical formulae of common risk assessment
tools such as standard deviation, value-at-risk and conditional value-at-risk are also
derived [34]. References on financial risk management in electricity markets are made
very selectively and stopped here. In the next section, valuation of generator in spot
market will be carried on.

1.4.2 Valuation of generator in spot market

As a broad classification, generator valuation can be separated into price-based
unit commitment approach [7][8][9] and real option approach [27][28][29]. The
application of option theory for generator valuation by Deng et al. has received most
attentions. In their work, generator profit is modeled by a spark spread option, in which
its value is solved to be the expected value of a derivative (the derivative takes a
probability distribution) according to the Black-Scholes theory. However, the key
argument in option pricing, i.e., the formation of riskless portfolio leading to risk-

21

neutrality is not valid for electricity price because electricity is non-storable. In view of
this, two works have been proposed. First, probability distributions of generator profit
based on two stochastic processes of electricity price, namely geometric Brownian
motion and geometric mean-reverting process [18], are analytically derived without
resorting to Black-Scholes theory [33]. Its simulation counterpart is also done [11].
Second, risk-adjusted valuation, instead of risk-neutral valuation of generator is
attempted [12]. These references collectively complete the literature review of the
development of generator valuation by real option approach.

1.5

Research Motivations
This thesis is concerned with wind power investment and development of its

analytical models. Yet, I start the thesis introduction by writing the conventional and
distributed generation planning, for a few reasons as below.

1.

Problem formulations of generation planning in the framework of mathematical

programming or optimization are more or less the same and saturated, merely adding
renewable components such as wind power does not bring too much breakthrough.

2.

Centralized generation planning has simply become obsolete in restructured

electricity markets. Generation investment should take place, yet how should
renewable investment be formulated remains as a research question. Renewable
investment could be a direct extension from conventional generation investment or
radically viewed from a new perspective.

3.

Mathematical programming or optimization for distributed generation (including

renewable) planning is reasonable, but its capability to analyze long-term DG
investment is questionable.

4.

Optimization itself has some inherent weaknesses in handling financial aspects of

generation planning. Discount rates (fuel and electricity price expected returns,
project required return, etc.) are usually exogenously assumed, without inferring

22

concrete reasoning to finance theories. In particular, fuel price volatilities are left
completely unattended, implying no measurement of financial risk.
5.

Generation projects are huge and carry large projected cash flows. Discount rate is
very sensitive to project valuation, but its importance is usually overlooked [14].

6.

It is believed that the financial approach, which can consistently accommodate

return and risk, is needed for renewable and wind power investment. Also, the
formulation has to be able to consider technical specific of renewable generation as
much as possible.

Generally speaking, there is no simple and readily available framework for wind
power investment to be analysed by optimization or financial model. The problem
formulation has to be vetted from its underlying scenario, which mostly depends on the
market and regulatory rules for wind power. For example, wind energy could be paid at
spot market price, fixed feed-in tariff or its tender price. Meanwhile, wind power is
driven by national renewable energy target to various extents. All these market and
regulatory factors determine the right type of investment models and subsequent
valuation result of wind project. In Chapter 2, details of the market scenarios for wind
power will be described.
One clear policy driver of wind power is the renewable portfolio or target so that
wind power project has to be deployed before a certain deadline. It is not impossible for
wind power investment having no profit if it is built for political and environmental
concerns rather than actual cost-benefit consideration. Nevertheless, the investment
timing of renewable projects to comply the renewable target deadline is flexible before
the deadline. It may be better to build later rather than now. Hence there is a distinct
motivation to model wind power investment as a real option because the flexibility of
investment timing can be captured. Real option evaluation of wind power project based
on an appropriate scenario is the main theme of this work.

1.6

Objective and Expected Contribution

23

The goal of this work is to create analytical models for wind power investment
analysis. Models would vary according to market scenarios, but the common objective is
to give assessment of the valuation of wind power projects, in course of the following
high-level considerations (not in order):

1.

Investment models are formulated in consistence to finance theories, which apply

for any but not limited to generation projects.

2.

Electricity market and regulatory rules should be properly addressed in the models.

3.

Benefit of wind power to the society as a whole should be assessed, in particular, its
value should depend on how much fossil fuel saved [6][13][22]

4.

Economic viability of wind power itself is more appealing than the subsidized case.

5.

Effects of wind variability can only be assessed in conjunction with the specifics of
individual power system where the wind farm is connected [4]. Generic investment
model applicable for different power system structures is preferred.

6.

Sufficient technical or power system considerations should be incorporated into the

wind power investment models.
Investment models fulfilling the above considerations are expected to contribute to

investors a set of comprehensive and accurate valuation tools for wind power projects in
various market scenarios.

1.7

Thesis Outline
This thesis composes of six chapters. As we have gone through, Chapter 1 is an

introduction of generation planning and investment, which serves as a platform for

renewable investment to come into discussion. Research motivations and objectives are
also stated in this chapter. Chapter 2 is an overview of market scenarios for wind power
investment. In particular, it highlights four scenarios in which each scenario shall lead to
unique investment methodology. Out of the four, two scenarios will be further explored
in subsequent chapters, which collectively explain the development of the proposed

24

analytical models of wind power investment. Chapter 3 is about the derivation of a

probabilistic wind power generation model. Chapter 4 is concerned with investment
analysis and capital structuring of wind power project under feed-in tariff. Chapter 5
offers a real option model of wind power investment from the perspective of utility.
Finally, conclusion is made in Chapter 6 to affirm accomplishment of the overall
objective of this research study. Every chapter, except this one, has a short summary or
future work. References are included at the end of each chapter. Derivations of formulae
are mostly kept in Appendix.

1.8

References

[1] Allen J. Wood and Bruce F. Wollenberg, Power Generation Operation and Control,
New York: Wiley, 1996.
[2] Andrew Keane and Mark OMalley, "Optimal Allocation of Embedded Generation on
Distribution Network," IEEE Trans. Power System, Vol. 20, No. 3, pp. 1640-1646,
Aug 2005.
[3] B. G. Gorenstin, N. M. Campodonico, J. P. Costa and M. V. F. Pereira, "Power
System Expansion Planning under Uncertainty," IEEE Trans. Power System, Vol. 8,
No. 1, pp. 129-136, Feb 1993.
[4] Bart C. Ummels, Madeleine Gibescu, Engbert Pelgrum, Wil L. Kling, and Arno J.
Brand, Impacts of Wind Power on Thermal Generation Unit Commitment and
Dispatch, IEEE Trans. Energy Conversion, vol. 22, No. 1, pp. 44-51, March 2007.
[5] Birger Mo, Jan Hegge and Ivar Wangensteen, "Stochastic Generation Expansion
Planning by means of Stochastic Dynamic Programming," IEEE Trans. Power
System, Vol. 6, No. 2, pp. 662-668, May 1991.
[6] Brendan Fox, Damian Flynn, Leslie Bryans, Nick Jenkins, David Miborrow, Mark
OMalley, Richard Watson, and Olimpo Anaya-Lara, Wind Power Integration,
Connection and System Operation Aspects, London: IET Power and Energy Series,
2007.

25

[7] Chung-Li Tseng and Graydon Barz, Short-Term Generation Asset Valuation, in
Proc. the 32nd Hawaii International Conference on System Sciences, 5-8 Jan 1999.
[8] Chung-Li Tseng and Graydon Barz, Short-Term Generation Asset Valuation: A Real
Options Approach, Operations Research, vol. 50, no. 2, pp. 297-310, Mar-Apr 2002.
[9] Erjiang Sun and Edwin Liu, Generation Asset Valuation under market
Uncertainties, in Proc. 2007 IEEE Power Engineering Society General Meeting,
Tampa.
[10]

Eva Tanlapco, Jacques Lawarree and Chen-Ching Liu, "Hedging with Futures

Contracts in a Deregulated electricity Industry," IEEE Trans. Power Syst., Vol. 17,
No. 3, pp. 577-582, Aug 2002.
[11]

Felix F. Wu, Jifeng Su, Hui Zhou and Yunhe Hou, Valuation of Generator Profit

from Spot Market: Simulation Approach, submitted to IEEE Trans. Power System.
[12]

Felix F. Wu, Yang, He Zhou and Yunhe Hou, Risk-adjusted Valuation of

Generator Asset, submitted to IEEE Trans. Power System.

[13]

Hannele Holttinen and Jens Pedersen, "The Effect of Large Scale Wind Power on

a Thermal System Operation," in Proc. the 4th International Workshop on Large

Scale Integration of Wind Power and Transmission Networks for Offshore Wind
Farms, pp. E1-E7, 20-22 Oct. 2003.
[14]

Hisham Khatib, Economic Evaluation of Projects in the Electricity Supply

Industry, IEE Power and Energy Series 44.

[15]

Hugo A. Gil and Geza Joos, Models for Quantifying the Economics Benefits of

Distributed Generation, IEEE Trans. Power System, Vol. 23, No. 2, pp. 327-335,
May 2008.
[16]

Hugo A. Gil and Geza Joos, On the Quantification of the Network Capacity

Deferral Value of Distributed Generation, IEEE Trans. Power System, Vol. 21, No.
4, pp. 1592-1599, Nov 2006.
[17]

Jinxiang Zhu and Mo-yuen Chow, "A Review of Emerging Techniques on

Generation Expansion Planning," IEEE Trans. Power System, Vol. 12, No. 4, pp.
1722-1728, Nov 1997.

26

[18]

Julio J. Lucia and Eduardo S. Schwartz, Electricity Prices and Power

Derivatives: Evidence from the Nordic Power Exchange, Review of Derivatives

Research, 5, pp. 5-50, 2002.
[19]

Min Liu, Felix F. Wu and Yixin Ni, A Survey on Risk Management in

Electricity Markets, in Proc. 2006 IEEE Power Engineering Society General

Meeting, Montreal.
[20]

Min Liu and Felix F. Wu, Managing Price Risk in a Multimarket Environment,

IEEE Trans. Power Syst., Vol. 21, No. 4, pp. 1512-1519, Nov 2006.
[21]

R. J. Kaye, H. R. Outhred and C. H. Barmister, Forwards Contracts for the

Operation of an Electricity Industry under Spot Pricing, IEEE Trans. Power Syst.,
Vol. 5, No. 1, pp. 46-52, Feb 1990.
[22]

R. N. Allan and Avella Corredor, Reliability and economic assessment of

generating systems containing wind energy sources, IEE Proc. C, Vol. 132, No. 1,
pp. 8-13, Jan 1985.
[23]

R. R. Booth, "Optimal Generation Planning considering Uncertainty," IEEE

Trans. PAS, Vol. PAS-91, No. 1, pp. 70-77, 1972.

[24]

Robert A. Collins, The Economics of Electricity Hedging and a Proposed

Modification for the Futures Contract for Electricity, IEEE Trans. on Power
Systems, vol. 17, no.1, pp. 100-107, Feb 2002.
[25]

Robert Dahlgren, Chen-Ching Liu and Jacques Lawarree, "Risk Assessment in

energy Trading," IEEE Trans. Power Syst., Vol. 18, No. 2, pp. 503-511, May 2003.
[26]

Roger Bjorgan, Chen-Ching Liu and Jacques Lawarree, "Financial Risk

Management in a Competitive Electricity market," IEEE Trans. Power Syst., Vol. 14,
No. 4, pp. 1285-1291, Nov 1999.
[27]

Shijie Deng, Financial methods in competitive electricity markets, Ph.D.

dissertation, University of California, Berkeley, CA, 1998.

[28]

Shijie Deng, Blake Johnson and Aram Sogomonian, Spark Spread Options and

the Valuation of Electricity Generation Assets, in Proc. the 32nd Hawaii

International Conference on System Sciences, 5-8 Jan 1999.

27

[29]

Shijie Deng, Blake Johnson and Aram Sogomonian, Exotic electricity options

and the valuation of electricity generation and transmission assets, Decision Support
Systems, 30, pp. 383-392, Jan 2001.
[30]

Suresh K. Khator and Lawrence C. Leung, "Power Distribution Planning: A

Review of Models and Issues," IEEE Trans. Power System, Vol. 12, No. 3, pp. 11511159, Aug 1997.
[31]

Walid El-Khattam, Kankar Bhattacharya, Yasser Hegazy and M. M. A. Salama,

Optimal Investment Planning for Distributed Generation in a Competitive Electricity

Market, IEEE Trans. Power System, Vol. 19, No. 3, pp. 1674-1684, Aug 2004.
[32]

William Rutz, Martin Becker, Frank E. Wicks and Stephen Yerazunis,

"Sequential Objective Linear Programming for Generation Planning," IEEE Trans.

PAS, Vol. PAS-98, No. 6, pp. 2015-2021, Nov/Dec 1979.
[33]

Yunhe Hou and F. F. Wu, Valuation of Generator Profit from Spot Market:

Analytical Approach, submitted to IEEE Trans. Power System.

[34]

Yunhe Hou and F. F. Wu, Risk Assessment of Generator Asset in Electricity

Markets: Analytical Approach, submitted to IEEE Trans. Power System.

28

Chapter 2

2 Market Scenarios for Wind Power Investment

Abstract
For any investment analysis, it is necessary to identify the relevant market
background before choosing the proper analytical tool or model. In this chapter, a number
of wind power investment scenarios are identified in accordance to modern electricity
market regimes. The two main scenarios are fixed tariff wind project by independent
power producers and wind power project undertaken by utility. Details of market
structure and regulation are discussed as far as investment modeling is concerned. It has
to be emphasized that different market scenarios would lead to different modeling
methodologies for best representing the reality. This chapter serves as introduction of the
rationale of modeling approaches chosen for the two highlighted scenarios that will be
further explored in subsequent Chapter 4 and 5.

2.1

Background and Scope

Chapter 1 has already outlined three topics in generation planning or investment,

namely, traditional generation expansion planning, distributed generation planning, and

generation investment in deregulated markets. They are problems corresponding to their
market regimes. It is important to recognize the type of market structure before inferring
to any planning or investment modeling methodology. Therefore in this chapter a survey
on modern electricity market rules and regulations for wind power development in some
major regions is first conducted. The survey is based on materials from a couple of
government regulatory issues, technical reports and internet resources rather than
academic papers, because the nature of materials is rather informative than research-

29

oriented. Most importantly, the survey is classified into four different scenarios of wind
power investment: feed-in tariff, obligation for wind generation, wind power in spot
market, and auctioning, as follow in the remaining sections of this chapter. Some
contents in the survey are not limited to wind power but apply to the more general
renewable energy. Based on feed-in tariff and obligation, wind power investment models
will be derived and presented in subsequent chapters.
Several points were recognized and paid attention to when this survey was written.
First, market rules and regulations could be detailed and have many special cases, the
merit is to capture essential and generic parts but avoid unnecessary extensions. Second,
proper scenarios are identified for investment modeling with emphasis on the subject of
making the investment, i.e. who the investor is. Third, it should be borne in mind that
wind power development is not new upon the restructuring of electricity market. Rules
and regulations for wind power were there for some time and have also been evolving in
parallel to electricity market restructuring. In short, I try to capture and consolidate the
links between wind power pricing and modern electricity market in order to create a
starting point for further investment analysis.

2.2

Feed-in Tariff
Tariffs for renewable generation are mostly feed-in and fixed. Feed-in could be

understood as dispatch with higher priority. Since wind and solar are intermittent and
their powers non-dispatchable, and also for the purpose of promoting renewable, they are
dispatched before conventional generation. Tariffs are usually fixed for many years of
operation of the renewable installations, and are differentiated among different renewable
technologies. Each unit of electricity generated is paid fixed throughout the whole period.
Such fixed tariffs usually have premium to provide guarantee return for the expensive
renewable investment, in which the premium is carefully assessed to balance between
investment incentive and consumer welfare.
Wind power development has a long history and was well before restructuring of
electricity markets in most regions. The wholesale generation bidding mechanism, on the

30

other hand, is designed for big conventional generators. For two reasons wind power
producers do not find spot market an attractive platform to sell electricity. First the spot
price is low compared with wind power initial cost. Second, the mechanism of bidding
does not favour intermittent wind power radically. Therefore, something different is
needed to promote wind power, and feed-in tariff is observed to be prevalent.
An introduction of feed-in tariffs is found in [42]. Apart from the two basic
features of feed-in tariff (fixed and priority dispatch), very often the later the renewable
installation is built, the smaller is the tariff, whereas the already existed installations are
not affected. Such mechanism is called tariff degression, which creates an incentive to
boost renewable investment early and at the same time takes into account the general
dropping trend of renewable technology costs. The rate of degression is in annual
percentage reduction.
Two advocates of feed-in tariff are Germany and Spain, in particular German
feed-in tariff has been very aggressive. I try to highlight the German Renewable Energy
Sources Act [47] on both wind power and photovoltaics, and extract some of their tariff
structures for discussion in the coming two sub-sections. Readers who do not need
specific figures may jump over directly to the concluding remark of the suitable
investment modeling approach for feed-in tariff wind power.

2.2.1 German wind tariffs

The level of degression for wind energy commissioning in 2000 and 2004 are 1%
and 2% respectively, to recognize the cost reductions in manufacturing of wind turbines.
However, for offshore wind energy, the tariff remains the same as in 2000 and degression
comes only in 2008 at 2%. Tariffs are different for onshore and offshore wind turbines,
and furthermore there are two levels of tariffs for each type of turbine. The basic tariff for
onshore wind turbine commissioning in 2007 is 5.17 cents/kWh for 20 years. If, in the
first 5 years, the wind farm generates more than expected and reaches 150% of a
reference installation, the tariff for the corresponding period is increased to 8.19
cents/kWh. The 150% reference is not a target but only a reference. For every 0.75% the
generation falls short of the reference, the increased tariff period will be extended by two

31

months. Such remedy tries to prevent excessive demand of windy site because less windy
site can enjoy longer period of higher tariff. However, there would be no fee if the
generation turns out to be less than 60% of the reference. After all, the degression still
applies. Table 2.1 summarizes the tariffs for both onshore and offshore wind farms.
Tariff ( cents /

Conditions for increased tariff

kWh)
Type

Start-up year

2004

2007

Onshore

Basic tariff

5.9

5.17

For the first 5 years if output reaches 150% of

Increased tariff

8.8

8.19

reference, yet the period is extended 2 months for

every 0.75% falls short of 150%

Offshore

Basic tariff

6.19

6.19

For the first 12 years if site is 3 nautical miles off

Increased tariff

9.1

9.1

the coast, extended half a month for every further

mile. Concurrently, high tariff period extended by
1.7 months for every metre in depth of water
deeper than 20m where turbines sit.

Table 2.1 Summary of German feed-in tariffs for land and sea wind power

2.2.2 German photovoltaic tariffs

Tariffs for photovoltaic (PV) installations are quite diversified, with classification
into installation methods (on buildings or open space) and capacities. Tariff degression
for open space PV (6.5%) is higher than that of building PV (5%). For easy reference,
Table 2.2 is a summary of feed-in tariffs for various photovoltaic installations.
Tariff ( cents / kWh)
2004

2005

2006

2007

<30kWp

57.4

54.53

51.8

49.21

30-100kWp

54.6

51.87

49.28

46.82

>100kWp

54

51.3

48.74

46.3

Faade bonus

Open space PV

45.7

43.42

40.6

37.96

Start-up year
PV on buildings

Table 2.2 Summary of feed-in tariffs for various photovoltaic installations

32

2.2.3 Concluding remark for investment modeling

Electricity market deregulation, as well as the environmental concern, provide
marketplace for independent renewable power producers. It is straightforward to analyze
wind power project by net present value (NPV) if the tariff structure is fixed and flat [38].
Apparently, the value of such a project is a simple annuity because both the price and
quantity of wind energy are passively fixed. There shall be no simple way to boost
project value in the level of operation. However, in terms of finance, capital structure has
something to do on firm value. Considering the minimal risk nature of such wind power
project, it should permit high leverage of the initial investment capital. I try to determine
the optimal debt level that maximizes firm value subject to operational characteristics of
wind power.
After all, evaluation of wind power investment under feed-in tariff is simple in
which NPV criterion is sufficient. By observing the regulatory and market factors for
wind power projects in contemporary electricity markets, it indeed leads to more
complicated investment scenarios. Specifically, I try to grasp the idea of large-scale wind
power project invested by distribution utility under certain types of policy obligations.
Policy obligations undertaken by two major parties, the US and the EU, will be explored
in coming sections.

2.3

Obligation American Experiences

American support on renewables can be referenced from a National Renewable

Energy Laboratory (NREL) technical report [40], which encompasses wide coverage of
experiences of wind power development in US. The experiences are in the context of
policy drivers and market factors state by state. I try to summarize those attributes as
renewable portfolio standard (RPS), integrated resource planning (IRP), green power
programs and tax credit & production incentives as follow. While the technical report
illustrates each attribute by real scenarios involving the actual utilities and states, I would

33

give generic descriptions for each attribute followed by some state-specific experiences if
necessary. The reason for this approach is to focus on the general characteristics of
attributes that are common across states.

2.3.1 Renewable Portfolio Standard

Renewable Portfolio Standard (RPS) is a very effective measure to promote
renewables as a whole. It is a state-level target mandating a certain percentage of its total
annual energy consumption to be supplied by renewables. The target percentage and
deadline vary state by state, and the target percentage is constituted by mixed renewables
according to their abundances. In turn utilities in the state have to procure renewable
power according to the percentages as part of the electricity sold. Overall RPS capacity
target is raised step by step in time. For example, California roughly achieved 15% in
2009 and aims at 33% by 2020 [36]. Texas also has strong compliance with RPS,
especially fulfilled by wind power [45]. Utilities may choose to build its own renewable
facilities or purchase renewable power from others to fulfill RPS. Detailed
implementation of RPS may also change state by state. In California, utilities are allowed
to use part of the public/system benefits funds to cover any above-market costs of
renewable electricity purchase required by the RPS. While in Texas, RPS leads to a subsystem of trading renewable energy certificates (RECs). RECs are granted to renewable
generators according to their electricity output. Renewable generators sell RECs for extra
revenues whereas utilities have to purchase those RECs as a measure of compliance to
the RPS, otherwise penalty would be put on. Up-to-date details of the Texas certificate
trading system can be found in [48].

2.3.2 Integrated Resource Planning

Integrated resource planning (IRP) is just as generation expansion planning of
both conventional and renewable generators. The least-cost expansion plan including
renewable energy is obtained subject to load forecasts. With the general advent of
deregulation and generation unbundling, IRP of renewables could run in parallel and is

34

not completely exclusive from competitive electricity markets. For example, Xcel Energy
in Colorado was once required by the state utility commission to build large-scale wind
farms because wind power was cost-effective compared with gas-fired generation. It also
supplied wholesale wind power to other utilities. Deregulation could limit the ownership
right of wind power projects by utilities. Still, there was case that, e.g. utility in Oregon
called for wind power projects and purchasing agreement to meet its load growth, amid
high wholesale electricity prices. After all, whether an utility directly owns or contracts
wind power project is not important, the point is the utility is given an option to procure
electricity other than in wholesale bidding market. In short, IRP remains a driver of wind
power in some states.

2.3.3 Green Power Programs

There are a lot of green power programs across states. Basically green power
programs are options given to customers to buy electricity, or attribute their electricity
consumptions from renewable. Nowadays, provisions of such options are increasingly
compulsory across states. Green power programs are primarily realized by wind power.
The options are usually fixed-tariff contracts for some years. For a few reasons end-use
customers would switch to green power programs. It is not surprising that customers are
willing to pay more simply because of their environmental awareness. In states where the
standard or base electricity rates are higher, at the same time with large wind resources,
wind power price is actually cheaper by itself. Or some consumers may find slightly
elevated but fixed green power prices are reasonable hedge over the volatile retail
electricity prices.
Ownership of wind power project is an issue of green power program. In most
cases, wind power projects are owned by independent power producers (IPPs) or utilities.
If wind power project is owned by utility, corresponding green power program can be
marketed by the utility itself, or make it non-discriminative with the base rate. If it is
owned by IPP, implementation of green power program could depend on the extent of
deregulation. Power retailers market green power programs if there is retail competition.
Or the utility contracts wind power by power purchasing agreements, in which the wind

35

capacity becomes part of its generation portfolio, and then offers its own green power
program.

2.3.4 Tax Credits and Production Incentives

1. Production tax credit
The production tax credit is a federal policy giving a tax credit for each unit of
electricity sold by qualified renewable facilities. The policy was originally created under
the 1992 Energy Policy Act, and has been extended and expanded quite a few times. The
up-to-date tax credit for wind power is (inflation-adjusted) $2.2 US/kWh and the wind
power facility has to be available by 31 Dec 2012 [46]. The duration of tax credit is 10
years counting from the facility in-service date. For renewable facilities owned by
utilities that do not have federal tax liabilities, the Renewable Energy Production
Incentive may support them alternatively.

2. Other tax and financial incentives

Other taxes, such as sales, investment and property taxes, may have abatements
subject to various states. Tax credit and financial incentive are the most straightforward
way to boost and subsidize renewable investment. Yet they are the least market-based
approaches and their extents may lack justification.

3. PURPA
Public Utility Regulatory Policies Act (PURPA) is another federal policy driver,
which was strongly implemented by California in 1980s. Under PURPA, the California
Public Utilities Commission (CPUC) required its utilities to procure electricity from
qualifying renewable facilities at the utilitys avoided cost. The purchasing contracts are
approved long-term and at high prices and included capacity as well as energy payments.
Properly because the offer is too generous, the PURPA contract was too popular and was
subsequently suspended in 1985. Mid-1990s was a short sluggish period of wind power
when previous PURPA contracts began to expire.

36

4. System/Public benefits funds

Following the sunset of PURPA contracts, Californias 1996 electric industry
restructuring legislation (AB 1890) mandated the three major investor-owned utilities to
collect surcharge on all electricity consumption to create a fund pool, known as
System/Public Benefits Funds, for supporting renewable development. Simply put, the
public benefits funds are like production incentive on each unit of electricity generated
from renewable facilities along a pre-defined period. Wind power is one of the qualifying
renewable facilities. The funds may need to be auctioned. Again the legislation has been
revised and extended, the latest version could be found in [44].

2.3.5 Concluding remark for investment modeling

The renewable portfolio standard leads to a unique scenario of utility wind power
investment. The investment decision is somehow passive because it has to conform to the
renewable target and deadline. Modeling the valuation of wind power project by
American real option is suggested. Furthermore, availability of green power program
allows distribution utility to attribute wind power production to other renewable IPP.
Although distribution utility may not directly own generation assets after deregulation,
one can still analyze the economic performance of a wind power project in terms of
bilateral contracts, as if the project an indirect investment of the utility.

2.4

Obligation European Experiences

The European commitments on climate change can be referenced from a series of

measures stipulated by the European Union. Similar to the US renewable portfolio

standard, EU also has a target percentage of renewable shares in total energy
consumption. Furthermore EU has a target on emissions reduction. Collectively the
targets are so-called the EU energy and climate package. Apart from setting targets, EU
introduced an Emissions Trading Scheme (ETS) in which the industrial and power
generating parties from all its member states trade European Union Allowances (EUAs),

37

commonly known as emission allowances to compensate for emissions produced.

Background of EU ETS is widely available on the Internet or in papers introduction such
as [50]. Nevertheless, both the EU energy and climate package and the ETS will be
briefly mentioned in the following sections.

2.4.1 EU Energy and Climate Package

The EU energy and climate package is a bundle of policy targets on energy and
emission issues binding on all its member states. The three targets, so-called 20-20-20,
are as follow.
1.

To reduce emission of greenhouse gases by at least 20% compared to the level in

1990 by 2020.

2.

To increase the share of renewables in the total energy consumption (including

transport systems) to 20% by 2020.

3.

To achieve energy efficiency of 20% improved over the current status by 2020.

For convenience, the following simplifications on those targets are assumed:

1.

The emission target means the same for carbon dioxide.

2.

How the renewable burden sharing on all EU member states are ignored.

3.

The transport sector is neglected.

4.

A fixed percentage of wind power is assumed out of the 20% target.

2.4.2 EU Emission Trading System

A. Background of Kyoto Protocol
The aim of Kyoto Protocol is to reduce global greenhouse gas (GHG) emissions
by a certain date. Governments who have ratified this treaty can be separated into two
categories: developed countries (Annex 1) and developing countries (Non-annex 1). As
of January 2008, Annex 1 countries have to reduce their GHG emissions by a collective
average of 5% below their 1990 levels by December 2012. The levels of reduction are
specified for each party who ratified the Protocol. This figure actually corresponds to

38

some 15% below the GHG emissions in 2008 for many EU member states. Among EU
member states, the actual emissions reduction may range from an average of 8%
(compared to 1990s) to an emission increase from some less-developed EU countries.
Non-annex 1 countries do not have any restrictions on GHG emissions, but may
participate in the clean development mechanism (CDM) in which when a GHG emission
reduction project is implemented within them, certified emission reductions (CER) would
be earned and can be sold to Annex 1 countries. Annex 1 countries could meet the
emissions caps by purchasing emission allowances from other parties (presumably one
allowance is granted for one permissible tonne of GHG emissions for all Annex 1
countries). Failing to comply will be penalized by having to submit 1.3 emission
allowances for every tonne of GHG emissions in the second commitment period starting
from 2013. International talks have started on matters of second commitment period.

B. European Union Emission Trading Scheme (EU ETS)

EU ETS is a trading system especially for EU member states to trade emission
allowances. Its existence is closely related to the fulfilment of Kyoto Protocol for EU
member states but in fact it had started before Kyoto Protocol was kicked off. European
Union Allowances (EUA), the formal name of emission allowances, are granted to plant
operators for free (grandfathering) according their historical emission levels with
reductions. The allowances are given out for a sequence of several years at once so that
plant operators can neutralize annual irregularities in GHG emissions. The first phase of
EU ETS ended in December 2007 and it was found that the verified emissions between
2005 and 2007 still experienced an increase. The reason is that individual countries
granted the allowances loosely and as a result the price of allowances also dropped to
nearly zero by the end of 2007. Working closely with the Kyoto Protocol, changes
proposed for 2013 onwards (second commitment period of Kyoto Protocol) include
centralized allocation, a migration to auctioning a greater share of allowances instead of
grandfathering and also potentially, a more stringent emissions cap.

2.4.3 Nordic Energy Perspectives

39

Nordic Energy Perspectives (NEP) is an interdisciplinary research project on the

Nordic energy systems. It has a number of energy system models, collectively known as
NEP model toolbox. The models are able to analyse relevant policy instruments and
market factors, then demonstrates their influence and impact on energy markets and
systems. Electrical system is the main subset of the broader sense energy system. The
targets of the EU energy and climate package constitute a few main scenarios for NEP
models to work on. The development of Nordic energy sector is analysed and forecasted
through these scenarios. Results of the NEP analysis are projected based on targets of the
package. The main objective of the NEP project is to demonstrate to stakeholders of the
Nordic energy sector any anticipated effect of following the EU as well as global energy
and climate policies.
The NEP project has gone through its first phase during 2005-2006. Results of its
second phase carried out during 2007-2010 have been recently released. The results
compose of three main documents. The first one is an offprint known as Ten
Opportunities and Challenges for Nordic Energy [51], the second one is a full report
called Towards a Sustainable Nordic Energy System [52] and the third one is about
model toolbox descriptions called Coordinated use of Energy system models in energy
and climate policy analysis [37].
The full report of NEP contains very comprehensive analysis and projected results
based on various scenarios. By no means I extract and compare any results here again.
Instead I am going to highlight a few parts of the modeling methodology in the NEP
project for discussion as follow.

2.4.4 NEP Modelling Methodology

Reference [37] is a standalone book written on the modeling methodology
employed by the NEP project. It serves the following important purposes. First, it
describes the energy-systems modeling methodology in general and the use of different
approaches in NEP in particular. Second, it presents how various models within the NEP
model toolbox are coordinated through synchronization of model assumptions and input
parameters. Third, it illustrates how the models of the NEP project function, their

40

performance and model output achieved. The aim of writing [37] is not to pinpoint model
result in itself; insights and model results of the broader sense energy issue should be
obtained from the main NEP report [52]. For simulation results specific to the Nordic
electricity market, such as electricity production, investments, electricity prices, crossborder transactions, carbon dioxide emitted from power industry, etc., [39] is a good
reference.
A. Energy-systems modelling
Energy-systems modelling deals with models on energy issues. It may cover the
entire energy system including transports and heating systems, or just a subset of the
energy system, notably the electrical system. The key function of modelling is the ability
to transform complex reality into simpler and yet representative enough model that is
suitable to analyse and able to predict, here matters in relation to the energy issues.
Energy-systems modelling can differ in a few ways. In terms of mathematical
formulation, models can be descriptive (simulating models) or normative (optimization
models). They can also be classified into bottom-up models and top-down models. For
bottom-up electrical system models, they are mostly technology-oriented and treat
demand forecast as exogenously given. Energy demand is supplied by various generation
technologies, and technological change takes place through phasing out of existing
technologies by new technologies according to cost performance. Effectively bottom-up
electrical system models belong to optimization problem. If the energy demand is a
function of other parameters, say electricity price, then the model may be regarded as
partial-equilibrium model. The original electrical system model becomes part of the
macro-economy only and the relationship between its energy demand and other economic
variables is governed by elasticity of substitution. It then becomes the so-called topbottom model, in which it endogenizes the macroeconomic development through changes
on parameters of the energy system. Naturally, top-down models have little technological
explicitness compared to bottom-up models.
Energy system models can also be grouped by modeling approaches in which two
main types are techno/engineering-economic model and general/partial equilibrium

41

model. Concise descriptions and comparisons of the two are included in [37]. Here I only
make a little supplement. In power system literature, generation expansion planning
problems of various complexities and depths are solved by optimization or linear
programming methods. They belong to the type of techno/engineering-economic model.
For energy system models as part of the macro economy, they could be dynamic with
time and belong to general/partial equilibrium model. Using equilibrium models to
describe energy systems generally contains less technological detail.
In NEP model toolbox, the MARKAL-NORDIC is a good example of
engineering-economic optimization model, whereas most others are equilibrium models
on the broader economy in which the electrical system is only part of it.

B. Synchronization of model assumptions and input parameters

Since the NEP project consists of a number of models of different modelling
approaches, it is necessary to run the models based on a common ground of assumptions.
In particular, assumptions to be synchronized include scenarios and selected input
parameters. Regarding the formation of some main scenarios, they are constructed upon
the EU energy and climate package as follow.
1. Reference scenario, only 20% reduction in carbon dioxide emission
2. EU policy scenario, 20% reduction in carbon dioxide emission as well as 20% energy
from renewable
3. Global policy scenario, 30% reduction in carbon dioxide emission as well as 20%
energy from renewable
The needs of synchronization of input parameters are case by case. On one hand,
if, say electricity price projections in all models are the same, then the differences in
model output results should be irrelevant from electricity price. On the other hand, if
model designs and formulations are largely different, then model results may probably
diverge due to structural differences among models even the input parameters are the
same in values. Input parameters synchronized among NEP models are, for example, fuel
prices, electricity price, renewable electricity generation, emission allowance price, etc.
In particular, the EUA prices for various scenarios are included in Table 2.3.

42

Year
EUA

price

Reference scenario

EU policy scenario

Global policy scenario

2020

2030

2020

2030

2020

2030

30

30

20

40

60

40

(euro $/ ton)

Table 2.3 EUA prices for various scenarios in Nordic Energy Perspectives

As long as input parameters are fixed, it implies they are exogenous variables, i.e.
determined ex ante. However, an input parameter to one model may be the output
variable of another. For instance, the EUA price may be fixed for a scenario as input
parameter to one model, in turn the model predict emissions reduction in the future. To
another model, EUA price can be endogenously determined from any emission reduction
target. So for the same variable, attention has to be paid to when it is a parameter and
when it is an output result.

C. Functionality of NEP model toolbox

It has been mentioned that models of the NEP model toolbox are built upon
different modelling methodologies, with a spectrum of input and output parameters. Input
data for one model may be the output result being looked for in another model. In the
analysis of NEP, models could be executed serially, i.e. output of one model serves as
input to another. A model result makes sense only if its scenario assumption and other
input data are specified. The purpose here is not to repeat the model results, rather, model
functionalities are identified. The attributes of selected but anonymous models (or group
of models) are summarized in Table 2.4, in which group means a group of similar
models, model output is the same as prediction, whereas model input refers to value
assumptions of main parameters.

43

Group

Model output

Model input

Remarks

Electricity

Fuel price, CO2 price

Generation

generation of each

capacity

additions

may

be

exogenous or endogenous

technology
2

Wholesale electricity

Load demand

The system price could depend upon time

price

resolution of models, short or long run margin

costs, any surplus in market, endogenous or
exogenous investments, etc.

CO2 emissions

CO2 price

CO2 emission reductions differ according to

sectors concerned

Electricity demand

Price

of

different

Demand is endogenous in equilibrium models,

GDP,

exogenous in optimization problem, could be

energies,

CO2 price

temperature

differentiated by industries and nations

Macroeconomic

CO2 price is determined from economic

variables,

equilibrium models. It is also affected by

political

targets

national commitments and availability of

carbon capture and storage

Table 2.4 Summary of attributes of groups of NEP models

2.4.5 Concluding remark for investment modeling

The NEP project has already contributed a number of models for energy systems.
Those models are reviewed briefly in terms of their modelling methodology rather than
specific results. Results of different models may not be comparable because the structural
designs of models are different. Nevertheless, the purpose of describing NEP modelling
methodology is to provide a contrast to the proposed real option investment model to be
presented in Chapter 4. The contrast is a list of expected similarities and differences
between the two modelling approaches. The very common attribute is that both models
are constructed under the theme of renewable energy target. But for the fact that the
research work of this thesis would become useless if the proposed model were very
similar to the NEP ones, there are actually a number of differences.
1.

NEP models are descriptive or normative; mine is for investment analysis.

2.

NEP model results serve many stakeholders, especially the government; ours is for
the investor.

44

3.

The NEP project models parameters on a macro scale, e.g. national energy mix and
emissions; ours is focused on individual participant is influenced by market
parameters.

4.

For NEP models working on price predictions, they are mostly equilibrium model
and less technically oriented; our model tries to capture power system constraints as
much as possible.

5.

NEP models preserve the use of exogenously assumed financial parameters of price
and discount rate, our treatment to price (its rate of return and volatility) and
discount rate is consistent to finance theories.

2.5

Wind Power in Spot Market

Wind power investment analysis in spot market depends heavily on how strict the

market rules and regulations are set. As wholesale competitive market is primarily
designed for and operated by conventional generators, it may not be a level playing field
for wind power generators, let alone the actual price level. Obviously, penalty of
scheduling deviations is a big hurdle for the inherently intermittent wind power output.
So in early time, wind power generators did not participate in spot market even they were
not banned [49].
Investment evaluation of wind power project is specific to market environment.
Rules and regulations not only vary market by market, but are also being updated
periodically. Some rules have been amended favorable to wind power generators in spot
market, but instead of listing all latest developments, which are better referenced from
respective market authorities, here only the key aspects are identified and summarized as
follow.

No penalty on scheduling deviations, or allow partial imbalance

Provision of very liquid real-time market

Partial capacity payment according to wind power capacity factor

Relax the penalty to be charged on net deviation over a month

45

Different markets have different paces to increasingly accommodate wind power.

Collectively from [40][41], for example, the PJM runs without penalty on scheduling
deviations and offers active real-time market. The NYISO has been working towards
partial capacity payments and only partial imbalance penalty to wind power generators.
The CAISO has been actively designing the monthly scheduling deviation scheme in
association with compulsory hour-ahead wind forecasts [53][54].

2.6

Auction and Tendering

The last category of wind project investment may be grouped as auction and

tendering. They are similar to feed-in tariff for the feature of priority dispatch, but most
importantly differ in the pricing mechanism. Feed-in tariffs normally have premiums,
whereas auction or tendering starts from the lowest bid. Furthermore, auction and
tendering are slightly different with each other. Usually for wind power auction,
government set a targeted capacity of wind power and a price cap for investors to bid.
The bids starting from the lowest price up to the last megawatt needed win the auction at
their submitted prices. Brazil is an example of running wind power auctions [43]. On the
other hand, China employs tendering for its wind power development. China releases
wind power project case by case. The single investor of the lowest bid of tariff wins the
project. However, extremely low or intentionally losing bids are reported, especially from
public power companies, which may discourage investments from others [35].

2.7

Summary
To summarize, this chapter provide four market scenarios for wind power

investment: feed-in tariff, obligation driven, spot market, auction and tendering. The first
two will be formulated as investment models in this thesis. It is important to recognize
that wind power investment analysis only makes sense when the scenario and the
evaluation method are matched.

46

2.8

References

[35]

Behind the Chilly Air: Impacts of China's New Wind Pricing Regulation.

[Online]. Available: http://www.worldwatch.org/node/3904

[36]

California

Public

Utility

Commission.

[Online].

Available:

http://www.cpuc.ca.gov/PUC/energy/Renewables/
[37]

Coordinated use of Energy system models in energy and climate policy analysis.

[Online]. Available: http://www.nordicenergyperspectives.org/modellrapport.pdf

[38]

Fotios E. Karagiannis, Wind Energy Investments in Greece an Economical

Approach, in Proc. The 10th Mediterranean Electrotechnical Conference, MEleCon

2000, Vol.3, pp. 1141-1144.
[39]

Insights

from

NEP

policy

scenario

simulations.

[Online].

Available:

http://nordicenergyperspectives.org/Insights_from_the_NEP.pdf
[40]

L. Bird, B. Parsons, T. Gagliano, M. Brown, R. Wiser and M. Bolinger (July

2003). Policies and Market Factors Driving Wind Power Development in the United
States. National Renewable Energy Laboratory, Golden, Colorado. [Online].
Available: http://eetd.lbl.gov/ea/emp/reports/53554.pdf
[41]

L. Bird, B. Parsons, T. Gagliano, M. Brown, R. Wiser and M. Bolinger, Policies

and Market Factors Driving Wind Power Development in the United States, Energy
Policy, vol. 33, issue 11, pp. 1397-1407, 2005.
[42]

M. Mendonca, Feed-in Tariffs, Accelerating the Development of Renewable

Energy, London: Earthscan, 2007.

[43]

Price falls in second Brazilian renewables auction. [Online]. Available:

http://www.windpowermonthly.com/news/1024555/Price-falls-second-Brazilianrenewables-auction/
[44]

Public Benefits Funds for Renewables, Database of State Incentives for

Renewable

and

Efficiency.

[Online].

Available:

http://www.dsireusa.org/incentives/incentive.cfm?Incentive_Code=CA05R&re=1&ee
=1

47

[45]

R. Wiser and O. Langniss (Nov 2001). The Renewable Portfolio Standard in

Texas: An Early Assessment. Lawrence Berkley National Laboratory. [Online].

Available: http://eetd.lbl.gov/ea/ems/reports/49107.pdf
[46]

Renewable Electricity Production Tax Credit, Database of State Incentives for

Renewable

and

Efficiency.

[Online].

Available:

http://www.dsireusa.org/incentives/incentive.cfm?Incentive_Code=US13F
[47]

Renewable

Energy

Sources

Act

Germany

2007.

[Online].

Available:

http://www.gtai.com/uploads/media/EEG_Brochure_01.pdf
[48]

State of Texas Renewable Energy Credit Trading Program. [Online]. Available:

http://www.ercot.com/content/mktrules/protocols/current/14-080109.doc
[49]

Steven M. Wiese and Terry Allison (Oct 2000). Wind Power in Californias

Restructured Electric Market: Wind and the California ISO. National Wind
Coordinating

Committee.

[Online].

Available:

http://www.nationalwind.org/assets/publications/Wind_Power_in_CA_s_Restructure
d_Electric_Market_2000-10.pdf
[50]

Tarjei Kristiansen, Richard Wolbers, Tom Eikmans and Frank Reffel Carbon

Risk Management, in Proc. 9th International Conference on Probabilistic Methods

Applied to Power Systems, KTH, Stockholm, Sweden, 11-15 June 2006.
[51]

Ten Opportunities and Challenges for Nordic energy. [Online]. Available:

http://www.nordicenergyperspectives.org/Sartryck.pdf
[52]

Towards

Sustainable

Nordic

Energy

System.

[Online].

Available:

http://www.nordicenergyperspectives.org/huvudrapport.pdf
[53]

Yuri Makarov, David Hawkins, Eric Leuze and Jennie Vidov. California ISO

Wind Generation Forecasting Service Design and Experience. California Independent

System

Operator

Corporation.

[Online].

Available:

http://www.repartners.org/pdf/CAISOWdForecastModel.pdf
[54]

Yuri Makarov, et al. (Apr 2005). Incorporation of Wind Power Resources into the

California

Energy

Market.

[Online].

http://www.caiso.com/docs/2005/04/05/2005040508370111356.pdf

48

Available:

Chapter 3

3 Probabilistic Wind Power Generation Model

Abstract
Wind power output is always uncertain but, in a sufficiently long time interval,
the output exhibits statistical behavior that is meaningful enough to be characterized by
probability distribution. The aim of this chapter is to develop a model for probabilistic
wind power generation. In particular, I successfully derive the analytical expression and
statistics up to the fourth order of the wind power density function. The work also
extends the modeling of wind power output up to a regional scale by Gram-Charlier
series. Model results are checked by empirical power data and Monte Carlo simulation.
This paper discusses some applications of the wind power statistics such as probabilistic
production costing and reliability evaluation in power system literature.

3.1

Introduction
No matter what scenario of wind power investment is about, the very first

attribute needs to be investigated is amount of wind energy able to be produced. Since

wind speed is uncertain, the only way to quantify random wind power is by statistical
characterization of its long-term behaviors. Output power of a wind turbine depends,
among others, on wind speed magnitude and its power curve. Wind speed is recognized
to follow Weibull distribution. If a special case of Weilbull, the Rayleigh distribution and
a linear power curve are assumed, analytical probability distribution of the wind turbine
output power can be obtained. Based on the probability distribution, analytical formulae
of cumulant up to the fourth order are derived. The statistics model and predict wind
power over a period of time. The formulae are verified by Monte Carlo simulation and

49

also compared with empirical results converted by real wind speed data. This work is not
on time series forecasting, therefore having no chronological information as wind speed
data are collapsed into probability density function.
The attractiveness of the wind power generation model is the analytical formulae
of any higher order cumulants. They allow wind power to be evaluated in production
costing and reliability in an efficient manner. The topic is briefly discussed in the rest of
this chapter.

3.2

Wind Speed Distribution

Several distributions other than Gaussian had been suggested early as distribution

for wind speed [57]. Of these distributions the Weibull has generally been recognized and
is employed as the wind speed model in engineering discipline [63][73]. It is also not
difficult to find other research papers producing similar Weibull results. The question is
how fit the distribution could maintain when the Weibull is restricted to the Rayleigh.
Weibull distribution is governed by two parameters: the scale and the shape parameter.
Rayleigh is a special case of Weibull in such a way that the Weibull shape parameter is
fixed at 2 [65][78]. As far as the literature review done in this paper, it reveals in most
circumstances the shape parameter so determined by Weibull fitting is within reasonable
range around 2 [63][73]. Furthermore, higher annual average wind speeds (>4.5 m/s) tend
to have shape parameters close to 2 [88]. Therefore it gives a sense of the appropriateness
of Rayleigh wind speed model.
The probability density function (PDF) of Weibull distribution is
f w~ (w) =

k w k 1 ( w )
( ) e

(3.1)

w0

where k is the shape parameter and is the scale parameter. It becomes the Rayleigh
~

distribution when k equals 2. Thus the Rayleigh density function of wind speed w is
f w~ (w) =

2w

w 2
( )

w0

And its cumulative distribution function (CDF) is

50

(3.2)

P{w w} =
0

2w

w
( )

dw = 1 e ( )

(3.3)

The mean of a Rayleigh random variable is also stated:

_

w=

(3.4)

The elevation at which wind speed is measured may be different from the actual
turbine hub, hence some sorts of interpolation are required. There are two common
mathematical models for quantifying vertical profile of wind speed, see descriptions in
[63]. One of them being used here is the so-called logarithmic law:
wh ln(

h
)
z0

(3.5)

where h is the height, z0 is the roughness length and wh is the wind speed at level h
concerned. Roughness length is defined as the height drops to where the mean wind
speed becomes zero. It is necessary to take hub height into consideration for adjusting
vertical profile of wind speed.

3.3

Wind Turbine

3.3.1 Ideal Power Curve

A wind turbine power curve tells how much output power is generated
corresponding to any wind speed. Wind speed domain can be partitioned into a few
ranges each with its own properties. Below the cut-in speed, the output power is briefly
zero. From cut-in speed to rated speed, the output power increases according to a number
of factors to be discussed later. From rated speed to cut-off speed, the output power is
made to level off. Beyond cut-off speed, wind turbine is forced to shut down and
therefore output power is again zero. It is roughly acceptable to assume an ideal power
curve with linear segments as shown in the Fig. 3.1. From the literature, linear power
curve frequently appeared [64][65] for the purpose of a generic model. There is also
attempt to approximate power curve in the cut-in-to-rated region by a general non-linear
relationship between wind speed and power [83].

51

Fig. 3.1 An ideal wind turbine power curve

An ideal power curve converting wind speed w to output power g, given gm as the
maximum or rated output power, is represented by the following expression:
0 w win or w wout
0
~

g = aw + b
win < w < w r
g
w r w < wout
m
gm
g win
where a =
and b = m
w r win
wr win

(3.6)

and win, wr and wout denotes cut-in, rated and cut-out wind speed respectively.
The obvious query to this linearized power curve would be how accurate it is,
particularly in the cut-in-to-rated region and rated-to-cut-off region. In reality the power
curve is not perfect with linear segments. While it is still reasonable to assign zero output
power below the cut-in speed, sub-rated and rated regions of the power curve depends on
aerodynamic principle, wind turbine generator type and blade control as briefly discussed
next. Capturing all these factors produce an empirical power curve which is usually
obtainable from the wind turbine manufacturer.

3.3.2 Aerodynamic principle

Aerodynamic principle states that the power in airflow follows a cubic
relationship with the free wind speed w:
Pair =

1
Aw3
2

52

(3.7)

where is the air density and A is the area swept by the rotor. Not all of this power is
extractable by a wind turbine as mechanical power, apart from the power loss during
conversion to electrical power. In fact if the turbine output power Pwt is measured and
divided by Pair, the ratio is referred as power coefficient CP:
CP =

Pwt
Pair

(3.8)

It is observed that no matter how the wind turbine is designed, there is a maximum value
for CP, known as the Betz Limit and CPMax equals 0.593. In other words one can never
extract more than 60% of the power in airflow. Meanwhile, the power coefficient is
further deteriorated according to the tip speed ratio, which is defined as
=

(3.9)

where R is the radius of the blade and is the angular speed of the rotor. To achieve the
maximum power coefficient, the tip speed ratio has to remain constant at a particular
value. It has implications: this only occurs at a particular wind speed for fixed speed wind
generator; for variable speed wind generator its rotational speed can be adjusted to track
the wind speed in order to operate the wind turbine at optimal tip speed ratio, hence
preserving maximum power coefficient [59].

3.3.3 Wind turbine generator type

Wind turbines can be classified according to the drives, either fixed speed or
variable speed. Fixed speed wind turbine consists of using induction (asynchronous)
generator, whereas variable speed wind turbine uses doubly-fed induction generator
(DFIG) or synchronous generator. Methods of control of output power are different for
the two types of turbine drives. The primary concern here is to explore how the two types
of drives affect the speed-to-power conversion capability of wind turbine at sub-rated
power. In other words, the emphasis is the cut-in-to-rated region of the power curve.
For asynchronous generator wind turbine, since the rotational speed is fixed, it has
only one occasion that the tip speed ratio is optimal to yield maximum power coefficient.
This moment will definitely be reached as the free wind speed increases, at where the
maximum power is extracted from the wind, but other than this moment the turbine

53

output power rises less than proportionately to the cube of free wind speed. Only if the
generator is DFIG or synchronous, then the rotational speed can be made controllable
such that at each level of wind speed, the rotational speed is chosen at the maximum tip
speed ratio and hence maximum power coefficient. Fig. 3.2 demonstrates the wind
turbine power curve in its cut-in-to-rated region for optimal tip speed ratio or maximum
power coefficient.

Fig. 3.2 Wind turbine characteristics for maximum power extraction (Courtesy of [59])

Then the power curve in the sub-rated speed region could also be predicted at the
maximum power extractable. Fig. 3.3 demonstrates the power curves of fixed speed wind
turbine and variable speed wind turbine in a general fashion.

Fig. 3.3 Comparison between power curves of fixed speed and variable speed wind
turbine generators (Courtesy of [65])

54

3.3.4 Power regulation

Power regulation of wind turbine is done by blade control. Of course, blade
control does not mean that the output power can exceed that of achievable from a given
wind speed level. It serves to reduce loads on the wind turbine at above rated speed. At
extremely high wind speed, yaw control may take place to turn the nacelle out of the
wind. Blade control takes two major forms: pitch control and stall control. Pitch control is
accomplished by rotating the blades about their longitudinal axis. Stall control simply
relies on the aerodynamic design of the blades such that they can stall at high wind
speeds. Both mechanisms are common to fixed-speed wind turbines whereas only pitch
control is available to variable-speed wind turbines in general. The goal of blade control
is to regulate output power at the maximum level even wind is beyond rated speed. Fig.
3.4 shows the power curves under pitch-controlled and stall-controlled blade. The power
curve is rather well defined for pitched-controlled wind turbines whereas there is a little
overshoot for stall control.

Fig. 3.4 Comparison between power curves of pitch control and stall control wind turbine
generators (Courtesy of [64])

3.3.5 Empirical power curve

55

A well-shaped power curve suggests the theoretical conversion capability of an

ideal wind turbine. Practically, the power curve is made up of empirical speed-to-power
data points provided by the manufacturer. The advantage of an empirical power curve is
that it has taken into account of all intermediate factors and most importantly, electrical
losses. But it is no longer a mathematically nice function.
Reference [72] gave a precise discussion on to obtain an empirical power curve.
The power curve is usually a set experimental of discrete values specified by the
manufacturer. These values are interpolated and linked by continuous piecewise-linear
segments. In view of the discussion about wind turbine generator type and blade control,
I also believe using empirical power curve is most suitable. An empirical power curve
basically maps any values of wind speed to wind power.

3.4

Wind Power Distribution

Based on the ideal power curve and Rayleigh distribution of wind speed, an exact

analytical expression of the probability distribution of wind turbine power was derived
[69]. Built upon this idea, I derive the distribution based on piecewise-linear power curve,
which is more accurate. An empirical power curve can be constructed from manufacturer
speed-power data so that piecewise-linear segments connect all the way up from zero to
cut-out wind speed. Here the focus is on the sub-rated region of a power curve as
represented by
0
a w+b
1
1
~
a2 w + b2
g=

an w + bn

g m

0 w win or w wout
win < w w1
w1 < w w 2

w n 1 < w < w r
w r w < wout

where

56

(3.10)

g w1 g 1win
g1 g 0
, b1 = 0

w1 win
w1 win
g g1
g w g w
a2 = 2
, b2 = 1 2 2 1
w 2 w1
w 2 w1

g g n 1
g w r g m wn 1
an = m
, bn = n 1
w r w n 1
w r w n 1
a1 =

(3.11)

and g0=0, g1, g2, , gn-1, gm correspond to the wind turbine output at win, w1, w2, , wn-1,
~

wr respectively. g is the random wind power, defined by constants ai and bi {i [0, n]} , gm
is the maximum net output, win, wr and wout are the cut-in, rated and cut-out wind speed.
Without loss of generality, consider breaking the ascending segment into only two
segments, with (w1, g1) a knee point freely chosen on the curve part. This is illustrated in
Fig. 3.5.

Fig. 3.5 Visual comparison between three-segment and four-segment power curve

The derivation of the cumulative distribution function (CDF) of the wind turbine
output power is included in the Appendix I. Here I write the corresponding probability
density function (PDF) f(x):

57


( win )2
( w out )2
(1 e + e ) ( x)
x=0

x b1 2
)

x b (
2( 2 21 )e a1

0 < x g1
a1

f ( x) =
x b2 2
g1 < x < g m
)

x b (
2( 2 22 )e a2

a2

w
r 2
( w out ) 2
( )

) ( x g m )
e
e
x = gm

(3.12)

where ( x) is the delta function at x=0 to maintain the derivative-integral relation

between the CDF and PDF. Furthermore, I have derived recursive formulae for the first
four order moments or cumulants of the wind power PDF and will state the results in the
next part.

3.4.1 Analytical Formulae of Wind Power Statistics

Knowing that moments and central moments can be deduced interchangeably, it is
more convenient to start with the mean m1 and then all higher order moments can be
generated. All the derivations are kept in Appendix, only final results are shown next.
Mean

x = m1 =

x f ( x)dx =

2
2
a1

{erf (l2 ) erf (l1 )} + g0 e l 1 g1e l 2

2

2
2
( r )2
(
a

+ 2
{erf (k2 ) erf (k1 )} + g1e k 1 g m e k 2 + gm (e e
2

wout

(3.13)
)2

Variance
~

The variance of x , or the second order central moment c2, is given by

(3.14)

2 = c 2 = m2 m12

where
2

m2 = {g 12 + (a2 ) 2 }e k 1 {g 2m + ( a2 ) 2 }e k 2 + a2b2 {erf ( k2 ) erf ( k1 )}

wr

( )
(

+ {g 02 + (a1 ) 2 }e l 1 {g 12 + ( a1 ) 2 }e l 2 + a1b1 {erf (l2 ) erf (l1 )} + g m2 (e e

2

Skewness and Kurtosis

58

wout

)2

(3.15)
)

Third cumulant:
3 = c3

(3.16)

= m3 3m2 m1 + 2m13

where
2
2
a 2 2
3
3

m3 = {g 13 + a22 2 ( g1 + b2 )}e k 1 {g 3m + a22 2 ( g m + b2 )}e k 2 + 3a2 ( 2 + b22 )

{erf (k2 ) erf ( k1 )}
2
2
2
2
2
2
a2 2
3
3

+ {g 30 + a12 2 ( g0 + b1 )}e l 1 {g 13 + a12 2 ( g1 + b1 )}e l 2 + 3a1 ( 1

+ b12 )
{erf (l2 ) erf (l1 )}
2
2
2
2

+ g m3 (e

( wr )2

( wout ) 2

(3.17)
Fourth cumulant:
4 = c4 3c22

(3.18)

= (m4 4m3 m1 + 6m2 m12 3m14 ) 3(m2 m12 ) 2

where
2

m4 = {g14 + 2( a2 ) 2 ( g12 + b2 g1 + a22 2 + b22 )}e k 1 {g m4 + 2( a2 ) 2 ( g m2 + b2 g m + a22 2 + b22 )}e k 2 + (3a23b2 3 + 2a2 b23 ) {erf (k2 ) erf (k1 )}
2

+ {g 04 + 2(a1 ) 2 ( g 02 + b1 g0 + a12 2 + b12 )}e l 1 {g14 + 2(a1 )2 ( g12 + b1 g1 + a12 2 + b12 )}e l 2 + (3a13b1 3 + 2a1b13 ) {erf (l2 ) erf (l1 )}

+ g m4 (e

( wr )2

( wout ) 2

(3.19)
By definition, we can also find skewness and kurtosis excess respectively as follow.
1 =
2 =

3.5

c3

(3.20)

3
c4

(3.21)

Wake Effect and Wind Direction

When many wind turbines are assembled in an array, downstream wind speed is

reduced by upstream wind turbines. This is called wake effect. Hence the total power of
wind farm is less than the proportional scaling of single wind turbine power. To account
for the speed deficit, a wake effect model consolidated in [72] is applied. Let wx denote
the reduced wind speed x meter behind the upstream turbine, w0 denote the undisturbed
wind speed, then

59

wx = w0 [1 (1 1 CT )(

R 2
) ],
R + kx

(3.22)

where R is the upstream turbine rotor radius, k is the so-called wake decay constant and
CT is the thrust coefficient. Parameters k and CT depends on wind farm terrain and wind
turbine type respectively. Downwind distance x is normally 12R to 24R.
Regarding the effect of wind direction, the Dutch wind power software (WAsP)
[91] reports that the improvements are small (up to 5%) for locations with distinct wind
directions, but are negligible for most cases even when wind direction is considered in
calculations. For simplicity and investment purpose wind direction is ignored.

3.6

Evaluating Production Cost and Reliability with Wind

Power
Electrical energy is increasingly produced from renewable sources worldwide for

a number of reasons such as environmental concern, elevating fossil fuel prices, political
motive, etc. Wind power is one of the most popular forms of renewable energy. Yet wind
power is intermittent and virtually non-controllable. Its large-scale deployment would
influence power system in unprecedented ways. High penetration wind power poses a
need of refinement to existing methodologies on production costing and reliability
evaluation, which is the aim of this section.
In power system literature, research on renewable energy could actually be dated
back to late 70s fostered by the oil embargo and price spike. Renewable facilities
concerned were primarily wind turbine and solar photovoltaics. They were called
variously as non-dispatchable, dependent generating, time dependent or unconventional
sources. The names share the same meaning but are all referred as dependent generating
sources for coherence here. This work focuses on only wind power, although the
proposed method also applies to photovoltaics.
Theoretically, reliability evaluation with dependent generating sources could take
the same approach of loss-of-load as conventional generators. The main step is to
represent the dependent generating source as a multi-state (capacity) unit. Early attempt

60

did not consider failure and repair characteristics of wind turbine [80]. It was improved to
incorporate wind turbine availability by discrete convolution [79][93]. In [64], wind
speed is modeled by Markov chain, then power curve and availability of individual wind
turbines are all taken into account to form the probability distribution of wind farm output
capacity. The capacity table is used to modify system load distribution by discrete
method and reliability is evaluated as usual [81][82]. Forced outage rate handled by
cumulant method is illustrated in reliability context [60] and that work was extended to
consider load dependence by clustering [61]. Separately, cumulant method in the context
of probabilistic production costing is found in [70][76].
It should be consensus that wind turbine availability has been properly addressed
in course of the abovementioned historical development. But it appears none of the
referenced work has brought into a wake effect model of wind speed successfully.
Furthermore, all previous works tended to be confined by modeling only one single wind
farm, but are not readily generalized to wind output from a large region for production
costing and reliability evaluation in a national perspective. A more recent work has
attempted to produce a regional wind power probability distribution by discrete
convolution [65]. Our approach is conceptually very simple. If wind power is treated as a
random variable, then any total regional wind power is obtained by recursively adding all
individual output distributions. There is no need to distinguish wind turbine out of a wind
farm in the summation process for the reason that, given any downwind turbine speed
reduced by wake effect, the waked speed profile could be modeled by a new, separate
distribution. Correlations due to intra-farm and inter-farm proximity are all efficiently
rendered by the correlated cumulant method [70], which is the blueprint of this work. It is
critically extended by analytical expressions of any higher order statistics of wind turbine
power output derived here. The resultant regional wind power distribution is anticipated
to be normal-shaped, suggested by the Central Limit Theorem.

3.7

Data Source

61

The demonstrative examples will be based on two sources of hourly wind data.
The first one is a multi-decade, speed-only database of tens of Dutch locations [84]. The
second one is less extensive, but with corresponding output from a real wind turbine [87],
which is most valuable. Brief descriptions of them are as follow.

3.7.1 Royal Netherlands Meteorological Institute

There are more than 50 measuring stations, onshore or offshore, each has a name
and ID. Every station measures its hourly wind speed, with figures of measuring height
and roughness length. Data entry may be void (missing) or negative (faulty). Out of all 31
stations are chosen that are most complete in the past 19 years (1991-2009). Void and
negative data entries are assigned a very small positive value. All 29th Feb in leap year
are ignored for easy programming. Effectively, each location has 8,760 x 19 equals
166,440 hour wind speed.

3.7.2 Vermont Small-scale Wind Energy Demonstration Program

In this dataset, wind speed and corresponding output of medium scale wind
turbine (nominal 10kW) [85] are available from a few sites. Data are recorded at 10-min
intervals. Within each hour, average wind speed and total energy production are known.
Maximum and minimum wind speed and output power per hour are also given. It
provides better insight because the average wind speed of an hour is weak to show any
intra-hour highs and lows. Most importantly, those speed-power pairs can be used
determine empirically the wind turbine power curve.

3.8

Data Pre-processing

3.8.1 Wind speed measuring height

62

Normally, wind speed is measured at a convenient height above ground. It has to

be interpolated to wind turbine hub level for power conversion. In general wind profile
follows a logarithmic relation with its height above ground [74]:
wh ln(

h
)
z0

(3.23)

where h is the elevation, wh is the wind speed at the elevation concerned and z0 is socalled the roughness length. Roughness length indicates surface friction and is defined as
the height drops to where the mean wind speed becomes zero. It is necessary to up-scale
measured wind speed to the hub level.

3.8.2 Wind speed partitions and the parameter lambda

Cut-in, rated and cut-out speeds of a wind turbine are designed to suit the
prospective wind regime where it is located. The rule of thumb of design is based on the
_

annual average wind speed w onsite: win equals 0.6 w ; wr equals 1.5-1.75 w and wout
_

equals 3 w [59]. Next, value of the scale parameter is needed for generating Rayleigh
random variates. Such value can be estimated from a set of historical wind speed data.
Alternatively, if the long-term average wind speed is known, the lambda could be
determined passively by the following equation
_

w=

(3.24)

which relates and the Rayleigh mean.

3.8.3 Empirical power curve

Power curve measures the conversion ability of a wind turbine. A simplified
power curve consists of linear segments is already described. In general the ramp
segment should rise nonlinearly because aerodynamic principle states that the power in
airflow follows a cubic relationship with free wind speed. Meanwhile, its actual
performance depends on generator type (fixed or variable speed) and control method

63

(pitch or stall controlled). In practice, power curve data are empirical and given by wind
turbine manufacturer. It consists of power values (and losses) across a range of operating
wind speed, e.g. in pp. 68 of [63], with rated capacity and net output of 3.2MW and 3MW
respectively. An empirical power curve is not smooth and may not have a proper
mathematical form.
Even wind speeds at different hours are the same, the energies produced could be
different. Arguably, the speed is only average of an hour and intra-hour fluctuations could
be significantly different. Wind turbine performance varies with wind gust, turbulence
and wind directions to different extents. It is reported in commercial software WAsP that
wind speed within a particular directional sector does not normally show Weibull
distribution, fortunately, improvement made to the estimation of power output by
considering directional effect is little in most cases.
The approach here is to run regression on a number of pairs of real speed-power
data to determine the empirical power curve [62]. Theoretically, any long run site
characteristic, e.g. terrain, wind direction etc., is included in the power curve so
determined. The regressed power curve is like an average performance measure. Fig. 3.6
shows a trial regression, which is divided into two parts: the ramp and the plateau. Their
underlying functions are believed to be cubic and constant function of speed respectively;
this subjective process is necessary for regression analysis. Then the fitted curves are
simply obtained by linear regression function built in Matlab.

Fig. 3.6 Empirical power curve determined by regressing real data

64

Specifications of the actual wind turbine are found in [85]. Data used (Mar 07 Feb 08)
are taken from the Harvest Hill Farm of the Vermont Program.

3.8.4 Wake effect

Though waked wind speed is different from the undisturbed one in magnitude, it
still demonstrates a reasonable shaped Rayleigh distribution. Fig. 3.7 shows a sample
waked wind speed profile for wind turbine with typical CT value [90].

Fig. 3.7 Waked wind speed density function (b) compared with its original Rayleigh
source (a)

65

According to (3.22), wake effect depends on a number of parameters, which in

turn depend heavily on site configuration. It is impossible to come up with a universal
answer of how much wind speed is reduced due to wake effect for different wind farms.
Assuming a wind farm of individual 3MW wind turbines in an array shape, I try to
enumerate the effective average power of wind turbine against the downwind distance
and number of wind turbines per square side, and for different wind regimes. As shown in
Fig. 3.8, the effective average power of a wind turbine decreases with increasing number
of turbines per array side and shortening wind turbine separation. The plot can be used as
a shortcut to check typical result of wake effect.

Fig. 3.8 Effective average power of waked wind turbine

3.9

Simulated and Empirical Results

66

3.9.1 Historical wind speed analysis

Comparison approach

For fair and synchronized comparison, only locations with complete records in the
past 19 years (1991-2009) are chosen (still, faulty data may be included)

29th Feb in leap year is ignored. So every year has 8,760 hours, and the total
number of hours for each location is 8760 x 19 = 166,440

Rayleigh and Weibull parameters based on global data are run for in Table 3.1.

67

Station Station
ID
location

Occurrence of Occurrence of
Rayleigh Weibull Weibull
zero speeds
faulty
k

measurements
1254
0
8.61
8.31
1.73
210 Valkenburg
49
0
9.70
9.83
2.14
225 IJmuiden
363
22
9.38
9.36
1.97
235 De Kooy
685
0
8.46
8.30
1.84
240 Schiphol
1285
0
5.51
5.40
1.83
260 De Bilt
961
0
7.50
7.37
1.85
269 Lelystad
656
0
8.06
7.94
1.87
270 Leeuwarden
701
122
7.40
7.27
1.85
273 Marknesse
2380
0
6.91
6.61
1.65
275 Deelen
4326
1874
6.24
5.69
1.38
278 Heino
952
4929
6.97
6.43
1.42
279 Hoogeveen
983
0
7.47
7.30
1.81
280 Eelde
1669
0
6.43
6.21
1.73
283 Hupsel
283
1
8.12
8.13
2.01
286 Nieuw Beerta
3981
0
6.51
6.14
1.55
290 Twenthe
120
0
7.54
7.57
2.03
310 Vlissingen
285
5885
9.14
8.65
1.49
312 Oosterschelde
292
984
8.54
8.49
1.94
316 Schaar
204
13
9.26
9.41
2.18
320 L.E. Goeree
437
0
8.04
7.95
1.91
323 Wilhelminadorp
131
130
9.86
10.05
2.21
330 Hoek van
Holland
385
2124
7.99
7.84
1.81
331 Tholen
163
678
8.82
8.90
2.10
343 R'dam
Geulhaven
2032
0
7.94
7.54
1.62
344 Zestienhoven
740
171
7.45
7.24
1.79
348 Cabauw
1635
0
6.65
6.48
1.78
350 Gilze-Rijen
1267
0
7.22
6.98
1.75
356 Herwijnen
1763
0
6.72
6.51
1.74
370 Eindhoven
1795
0
6.65
6.37
1.68
375 Volkel
338
0
7.38
7.30
1.91
380 Beek
3372
0
6.08
5.74
1.57
391 Arcen

Table 3.1 Summary of distribution parameters of selected wind speed data

Result highlights

Wind speeds are all scaled to the same height for calculating wind power

Results are sorted in descending order of Weibull shape parameter k

Simulated mean power refers to the mean generated by a random variable of the
corresponding Rayleigh scale parameter

68

Empirical mean power is based on historical wind speed data

For each location, historical wind speed data are partitioned into years and then
statistics are run annually. For instance, Station 225 is selected and its results are
included in Table 3.2.

In Table 3.3 for each location, simulated average power and empirical average
power (global mean) are calculated and compared. Furthermore, the empirical
wind speeds are partitioned into years to compute annual averages. Standard
deviations of these averages about respective global means are calculated and
expressed in per unit of the global mean.

Apart from annual average power, it may also be meaningful to check the power
averaged on the same month along all years. Fig. 3.9 shows the mean powers
along specific months for the same selected wind sites.

The trend of a sample Dutch location is shown in Fig. 3.10. Intuitively, annual
averages are pretty much close to a global mean. Within each year, the power
shows large fluctuations (standard deviations).

Annual average wind powers of a number of Dutch locations are plotted in Fig.
3.11. A minor downtrend is observed over last two decades.

69

Empirical average
power (kW)

Year
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
1999
1998
1997
1996
1995
1994
1993
1992
1991

1090
1256
1145
1239
1304
1204
1140
1408
1360
1346
1295
1201
999
1156
1141
1216
1303
1405
1209

Standard deviation within

each year
Weibull k (partitioned)
1201
2.04
1217
2.26
1172
2.17
1218
2.14
1212
2.23
1218
2.13
1192
2.09
1236
2.17
1233
2.17
1232
2.16
1218
2.20
1217
2.02
1139
2.15
1228
2.01
1176
2.13
1201
2.16
1228
2.16
1232
2.25
1190
2.20

Table 3.2 Annual figures of the 3.2M wind turbine placed at Station 225 IJmuiden

70

Station
ID
Weibull k
330
2.21
320
2.18
225
2.14
343
2.10
310
2.03
286
2.01
235
1.97
316
1.94
380
1.91
323
1.91
270
1.87
273
1.85
269
1.85
240
1.84
260
1.83
280
1.81
331
1.81
348
1.79
350
1.78
356
1.75
370
1.74
283
1.73
210
1.73
375
1.68
275
1.65
344
1.62
391
1.57
290
1.55
312
1.49
279
1.42
278
1.38

Standard deviation of
Simulated Empirical
empirical annual
Deviation between
average
average power averages along all
simulated and
power
(global mean)
partitions (p.u.)
global means (%)
1314
1325
0.11
0.8%
1190
1201
0.07
0.9%
1281
1232
0.09
3.9%
1090
1060
0.06
2.8%
781
699
0.16
11.7%
925
838
0.08
10.4%
1216
1136
0.08
7.0%
1026
983
0.10
4.3%
740
659
0.12
12.3%
904
800
0.09
13.0%
909
827
0.09
10.0%
745
675
0.09
10.3%
771
692
0.08
11.4%
1006
893
0.08
12.7%
299
244
0.29
22.2%
762
678
0.09
12.5%
893
838
0.12
6.5%
757
653
0.09
16.0%
558
503
0.10
10.9%
699
611
0.10
14.4%
576
514
0.10
12.0%
504
445
0.12
13.2%
1042
941
0.11
10.7%
557
493
0.14
13.0%
623
572
0.09
8.9%
880
791
0.10
11.4%
423
377
0.19
12.1%
525
476
0.13
10.3%
1162
1157
0.18
0.4%
637
568
0.17
12.2%
459
408
0.14
12.4%

Table 3.3 Comparison between simulated and empirical average powers of the 3.2MW
wind turbine placed at various locations.

71

210 Valkenburg

1800

225 IJmuiden
235 De Kooy

1600

240 Schiphol
260 De Bilt
269 Lelystad

1400

Average power (kW)

270 Leeuwarden
273 Marknesse

1200

275 Deelen
278 Heino

1000

279 Hoogeveen
280 Eelde

800

283 Hupsel
286 Nieuw Beerta
290 Twenthe

600

310 Vlissingen
312 Oosterschelde

400

316 Schaar
320 L.E. Goeree

200

323 Wilhelminadorp
330 Hoek van Holland

331 Tholen
343 R'dam Geulhaven
344 Zestienhoven

ba

lm

ea
n

G
lo

348 Cabauw
Month of year

350 Gilze-Rijen
356 Herwijnen
370 Eindhoven

Fig. 3.9 Average powers along specific months of all years

72

Fig. 3.10 Annual average wind power, Station I.D. 210, Valkenburg.

Fig. 3.11 Annual average wind power, various Dutch locations.

73

3.9.2 Mean and standard deviation of annual average wind power

Arbitrarily, average wind powers of two sites are referenced from empirical data
and expressed in per unit of nominal capacity in Table 3.4. They are used to calculate the
average annual wind energy production of wind power project. Annual averages of
different years are inevitably fluctuating. In order to quantify the fluctuations, standard
deviation of annual average powers is also calculated. It is observed that deviation in sea
wind power is less than land wind power, in this occasional example.
Station

Type

ID

Average

Average power in p.u.

Standard deviation of annual

power (kW)

nominal

average power in p.u. nominal

capacity

of

1000kW

capacity (kW)

210

Land

941

294

32

320

Sea

1201

375

26

Table 3.4 Average wind power and its standard deviation

Next, I assert that the global mean power (estimated function value) is a
reasonably good fit of the annual averages (samples). The difference between the
estimated function value and the sample is called residual. In our case, residual is
calculated by subtracting the global mean from every annual average power of a Dutch
station. Residuals are then normalized by their respective global means so that the
residuals of all the Dutch stations can be grouped together. The goodness of fit of the
proposed model is checked by normality test of residuals and the result is shown in Fig.
3.12. The distribution of the residuals is reasonably closed to symmetrical around a mean
value at zero. Statistical estimations of parameters for annual and monthly resolutions are
shown in Table 3.5 and Table 3.6 respectively.

74

Fig. 3.12 Normality test for residuals (differences between global mean and annual
averages)
Mean

95% C.I. of mean

Standard deviation

95%

C.I.

of

deviation
0

-0.9756

12

0.9756

11.4037
12.7858

Table 3.5 Statistical properties of residuals

75

standard

Mean

95% C.I. of mean

Standard deviation

95%

C.I.

of

standard

deviation
0

-1.1938

51.2

50.3678

1.1938

52.0563

Table 3.6 Statistical properties of residuals (monthly basis)

3.9.3 Monte Carlo simulation for wind power statistics

The formulae of four wind power statistics are checked by Monte Carlo
simulation. Result variations due to power curve (three or four segments) and simulation
runs are demonstrated in Table 3.7. It can be seen that deviations between corresponding
analytic and simulated values for the case of 3-segment power curve is larger than those
of 4-segment power curve, indicating the latter is a more accurate model.
Rated power = 3000kW

mean

standard

skewness

kurtosis excess

deviation
3-segment
power curve

4-segment
power curve

Monte Carlo 10000 runs

1190.7

1131.9

0.4209

-1.3377

Analytic

1181.7

1127.7

0.4358

-1.316

Deviations (%)

1.86

1.07

5.73

3.26

Monte Carlo 2500 runs

1069

1117.8

0.6963

-1.0321

Monte Carlo 10000 runs

1041

1106.2

0.7297

-0.964

Analytic

1053.1

1103.4

0.7094

-0.9861

Deviations (%)

1.15

0.26

2.86

2.24

Table 3.7 Wind power statistics: analytic vs simulation

Computation is implemented in Matlab 7.0 on a common desktop computer with
2.66GHz Intel Core2 processor and 2GB RAM.
Wind power PDFs are also plotted for visual inspection. Fig. 3.13 shows the case
of 3-segment (simple) power curve, for which the PDF is smooth in the middle. Fig. 3.14
shows the case of 4-segment (improved) power curve, the PDF shows a jump.
Theoretically, the model accuracy increases as the power curve is partitioned more.

76

Fig. 3.13 Wind power PDF synthesized from simple power curve

Fig. 3.14 Wind power PDF synthesized from improved power curve

3.9.4 Comparison between analytical and empirical wind power PDF

The fact that analytical and simulated results are matched only means the
formulae are correctly derived, it does not guarantee the same for analytical and any
empirical data. Analytical model uses only parameter of wind speed distribution and
simplified power curve to calculate wind power. Empirical data are what actually be

77

measured from the wind turbine under the same speed distribution. Accuracy of the
analytical model lies on the fitness of distribution envelope and precision of power curve.
In Fig. 3.15, analytical and simulated wind power PDFs are presented as control (top), the
critical point is whether the analytical PDF models the empirical power data well
(bottom). In this case the true power data are again obtained from Harvest Hill Farm.

Fig. 3.15 Modeling empirical wind power by analytical PDF

The analytical model does not predict distributions year after, though. Apparently
there is no causal relation between inter-year wind distributions. The analytical and
empirical values of annual wind energy production as in Fig. 3.15 are 2769.4 kWh

( g 8760 ) and 2610 kWh respectively, close enough as they originate from a common set
of parameters. Actual production the year after is 3239.7 kWh. Higher order statistics of

78

future wind power distribution are expected to be even more irregular compared with the
current one. For practical purpose, only the annual average estimation may be utilized.

3.9.5 Regional wind power distribution

An extended numerical example of this chapter is to synthesize the probability
distribution of regional wind power output for production costing and reliability
evaluation in large scale. As worked out, individual wind turbine output is a random
variable characterized by probability distribution. Mathematically, any total wind power
is the convolution sum of wind turbine random outputs. But obviously it is too
overwhelming if thousands of wind turbines, each with numerous capacity states, are
convoluted to yield the resultant distribution. I seek to employ the Gram-Charlier series
to approximate the resultant wind power density function, but soon the substantial
correlation among wind speeds is realized and the independence assumption of GramCharlier approach is potentially violated. Fortunately, the correlated cumulant method [70]
is readily applicable, which is the blueprint of this section. The cumulant method requires
only one-off calculation of cumulants, and the value of analytical statistics derived earlier
comes in place. No matter how individual wind turbines are different in terms of
availabilities, ratings and wind profiles, etc., each wind turbine possesses a set of
characteristic cumulants for the summation process.
Previously, it is shown that the power density function of one wind turbine has
concentration at zero output. When many of such PDFs of different capacities are added
together, one would anticipate by Central Limit Theorem the resultant density function to
be bell-shaped. I verify this belief by enumeration of real data. The resultant wind power
density function can be obtained by two ways. On one hand, it is done by successive
convolution of individual outputs, which is rendered by Matlab subroutine. On the other
hand, the Gram-Charlier series gives an analytical approximation to the resultant wind
power PDF by processing the cumulants of all individual outputs. The two approaches
should match in outcome when sufficient quantities of wind turbine are grouped together.
However, extensive real wind power data could be difficult to obtain as an open source.

79

For research trial and illustration, I build a hypothetical scenario with reference to real
data as much as possible.
Denmark is used as the backdrop of our hypothetical scenario. Denmark has over
3,000 MW wind power installed capacity in 2008 [92], shared by slightly more than
5,000 wind turbines in 2009, with breakdown by ratings as grouped in Table 3.8 [90].
The ideal exposition of this numerical example would be to synthesize a power density
function of all wind turbines using empirical wind power data. As a compromise, wind
power data are converted by wind speed, and hourly wind speed records of tens of Dutch
locations for any single year are available. But obviously there are not enough wind speed
profiles compared with the number of wind turbines, and by no means these wind speed
profiles are matched with the actual wind turbine locations. Therefore we only conduct
test example.
Ratings (kW)

0-225

226-499

500-999

1,000+

Total

Numbers

1427

306

2596

749

5078

Table 3.8 Wind turbine breakdown by capacities in Denmark 2009

Specific formulations of the test example follow. Thirty one wind speed profiles
for a particular year from the Dutch database are extracted. Each profile of hourly wind
speed is tagged with a wind turbine of randomised rating to enumerate the respective
annual wind power distribution. All these power distributions exhibit randomness to
various degrees, and are mildly correlated due to wind speed correlation. It is interesting
and meaningful to check how closed to bell shape the resultant density function would be
when individual wind power PDFs are convoluted successively. To visualize the
improvements of large number, Fig. 3.16 plots the wind power PDFs generated by an
occasion of six successive convolutions stage by stage. It is demonstrated that the
resultant density function approaches bell shape when the number of sites convoluted
increases. The corresponding cumulative wind power capacity increases along the
horizontal axis.

80

Fig. 3.16 Successive convolution of individual wind turbine outputs

Apart from numerical convolution, the resultant wind power, in terms of
standardized density function, can be approximated by the correlated cumulant method.
Fig. 3.17 shows the analytical approximation of the resultant wind power PDF to the
same setting as by convolution. The envelope is laid on a histogram of simulation of a
standard normal random variable for visual checking. It could be seen that the analytical
PDF and the Monte Carlo simulation result are similar, but not yet matched for the fact
that only seven sets of cumulants are available to synthesize the curve. The result is
improved considerably when the number of wind turbines increases, say up to 31, as
shown in Fig. 3.18. Then the analytical envelope is very much alike a standard normal
PDF.

81

Fig. 3.17 Standardized PDF of correlated cumulant method of 7 variables

Fig. 3.18 Standardized PDF of correlated cumulant method of 31 variables

82

As an attempt I have compared resultant wind power PDF by numerical

convolution and by analytical approximation. Some possible reasons of mismatch are
aware of. Apparently, faulty wind speed data explain partially. The fact that only
cumulants up to the fourth order are used also leads to less precision.

3.10 Remarks
In this section, I would like to say something about the wind distribution. For
investment purpose, average wind power is used, which is derived from long-term
probability distribution of wind speed. Long-term probability distribution of wind speed
is commonly regarded as Weibull or Rayleigh and can be obtained by sufficient amount
of samples. Since wind speed evolution is time dependent, theoretically it should be
modeled by stochastic process. For stochastic process in which the state resident times
follow exponential distribution, it is said to be a Markov process. The corresponding state
probabilities are described by stochastic differential equation; in turn the limiting or
steady state probability equals the long-term average availability. A prominent work on
long-term average wind power or availability by Markov process is [64]. There, it is
known that wind speed state residence time is not necessarily exponentially distributed,
depending on sampling interval. But the long tern average value is true even the residence
time distribution is not exponential because actual resident time distribution only matters
when time dependent characteristic of the stochastic process is concerned. Hence,
Markov chain is still applicable, given that we are interested in the limiting state
probabilities only by using modified transition rates. Here, I apply probabilistic analysis
on wind speed samples to arrive to the long term average. Understandably, a proper
stochastic wind speed model is needed, if one is concerned with the conditional
probability of wind time series. Such model is expected to be necessary for works like
simulating wind power profit in spot market.

3.11 References

83

[55]

A. Stuart and J. K. Ord, Kendalls Advanced Theory of Statistics: Vol. 1.

Distribution Theory, sixth edition, London: Edward Arnold, 1994.

[56]

B. G. Gorenstin, N. M. Campodonico, J. P. Costa and M. V. F. Pereira, Power

System Expansion Planning under Uncertainty, IEEE Trans. on Power Systems, vol.
8, no. 1, pp. 129-136, Feb 1993.
[57]

Barbara G. Brown, Richard W. Katz and Allan H. Murphy, Time Series Models

to Simulate and Forecast Wind Speed and Wind Power, Journal of Climate and
Applied Meteorology, Vol. 23, pp. 1184-1195, May 1984.
[58]

Birger Mo, Jan Hegge and Ivar Wangensteen, Stochastic Generation Expansion

Planning by Means of Stochastic Programming, IEEE Trans. on Power Systems, vol.

6, no.2, pp. 662-668, May 1991.
[59]

Brendan Fox, Damian Flynn, Leslie Bryans, Nick Jenkins, David Miborrow,

Mark OMalley, Richard Watson, and Olimpo Anaya-Lara, Wind Power Integration,
Connection and System Operation Aspects, London: IET Power and Energy Series,
2007.
[60]

C. Singh and A. Lago-Conzalez, Reliability Modeling of Generation Systems

Including Unconventional Energy Sources, IEEE Trans. PAS., Vol. PAS-104, No. 5,
pp. 1049-1056, May 1985.
[61]

C. Singh and Y. Kim, An Efficient Technique for Reliability Analysis of Power

Systems Including Time Dependent Sources, IEEE Trans. Power Syst., Vol. 3, No.
3, pp. 1090-1096, Aug 1988.
[62]

Cameron W. Potter, Hugo A. Gil and Jim McCaa, Wind Power Data for Grid

Integration Studies, in Proc. 2007 IEEE PES General Meeting, Tampa, 24-28 June
2007.
[63]

David A. Spera, Wind Turbine Technology, Fundamental Concepts of Wind

Turbine Engineering, NY: ASME Press, 1994.

[64]

F. Castro Sayas and R. N. Allan, Generation Availability Assessment of Wind

Farms, IEE Proc.-Gener. Transm. Distrib., Vol. 143, No. 5, pp. 507-518, Sept. 1996.
[65]

Francois Valle, Jacques Lobry and Olivier Deblecker, System Reliability

Assessment Method for Wind Power Integration, IEEE Trans. Power System, Vol.
23, No. 3, pp. 1288-1297, Aug. 2008.

84

[66]

G. Papaefthymiou, P. H. Schavemaker, L. van der Sluis, W. L. Kling, D.

Kurowicka and R. M. Cooke, Integration of Stochastic Generation in Power

Systems, International Journal of Electrical Power & Energy Systems., Vol. 28, No.
9, pp. 655-667, 2006.
[67]

Henry M. K. Cheng, Yunhe Hou and Felix Wu, Probability Distribution of the

Output Power of Wind Turbine, in Proc. The 16th International Conference on

Electrical Engineering, Busan, Korea, 11-14 Jul, 2010.
[68]

Hugh Sharman, Why Wind Power Works for Denmark, ICE Proc.-Civil

Engineering, Vol. 158, May 2005, pp. 66-72.

[69]

Hui Zhou, Yunhe Hou, Yaowu Wu, Haiqiong Yi, Chengxiong Mao and Gonggui

Cheng, Analytical Assessment of Wind Power Generation Asset in Restructured

Electricity Industry, in Proc. the Universities Power Engineering Conference,
UPEC, 2007, pp. 1086-1092.
[70]

Jeremy A. Bloom, Probabilistic Production Costing with Dependent Generating

Sources, IEEE Trans. Power App. Syst., Vol. PAS-104, No. 8, pp. 2064-2071, Aug.
1985.
[71]

J. P. Stremel, R. T. Jenkins, R. A. Babb and W. D. Bayless, Production Costing

Using The Cumulant Method Of Representing The Equivalent Load Curve, IEEE
Trans. on PAS, vol. PAS-99, no. 5, pp.1947-1956, Sept/Oct 1980.
[72]

Julija Matevosyan and Lennart Soder, Minimization of Imbalance Cost Trading

Wind Power on the Short-Term Power Market, IEEE Trans. on Power Systems, vol.
21, no.3, pp. 1396-1404, Aug 2006.
[73]

L. L. Freris, Wind Energy Conversion Systems, NJ: Prentice-Hall, 1990.

[74]

Lars Landberg, A mathematical Look at a Physical Power Prediction Model,

Wind Energy, 1, pp. 23-28, 1998.

[75]

M. Mendonca, Feed-in Tariffs, Accelerating the Development of Renewable

Energy, London: Earthscan, 2007.

[76]

Michael C. Caramanis, Richard D. Tabors and Kumar S. Nochur, The

Introduction of Non-dispatchable Technologies as Decision Variables in Long-term

Generation Expansion Models, IEEE Trans. on PAS, vol. PAS-101, no. 8, pp.26582667, Aug 1982.

85

[77]

N. S. Rau, P. Toy and K. F. Schenk, Expected Energy Production Costs By The

Method Of Moments, IEEE Trans. on PAS, vol. PAS-99, no. 5, pp.1908-1917,

Sept/Oct 1980.
[78]

Nicola Barberis Negra, Ole Holmstrm, Birgitte Bak-Jensen and Poul Srensen,

Model of a Synthetic Wind Speed Times Generator, Wind Energy, 11, pp. 193-209,
Sept. 2007.
[79]

P. Giorsetto and K. F. Utsurogi, Development of a new procedure for reliability

modeling of wind turbine generators, IEEE Trans. PAS., Vol. PAS-102, No. 1, pp.
134-143, Jan 1983.
[80]

R. G. Deshmukh and R. Ramakumar, Reliability analysis of combined wind-

electric and conventional generation systems, Solar Energy, Vol. 28, No. 4, pp. 345352, 1982.
[81]

R. Billinton and L. C. H. Cheung, Load modification: a unified approach for

generatingcapacity reliability evaluation and production-cost modeling, IEE Proc.

C, Gen. Trans. & Distrib., Vol. 134, No. 4, pp. 273-280, Jul 1987.
[82]

R. Billinton and A. A. Chowdhury, Incorporation of wind energy conversion

systems in conventional generating capacity adequacy assessment, IEE Proc. C, Vol.

139, No. 1, pp. 47-56, Jan 1992.
[83]

Rajesh Karki, Po Hu and Roy Billinton, A Simplified Wind Power Generation

Model for Reliability Evaluation, IEEE Trans. on Energy Conversion, vol. 21, no.2,
pp. 533-540, Jun 2006.
[84]

Royal

Netherlands

Meteorological

Institute.

[Online].

Available:

http://www.knmi.nl/samenw/hydra/
[85]

Specifications of a

Bergey xls wind

turbine. . [Online]. Available:

http://www.windenergydirect.org/Windgenerators-bergy-xls10kw.php
[86]

Vestas. [Online]. Available: http://www.vestas.com/

[87]

Vermont Small-scale Wind Energy Demonstration Program. [Online]. Available:

http://www.vtwindprogram.org/
[88]

W. C. Cliff, The Effect of Generalized Wind Characteristics on Annual Power

Estimates from Wind Turbine Generators, PNL-2436, Richland, Washington:

Battelle Pacific Northwest Laboratory.

86

[89]

William Rutz, Martin Becker, Frank E. Wicks and Stephen Yerazunis,

Sequential Objective Linear Programming for Generation Planning, IEEE Trans. on

PAS, vol. PAS-98, no. 6, pp. 2015-2021, Nov/Dec 1979.
[90]

Wind Turbines in Denmark. Danish Energy Agency, Amaliegade, Copenhagen,

Nov 2009.
[91]

Wind

Atlas

Analysis

and

Application

Program.

[Online].

Available:

http://www.wasp.dk/
[92]

World Wind Energy Report 2008. World Wind Energy Association, Bonn,

Germany.

[Online].

Available:

http://www.wwindea.org/home/images/stories/worldwindenergyreport2008_s.pdf
[93]

X. Wang, H. Dai and R. J. Thomas, Reliability modeling of large wind farms and

electric utility interface system, IEEE Trans. PAS., Vol. PAS-103, No. 3, pp. 569575, Mar 1984.

87

Chapter 4

4 Fixed Tariff Wind Power Investment Model

Abstract
Viability of business project is typically measured by net present value (NPV) of
the projects total future cash flows. This chapter follows the usual practice in finance and
accounting to evaluate an investment of independent wind power project under fixed
tariff. Since debt financing generally increases firm value, I investigate the effect of
optimal capital structure to such wind power project by proposing a stochastic
optimization model to maximize the project NPV subject to constraint on project debt
permissible. Model considerations include, among other, debt type, wake effect, onshore
or offshore wind energy production characteristic, etc. Insights of the model are
illustrated by sensitivity analysis.

4.1

Introduction
Traditionally, monopolistic and regulated electric utility is a very stable business.

Its investment return is usually regulated to a fixed level. Revenue grows as a result of
tariff increment and natural load growth, in which tariff is mostly linked with inflation
and fuel costs. Basically utility earns a fixed return over its asset size by providing
electricity, and there is not much room for active management to exploit extra profit. It is
a kind of attractive business to conservative investor because its return is well predictable
and the risk is very low.
With electricity market restructuring, generation sector becomes competitive, only
the network business is left regulated. Meanwhile, electric utility is no longer confined by
geographical area because deregulation offers expansion opportunities for power holding

88

companies to acquire both generation and network assets all over the world. There are
various types of assets available, such as large conventional power plants, small
combined heat and power plants, distribution network, renewable generators, etc.
Regarding the topic of this chapter, it is logical to ask how attractive a fixed tariff wind
power project is. It is most likely a stable investment because the tariff is fixed for
prolong period of time and apparently, wind energy productions year on year are stable.
Empirical evidence on constant annual wind energy production is already discussed in
chapter 3. So apart from the tariff, the quantity of wind energy produced annually is also
assumed fixed or an average. Then the project valuation would be similar to an annuity of
profit cash flows. This chapter is going to evaluate such investment project.
There have been a number of policy and market factors for the historical
development of wind power in U.S. [96], notably production cost credit (PTC),
renewable energy production incentive, public benefits fund, etc. Some are already
expired; for those kept until now, the most up-to-date provisions can be found in [95].
Booming and sluggish periods of investment cycle in wind farms were observed, partly
due to the fact that the policy incentives are nominally short-term, with availability
subject to periodic renewal, and have uncertainty in provisions to be extended. A very
sustainable example would probably be Germany who has been pushing feed-in tariff
(FIT) for renewable. The key components of FIT are fixed rate per unit energy sold in
long-term, and priority access to the grid. For example, latest German FIT for wind is
basically 5.02/kWh for 20 years of operation [98]. The price is set by scientific studies
and the merit is to provide cost-based compensation to renewable investors. It is
recognized that such compensation scheme provides price certainty and long-term
coverage to renewable investors that are essential to project financing, which determines
the success of FIT.
Since debt interest is tax-deductible, debt financing not only clears the hurdle of
initial capital, but also boosts the value of project according to the M&M Proposition I on
capital structure [99]. Rational investor would then seek an optimal capital structure in
order to achieve the maximum, levered value for its business project because the levered
value radically represents the present value of all future cash flows obtainable from the
firm. However, over-borrowing would pose the firm to default if any periodic revenue in

89

future is not enough to repay debt instalment. For wind power investment, the annual
wind energy production year-on-year is the direct cause of randomness in annual revenue,
given the tariff is fixed. If the wind energy production is characterized by probability
distribution in annual horizon, which matches with the usual time resolution for
investment evaluation, then theoretically the objective of wind power investor could be
regarded as stochastic optimization (maximization) of the levered net present value (NPV)
of the project, subject to constraint on the initial debt level as decision variable. In other
words, I propose an optimization model to find the maximum debt permissible to yield
the most profitable wind power investment project under typical FIT provisions and
financing terms.

4.1.1 Scope
Analysis of any project investment has to pay attentions to two aspects. First, the
evaluation itself has to comply with finance theories. Second, the project technical
characteristics have to be addressed in financial fashion. Wind power project is no
different from others in a sense that its evaluation also needs to consider initial cost,
operating cost, future revenue, among others. Furthermore, its characteristics such as
energy production, tariff structure, dispatch priority, etc. theoretically shall be considered
as much as possible. Building a single investment model covering all technical and
regulatory aspects could be difficult, especially when the aspects are cross-disciplinary.
This section briefly explains the tradeoff between model complexity and practical
significance of various aspects.
1. Tariff structure
A survey of feed-in tariffs for renewable energy is [97]. In simple sense, fixed
tariff means that every unit of electricity generated is paid a single price forever. This
basic structure can be diversified into progression, degression, step-change or even other
tailor-made features. The theme in this chapter is to construct a wind power investment
model under a generic single fixed tariff.

90

2. Time resolution
Like most other generation projects, wind farm is a long-term investment of at
least fifteen years or more. If the time horizon is broken down into periods, normally the
resolution is taken as year, which implies annual figures of energy production, revenue
and cost to be used. Any fluctuations within a year are masked by the annual averages.
Annual averages for long-term investment are reasonable. In particular, it is apparently
true that wind speed at any parts of the globe would not increase or decrease substantially.
When sufficiently long wind speed time series are averaged, the mean tends to the global
mean. The minimum length of such time series to be representative has to be a year
because it covers one complete cycle of all seasons. Hence annual average energy
production suffices. Although annual energy productions of different years are not equal,
the global mean is arguably the best one can do for investment evaluation purpose.

3. Average energy production

As discussed in chapter 3, annual time resolution is probably most appropriate for
the representation of probabilistic wind energy production. Since annual figures are
inevitably fluctuating around a global mean, the global mean value is arguably the best
one could refer to. Fig. 3.11 suggests that annual average wind power along multi-year
horizon is reasonably flat.

4. Wake effect and scaling

Wind speed measurements in big site are usually partitioned for improving
accuracy. At best, proprietary simulating software such as software WAsP can be used to
simulate the dynamic energy production of a wind farm subject to turbine sitting,
complex terrain and local wind regime. That is definitely beneficial to operational needs,
but may not help too much on investment planning. Here for simplicity, wake effect and
scaling are handled in simple sense by referring to Fig. 3.8.

4.2

Accounting Preliminaries

91

Financial theories dealing with capital investment decisions are already well
established. The very basic, discounted cash flows analysis or net present value (NPV)
method is the golden rule. The fixed tariff wind power investment project will also be
mapped under the NPV framework. While financial theories may be very sophisticated,
the norm is to apply only essential elements that are adequate for representing
engineering ideas in an investment model.

4.2.1 Definition of cash flow

The very first question before invoking NPV method is how much a new project
or investment can earn. In practice, earnings are measured by cash flows. Furthermore,
only the change in cash flows brought by the project is relevant to the question. The
change in cash flows to be contributed by the project is referred as incremental cash flows,
in which NPV method is based on.
While incremental cash flows are the actual money created by the project, other
indirect consequences of adopting this project may also count. Among them are sunk
costs, opportunity costs, erosion and net working capital (NWC), see description in [99].
NWC is conveniently assumed as the difference between accounts receivable and
accounts payable. Practically, sunk cost, opportunity cost and erosion are difficult to be
quantified and thus they are not considered in a simple model. Finance cost is left in the
project cash flows because it does not affect the underlying profitability of the project. On
the other hand, any interest, dividend or principal paid has to be deducted from the cash
flows.
A project generates a stream of periodic cash flows. At each period, the project
cash flow is composed of three elements: operating cash flow (OCF), capital spending I
and change in NWC [99].
Project cash flow = OCF NWC I

(4.1)

In the following I will trim down the project cash flow expression for an essential
investment model. The simplification aligns with the usual financial practice and the
reality.

92

Net working capital, as stated previously, is accounts receivable minus accounts

payable. Change in NWC is the year on year fluctuation in NWC that would affect the
periodic project cash flow. However, normally most of the working capital injected
initially could be recovered at the project end. If it is assumed 100% recoverable, then
working capital (and hence any NWC) can safely be omitted without missing the
concept and accuracy.
Capital spending is taken only at the beginning. Any major upgrade of the project
that needs substantial capital spending is ignored.
Operating cash flow is merely revenue minus operating cost from the point of
view of project operation. If an accounting profit and loss table is given, OCF is
determined by its own definition in terms of earnings before interest and tax (EBIT),
depreciation (Dp) and corporate tax (Tc) by the following equation:
OCF = EBIT + Dp Tc

(4.2)

Depreciation is an accounting concept. No matter how the initial capital is depreciated,

the periodic depreciation amounts are added back to EBIT for arriving to the OCF. On
the other hand, tax should be deducted from EBIT. In fact, depending on what accounting
figures are given, OCF can be obtained by three approaches, namely Bottom-Up
approach, Top-Down approach and Tax Shield approach [99]. After all, these are just
accounting manipulations; financial (NPV) analysis is concerned with project cash flows
or equivalently OCF when NWC is ignored.

4.2.2 Net present value

NPV of a project is the present value of all future cash flows minus initial capital
spending. The cash flows concerned should be OCF. To allow for sufficient generality, a
growing annuity is taken as the format of cash flows, hence, the NPV looks like
NPV =

OCF
[1 e ( g )T ] I
g

(4.3)

where T is the life of the project in years, g is the growth rate of annuity and is the
project rate of return. Rate is generally called project discount rate. If NPV is positive
given the value of , it means the project can safely deliver a return of , therefore is

93

also called require return. On the other hand, a negative NPV at a given discount rate
does not really mean the project will incur a loss. It only means such a project cannot
deliver a return as high as required.
Note that NPV is not project valuation. Firm value could be denominated by the
market capitalization of a listed company. The stock price constitutes a dynamic
valuation of the firm or the present value of future cash flows of the firms underlying
business.

4.2.3 Other discount rates

1.

Internal rate of return

There are many arbitrary discount rates that yield positive NPV. The discount rate

that just breakevens an investment, i.e. making NPV zero, is more meaningful. Such
discount rate is called internal require rate of return (IRR).

2.

Cost of capital
The required return is an important concept in finance and deserves further

elaboration. Investors demand a return for whatever project. The size of return demanded
may be arbitrary but conceptually it should be just enough to compensate the investors
for financing the capital invested in that project. Hence the required return of a project
can also be interpreted as the cost of capital of such project. The concept of return is not
confined to only project but also applies to listed company. Payouts to shareholders are
cost of the stock of the issuing company. If a company cannot generate sufficient return
to meet its cost of capital, the consequence is a reduction in market value of its stock.
By now it is clear that rate of return, required return, discount rate and cost of
capital can all be used more or less interchangeably.

3.

Cost of equity
Cost of equity of a firm is difficult to judge because different investors have

different required returns on the firm. One way to estimate it is by observing the rate of
return of the company stock as collectively produced by a pool of investors in the market.

94

Without going into details, two approaches are named: the dividend growth model and
capital asset pricing model (CAPM) [99].

4.

Cost of debt
Cost of debt is relatively simple. The coupons paid to bondholders are expenses of

the company. Hence the cost of issuing bonds is the coupon rate when the bond is
initially issued at par value. However, bond price in the past does not reflect the
credibility of a company now. Therefore it is the market bond yield that tells the present
cost of debt of the company.

5.

Weighted average cost of capital (WACC)

If the market values of equity and debt are denoted as E and D respectively, then

the combined market value of equity and debt, V, is

V=E+D

(4.4)

V could be viewed as the asset value, whereas equity is what left behind after debt is
deducted from asset. Market values of equity and debt should be referred instead of the
book (accounting) values. Market value of equity can be obtained by multiplying share
price and the number of outstanding shares. Market value of debt is calculated by
multiplying bond price and the number of outstanding bonds. Having the return on equity
RE and return on debt RD estimated, it is straightforward to take the weighted average of
them and get the weighted average cost of capital (WACC) as follows.
WACC =

E
D
RE + RD
V
V

(4.5)

where E/V and D/V are often called capital structure weights. Practically corporate
earnings are always taxed. Since it is the after-tax cash flows that matter, the
corresponding discount rate has to be expressed on after-tax basis as well. In particular,
debt interest is tax deductible, so the cost of debt should be smaller in the presence of
corporate tax. Effectively government pays some of the interest expenses. If tC is the
corporate tax rate, then WACC is modified to
WACC =

E
D
RE + RD (1 tC )
V
V

95

(4.6)

The return of a firm has to compensate a mixture of returns of its debt-holders or

creditors and its shareholders. In other words, the WACC reflects the cost of asset (equity
plus debt) as a whole.

4.2.4 Capital structure

Capital structure affects firm value. As the manager of a firm, the objective is
always to maximize the firm value. As far as capital structure is concerned, optimal
allocation of equity and debt can maximize firm value. Previous discussion on NPV did
not specify where the initial capital comes from. It is important to check for any effect on
firm value if part of the investment capital is borrowed as debt. This is capital structure
matter and appears as the famous Modigliani-Miller (M&M) Propositions in finance
literature, which is extracted as follow.
A. M&M Propositions I & II without corporate taxes
It ought to start with the M&M Propositions without corporate taxes. M&M
Proposition I simply states that the value of a firm does not depend on its capital structure
in a world without corporate taxes. If an all-equity firm borrows some debts, its intrinsic
value does not increase. In accounting terms, only the firms asset value is boosted up but
not its equity. M&M Proposition II says the firms cost of equity is directly proportional
to the debt-equity ratio, but the WACC remains unchanged at the original level of cost of
asset RA even if debt-equity ratio increases. It then follows
WACC =

E
D
RE + RD = RA
V
V

RE = RA + ( RA RD )

D
E

(4.7)
(4.8)

Equation (4.8) gives a theoretical exposition of RE. As debt-equity ratio increases, so as

the risk of equity, then the cost of equity has to be raised accordingly. However, WACC
is flat, i.e. no reduction in WACC and hence no advantage by borrowing. In other words,
debts do not increase firm value.
I use the NPV equation to illustrate the concept of M&M Propositions II. When
the project is all equity-financed, it does not really matter whether the discount rate is RE

96

or WACC because there is no debt. Suppose half of the project initial capital is borrowed
at interest rate RD, the debt-equity ratio would then be one and RE is given by
RE = RA + ( RA RD )(1) = 2 RA RD

(4.9)

where RA is the cost of asset of the originally all equity-financed project. Whereas WACC
is shown to be unchanged at the level of RA as follows.
WACC = 0.5(2 RA RD ) + 0.5( RD ) = RA

(4.10)

The equation of WACC is true for all debt-equity ratios.

B. M&M Propositions I and II with corporate taxes

What matters to borrowing is corporate tax. The tax-version of M&M Proposition
I states that the value of a levered firm VL equals the unlevered firm value VU plus its tax
shield tCD [99], as represented by the following equation:
VL = VU + tC D

(4.11)

Implicitly, the setting is based on perpetual cash flow. Unlevered firm value is arrived at
discounting the OCF, OCF by the so-called unlevered cost of capital RU:
VU =

OCF
RU

(4.12)

Note that RU is an exogenous value. VU is the present value of future cash flows and
literally has no guarantee from the initial capital spent. M&M Proposition II with
corporate taxes modifies the equations of the WACC and RE as follow:
WACC =

E
D
RE +
RD (1 tC )
VL
VL

(4.13)

D
(1 tC )
E

(4.14)

RE = RU + ( RU RD )

The cost of equity RE still rises linearly with debt-equity ratio, indicating that as debt
increases, so as the risk of equity and so a higher return on equity is demanded. On the
other hand, debt can bring down the overall cost of capital and is therefore advantageous.
The characteristics exhibited by (4.13) and (4.14) lie on how M&M Proposition I with
corporate taxes defines firm value: Without debt, the firm value is VU, which is also the
market worth of the equity. After borrowing of an amount of debt D, the firm value
increases by tCD to become VL. The corresponding new market worth of equity

97

(4.15)

E = VL D

In other words, equity is calculated ex post. What (4.15) really says is not a reduction of
the worth of equity, it means a smaller amount of equity is enough for the same project.
To illustrate how WACC drops as debt raises, one can substitute (4.14) into (4.13)
to write WACC in terms of unlevered cost of capital RU. Details are given in Appendix
IIA. The result gives Corollary 1 of WACC:
WACC = RU (1

D
tC )
VL

(4.16)

Since discounting perpetual OCF by RU gives VU, it is logical to think that discounting the
same OCF by WACC should yield VL. This gives Corollary 2 of WACC and the
derivation is in Appendix IIB:
OCF
VL
WACC

(4.17)

The result agrees with the knowledge. WACC at any time is just a generalized form of
discount rate. If there is debt, discounting OCF by WACC gives VL. In the absence of
debt, WACC itself reduces to the unlevered cost of capital RU, consequently discounting
OCF by such WACC yields VU.
The last thing I would like to investigate is the relationship between equity, RE,
and a modified operating cash flow. We initially guess what equity holders should get is
only OCF net of debt interest iD, whereas debt interest of course goes to debt holders. But
since debt interest is tax-deductible, there is an extra term iDtC created and it goes to the
pocket of equity holders. So discounting this modified OCF by RE should give the equity
back. The steps are included in the Appendix IIC and the result gives Corollary 3:
OCF iD (1 tC )
E
RE

(4.18)

It matches very much with the intuition.

4.3

Model Formulation for FIT Wind Power Investment

Recall from the M&M Proposition I that a firm can increase its value by leverage

in the presence of corporate tax. The additional value is the tax shield. The Proposition

98

result is best illustrated in infinite time horizon as

(4.19)

VL = VU + tC D

where VL, VU, tC and D are the levered firm value, unlevered firm value, corporate tax
rate and debt principal respectively. The unlevered firm value is interpreted as
discounting the operating cash flow OCF by the so-called return of unlevered capital RU:
VU =

OCF
RU

(4.20)

Note that VL is not the value solely attributable to the project investor. It has to be
deducted by the initial capital cost to form the net present value (NPV). Our objective is
to maximize the levered wind power project NPV with debt as decision variable subject
to constraint. In basic formulation of constrained optimization, the problem is:
NPV = max {VL I }
0 D < I

(4.21)

where I is the initial capital cost of wind turbine.

The formulation has to be modified in a number of ways to fit the reality. First,
the time horizon should be finite. Second, the form of debt (bond or amortized loan) has
to be defined because it affects the portion of debt interest in each debt installment, hence
amount of tax to be calculated. Third, the term OCF has to be expanded in terms of an
operating income Oi and tax TC by accounting principle as follow.
~

O CF = O i (tW , EW , C ) TC

(4.22)

Here the operating income is written as a function of the product of wind tariff tW and
annual wind energy production EW, then minus total operating cost C. And I recognize
~

that EW is a random variable as discussed in chapter 3. Therefore,

~

O i (tW , EW , C ) = tW EW C

(4.23)

In this setting, total cost has excluded depreciation, i.e., no deprecation has been deducted.
Also, the total cost is assumed to inflate linearly at a rate gC. Finally, corporate tax TC is
calculated for two cases: with and without depreciation as follow:
~
~
[Oi (tW , EW , C ) D p iD ]tC
TC =
~
~
[O (t , E , C ) i ]t
i W
W
D C

where Dp is periodic depreciation, iD is debt interest and tC is the corporate tax rate.

99

(4.24)

A. Debt type

Since the decision variable of the optimization is debt principal in which the
periodic debt interest is conditional on, debt interest has to be expressed in terms of the
debt principal. But the conversion depends on the type of debt concerned. If the debt is a
bond with coupon rate yD, then the debt installment d and also the debt interest iD would
mean the same thing, and so
iD = d = DyD

(4.25)

Obviously, the annual operating income at least has to be enough to repay debt
installment, below which the firm is considered at default. But practically it cannot drive
to that limit but only to a level capped by the debt service coverage (DSC) ratio, rDSC.
Since the operating income is random, I propose a constraint on debt in probabilistic
setting to find the maximum allowable debt or value-at-risk (VaR) debt level, DVaR such
that the probability of default is at most a,
~

DVaR = max[ D : P ~ {Oi DyD rDSC } = a]

(4.26)

Oi

Note that a is a subjective probability level.

B. Depreciation

Depreciation makes tax payable smaller. The annual deductions and time span
depend on the tax depreciation system. In U.S., they use the Modified Accelerated Cost
Recovery System (MACRS). Under its Internal Revenue Code, wind turbine is classified
as 5-year property, so the annual deductions last for six year. I try to represent the
sequence of depreciation shield over the entire period as a function f Dp (.) ,
f Dp ({D p }t : t [1, k + 1])

where k+1 is the number of years where depreciation takes place.

C. Stochastic optimization of optimal capital structure

100

(4.27)

After considering all the above particulars, the levered firm value from (4.19),
~

which is theoretically a function of debt D and random wind energy production EW , is

expanded into the following comprehensive form:
~

VL ( D, EW ) = f PV (OCF )
~

= f PV (Oi TC )
~

t E (1 tC )
C (1 tC )
(1 + gC )i
1
= W W

[1
]
[1
]
RU
RU gC
(1 + RU )i
(1 + RU )i
+

(4.28)

yD DtC
1
[1
] + f Dp ({D p }t : t [1, k + 1])
RD
(1 + RD ) j

Finite horizon is worked on, where i is the span of FIT, j is the term year of bond, and
k+1 is the number of years depreciation lasts. Typically, k<j<i. gC is the growth rate in
~

operating cost. The discounted sum of OCF is represented by a function f PV (.) . Note that I
assume RD as the cost of debt at par value and therefore it is also equal to yD.
Furthermore, I propose using the project NPV as the objective function, which is
~

defined as G ( D, EW ) for two scenarios. It takes the form of levered firm value minus
investment cost, i.e. healthy state, if the operating income divided by debt installment is
bigger than the DSC ratio. Otherwise, the investor is forced to withdraw but recovers
initial capital less all debt repaid.
~

VL ( D, EW ) I
G ( D, EW ) NPV =

I D

tW EW C > rDSC d
~

(4.29)

tW EW C rDSC d

For the maximization G( D, EW ) which is understood to be NPV, it is meaningful to

~

optimize the expected value of NPV given EW is random. Hence if the probability
~

distribution of annual wind energy production is F~ ( x) P( EW x) , then the expected

EW

value of NPV is [G( D, EW )] = G( D, x)dF ( x) . This leads to stochastic optimization of

0

levered NPV as
~

max { g ( D) [G ( D, EW )] }

0 D < I

(4.30)

s.t.
D DVaR

101

Without priori information about the probability distribution of EW it is still possible to

present a closed form expression for the optimization problem. Expand the term
~

[G ( D, EW )]

as follow:
g ( D) =

rDSC d + C
tW

( I D)dF ( x) +

r d +C
= ( I D) F ~ ( DSC
)+
EW
tW

If we focus on the term

~

VL ( D, EW ) , it is obvious

rDSC d + C [VL ( D, x) I ]dF ( x)

tW

rDSC d + C
)]
rDSC d + C V ( D, x)dF ( x ) I [1 F ~ (
EW
tW
t

(4.31)

rDSC d + C V ( D, x )dF ( x)
tW

, in which from (4.28) there is an EW in

to expand it by integration by parts as below:

rDSC d +C
tW

xdF ( x)
rDSC d + C
tW
xdF ( x)
0
rDSC d + C
tW
xdF ( x)

xdF ( x)

= ( EW )

rDSC d +C
tW

= EW [ xF~ ( x)]0
EW

= EW (

(4.32)

rDSC d + C
tW

F ( x)dx

rDSC d + C
r d +C
) F~ ( DSC
)+
E
tW
tW
W

rDSC d +C
tW

F ( x)dx

So I have derived the whole expression of the objective function. It can be readily
programmed in Matlab. The final step is to make reference to the empirical distribution of
~

EW , i.e. the annual wind energy production in chapter 3. A discrete cumulative

distribution from empirical wind power data would be useful for the optimization process.

4.4

Trial Data for the Model

The fixed tariff wind power investment model follows an application of stochastic

maximization of the levered project NPV. A comprehensive set of contemporary data is

needed to illustrate specific model results. However, only nominal base case parameters
are used as a trial, followed by sensitivity analysis. Moreover, some simplifications are

102

taken for programming convenience. First, the residual project value (I-D) when
bankruptcy occurs is not counted. Second, wake effect is not considered because the
result is based on per unit wind capacity. Third, normal distribution is used for the
random annual wind energy production. The historical mean value of wind power is
~

taken as the mean of random variable EW . The simplifications would not sacrifice the
insights of the model.

4.4.1 Base case financial parameters

The rest of the model is simply plug-in of various parameters, as summarized in
Table 4.1. For instance, per unit cost data of wind turbine are classified by onshore or
offshore as referenced from [94]. The default probability a corresponding to a VaR debt
level is a pre-defined value, commonly at 5% or 1%. All other parameters are nominal
values subject to actual scenario. Debt level is a decision variable in the optimization
algorithm ranging from any percentage (0-100%) of investment cost. Model outputs are
levered wind power project NPV and VaR debt level in the example section.

Corporate tax rate tC

0.34

Return on unlevered equity RU

0.1

Default probability a

0.05

Wind turbine lifetime (years)

20

Annual coupon bond lifetime (years)

10

Debt coupon rate yD

0.07

Euro dollar : US dollar

1:1.4

Land wind power tariff ( cents/kWh), offshore data in parenthesis

5.17 (6.19)

Land wind turbine investment cost I (US$), offshore data in parenthesis

1.4M (2.1M)

Land wind turbine maintenance cost C (US$), offshore data in parenthesis

30K (60K)

Depreciation schedule

5-year property

rDSC

1.2

gC

0.01

Wake effect

nil

Assumed distribution for EW

Normal distribution

Table 4.1 Financial parameters for fixed tariff wind power project

103

4.5

Numerical Example

4.5.1 Base case

The emphasis of this example section is to illustrate what the proposed debt
optimization model can do instead of pinpointing particular model results. Discussions
are based on per MW nominal capacity of an onshore wind farm under fixed tariff. The
first model results presented are the wind power project NPV and levered NPV, as shown
in Fig. 4.1. The message conveyed is very straightforward. Both NPV and levered NPV
decrease with increasing investment and maintenance costs to various extents. The drop
of levered NPV is buffered initially by the beneficial effect of leverage. The second
highlighted result is the maximum attainable value of the decision variable, VaR debt
level, for a range of investment cost and maintenance cost, as shown in Fig. 4.2. For the
lower end of maintenance cost, VaR debt tracks the investment cost nicely until a level
beyond which the VaR debt is flat. It implies that for investment cost below that critical
level, the optimality is actually all-debt, whereas beyond the critical investment value the
amount of debt is capped at where it was. For the higher end of maintenance cost, the
VaR debt stays constant below any investment cost. The overall picture is that the
optimized VaR debt level increases with decreasing operating cost.

104

Fig. 4.1 Project NPV and levered NPV of one MW onshore wind capacity investment

Fig. 4.2 The VaR debt level of one MW onshore wind capacity investment

105

4.5.2 Sensitivity analysis

Sensitivity analysis is conducted to demonstrate how the VaR debt level and
maximized levered project NPV are affected by other parameters. The effects on DVaR
level are illustrated by changes in debt coupon rate and default probability, for land and
sea wind power in Fig. 4.3 and Fig. 4.4 respectively. The effects on maximized levered
valuation are illustrated by changes in debt coupon rate and return of unlevered equity,
for land and sea wind power in Fig. 4.5 and Fig. 4.6 respectively.

Fig. 4.3 Sensitivity analysis of VaR debt to debt interest rate and default probability for
onshore wind farm

106

Fig. 4.4 Sensitivity analysis of VaR debt to debt interest rate and default probability for
offshore wind farm

Fig. 4.5 Sensitivity analysis of maximum levered NPV to debt interest rate and return on
unlevered equity for onshore wind farm

107

Fig. 4.6 Sensitivity analysis of maximum levered firm value to debt interest rate and
return on unlevered equity for offshore wind farm

4.6

Summary
In this chapter, a novel view on the valuation of fixed tariff wind power project is

introduced. Due to the fact that such project has very limited risk, debt financing is
desirable. The risk is assessed by the probability that annual revenue is not enough to
cover debt installment. Associated with a subjective value of default probability is the
value-at-risk debt level. Optimal capital structure suggests maximized project value. This
chapter proposes an optimization framework to maximize wind power project valuation
with debt as decision variable, subject to the value-at-risk debt constraint.

4.7

References

108

[94]

Brendan Fox, Damian Flynn, Leslie Bryans, Nick Jenkins, David Miborrow,

Mark OMalley, Richard Watson, and Olimpo Anaya-Lara, Wind Power Integration,
Connection and System Operation Aspects, London: IET Power and Energy Series,
2007.
[95]

Database of State Incentives for Renewables & Efficiency. [Online]. Available:

http://www.dsireusa.org/
[96]

L. Bird, B. Parsons, T. Gagliano, M. Brown, R. Wiser, and M. Bolinger, Policies

and market factors driving wind power development in the United States, Energy
Policy, vol. 33, issue 11, pp. 1397-1407, 2005.
[97]

M. Mendonca, Feed-in Tariffs, Accelerating the Development of Renewable

Energy, London: Earthscan, 2007.

[98]

Renewable

Energy

Sources

Act

2009.

[Online].

Available:

http://www.erneuerbareenergien.de/files/english/pdf/application/pdf/eeg_2009_en_bf.pdf
[99]

Stephen A. Ross, Randolph W. Westerfield and Bradford D. Jordan, Corporate

Finance Fundamentals, New York: McGraw-Hill, 2006.

109

Chapter 5

5 Real Option Wind Power Investment Model

Abstract
This chapter presents a real option investment model for utility-owned wind
power project. Total fuel cost saving from conventional units is regarded as the benefit of
wind power generation. In particular, the fuel cost is determined by production costing
technique by system load net of wind power. Given a renewable energy target and
deadline as analogy to option strike price and maturity date respectively, the wind power
project is then formulated as a real option. The proposed model is able to determine the
option value of the wind power target capacity and its optimal investment timing. It also
captures parameters such as target date, cost trends of fuels and wind turbine, interest
rates and fuel mix. Any shortfall in project reward suggests indemnification by
appropriate level of carbon price or renewable credit. This model systematically
characterizes required return, price volatilities and risk.

5.1

Introduction
Restructuring of electricity market has brought research opportunities for the

subject of generation planning and investment. In the past when utilities were vertically
integrated, optimization approach, say mathematical programming or other heuristic
methods were used. Relatively new methodologies for the evaluation of conventional
generation investment in spot market have also been developed, e.g. price-based unit
commitment bidding [108][109], real option valuation [149][150][151] and more recently,
analytical calculation of generator profit [158], all are based on some stochastic processes
of electricity price. Nowadays, however, attentions are focused on renewable, in

110

particular non-dispatchable renewable such as photovoltaic and wind power, hence

renewable investment model is needed. Unlike conventional generation investment in
which electricity price is a key component, photovoltaic and wind investment depends
heavily on policy drivers as well as market regimes. A straightforward scenario is net
present value (NPV) analysis of wind power project under fixed feed-in tariff [116].
Although individual investors can own wind turbine generators, many policy and
market factors driving wind power development in U.S. are actually put on utility side
[130]. Three highlighted factors are Renewable Portfolio Standard (RPS), Integrated
Resources Planning (IRP) and Green Power Programs. The generic investment scenario
starts with the RPS, assuming the total consumption needs to be satisfied by a certain
percentage of wind power. For example, California aims at 33% renewable by 2020 [106],
Texas has strong compliance with RPS especially fulfilled by wind [142]. On the other
hand, utilities would include wind generation as an IRP option in response to elevating
fuel costs and wholesales electricity price even without the RPS obligation. Green power
programs promote coverage of wind power by the generation portfolios of utilities, either
through direct ownership or purchasing agreement.
RPS implies that utilities have either to build enough wind turbines on their own
or purchase from other wind power producers by a certain deadline. In practice, utilities
may even invest wind power earlier than the deadline if conditions are favorable or
optimal. The conditions considered are fossil fuel prices. If fuel prices are high enough,
substituting the energy demand by wind generation is profitable, vice versa. Of course,
this is a one-off, irreversible investment decision based on fuel price expectation. The
investment decision is analogous to the premature exercise of an American styled option,
in which the wind power project is modeled as a real option.
My proposed real option model serves utility a number of decisive questions for
investing wind power. Given a target of wind power capacity, the model addresses the
optimal time to build it. The value of wind power project is assessed through its benefits
brought to the utility or the corresponding system, specifically how much fossil fuels it
can save, hence total fuel cost saving. The assessment can be realized directly in fuel
prices or passively in electricity contract prices. Being conditioned on fuel prices, the
valuation is more intrinsic than subsidizing policy. Furthermore, power system

111

characteristics in terms of generator type and availability are captured. This model also
provides an alternative assessment on carbon dioxide price. Right now CO2 price is an
exogenous variable practically denominated in European Allowance (EUA) price. In this
work we try to internalize or endogenize the price of carbon emissions. CO2 price adds on
the cost of conventional electricity, a closely related instrument is the renewable credit
that serves as extra revenue to wind power or other renewable generator. The price level
of CO2 or renewable credit could be determined such that the wind power project, when
purposely built for fossil fuel substitution, has to breakeven.
It becomes clear that the previous chapters follow logically to arrive to the
establishment of real option modelling of wind power project in this chapter. In chapter 1,
it is observed that there is a transition in electricity market structure from monopolistic
regime to competitive market. Accompanying this trend is a shift of analytic mindset
from generation planning to generation investment. Meanwhile, renewable generation has
also emerged, but its pace of development is strongly related to policy drivers. And more
importantly the analysis of renewable investment depends heavily on the accommodation
specifically designed for renewable. It then comes to chapter 2 explaining one of the key
drivers for renewable deployment nowadays is the renewable energy target. Utility has to
have certain renewable capacity by a deadline and such renewable investment, say wind
power investment, is strikingly similar to an American real option, which is the basis of
this chapter. On the other hand, chapter 3 provides probabilistic analysis of wind power
generation so that the results are capable to be utilized by the probabilistic production
costing technique on the same platform. The overall aim of this chapter is to develop
financially consistent investment models for wind power that are able to account for
relevant market mechanisms and address technical considerations.

5.2

Literature Review and Comparison

5.2.1 Review of selected real option applications in energy sector

112

There are many real option applications in energy-related topics. For the real
option method to be initiated there must be some real operational flexibilities to be
modeled, hence the name real option refers. It is not difficult to identify such real world
scenarios for real option modeling to come in place.
For example, renewable is generally perceived as more competitive if fossil fuel
prices are high. A R&D programme of renewable technology can have three possible
states, subject to future price levels of fossil fuels. Say, if fuel prices are high, the R&D
programme can be deployed to launch the new product (renewable installation). If fuel
prices are medium, the R&D programme can be carried on until market for renewable is
favorable. If fuel prices are very low, renewable has no competitive edge and therefore its
R&D programme can be abandoned at all. Hence, subject to the evolution of fuel price, in
which it is normally a stochastic process but practically represented in lattice, the R&D
program is like a real option and its valuation can be obtained [101].
Similarly, real option can be used as planning model for renewable generation
technologies. Renewable generation technologies are treated in a similar fashion with
other conventional generators in generation planning model formulated in dynamic
programming. Price uncertainties are wholesale electricity price and fuel prices whereas
the real option or flexibility is the timing of renewable investment [118].
One of the demand-side management facilities, the curtailable load service offered
by utilities, can also be viewed as a real option and subsequently valuated by customers
[152].
More specifically to wind power, switchable tariff for wind power in Spain can be
modeled as a real option [156]. Wind power generators are encouraged to participate in
spot market to echo with deregulation. Since spot price has highs and lows but wind
power bidding is very passive, the switchable tariff has some additional benefits for
attracting wind power operators to join. On the other hand, the switchable tariff also
allows wind power generators to switch back to and forth from fixed tariff, thus
enhancing the overall value of this switchable tariff.
I am not going to do an exhaustive review of real option applications, as there are
many merely in the energy sector [111]. In fact, prior to all those industry-specific
applications of real option, the idea of accounting firm value or asset valuation by the real

113

options to make future investment actually appeared as early as in [148].

Abovementioned references give a sense of how the problems are formulated in general.

5.2.2 Comparison with existing works

While it has been suggested the true value of wind power should be measured by
fossil fuels saved [140], I believe this thesis chapter is the first to explicitly formulate
such idea into an investment problem. The real option framework is applicable to wind
power project as long as the fuel cost saving is considered as the underlying process. Yet,
the real option framework, together with the concept of fuel substitution, has been applied
in other generation investment analysis, notably the decision of end users in deregulated
market to invest in micro gas turbine (DG) in substitution of electricity purchase from
utility [100]. Given the utility electricity rate as assumed initially fixed, an end user may
find self-generation by on-site micro gas turbine more economical if the gas price is low
enough. The DG investment is an investment option in which if the end user holds it, he
receives no incremental electricity saving; if the end user exercises it (by means of
investment cost), he would receive a present value of total cost saving which is equal to
electricity cost less gas cost. Gas cost is modelled in GBM and the investment option is
solved by differential equations. The model is also able to determine the gas cost
threshold below which micro gas turbine is optimal to be invested. The work of [100] has
two major extensions, first, to consider operational flexibilities of gas turbine and second,
to allow stochastic process of electricity price. The second extension is similar to the
spark spread option described by Deng et al. in which both electricity and fuel prices are
modelled by stochastic process. Yet such option was actually first explored in [144].
Nevertheless, there are a number of differences, in fact, improvements in the proposed
model of this thesis over [100], listed as bullet points below:

6.

Finite equipment life versus unrealistic infinite horizon

7.

An actual renewable target deadline associated with option maturity date

8.

Bivariate binomial model for two fuels simultaneously

9.

Both investment and maintenance costs have drift rates

114

10. Risk-adjusted project discount rate

Last but not the least, another real option modelling of wind power project is also
mentioned [123]. However, it allows evolution of the wind project NPV as the underlying
process; a direct reference from the case of stock. For stock option, value of the share (or
total capitalization if multiplied by total number of shares) and the present value of all
future cash flows of the company are thought to be equivalent. Hence the evolution of
stock price in any stochastic process implies the same movement for the present value of
cash flows. However, it does not mean net present value in the case of real option can be
modelled in the same way. Only the underlying price variable can be modelled by
stochastic process and the corresponding NPV should be calculated period by period.

5.2.3 Preliminaries of option pricing theory

It may be a little bit abstract if the option formulation of a real wind power
project is kicked off without elaborating the original meaning of the stock option pricing
theory. Therefore I first discuss the classical derivation of a call option on stock, in which
the stock price is assumed to take only two possible values one period from now.
Suppose the current stock price is S0; one period after, it takes probability pu to increase
by a factor of u to Su, or takes probability pd to decrease by a factor of d to Sd. This is
called one-step binomial model of S as depicted in Fig. 5.1:
pu

Su = uS0

S0

pd

Sd = dS0

Fig. 5.1 A time step of a binomial model

Suppose further that the call option will be matured or exercised at the end of this
period with strike price K. The call option will take two possible prices, max[Su-K,0] or
max[Sd-K,0] at the two nodes, which are deterministic. The logic question to ask is what
would be the rational price for the option now, c0. It turns out that if the issuer of the call

115

can construct a portfolio consisting the call option and some quantities of the underlying
stock, and further if the issuer can develop a strategy to make the terminal values of this
portfolio indifferent no matter what S realizes at the end of the period, then there should
be an unbiased way to calculate c0. Let the issuer long delta number of share for every
call issued, the current portfolio value is S0-c0. At the end of the period, the portfolio
value is Su-cu if S goes up or Sd-cd if S goes down. Theoretically, the issuer can
determine the required to make Su-cu = Sd-cd, so that

cu cd
(5.1)
Su S d
In this case the portfolio change is riskless. Then by no-arbitrage, the portfolio should
=

earn risk-free return, hence (S0-c0)(1+r)= Su-cu, giving the formula for c0:
c0 =

qcu + (1 q )cd
1+ r

(5.2)

where q is referred as the risk-neutral probability:

q=

1+ r d
ud

(5.3)

From this example, the call option value is determined without relying to the
probabilities pu and pd, i.e., we do not need to predict movements of stock price. What
being required by the model is the prerequisite to determine a correct delta in order for
hedging of the portfolio successful. This seems to be straightforward for the present
example because it only has one period. For practical time span of an option in which its
underlying process is modeled by geometric Brownian motion (GBM), the portfolio sum
has to be rebalanced, or the delta recalculated continuously. In infinitesimal time scale,
the example goes back to the original description of the Black-Scholes option pricing
theory. The expression of delta then becomes
=

c
S

(5.4)

In other words, delta is the rate of change of option value to the underlying stock price.
To explore the meaning carried by the call option value, also known option
premium, consider the flexibility granted to the holder of call option. The call allows the
holder the right but not obligation to buy a stock at a pre-specified price in the future.
Clearly, this flexibility is on the attractive side to the option holder because he can wait

116

and see what the stock price would be before exercising the option. He will certainly not
exchange for the underlying stock if the stock price turns out to be lower than the option
strike price. Therefore he needs to pay a premium for such an advantage. On the other
hand, the issuer providing such option bears the risk. He has to deliver the underlying
stock if the call option is exercised. Therefore, the issuer must keep a stock of share to
hedge the call option he issues, and he also needs to adjust the portfolio continuously for
the call option to be dynamically priced. The above discussion is the underlying meaning
of hedging and option premium.

5.3

Contingent Claim and Real Option

The derivation of Black Scholes differential equation relies on the setting that the

underlying asset modelled by geometric Brownian motion. The solution of the

differential equation not only solves for option, but also prices for other derivatives
depending on what the boundary condition is. Actually the technique of replicating
portfolio and delta hedging can be used to valuate a real project, here the wind power
project value. The generalized approach is called contingent claim analysis. A contingent
claim can be regarded as the present value of all future net cash flows of a project. The
valuation of the claim is solved by another differential equation, which is very similar to
the BS one. For the wind power project defined in this chapter, the net cash flow is fuel
cost saving less wind turbine operating cost, so the project valuation is actually the
present value of all future net cost saving to the utility system. And this claim value is
contingent on the dynamic of fuel price.
In the sub-sections follow, I will first go through the steps for deriving the
contingent claim differential equation for the wind power project value, using the same
technique of delta hedging and no-arbitrage assumption. Then the project value is solved
by proper boundary condition, which is the typical cost-benefit representation of a project.
Finally, I extend the formulation to real option modelling of the flexible investment
timing of the wind power project by a certain deadline.

117

5.3.1 Contingent claim and justification for delta hedging

Suppose there is only one kind of fossil fuel saved, its price S follows a geometric
Brownian motion (GBM)
~

(5.5)

d S = S dt + S d z

where is the drift rate of price,

is the volatility rate of the return of price and d z is the

~

standard Wiener process. Generically, project revenue is equal to price multiplies output
per unit time. Here for the wind power project, output wind energy per unit time is
~

initially taken as one, so without further loss of generality, S itself denominates as fuel
cost saving.
Black Scholes (BS) option pricing theory starts with the simplest kind of option
that gives its holder the right to buy a share of stock. That is a call option and, based on a
number of assumptions, BS formula provides valuation for such option. The technique is
generalized here to price a wind power project as a contingent claim. Denote the wind
~

power project value in V ( S , t ) , which is a function of fuel price S and time t. Before
~

deriving the differential equation for V ( S , t ) , we revisit some of the BS theorys

assumptions and elaborate how they are satisfied when the theory is applied to contingent
claim or real option. Similarities and differences for the two cases are discussed briefly.
The most obvious difference is the subject to be priced. BS formula valuates stock
option while wind power project value is needed here. An implicit assumption in BS
theory is that the underlying asset (stock) is tradable in fractions of shares. A wind power
project does not readily have common stock if it is solely owned by a utility. Still, the
utility can make V tradable, not necessary in open market, by partially selling the project
to other investors. In this case the wind turbine generator becomes a separate entity and
serves the utility through purchasing agreement. The purchasing agreement which
denominates the revenue of wind turbine generator is taken as the original fuel cost
saving when market equilibrium is assumed. Another Black Scholes assumption is no
transaction cost, which exhibits as a single interest rate for both deposit and loan, and
unlimited deposit and withdrawal of fund. This assumption is acceptable for big electric

118

utility as it can secure source of fund from banks relatively easily, and the fund size is
usually big enough to drive interest rate very low.
Upon satisfaction of the BS assumptions, the wind power project value should
depend only on the price of fuel and time. Suppose that the utility owns one wind power
~

~
V ( S , t )
project and at the same time shorts
units of fuel contract. Define the value of
S

such portfolio as
~

V ( S , t ) ~
=V
S
S

(5.6)

The change is portfolio value is

~

d = dV

V ( S , t ) ~
dS
S

(5.7)

Fuel contracts are traded in commodity as well as purely financial market. The portfolio
value changes as fuel price changes with time. But as far as what wind turbines do in
terms of fuel saving, should the conventional generation turned out to be cheaper amid
fuel price slump, the utility can hedge downside price movement by short-selling fuels.
Assuming the position in fuel contracts can be continuously rebalanced, here in annual
resolution, so that any gain or loss of the wind power project can always be offset by the
contract positions. This is called hedging and the return of the portfolio should be certain
and independent of change in fuel price. Meanwhile, by Itos lemma,
~

~ 2
~ V ( S , t ) 1
~ V (S , t ) ~
V (S , t )
V (S , t )
dV = [
+ S
+ 2S2
]dt + S
dz
2
t
S
2
S
S
~

(5.8)

Substituting d S and d V into (5.7) gives

~

V ( S , t ) 1 2 ~2 2V ( S , t )
d =(
)dt
+ S
t
2
S 2
~

(5.9)

As d z is already eliminated, the above equation becomes deterministic and therefore the
tildes on S are no longer required. Apart from capital gain of the portfolio, the wind
~

project has net fuel cost saving per unit time, denoted in ( S , t ) . However, short-selling
fuel contracts (or reducing the physical stock of fuels) lower the convenience of operation
theoretically, as the utility has to nonetheless burn fuels for conventional generation. If

119

the fuel convenience yield is represented by , then in a short period of time, the extra
~

V ( S , t ) ~
flow would actually be (S , t )dt
S dt . The two terms are stochastic, but the
S

magnitude is in the order of dt 2 , so can be safely reduced to deterministic setting. The

resultant change in capital gain and cash flow of the portfolio is
d =[

V ( S , t ) 1 2 2 2V ( S , t )
V (S , t )
+ S
S
+ (S , t )]dt
t
2
S
S 2

(5.10)

Since this investment is riskless, it should earn as much as the original portfolio earns for
risk-free return in the same period of time, i.e.
[

V ( S , t ) 1 2 2 2V ( S , t )
V ( S , t )
V ( S , t )
+ S
S
+ ( S , t )]dt = r (V
S ) dt
t
2
S
S
S 2

(5.11)

Re-arranging to yield the partial differential equation:

V ( S , t )
V ( S , t ) 1 2 2 2V ( S , t )
+ (r )S
+ S
+ ( S , t ) = rV ( S , t )
t
S
2
S 2

(5.12)

Now the solution of (5.12) is the project valuation (not NPV) in infinite horizon,
what needs to be done for the value to be describing wind power project is the definition
~

of ( S , t ) . First, we relax the quantity of project output as the annual average wind energy
production, for the reason that annual average is the most stable and assumed constant. It
is realized in terms of fuel heat saved H (Mbtu). Then it maintains the directly
proportional relationship between fuel price S ($/MBtu) and fuel cost saved. Second, the
profit function (net fuel cost saving) is defined as fuel cost saved minus operating cost
C of wind turbines.
~

(5.13)

(S , t ) = H S C

For applications that model firms flexibility to shut down operation when market
condition is bad [144], the profit function would alternatively take an max[.] operator.
This also laid down the basis for the work of electricity spark spread option [151].
However, here in this work the emphasis is not the flexibility to idle wind turbine because
it has no fuel cost. On the other hand, this work models the flexibility or option to delay
investment, i.e. investment timing of wind power project. The wind power investment
~

option will be denoted in F (V ) as a contingent claim on V .

120

5.3.2 Solution of contingent claim as project valuation

If we assume the project V to be infinitely lived, equation (5.12) becomes
independent of time and its time derivative can be omitted. In other word, the project
value one unit of time later looks exactly the value now, except with a new start state S.
The simplified differential equation is
(r ) S

V ( S ) 1 2 2 2V ( S )
+ S
rV ( S ) + ( S ) = 0
S
2
S 2

(5.14)

It is an ordinary second order differential equation having closed form solution. The
complete solution of V is found to be [103]:
V ( S ) = A1S 1 + A 2 S 2 +

HS

C
r

(5.15)

where A1 and A2 are constants equal to zero in most common applications [103]. The
solution of valuation by contingent claim analysis is then completed.
Equation (5.12) describes the project valuation by embedding the profit stream in
the differential equation. On the other hand, firms operation can be evaluated by real
option and the resultant project value is the aggregate of all real options. For the same
underlying GBM process for fuel price, consider the Black-Scholes differential equation
~

(BSDE) for basically solving the present value of any derivatives f ( S , t ) and an improved
case containing dividend rate [115][139][122]:
f
f 1 2 2 2 f
+ (r ) S
+ S
= rf
t
S 2
S 2

(5.16)

where r, and carry the same meanings as before. also equals minus , where is
the required return of the wind power project. Hence can be interpreted as the
difference in return between holding the wind project and simply speculating fuel price of
return .
Suppose the derivative is a forward contract with strike price C. The forward
contract models periodic wind project net fuel cost saving . Let the current time be t
and the operating time moment be T. Thus the forward payoff at time T is
f ( S , T ) = HS C

121

(5.17)

which is the boundary condition of the BSDE. Solving it gives the forward price at any
time t:
f ( S , t ) = HS e (T t ) e r (T t )C

(5.18)

This is the present value of wind project operation at one future point in time. But the
forward payoff is a one-shot value, whereas firm valuation is an aggregate of all payoffs
along its lifetime, so we need to sum all the forwards and define the project valuation V
by [144]
V

(5.19)

f ( S , 0)d

Now T represents the lifespan of the project. Valuation V defined above should match
with the solution (5.15). Substituting (5.18) into (5.19) and summing all the f(S,0) for
{ [0, T ]} (integrating over T) yields
V

( HS e C e r )d =

HS

C
r

(5.20)

as T goes to infinity. So the equivalence is shown. The solution is simply the expected
present value of the infinite stream S discounted at risk-adjusted discount rate minus the
riskless perpetuity of the operating cost.
For a few reasons the contingent claim approach is preferred: 1) its derivation is
more robust; 2) its closed form solution V (of infinite case) is more intuitive; 3) it makes
modeling of investment delay possible and finally 4) it facilitates the use of American
option formulation and numerical computation of option lattice. The third point will be
discussed right ahead whereas the fourth point will be elaborated in later sections.

5.3.3 Real option accounting delay of investment

Now we go back to interpret the valuation V in (5.15) or (5.20). It is theoretically
a claim to all future profits of the project or the present valuation of these profits. The
meaning of V has two ways to go. If the project is already running, V is its market price.
Any new investors who want to purchase it would judge whether such price (valuation)
will be recoverable by future profits of the project, but not its initial investment cost made
by somebody else. If the project is yet to implement, V has to be analyzed in conjunction

122

with the initial cost I to form the NPV. The project is worthwhile to invest if NPV is
greater than zero, vice versa. But NPV is only a now-or-nothing concept. If NPV is
negative, it does not tell whether waiting would allow favorable conditions to come back
and investment by then would be good again.
In view of this limitation, I invoke real option methodology to valuate the wind
power project. The flexibility associated is the investment timing. Recall from the
example of simple call option, the option price is actually a premium its holder needs to
pay in order to own the right but not obligation to buy the underlying stock in future.
Here if the investment timing of wind power project is allowed flexible, the real option
acts as a choice for its investor to acquire the wind power project sometime in the future.
The real wind power investment option, call it F, prices the flexibility of delay.
The hedging strategy for this real option is more complicated and would require
more restrictive assumption. Recall that the issuer of a stock call option needs to long the
underlying stock for hedging. The share price is public information from the market,
therefore the hedging strategy is independent of the option holder. On the other hand, the
data of the amount of fuel savable is dependent on the utility generation mix and wind
power generation characteristic, which are not easily and accurately known. Furthermore,
the hedging strategy is utility specific so that the exotic option is tailor-made to that
utility only. For this real option to be hedged and properly priced, a market maker may be
assumed present. Suppose that a utility is planning a wind power project in the future,
which will save multi-year fuel cost once it is built. The consequence can also be
accomplished financially if the utility longs the exotic option and by the time it is
exercised, it entitles the utility to receive a present value sum of total net fuel cost saved
by the prospective wind capacity. This would require the market maker to hedge a
combination of fuel contracts synthesizing the total net fuel cost saving for the utility.
Previously I have solved the wind power project value V(S) as a contingent claim
to all future net fuel cost savings. Now this V is treated as an underlying process for the
exotic real option F(V) to build upon. By considering the market maker able to long

=F(V) number of underlying asset for every real option issued, and assuming
continuous hedging and no-arbitrage in the same manner, a differential equation in F(V)
can be derived:

123

1 2 2
V F "(V ) + (r )VF '(V ) rF (V ) = 0
2
~

(5.21)
~

This derivation requires the underlying process V following the same GBM as S . But
~

from (5.20), V is not strictly a GBM because of the term C/r. Nevertheless I try to
validate the case because the term C/r is less significant compared with HS/

dominating the GBM. Another remark is since V(S) is spanned over S, i.e. the value of
the underlying asset S is perfectly correlated to the claim V(S) (and also the option
F(V(S))) in any short period of time, therefore it is more desirable to write F(S) instead of
F(V) [103]:
1 2 2
S F "( S ) + ( r ) SF '( S ) rF ( S ) = 0
2

(5.22)

Corresponding solution of F is
F ( S ) = B1 HS 1

(5.23)

where
1 =

(r ) 1 2 2r
1 (r )
] + 2

+ [
2
2
2
2

(5.24)

and B1 is a constant determined by boundary conditions [103]:

F (0) = 0

(5.25)

F ( S *) = V ( S *) I

(5.26)

F '( S *) = V '( S *)

(5.27)

where I is the investment cost. Equation (5.26) is the value matching condition. It makes
F evaluated at a particular price S* such that the investment option is indifferent from the
original NPV. Equation (5.27) is the smoothing-pasting condition. Now the constant B1 is
solved with these boundary conditions, given the general functional form of V(S) in
(5.20).
B1 =

( 1 1) 1 1

(C / r + I ) 1 1( 1) 1

S* =

1 (C / r + I )
1 1

(5.28)
(5.29)

In other words, the wind power investment option F is solved with a specific form of V(S).
A side product is the optimal investment price S*. An implicit assumption of infinite

124

lifespan of a project means there is no deadline of making the investment. This

assumption will be forgone automatically if finite lifespan is considered in practice. Up to
this point the real option framework for wind power investment is completed.
Finally I offer a critical remark on the hedging of this exotic real option. The
hedging for stock option can be done relatively easily because the subject to be hedged is
stock price. However hedging the exotic wind power investment option is actually much
more difficult because now the subject to be hedged is V(S), the present value of total net
fuel cost saving, which in turn depends a number of hard-to-predict parameters such as
multi-year fuel price expectations, change in investors risk preference (discount rate),
and annual wind energy production, etc. They would certainly show discrepancies from
initial forecasts after the wind farm is built. Hence perfect hedging could be impractical.
As a remedy, what I have assumed the existence of market maker is not necessary
because the intention of designing such an exotic option is not really to substitute real
wind power investment. It merely provides a reference valuation of the same project if it
could be delayed. The value quotation is for business decision only rather than
speculation. Therefore I argue that inaccuracy can be tolerated for this purpose.

5.4

Bivariate Binomial Lattice for two Fuel Prices

The composition of fuels saved by wind generation depends on the utility

generator mix, load profile and dispatch priority of wind. In general, more than one type
of fuel is displaced. Here two assumptions are made for simplicity, wind is right above
nuclear and large hydro in the generation stack, and the case of two fuel savings is
illustrated. A generalized two-state real option formulation then follows.
Suppose for a case of coal price Sc and oil price So, their stochastic processes are
in correlated GBMs:
~

d Sc = c Sc dt + c Sc d zc
~

d So = o So dt + o So d zo

(5.30)
(5.31)

where [d z c2] = dt , [d z o2] = dt , [d z cd z o] = dt , is the correlation between the two prices, and

125

 c = c

(5.32)

o = o

(5.33)

The differential equation governing wind project value V with two underlying processes
is:
( r c ) Sc

V ( Sc , So )
V ( Sc , So )
+ ( r o ) So
Sc
S o

2V ( Sc , So )
V ( Sc , So )
2V ( Sc , So )
1
+ [ c2 Sc2
+ 2 c oSc So
+ o2 So2
]
2
2

S
Sc
So2
c
o

(5.34)

+ ( Sc , So ) rV ( Sc , So ) = 0

Define wind benefit as total fuel cost saving minus its operating cost by
(5.35)

( S c , S o ) = H c Sc + H o S o C

where Hc and Ho are the coal and oil saved by wind respectively, in terms of heat (MBtu)
and to be computed by probabilistic production costing. Furthermore, H c : H o is assumed
fixed. Equation (5.34) is nicely solved by simple substitution to get the particular solution
V ( Sc , So ) =

H c Sc

H o So

C
r

(5.36)

which is a valuation of wind power project with infinite cash flows. Discrete valuation of
finite life j is
V j ( Sc , So ) = (1 e j c )

H c Sc

+ (1 e jo )

H o So

(1 e jr )

C
r

(5.37)

The final step is to compute the corresponding investment option F. Only with F
the investment cost I is incorporated as a boundary condition. But soon it reveals that
closed form solution is not available because now the problem has two price variables.
For this reason and also as equipment life is inherently finite, discrete approach is more
preferable. And I come to binomial lattice because it is excellent for handling early
exercise of American option, as to the case of optimal investment timing. The
discretization procedure follows as below.

5.4.1 Univariate binomial model

Equation (5.14) is an example of ordinary second order differential equation that
has closed form solution. Partial differential equations can be in general computed by
finite difference approach. The celebrated Black-Scholes formula provides an analytic

126

solution to a partial differential equation which describes the option theory. Numerical
methods developed for solving the Black-Scholes formula includes binomial model [126]
and Monte Carlo simulation [136] and they both give very satisfactory approximation.
Subsequent developments relating to numerical methods for options had built a vast
literature but that may be a too mathematically-oriented research. For engineering
applications, we choose the binomial model which is simple, effective and intuitively
appealing. In particular, it has two attractive advantages: the distinctive ability to cater
early exercise in a recursive procedure and extendable to options with two underlying
variables.
The univariate binomial model is begun with. Its key idea is to discretize the
GBM of the underlying price variable into many infinitesimal time steps, t. For each
time step, the value of S can either go up by a multiple of u with probability p, or down
by a multiple of d with probability 1-p. Graphically, it is the same as in Fig. 5.1. Our aim
is to determine parameters u, d and p such that the binomial model can approximate the
~

GBM satisfactorily. Recall d S = S dt + S d z or equivalently

~

d x = vdt + d z

(5.38)

where
~

x = ln S

(5.39)

v = 0.5 2

(5.40)

Consider the natural logarithm of the one plus rate of return of S over the time step t:
~

log(

log u with probability p

S (t + t )
) = v t =
S (t )
log d with probability 1 p

(5.41)

It is indeed the continuously compounded rate of return of S taking on two possible

values. The limiting mean and variance of the continuously compounded rate of return of
price defined by the binomial lattice can be matched with that of the actual price process
in GBM by the following prescribed parameters:
u = e

d = e

(5.42)

1
v
p = [1 +
2

t ]

127

(5.43)

Verifications of the limiting cases of mean, variance and probability distribution of the
continuously compounded rate of return are left in [126].

5.4.2 Bivariate binomial model

A bivariate binomial model is needed in this chapter. Among the early attempts,
Boyle modified the original univariate binomial model to a trinomial one and then extend
it to a five-point lattice for two state variables [137]. In that work the limiting mean and
variance of the lattice process are matched with that of the price itself but not the
logarithm, so the algebra is quite complex. The matching is much easier if logarithmic
price is worked on. We follow [135] for a bivariate binomial lattice on two correlated
GBM processes collectively from (5.30) to (5.33). Let puu be the probability that both
prices go up, pud be the probability that Sc goes up and So goes down, vice versa for pdu,
and pdd be the probability that both go down. Moreover, the substitution (5.42) remains
the same for each of the two processes. By similar consideration of logarithms of Sc and
So as from (5.38) to (5.40) and matching limiting means and variances, the probabilities p
are determined as follow:
1
p uu = [1 +
4
1
p ud = [1 +
4
1
p du = [1 +
4
1
p dd = [1 +
4

t (

vc

vo

vo

c o

t (

vc

c o

t (
t (

vc

) + ]
) ]

vo

vo

c o
vc

c o

(5.44)
) ]
) + ]

Hence the bivariate binomial lattice is completely defined.

5.4.3 Risk neutral valuation

The final step is to compute the value of F by a recursive procedure under the
concept of risk-neutral valuation. Risk-neutral valuation of derivatives says even without
differential equations like BSDE, as long as delta hedging is known feasible, it is
equivalent to price a derivative (here an option) in a risk-neutral world. There is no risk

128

premium for both the rate of return of the underlying asset and the discount rate of option
payoff, hence both are simply the risk-free rate r. This is particularly important because
the probabilities p in (5.44) for the lattice underlying variables can be modified to depend
on the rate of return of those variables by collectively (5.32), (5.33) and (5.40).
Subsequently substituting by r yields the risk-neutral probabilities q:
1
r c 0.5 c2 r o 0.5 2o
) + ]
+
q uu = [1 + t (
4
c
o
1
r c 0.5 c2 r o 0.5 o2

) ]
q ud = [1 + t (
4
c
o

(5.45)

1
r c 0.5 c2 r o 0.5 o2
+
) ]
q du = [1 + t (
4
c
o
1
r c 0.5 c2 r o 0.5 2o

) + ]
q dd = [1 + t (
4
c
o

Risk-neutral probabilities are used to weigh option payoff.

The recursive procedure of calculating discounted option payoff is illustrated
together with the lattice in Fig. 5.2. Suppose the renewable deadline in n years from now
and assume no construction lead time. At year n the project must be invested but it may
be optimal to even invest earlier depending on the fossil fuel prices. The scenario is
exactly modeled by an American option in which the renewable deadline is the option
expiration date. The recursive procedure is summarized by the following steps:
1. Construct the bivariate lattice for the revenue function (Sc, So) with n periods
2. Starting from the end of the lattice, the investment must be exercised at the nth
period. Option payoff F is the valuation of the finitely-lived wind farm Vj minus
investment cost I or simply the NPV
3. One period earlier, the investor has the choice to either invest at the current period
to get a contemporary NPV or wait into the next one to get a discounted option
payoff. The decision depends on which one is bigger
4. While the lattice evolves by probabilities p, the option payoffs are weighted by
risk-neutral probabilities q. Hence future option payoffs are discounted by riskfree rate r
5. Recursively compute backward until F0

129

 ( Sc , So )

puu

(uSc , uSo )

(u n 1Sc , u n 1So )

puu

(u n Sc , u n So )

F0

quu

F u ,u

F u n1 ,u n1 = max[a, b]

quu

n
n
F u n ,u n = V j (u Sc , u So ) I

pud

(uSc , dSo )

(u n Sc , u n 1dSo )

pud

qud

F u,d

F u n ,u n1d

qud

pdu

(dSc , uSo )

qdu

F d ,u

pdd

(dSc , dSo )

qdd

F d ,d

(u n 1dSc , u n So )
Fu

n1

d ,u

(u n 1dSc , u n 1dSo )
F u n1d ,u n1d

(u n Sc , u n 1dSo )
n
n 1
F u n ,u n1d = V j (u Sc , u dSo ) I

pdu
qdu
pdd
qdd

(u n 1dSc , u n So )
n 1
n
F u n1d ,u n = V j (u dSc , u So ) I

(u n 1dSc , u n 1dSo )
n 1
n 1
F u n1d ,u n1d = V j (u dSc , u dSo ) I

i
i
i

i
i
i

(ud n 1Sc , ud n1So )

F ud n1 ,ud n1

(ud n 1Sc , d n So )
F ud n1 , d n

(d n Sc , ud n 1So )

Note

F d n ,ud n1

a = V j (u n 1Sc , u n 1So ) I

( NPV )

b = e r (q uuF u n ,u n + q ud F u n ,u n1d + q duF u n1d ,u n + q dd F u n1d ,u n1d ) (discounted future option)

( d n S c , d n So )
n
n
F d n , d n = V ( d Sc , d So ) I

Fig. 5.2 Bivariate binomial lattice and iteration of its option value

5.4.4 Extension to multi-fuel displacements

Readers may be triggered to ask what would be the case if there were more than
two fuels saved by the wind power project, and how would the real option be formulated.
First of all, the proposed application of bivariate model is for demonstration only.
Certainly a typical power system could use more than two types of fossil fuels. However,
attention is paid to the notion that not all types of generation might be displaced by wind
power. It depends on priority. In fact, one would not prefer prospective wind capacity to
substitute base generation such as nuclear or large-scale hydro generation. So the critical
point is the number of fuel type actually substituted by wind. Furthermore, the current
setting of two-fuel substitution reconciles with the usual classification of peak load and
cyclic load. If the backdrop of the model is changed to an electricity market in which
peak and cyclic load bilateral contracts are prevailing, then the bivariate model still
accommodates the two contract price processes readily.

130

Nevertheless, a multi-variate binomial approximation is still desirable. A method

able to approximate a multi-variate lognormal distribution by a multi-variate binomial
process is proposed [155]. I have yet to extend my model to multi-variate cases with this
discretizing methodology but there shall be no obstacle to the extension work.

5.5

Categorization of Parameters
In the previous section, the real option formulation for the wind power project is

mostly completed. It is analytical and generic. The framework matches with a wind
power project undertaken by utility for complying any renewable energy target and
deadline. Once the wind power project is commissioned, it saves fossil fuels that would
otherwise be consumed by conventional generation. Total fossil fuel cost saving is
defined as the benefit of wind power. The real option methodology is able to determine
the optimal investment timing and the value of wind power project as a real option,
accounting various uncertainties.
For the model to be practically relevant, real world parameters have to be
considered. In particular, the time horizon is reduced to finite. For example, the target
deadline is 10 years from now on, while wind turbine equipment life is set at 15 years.
Other flexible parameters include wind power target capacity and wind turbine type
(onshore or offshore), implying different initial and maintenance cost considerations. The
detailed categorization of data and parameters is broken down into various building
blocks as explained as follow.

5.5.1 Annual average wind energy production

For any investment project, its revenue is output price multiplies quantity of
output per unit time. Time resolution of investment planning is typically a year. Hence,
revenue and cost are reported on annual basis. For wind power project, the desired output
per unit time is annual wind energy production because annual figures throughout
different years are mostly constant. Empirical evidence of this belief has been discussed

131

in chapter 3. The setting facilitates a linear relationship between fuel cost saving and fuel
price so that they share the same stochastic process, which is necessary for the problem
formulation.

5.5.2 Fuel displacement

One of the key ideas of this work is to measure the benefits of wind power in
terms of fossil fuel saving that would otherwise be incurred if the wind power project is
not built [104][120]. Fuel saving is the product of unit fuel cost and fuel consumption
displaced. While unit fuel cost is external factor (modeled by stochastic process), fuel
reduction depends on the characteristics of individual power system such as load profile,
fuel mix and unit characteristics, as well as the wind regime. It has been suggested that
long-term wind power probability distribution can be subtracted or de-convoluted from
load probability distribution to form a net load duration curve [145]. Extended from this
idea, I apply probabilistic production costing to determine expected values of fuel
reductions after wind power is added to a system.
Probabilistic production costing (PPC) is a mature topic in power system literature,
which is mainly used to calculate expected generation costs and fuel consumption. It
serves electric utility for purposes such as long-term generation planning, fuel budgeting
as well as system operation. Annual fuel budgeting is of interest here. Development of
production cost models was deterministic in very old days, without considering unit
forced outage probability. Load samples, say hourly loads for a year, are represented by
the load duration curve (LDC), which is then block-loaded by generating units with equal
incremental cost to simulate economic dispatch. Expected costs are calculated
accordingly. Simple solution method such as enumeration is used. Transition of
production costing to probabilistic one was due to the breakthrough of considering unit
forced outage probability by numerical convolution [119][140]. Generating capacities are
then de-convoluted from the load duration curve piece by piece to form an equivalent
LDC until a certain degree of reliability or un-served energy is reached.
Numerical convolution is troublesome and computationally heavy. The discrete
algorithm can be improved but its nature cannot radically change, until the development

132

of analytical representation of the equivalent LDC by cumulant/moment [124][147]. The

cumulant method [124] is applied because it has many advantages in determining
expected fuel consumptions. Cumulants are used to describe probability distribution of
generation outage, total or partial, of any unit step sizes. Similarly cumulants work for the
load probability distribution. Since the cumulants of the sum of random variables are the
sum of the individual cumulants of corresponding order, the cumulant sums of all outage
and load generate the cumulants of the equivalent load. Then the equivalent load
distribution can be recovered by Gram-Charlier approximation using the resultant
cumulants. Details of moment/cumulant and the Gram-Charlier approximation can be
referred from the Appendix or directly from the two references.
To estimate the expected fuel reductions by wind power, simply amend the
cumulants of equivalent LDC by the set of cumulants of wind power probability
distribution of the corresponding time scale. Then the differences in expected fuel burns
before and after wind power addition are attributable to wind. Pursuing to closed-form
expression, analytical forms of wind power cumulants were derived in Chapter 3, but
practically simulated values of wind power cumulants are adequate.
The type and amount of fuel wind power can displace is system dependent. When
hourly load samples of a year are condensed into a probability distribution, the load can
be roughly divided into base, cyclic and peaking. Generating units that operate all the
time a year below the minimum system demand is regarded as serving the base load [121].
Units at the middle and top of the generation stack are referred as cyclic unit and peaking
unit respectively, but normally there is no concrete definition of how short the operating
time to be a peaking unit. A typical system would have a mix of generators, e.g. nuclear
and/or large hydro plants for base load; steam units for cyclic load; combustion turbines
and/or pump storage plants for peak load. When wind power is inserted into the stack, it
is assumed to be just above the base, in other words, below any thermal generation in
order to achieve maximum fossil fuel displacement. Therefore wind power saves a
combination of coal, gas and oil.
However, a few assumptions are needed for problem simplification. First, only
two fossil fuels are assumed above the base generation stack. Primarily, cycling units are
coal-fired, whereas peaking units can be either oil or gas turbines. The implications are

133

that the value of wind power project is limited to contingent on two fuel prices instead of
three and for synchronization with the IEEE Reliability Testing System that only data of
coal-fired unit and oil combustion turbine are present for illustration. Second, the
expected heats saved for any two fuels are assumed stable year on year. It has a number
of implications as well. 1) The annual wind energy production has to be stable, which
was demonstrated by empirical data of annual averages of wind power in chapter 2. Yet
the wind power distribution and its correlation with load demand, hence any
chronological variation in fuel saving is inevitably ignored. 2) Effects of load growth and
generation mix are ignored as total fossil fuel saved is assumed to be dependent on the
wind power capacity only. What I am trying to assert is a fixed ratio of coal and oil/gas
saved, accompanied by a relatively static generation stack and time-invariant load
probability distribution. 3) Since the time resolution for investment planning is year, it is
rational and necessary to take annual expected heats saved as constant. Only if this is
assumed, the wind project benefit as defined is linearly related to annual fuel prices. To
conclude, the constituents of fossil fuels saved by wind power are assumed static and
system specific.

5.5.3 Wind turbine capital and maintenance costs

Two sets of capital and maintenance costs are allowed in general: for onshore and
offshore wind turbines. Obviously costs for sea wind turbines are much higher.
Furthermore, turbine capital cost is assumed to go down due to technological learning
whereas maintenance cost is assumed to go up, both at deterministic rates.

5.5.4 Choice of discount rate

Valuation of a project not only depends on the size of future cash flows, but is
also sensitive to the discount rate used. In this sub-section, I apply the famous financial
method, the Capital Asset Pricing Model (CAPM), to determine the appropriate discount
rate ought to be used for generation planning or investment. In my opinion, the choice of
discount rate in power system applications has been obscure. Discount rate determined by

134

the CAPM may be regarded as less biased and more endogenous from the financial
market. Discussions on CAPM, and the more general mean-variance portfolio, can be
found in many standard finance textbooks, for example [110][131][153] under the topic
of corporate finance and investment. Therefore I do not repeat them.
CAPM is one theoretical way to determine the appropriate required return of an
asset, given a portfolio of diversified assets. It says that an individual asset return depends
on the expected return of the overall market, return of a risk-free asset rf and the assets
sensitivity to systematic risk. Overall market itself is actually composed of all assets
traded in that market, therefore market return is the return of the portfolio of all the assets.
Usually, a market index is the good proxy of market portfolio. For risk-free asset, U.S.
treasury securities are regarded as the most secure; the yield is practically taken as riskfree rate. The risk of an asset is not measured alone but compared with the broad market
risk, known as systematic risk. Systematic risk is inherent in the market and affects every
asset. Risks other than systematic risk are collectively called diversifiable risk, i.e. risk
that can be eliminated through diversification. Only systematic risk is compensated and
taken into account of asset return. The asset sensitivity to systematic risk is represented
by a coefficient beta . The expected return E(ri) and beta i of any asset i are related by a
security market line, which has a slope equals to the so-called reward-to-risk ratio for any
security. Then CAPM says, in market equilibrium, the reward-to-risk ratio of any
individual security should be equal to the market reward-to-risk ratio, thus
~

E ( ri ) r f

E (rm ) r f
1

(5.46)

where m=1 by definition as it is the risk of the market relative to itself. Rearranging to
give the CAPM equation
~

E (ri ) = r f + i [ E (rm ) r f ]

(5.47)

In other words, expected return of any asset is the sum of the risk-free return and a term
multiplied by the beta of that asset. In the remaining of this work, usually is used
~

instead of E (ri ) to mean the appropriate discount rate. The term

~

E (rm ) r f

is called the market risk premium.

135

(5.48)

In order to know the appropriate project discount rate, the project risk
characteristic, i.e. its beta has to be known beforehand. Beta can be determined by two
methods. By equation, beta is the correlation of the historical return of the project with
the market.
~

i =

cov(ri , rm )
~

(5.49)

var(rm )

Econometrically, beta is also the regression coefficient of the following equation

~

ri = i + i rm

(5.50)

Therefore the key step is to choose the right proxy of return. Since we want to check the
desired return for a wind power project, one way is to choose exactly the wind power
company. However, considering industry similarity, the Dow Jones Utility average is
deemed a good choice. The market return is the broadest S&P 500 index. By regressing
the industry specific beta, the discount rate can be simply obtained by the CAPM
equation.

5.5.5 Carbon price and renewable credit

In this section I would like to elaborate on how the proposed wind power
investment model is extended to consider carbon price. Carbon (dioxide) price is usually
denominated in Euro per ton of CO2, equivalently the price of one EUA. Carbon price
adds cost on the conventional generators and as a result ties up the costs of conventional
and renewable generators. In my opinion, the effect of carbon price would otherwise be
the same if renewable generators were granted with some renewable credits. In an ideal
situation, a renewable generator should be equally better off if either conventional
generator outputs are inflated by carbon price or itself receiving equivalent amount of
renewable credit. In this section, I try to consider the effect of carbon price/renewable
credit on wind power investment.
A review of carbon price dynamics in various exchanges could be found in [154].
There is also experimental work on simulating the dynamics of renewable credit
depending on the context of system, e.g. tradable green certificates in Sweden [128]. It

136

seems there is no definite answer how high or low should renewable credit be. As a
matter of fact, carbon price is usually exogenously assumed by expert knowledge for
investment analysis whenever emission cost is invoked. Earlier in this chapter, I have
already argued the intrinsic value of wind power should depend on how much fuel it can
save for individual power system. Probabilistic production method calculates expected
fuel reductions and accordingly, emissions saved. Another simplifying method to
estimate the average amount of emissions displaced by wind power is also found [121],
though outages are completely ignored and only emission from the topmost generation
stack is modeled. Here I treat emissions mainly as carbon dioxide, and propose a way to
internalize the level of carbon price. Carbon price, a penalty to conventional generators,
can be equally regarded as subsidy to renewable generators. The idea is to attribute
carbon saving, in the equivalence of renewable credit, to the valuation of wind power
project. The analysis is based on two hypothetical scenarios of emission policy that are
intuitively appealing: carbon price is already there or comes suddenly. Their implications
to modeling are presented next.

A. Carbon price exists

Denote Sco2 (/ton) as the carbon price, Cco2 (lbs/MBtu) as the carbon dioxide
content of fossil fuel i, where i {coal , oil , gas} . Figures of Cco2,i are roughly taken to be
205, 150 and 117 for coal, oil and natural gas respectively [111]. I define the annual
saving due to carbon revenue Rco2 as the product of renewable credit price SRC
(US$/MWh) and annual energy production of the wind farm EW (MWh):
fx Sco 2 Cco 2,i H i
Rco 2 =

ic, o, g

2204

S RC EW

(5.51)

where fx is the exchange rate of dollar per euro, H is fuel heat saved, and 1 ton
approximately has 2204 lbs.
Suppose carbon price is already there and its level is constant. For simplicity,
consider infinite horizon of the stream of carbon saving, then its present value is a
perpetuity, denoted by V* equals to

137

V*=

Rco2
r

(5.52)

This term amends to the valuation of wind power project (5.31) and becomes
V ( Sc , So ) =

H c Sc

H o So

C Rco 2
+
r
r

(5.53)

With effectively the same form of V, closed form F can again be obtained.

B. Carbon price comes as a sudden policy

Suppose the implementation time of emission policy is unknown and once
arrived, the carbon price takes a step function of constant level. The value of wind power
project would also be increased, but to the extent at which reduced by the uncertain
arrival time of carbon price. Naturally, I try to model this policy jump by a Poisson
process.
~

Let T be the time to occur of an event. The usual assumption is T having an

~

exponential distribution with parameter , i.e. P{T > t} = e t so that the probability of the
event to occur at a very small time interval is constant and equal to . A constant is an
important assumption for the probability of the event to occur at any time interval to be
equal to t and independent of time. This is a necessary condition for the random event
to have homogenous Poisson distribution. A natural way to estimate probability would
be the number of occurrence per unit time [132], notably the rate function. Define the
Poisson process by analogy with the Wiener process as follow:
~
0 with probability 1 t
dq =
1 with probability t

(5.54)

Note that the probability is not the probability distribution of exponential distribution.
The Poisson process is defined as a complete ruin to zero if an event occurs with a
probability t. It may be counter-intuitive that why an upward jump was not used
directly. To explain it, lets go back to equation (5.52) as an undisturbed perpetuity and
abuse its notation as V. If there is possibility for the emissions policy to be withdrawn in
the future at a rate , then V would follow the following process:

138

~
dV
= dq
V

(5.55)

By [103], rewriting the instantaneous asset return equation for V as

rVdt = Rco 2 dt + E (dV ) = Rco 2 dt Vdt

(5.56)

Hence,
V=

Rco 2
r+

(5.57)

It can be seen that the sudden lapse of the emissions policy can be treated as the original
perpetuity but with a larger discount rate r+ [103][138]. Now what is needed is the flip
side of this scenario: from no policy to a sudden arrival. Observing the undisturbed
perpetuity given by (5.52), if V* is the present value of the step function in question, then
V *+

Rco 2 Rco 2
,
=
r +
r

(5.58)

V *=

Rco 2
r (r + )

(5.59)

implying

Hence the close-form solution of the step function is also available. One important point
to note is, , becomes the arrival rate of policy. The mean time to arrive is 1/ .
Depending on how the rate is represented, say, in occurrences per total number of years,
then the unit of time would be year.
Note that V* of the step function of carbon price is based on infinite horizon, so it
cannot be directly added to the finite valuation of wind power project, but V* is the upper
bound. Also, it is expected that the contribution of uncertain arrival of carbon price is
much smaller than that of an existing emission policy.

5.6

Parameters Estimation

5.6.1 Fossil fuel price drift and volatility

Two key parameters of any real option problem are the drifts and volatilities of
underlying processes. The benefit of wind power project is measured in terms of fossil

139

fuel cost saving, hence fossil fuel prices are the underlying processes. Historical fuel
prices are used to derive long-term average logarithmic return (drift) and volatility, which
are then used as future projections. The price data of natural gas, coal and petroleum back
to 90s are obtained from the U.S. Energy Information Administration (EIA). Note that
specifically the fuel prices for electric power use are concerned because utility fuel
contract prices should be less volatile than the corresponding quoted commodity prices in
spot market. Table 5.1 contains the historical drift and volatilities of the three fuel prices,
whereas Fig. 5.3 plots the fuel prices for electric power use in U.S. in the last decade.

Natural gas [114]

1

Coal [113]

Petroleum [113]

2.28

13.19

2010 Price (US$/MBtu)

5.29

2009 Price (US$/MBtu)

4.89

2.21

9.95

Average logarithmic return (YoY %)

4.9

4.1

10.2

Annual volatility (%)

29.1

5.3

27.8

Average logarithmic return (MoM %)

0.47

0.57

0.72 2

Monthly volatility (%)

10.5

1.8

8.9 2

Coverage

Remark: 1 converted at 1 cubic foot = 1000 Btu

2

heat oil data as proxy

Table 5.1 Drifts and volatilities derived from fossil fuel prices

140

Fuel prices for electric power in U.S.

Gas price (US$/Mbtu)

18
16

Logarithmic return
(gas)

14

US$/MBtu

12

Coal price (US$/Mbtu)

10
8

Logarithmic return
(coal)

6
4

Petroleum price
(US$/Mbtu)

Logarithmic return
(petroleum)

20

09

07
20

05
20

03
20

20

99
19

19

97

-2

01

Fig. 5.3 Fuel prices for electric power use in U.S.

5.6.2 Fossil fuel price correlation

For real option contingent on two processes, the correlation of the two price
processes has to be known. Three combinations of correlation between coal, natural gas
and oil prices are regressed as shown in Table 5.2.

gas & coal

gas & oil

oil & coal

on annual basis

-0.00444

0.845511

0.299572

on monthly basis

-0.13898

-0.03074

0.159599

Table 5.2 Correlations between three pairs of fossil fuel

5.6.3 Risk-free rate

141

The risk-free rate is referenced from the yields of U.S. long-term Treasury bond.
The long-term (>10 years) composite yields in Dec. 2010 are shown in Table 5.3. The
average 3.99% is taken as the risk-free rate.
DATE

LT Composite yield (>10 yrs)

12/1/2010

3.78

12/2/2010

3.81

12/3/2010

3.85

12/6/2010

3.77

12/7/2010

3.95

12/8/2010

4.02

12/9/2010

12/10/2010

4.04

12/13/2010

4.01

12/14/2010

4.18

12/15/2010

4.23

12/16/2010

4.17

12/17/2010

4.04

Table 5.3 U.S. Treasury bond yields (Dec 2010)

5.6.4 Risk-adjusted discount rate by CAPM

The logarithmic returns of Dow Jones Utility Average and S&P 500 from year
2000 are regressed to determine the appropriate beta for a utility project in general,
which is used as proxy for wind power project. The regressed and calculated values of
beta are given in Table 5.4. By the CAPM equation (5.47), risk-adjusted return shall be
0.13 + 0.477(0.66 0.13) = 0.38%

on a monthly basis and,

4.19 + 0.662(6.34 4.19) = 5.61%

on an annual basis.

142

Monthly
By regression By formula

0.000984

cov(ri , rm )

var(rm )

Annual

Monthly

Annual

Monthly Annual

0.477

0.662

0.018932

0.002062

0.028584

0.477

0.662

Table 5.4 Beta for wind power project

For reference, the two indices are also plotted in Fig. 5.4.
S&P500 and DJ Utility Average
1800
1600
1400
S&P 500, closing of the first
day of each month

1200
1000

Dow Jones Utility Average,

closing of the first day of
each month

800
600
400
200
12/3/2009

12/3/2006

12/3/2003

12/3/2000

12/3/1997

12/3/1994

12/3/1991

12/3/1988

12/3/1985

12/3/1982

12/3/1979

Fig. 5.4 S&P 500 and Dow Jones Utility Average since 1980

5.6.5 Wind turbine costs

Cost data assumed for onshore and offshore wind turbines are shown in Table 5.5.

Turbine type

Investment

cost

(US$/MW) [104]

Maintenance cost C

Investment cost

Maintenance

(US$/MW) [104]

trend (%/yr)

trend (%/yr)

Land

1.9M

30k

-1

Sea

2.1M

60k

-2

Table 5.5 Assumed cost data for wind turbines

143

cost

5.6.6 Fuel consumption by PPC

A. System load
A typical probability distribution of annual system load is constructed from [105]
(Table 2, 3 and 4). Only one area is considered and its peak is taken to be 2850 MW. The
first four load cumulants LKi (MW) are calculated to be:

LK1 = 1751

LK2 = 1.0645 x 105

LK3 = 1.4232 x 107

LK4 = -2.3544 x 1010

Load cumulants, together with cumulants of generators and wind turbines will be needed
in production costing.

B. Generator data
The generator data from [105] (Table 6) are employed as in Table 5.6. The
quantities of each type of unit are included for a one-area system. Unit heat rates are also
tabulated, but only at full outputs for simplicity.

Unit
U50
U350
U197
U155
U100
U400
U12
U76
U20

Quantity
6
1
3
4
3
2
5
4
4

Type
Hydro
Coal
Oil/steam
Coal
Oil/steam
Nuclear
Oil/steam
Coal
Oil/CT

Size
(MW)
50
350
197
155
100
400
12
76
20

FOR
0.01
0.08
0.05
0.04
0.04
0.12
0.02
0.02
0.1

Table 5.6 One-area generator data

144

Unit heat rate

(Btu/kWh)
0
9500
9600
9600
10000
10000
12000
12000
14499

Corresponding outage cumulants OKi (MW) are calculated in Table 5.7.

Unit
U50
U350
U197
U155
U100
U400
U12
U76
U20

OK 1

OK 2
0.50
28.00
9.85
6.20
4.00
48.00
0.24
1.52
2.00

24.75
9016.00
1843.43
922.56
384.00
16896.00
2.82
113.21
36.0

OK 3
OK 4
1.213E+03
5.820E+04
2.651E+06
6.167E+08
3.268E+05
5.115E+07
1.316E+05
1.706E+07
3.533E+04
2.955E+06
5.136E+06
9.905E+08
3.251E+01
3.586E+02
8.260E+03
5.770E+05
5.760E+02
6.624E+03

Table 5.7 Generator outage cumulants

C. Wind Generation Data

Statistical characterization of wind power distribution in a long time interval is
also needed. This is derived by common wind turbine features (cut-in, rated and cut-out
speeds) and wind speed distribution parameters. Here only two broad classes of wind
regime are considered, onshore and offshore, by differentiation in the parameter lambda
of the Rayleigh distributed wind speed. To align with outage cumulant for the purpose of
constructing equivalent load duration curve in probabilistic production costing, wind
power cumulants are also expressed in the sense of outage, i.e. the cumulant of generation
falls short of nominal capacity. However, for simplicity, wind turbine equipment failure
is not considered. Table 5.8 shows the so-called outage cumulants, WKi (MW) of a
wind farm of nominal capacity 28.5MW, which is the consequence of a hypothetical RPS
of 1% of peak system load.

Turbine type

Rayleigh

WK1

WK2

WK3

WK4

Land

21.2439

58.9723

-456.59

589.5859

Sea

17.1635

90.9248

-368.29

-8991.81

Table 5.8 Wind power under-capacity cumulants

145

D. Production costing result

The purpose of running production costing in this investment model is to estimate
amounts of fuel saved by wind power. The cumulants obtained above, together with the
analytical production costing technique [124], serve the purpose. It is run on the IEEE
Reliability Testing System (RTS-96) single area testing system. The annual expected
energy productions by all system generators before and after wind power addition are
recorded in Table 5.9, which is self-evident. Total fuel reductions annually, in terms of
Btu, are given in Table 5.10. These estimated fuel reductions are in turned used as inputs
to the investment model.

Annual expected energy Annual expected energy Annual expected energy

Unit

heat

rate (MWh) without wind (MWh) with land wind (MWh) with sea wind

Unit group (Btu/kWh)

power

power

power

U20

14499

187.1336

178.0757

173.296

U20

14499

211.9111

201.6862

196.2889

U20

14499

239.8532

228.3243

222.2364

U20

14499

271.3255

258.3424

251.4838

U76

12000

1582.211

1507.92

1468.62

U76

12000

2584.761

2467.331

2405.083

U76

12000

4141.671

3961.299

3865.475

U76

12000

6487.993

6219.324

6076.26

U12

12000

1294.054

1241.876

1214.059

U12

12000

1383.653

1328.379

1298.9

U12

12000

1478.479

1419.974

1388.76

U12

12000

1578.755

1516.88

1483.856

U12

12000

1684.707

1619.322

1584.411

U100

10000

18604.73

17921.07

17555.32

U100

10000

29837.55

28833.56

28294.83

U100

10000

45848.28

44446.73

43692.5

U155

9600

116916.1

113816.1

112141.9

U155

9600

193108.5

188781.2

186435.6

U155

9600

294329.4

288794.6

285785.5

U155

9600

418689.7

412101.6

408510.8

146

U197

9600

729831.9

720511.7

715418.3

U197

9600

975528.1

965864.8

960565.4

U197

9600

1213736

1205003

1200191

U350

9500

2483642

2473893

2468483

Wind

61404.18

95051.62

U50

419486.4

419486.4

419486.4

U50

423719.7

423719.7

423719.7

U50

426971.6

426971.6

426971.6

U50

429384.6

429384.6

429384.6

U50

431101.7

431101.7

431101.7

U50

432259.8

432259.8

432259.8

U400

10000

3078616

3078616

3078616

U400

10000

3077497

3077497

3077497

Table 5.9 Expected energy productions before and after wind capacity addition

Turbine type

EENS

without

EENS with wind

wind

Coal

saved

Oil saved (MBtu)

(MBtu)

Land

29.7715

28.3042

2.9191 x 105

3.0523 x 105

Sea

29.7715

27.5404

4.4940 x 105

4.7005 x 105

Table 5.10 Annual fuel reductions to an IEEE-RTS96 area by 28.5MW wind capacity

5.7

Numerical Example

5.7.1 Base case results

The primary result of the model is the valuation of a 28.5MW (1% of system peak
load) wind capacity, by both NPV and real option method. Along with the discussion in
this chapter, valuation so derived is based on annual time horizon. As a countercheck,
results based on monthly resolution are also offered, provided that simple division of
annual wind energy production and load demand to monthly figure is assumed valid, and
all financial parameters are transformed to their monthly equivalence. Table 5.11
summarizes the financially appealing results. The close match between NPV results of
monthly and annual cases verifies the correctness of the model. However, option results

147

by different resolutions are very largely apart. Option value increases with price volatility.
It is found that annual volatility synthetic from its monthly data are mildly bigger than the
volatility directly obtained from annual price, for the case of every fuel price. Also,
monthly horizon provides many more time steps for the lattice evolution, so the option
value is potentially inflated. Therefore, annual figures are referred to.

Land

Sea

54.15

59.58

monthly

134.89

196.24

annual

144.79

215.49

monthly

1393.8

2150.0

annual

543.52

838.54

monthly

1.3

1.1

annual

16.6

15.5

Investment cost I (MUS$)

NPV (MUS$)

Investment option F (MUS$)

IRR (%)

Table 5.11 Base case valuation of a 28.5MW wind power project

The model results can be interpreted in the following ways. As long as NPV is
positive, it signifies that the wind power project is profitable. But NPV itself may not be a
convenient appraisal of project, so internal required rate of returns (IRR) are also stated.
Note that the IRR for offshore wind farm is a little bit smaller than that of onshore, as far
as base case is concerned. It is because the benefits of both onshore and offshore wind
power are measured by fuel prices only; offshore wind power does not enjoy tariff
advantage and the increased offshore production is not enough to compensate higher
offshore wind turbine costs.
Since the computation involves iterative process of a bivariate binomial model,
which is fairly complicated, it would be informative to mention the running time required
for the program code. The time required was found to be in the order of 10-4 seconds only.
It is expected that the time burden would still be light even the binomial model is
advanced to multi-variate case.

5.7.2 Sensitivity analysis

148

The proposed model can evaluate wind power investment subject to a group of
parameters, namely, target deadline, trends in fuel prices and wind turbine costs, discount
rate, carbon price, etc. It is a powerful decision making tool, but all parameters may be
illustrated one by one. Other than the base case results, the rest of the model capability is
demonstrated by sensitivity analysis on highlighted major parameters. The case of
onshore wind turbine is described first, with the case of offshore wind turbine
understandably the same.
The first sensitivity analysis determines the optimal investment timing. The
optimal investment time is realized when the NPV, calculated based on projection of
current fuel prices, is equal to the F value. NPV varies as the fuel prices vary. Hence,
there shall be sets of fuel price combination that NPV and F are equal. This is illustrated
in Fig. 5.5 , where the F value is plotted as a horizontal plane (543.52 MUS$) as
reference, whereas the NPV is an inclined plane subject to two fuel prices simultaneously.
Their intersection is the optimal investment time, expressed in terms of fuel prices.

Fig. 5.5 Sensitivity analysis of land wind project NPV over fuel prices

The second sensitivity analysis picks the carbon price. Previously in Table 5.11,
IRR is computed to be 16.6% for the onshore wind farm. If carbon price is introduced, it
improves IRR way ahead as shown in Fig. 5.6. Equivalence of renewable credit is also in
the plot for comparison. It is estimated that a carbon price of $40 euro/ton is substantial.

149

Fig. 5.6 Sensitivity analysis of land wind project IRR over carbon price

The effects of carbon price and lambda on the project NPV are plotted in Fig. 5.7.

Fig. 5.7 Sensitivity analysis of land wind project NPV over carbon price and emission
policy arrival rate

Finally, the sensitivity analysis of offshore wind power follows as the captioned
figures.

150

Fig. 5.8 Sensitivity analysis of sea wind project NPV over fuel prices

Fig. 5.9 Sensitivity analysis of sea wind project IRR over carbon price

151

Fig. 5.10 Sensitivity analysis of sea wind project NPV over carbon price and emission
policy arrival rate

The third sensitivity analysis tries to evaluate the movement in trading values of
the synthetic wind power investment options against time. The real option values in Table
5.11 are contemporary to the year 2010. It would be informative to know how the values
have evolved from the recent past. Fig. 5.11 plots the values of the two onshore and
offshore wind project investment options for the recent five years. The movements are
mainly due to change in realizations of fuel prices and risk-adjusted discount rate that all
depend on market conditions. For simplicity, I use the single most recent data instead of
roll-over figures of historical volatility, project beta, and wind turbine cost to calculate
the two sets of chronological option values. Whereas for fuel prices, interest rate and
market indices, their rolling figures are used because those data are expected to be more
volatile in the same time span.

152

Fig. 5.11 Synthetic trading values of wind power investment real options

5.8

Summary and future works

Up to this point I have completed the presentation of the proposed wind power

investment model. It is mathematically completed and aligns with the background of

renewable policy. One key part of this model, which is the calculation of amounts of fuels
saved by wind power, has been illustrated by probabilistic production based on IEEE
reliability testing system. Financial parameters are determined in consistence with
economic and financial theories. The model generates investment recommendation to
utility that is the investor concerned, including the valuation of wind power project, its
optimal investment timing, and the appropriate level of carbon price or renewable credit.
The model is generic to input parameters; only base case results are presented, with
sensitivity analysis to selected parameters. Finally, a few concluding remarks and ideas
for future work are briefly discussed.
First, regarding the wind power capacity addition in response to any renewable
target, typically it is step-by-step. Utility has flexibilities of when to build, in whole or in

153

part, as long as the cumulative capacity of renewable meets any interim and final target.
The multi-phase wind power project is then an American style, real option-on-option
investment. However, such complex option modelling is shown to be unnecessary
because it would otherwise be the same as a single option to invest the whole capacity at
any moment [103].
Second, right now the benefit of wind power project is measured by fuel cost
saving from other units. For distribution utility or load serving entity in deregulated
market, the cost and revenue of such a project would be realized in terms of power
purchasing contracts and green power tariff respectively. The spot market transaction is
usually small compared with bilateral contract, so for simplicity it is ignored. By the
same way of probabilistic production, the wind power would save in general a
combination of base and peak loads. Base load contract covers part of the system load
that is on all a year, whereas peak load contract serves selected periods of time when the
load ramps up. Theoretically the benefit of wind power project could be inferred from a
combination of base and peak load bilateral contracts saved due to its deployment. In that
case, electricity contract prices are needed instead of fuel prices. It is envisaged that such
contract price data, physical or financial, can be referenced from power exchange [146].
Separately, there are works on price modelling of block loaded electricity contracts [157].
Third, while the subject of this renewable investment model is wind power, either
directly owned or indirectly controlled by utility, the real world relevance of the proposed
model would be even more prominent if the renewable is solar photovoltaic. It is because
recently utility-owned PV programs have been initiated in California [107] and
Massachusetts [133]. This type of program poses a fundamental change to the distribution
utility business in deregulated market in a sense that utility can own generation assets
(again) and reduce power purchases from conventional generators legitimately. However,
a modification to the equivalent load duration curve is more desirable, which is the
accommodation of statistical correlation between system load and renewable output.
Solar output is deemed more correlated with human activities than wind speed. The work
on correlated cumulants in equivalent LDC was first explored in [134], and then made
simpler in [125]. Their models would serve as building block in the suggested PV
investment model.

154

5.9

References

[100] Afzal S. Siddiqui and Chris marnay, Distributed generation investment by a

microgrid under uncertainty, Energy, vol. 33, no. 12 pp. 1729-1737, Dec 2008.
[101] Afzal S. Siddiqui, Chris Marnay and Ryan H. Wiser (Mar 2005). Real Options
Valuation of US Federal Renewable Energy Research, Development, Demonstration,
and Deployment. Ernest Orlando Lawrence Berkeley National Laboratory. [Online].
Available: www.osti.gov/bridge/servlets/purl/860783-3a2DPb/
[102] Allen J. Wood and Bruce F. Wollenberg, Power Generation Operation and
Control, New York: Wiley, 1996.
[103] Avinash K. Dixit and Robert S. Pindyck, Investment Under Uncertainty, New
Jersey: Princeton, 1994.
[104] Brendan Fox, Damian Flynn, Leslie Bryans, Nick Jenkins, David Miborrow,
Mark OMalley, Richard Watson, and Olimpo Anaya-Lara, Wind Power Integration,
Connection and System Operation Aspects, London: IET Power and Energy Series,
2007.
[105] C. Grigg, et. al, The IEEE Reliability Test System 1996, IEEE Trans. on
Power System, vol. 14, no. 3, pp.1010-1020, Aug 1999.
[106] California

Public

Utility

Commission.

[Online].

Available:

[Online].

Available:

http://www.cpuc.ca.gov/PUC/energy/Renewables/
[107] California

Public

Utility

Commission.

http://www.cpuc.ca.gov/PUC/energy/DistGen/
[108] Chung-Li Tseng and Graydon Barz, Short-Term Generation Asset Valuation, in
Proc. the 32nd Hawaii International Conference on System Sciences, 5-8 Jan 1999.
[109] Chung-Li Tseng and Graydon Barz, Short-Term Generation Asset Valuation: A
Real Options Approach, Operations Research, vol. 50, no. 2, pp. 297-310, Mar-Apr
2002.
[110] D. G. Luenberger, Investment Science, Oxford U Press, 1998.

155

[111] Ehud I. Ronn, Real Options and Energy Management: Using options
methodology to enhance capital budgeting decisions, London: Risk Books, 2002.
[112] Energy

Information

Administration.

[Online].

Available:

http://www.eia.doe.gov/cneaf/coal/quarterly/co2_article/co2.html
[113] Energy

Information

Administration.

[Online].

Available:

http://www.eia.doe.gov/cneaf/electricity/epm/table4_1.html
[114] Energy

Information

Administration.

[Online].

Available:

http://www.eia.gov/dnav/ng/ng_pri_sum_dcu_nus_m.htm
[115] F. Black and M. Scholes, The Pricing of Options and Corporate Liabilities,
Journal of Political Economy, 81, 637-659, May-Jun 1973.
[116] Fotios E. Karagiannis, Wind Energy Investments in Greece an Economical
Approach, in Proc. The 10th Mediterranean Electrotechnical Conference, MEleCon
2000, Vol.3, pp. 1141-1144.
[117] Frank H. Page, Jr. and Anthony B. Sanders, A General Derivation of the Jump
Process Option Pricing Formula, The Journal of Financial and Quantitative
Analysis, vol. 21, no. 4, pp.437-455, Dec 1986.
[118] Gurkan Kumbaroglu, Reinhard Madlener and Mustafa Demirel, A real options
evaluation model for the diffusion prospects of new renewable power generation
technologies, Energy Economics, vol. 30, no. 4, Jul 2008, pp. 1882-1908.
[119] H. Baleriaux, E. Jamoulle, Fr. Linard de Guertechin, Simulation de
Lexploitation dunparc de machines thermiques de production delectricite couples a
des stations de pompage, Revu E, (edition S. R. B. E.), vol. 5, No. 7, pp. 3-24, 1967.
[120] Hannele Holttinen and Jens Pedersen, "The Effect of Large Scale Wind Power on
a Thermal System Operation," in Proc. the 4th International Workshop on Large Scale
Integration of Wind Power and Transmission Networks for Offshore Wind Farms, pp.
E1-E7, 20-22 Oct. 2003.
[121] Hugo A. Gil and Geza Joos, Generalized Estimation of Average Displaced
Emissions by Wind Generation, IEEE Trans. on Power System, vol. 22, no. 3,
pp.1035-1043, Aug 2007.
[122] J. C. Hull, Options, Futures and Other Derivatives (6th edition), New Jersey:
Pearson Education, 2006.

156

[123] J. I. Mounoz, J. Contreras, J. Caamano and P. F. Correia, Risk Assessment of

Wind Power Generation Project Investments based on Real Options, in Proc. the
IEEE Bucharest Power Tech Conference, Romania, 28 Jun 2 July 2009.
[124] J. P. Stremel, R. T. Jenkins, R. A. Babb and W. D. Bayless, Production Costing
Using The Cumulant Method Of Representing The Equivalent Load Curve, IEEE
Trans. on PAS, vol. PAS-99, no. 5, pp.1947-1956, Sept/Oct 1980.
[125] Jeremy A. Bloom, "Probabilistic Production Costing with Dependent Generating
Sources," IEEE Trans. PAS, Vol. PAS-104, No. 8, pp. 2064-2071, Aug 1985.
[126] John C. Cox, Stephen A. Ross and Mark Rubinstein, Option Pricing: A
Simplified Approach, Journal of Financial Economics, 7, 229-263, Sept. 1979.
[127] Kaushik I. Amin, Jump Diffusion Option Valuation in Discrete Time, The
Journal of Finance, vol. 48, no. 5, pp.1833-1863, Dec 1993.
[128] Klaus Vogstad, Combining system dynamics and experimental economics to
analyze the design of Tradable Green Certificates, in Proc. the 38nd Hawaii
International Conference on System Sciences, 2005.
[129] L. Bird, B. Parsons, T. Gagliano, M. Brown, R. Wiser and M. Bolinger (July
2003). Policies and Market Factors Driving Wind Power Development in the United
States. National Renewable Energy Laboratory, Golden, Colorado. [Online].
Available: http://eetd.lbl.gov/ea/emp/reports/53554.pdf
[130] L. Bird, B. Parsons, T. Gagliano, M. Brown, R. Wiser and M. Bolinger, Policies
and Market Factors Driving Wind Power Development in the United States, Energy
Policy, vol. 33, issue 11, pp. 1397-1407, 2005.
[131] Mark Grinblatt and Sheridan Titman, Financial Markets and Corporate Strategy,
New York: McGraw-Hill, 2002.
[132] Marvin Rausand and Arnljot Hyland, System Reliability Theory: Models,
Statistical Methods, and Applications, New Jersey: Wiley, 2004.
[133] Massachusetts

Department

of

Energy

Resources.

[Online].

Available:

http://sites.google.com/site/massdgic/
[134] Michael C. Caramanis, Richard D. Tabors, Kumar S. Nochur and Fred C.
Schweppe, "The Introduction of Non-dispatchable Technologies as Decision

157

Variables in Long-term Generation Expansion Models," IEEE Trans. PAS, Vol. PAS101, No. 8, pp. 2658-2667, Aug 1982.
[135] Paolo Brandimarte, Numerical Methods in Finance and Economics, New Jersey:
Wiley, 2006.
[136] Phelim P. Boyle, Options: A Monte Carlo approach, The Journal of Financial
Economics, Vol. 4, No. 3, pp. 323-338, May 1977.
[137] Phelim P. Boyle, A Lattice Framework for Option Pricing with Two State
Variables, The Journal of Financial and Quantitative Analysis, Vol. 23, No. 1, pp. 112, Mar. 1988.
[138] R. C. Merton, Option Pricing When Underlying Stock Returns Are
Discontinuous, Journal of Financial Economics, 3, 125-144, Jul. 1975.
[139] R. C. Merton, Theory of Rational Option Pricing, Bell Journal of Economics
and Management Science, 4, 141-183, Spring 1973.
[140] R. N. Allan and Avella Corredor, Reliability and economic assessment of
generating systems containing wind energy sources, IEE Proc. C, Vol. 132, No. 1,
pp. 8-13, Jan 1985.
[141] R. R. Booth, Power System Simulation Model Based on Probability Analysis,
IEEE Trans. on PAS, vol. PAS-91, no. 1, pp.62-69, 1972.
[142] R. Wiser and O. Langniss (Nov 2001). The Renewable Portfolio Standard in
Texas: An Early Assessment. Lawrence Berkley National Laboratory. [Online].
Available: http://eetd.lbl.gov/ea/ems/reports/49107.pdf
[143] Richard L. Shockley, An Applied Course in Real Options Valuation, Mason:
Thomson, 2007.
[144] Robert L. McDonald and Daniel R. Siegel, Investment and the Valuation of
Firms When There is an Option to Shut Down, International Economic Review, vol.
26, no. 2, pp. 331-349, Jun 1985.
[145] Ronan Doherty, Hugh Outhred, Mark OMalley, "Establishing the Role that Wind
Generation May Have in Future Generation Portfolios," IEEE Trans. Power System,
Vol. 21, No. 3, pp. 1415-1422, Aug. 2006.
[146] N2EX. [Online]. Available: http://www.n2ex.com/

158

[147] N. S. Rau, P. Toy and K. F. Schenk, Expected Energy Production Costs By The
Method Of Moments, IEEE Trans. on PAS, vol. PAS-99, no. 5, pp.1908-1917,
Sept/Oct 1980.
[148] S. C. Myers, Determinants of Corporate Borrowing, Journal of Financial
Economics, 5 (2), 1977, pp. 147-175.
[149] Shijie Deng, Financial methods in competitive electricity markets, Ph.D.
dissertation, University of California, Berkeley, CA, 1998.
[150] Shijie Deng, Blake Johnson and Aram Sogomonian, Spark Spread Options and
the Valuation of Electricity Generation Assets, in Proc. the 32nd Hawaii
International Conference on System Sciences, 5-8 Jan 1999.
[151] Shijie Deng, Blake Johnson and Aram Sogomonian, Exotic electricity options
and the valuation of electricity generation and transmission assets, Decision Support
Systems, 30, pp. 383-392, Jan 2001.
[152] Shmuel S. Oren, Integrating real and financial options in demand-side electricity
contracts, Decision Support Systems, 30, 2001, pp. 279-288.
[153] Stephen A. Ross, Randolph W. Westerfield and Bradford D. Jordan, Corporate
Finance Fundamentals, New York: McGraw-Hill, 2006.
[154] Tarjei Kristiansen, Richard Wolbers, Tom Eikmans and Frank Reffel Carbon
Risk Management, in Proc. 9th International Conference on Probabilistic Methods
Applied to Power Systems, KTH, Stockholm, Sweden, 11-15 June 2006.
[155] Teng-suan Ho, Richard C. Stapleton and Marti G. Subrahmanyam, Multivariate
binomial approximations for asset prices with nonstationary variance and covariance
characteristics, The Review of Financial Studies, Vol. 8, No. 4, 1995, pp 1125-1152.
[156] Wang Yu, Gerald B. Sheble, Joao A. Pecas Lopes and Manuel Antonio Matos,
Valuation of Switchable Tariff for Wind Energy, Electric Power Systems Research,
76, 2006, pp. 382-388.
[157] Xian Zhang, Xifan Wang and Y. H. Song, "Modeling and Pricing of Block
Flexible Electricity Contracts," IEEE Trans. Power System, Vol. 18, No. 4, pp. 13821388, Nov. 2003.
[158] Y. Hou and F. F. Wu, Valuation of Generator Profit from Spot Market:
Analytical Approach, submitted to IEEE Trans. Power System.

159

Chapter 6

6 Conclusion
It comes to the end of this work and I conclude the whole thesis by summarizing
in three main portions. First, the key contribution of each chapter is restated. Second, the
logical relationships chapter by chapter are reiterated. Third, I emphasize that this thesis
has adopted critical approaches to analyze wind power investment problems.
Chapter 1 reviews the historical development of generation expansion
methodology that showed a shift in mindset from planning to investment-oriented
analysis along with the transition of market regime. The rationale of studying wind power
investment problems is initiated. Chapter 2 is a survey of contemporary market rules for
wind power in which two major scenarios are identified for subsequent investment
modelling. Chapter 3 furnished a probabilistic wind power generation model with
derivation of its higher order statistical formulae. It has conducted empirical analysis on
wind speed wake effect and multi-year production trend that are necessary for subsequent
investment analysis. Then a stochastic optimisation framework for wind power project
under fixed tariff is proposed in chapter 4. The model is able to determine the optimal
capital structure to maximize levered wind power project value. Another real option
investment model is proposed in chapter 5 for analyzing wind power project owned by
utility. This model incorporates power system production costing result and is able to
determine the optimal investment timing subject to fossil fuel cost saving.
The organization of these chapters forms a logical sequence of investigation to the
research question. It started with an observation that electricity market structure is mostly
changing towards competitive market so that analytical tools have to accommodate the
shift in climax. In the meantime, many policy drivers for renewable have emerged and
dictated the pace of wind power deployment. I emphasized that the analysis of wind
power investment only makes sense when its corresponding market and policy scenario is
referred to. Such requirement leads to writing the survey in chapter 2, which forms the

160

basis of various investment models. Meanwhile, wind energy production characteristics

are expected to have direct impact on the profitability of wind farm. Therefore, chapter 3
offers such background information, in particular the wind power cumulants to
incorporate with probabilistic production costing. Finally the two investment models
proposed in chapter 4 and 5 follow logically from the two wind power market scenarios
identified in chapter 2.
The thesis aims at developing financially consistent investment models for wind
power that are able to account for relevant market mechanisms and address technical
considerations. The overall aim has been fulfilled by adopting critical analysis.
Specifically, both finance and power system theories are applied to the outmost extent in
the two investment models. First, the optimal capital structure of a feed-in tariff wind
power project is determined by ModiglianiMiller theorem, and its maximum levered
project value is solved by stochastic optimization subject to value-at-risk debt constraint.
The model can tell independent power producer the optimal leverage and maximum
achievable valuation of prospective wind power project. Second, fuel cost saving
capability of wind farm is computed and its long-term, time-dependent benefit to a utility
is assessed by Black-Scholes option pricing theory. The real option model can advise the
utility the optimal time to invest wind capacity in view of a renewable energy target and
date. The analytical approaches presented in this thesis are deemed contemporary to wind
power investment.

161

7 Appendices

I. Wind Power Probability Distribution

The power curve of a wind turbine is empirical in practice, particularly for the
region between win and wr. The region consists of discrete data points stating the
conversion ratio from wind speed to power. Instead of representing these points by
straight line or any curve fitting, they can be modeled by piecewise-linear segments.
Theoretically we can represent an empirical power curve of wind turbine by many
piecewise-linear segments as follow.
0
a w+b
1
1
~
a2 w + b2
g=

an w + bn

g m

0 w win or w wout
win < w w1
w1 < w w 2

w n 1 < w < w r
w r w < wout

(I.1)

where
g w1 g 1win
g1 g 0
, b1 = 0
w1 win
w1 win
g2 g 1
g 1w 2 g 2 w1
a2 =
, b2 =
w 2 w1
w 2 w1

g g n 1
g w r g m wn 1
an = m
, bn = n 1
w r w n 1
w r w n 1
a1 =

(I.2)

and g0=0, g1, g2, , gn-1, gm correspond to the wind turbine output at win, w1, w2, , wn-1,
wr respectively.

Wind Power Probability Distribution

For demonstration, suppose there is only one data point (w1, g1) in between (win, 0)
and (wr, gm). Hence the curve part is approximated by two piecewise-linear segments. By
considering different regions of wind speed in expression (I.1) and integrating, we obtain
the cumulative distribution function (CDF) of wind turbine output power:

162

P{g = 0} =

win

f ( )d +

wout

f ( )d = 1 e

( win)2

+e

( wout )2

For 0 < x g1
~

P{g x} = P{g = 0} + P{0 < g x}

= 1 e
= 1+ e

( win)2

+e

( wout )2

( wout )2

+ [e

( win)2

x b1 2
)
a1

x b1 2
)
(
a1

For g1 < x < g m

~

P{g x} = 1 + e
= 1+ e

( wout )2

( wout )2

e
e

g1 b1 2
)
a1

+ [e

( w1)2

x b2 2
)
a2

x b2 2
)
(
a2

(I.3)

P{g g m } = 1

The corresponding probability density function (PDF) is obtained by differentiating the

CDF:

( win )2
( w out )2
(1 e + e ) ( x)
x=0

x b1 2
)

x b (
2( 2 21 )e a1

a1
0 < x g1

f ( x) =
x b2 2
g1 < x < g m
)

x b (
2( 2 22 )e a2

a2

wr 2
( w out ) 2
( )
x = gm
e ) ( x g m )
(e

(I.4)

where ( x) is the delta function at x=0 to maintain the derivative-integral relation

between the CDF and PDF.

Statistics of the Probabilistic Wind Power

Suppose f(x) is the wind power density function, its rth moment (about zero) and
rth central moment are given respectively as

mr = x r f ( x)dx

and

(I.5)

cr = ( x m1 )r f ( x)dx

163

(I.6)

In particular, m1 is the mean and c1=0. For completeness we may also define m0=c0=1.
By binomial theorem, any rth moment can be expressed in terms of the rth and lower order
central moments. Therefore,
r
r
mr = cr j m1j
j =0 j

(I.7)

or vice versa,
r
r
cr = mr j (m1 ) j
j =0 j

(I.8)

They can be deduced interchangeably and it is more convenient to start with the mean m1
and then all higher order moments can be generated. Consider (I.5),

x r f ( x)dx =

+0

0r (1 e

g1

gm

x 2(

g0

+
=

g1

g0

gm

win

x b1
a12 2

gmr (e

x r 2(

wr

x b1
a12

)e
2

)2

+e
(

)e

wout

x b1 2
)
a1

)2

) ( x)dx

dx +

gm

x b2

x 2(

a22 2

g1

)2

x b1 2
)
a1

wout

)2

dx +

)e

x b2 2
)
a2

dx

(I.9)

) ( x gm )dx

gm

x r 2(

x b2

g1

a22 2

)e

x b2 2
)
a2

dx + g mr (e

wr

)2

wout

)2

The first two terms are the same except with different parameters for different segments,
implying that the formula derived for one segment should work for others. For simplicity,
we abuse the notation of a and b by omitting their subscripts. Let y = x b , then
a

gm

g1

x 2(

x b
2 2

)e

x b 2
)
a dx

=
=

g m b
a
g1 b 2 y ( ya
a

k2

+ b)r e y dy

(I.10)
r y2

2 y( ya + b) e

dy

k1

where the lower and upper limits of the integration domain are denoted by k1 and k2
respectively for convenience (for the segment g0 to g1, let the limits be l1 and l2
respectively). For an integral of the form
g=

2
y n 1 dh
, = 2 ye y and
2 dx

y n e y dy

, denote it by I n . Let

consider integration by parts, then recursively

In =

y n 1 y 2 n 1
e +
I n2
2
2

For n=1, it is observed that

164

for n 2

(I.11)

1 2
I1 = e y
2

(I.12)

For n=0,
( I0 )2 =

( u 2 + v2 )

dudv =

r 2

rdrd

(I.13)

After considering the integration domain,

1 2
1 2 2
I1 = [ e y ]kk2 = (e k 1 e k 2 )
1
2
2

I0 =

k2

e y dy =

k1

e y dy +

k1

k2

(I.14)
2

e y dy

(I.15)

Hence I 0 can be expressed in terms of error function erf(.):

I0 =

(I.16)

{erf (k2 ) erf (k1 )}

Although an error function does not have closed form solution, one can expand the
2

integrand e y by Taylor series and integrate term by term to obtain an approximation. In

modern computer programs, it is readily generated by numerical integration.
By now we have obtained a recursive formula of any higher order of the
integral I n once I1 and I0 are generated. Let us show the derivation up to I3 for the
calculation of the mean and variance.

A. Mean
The mean of the random wind power is obtained by setting r=1 in (I.5). From
(I.10),

k2

2 y ( ya + b)e y dy

k1

= 2 a

k2

y 2 e y dy + 2b

k1

k2

ye y dy

k1

(I.17)

= 2a I 2 + 2bI1
2
1

2
2

= a ([ ye y ]kk2 + I 0 ) + b(e k e k )
1

2
2
a

=
{erf ( k2 ) erf ( k1 )} + g1e k 1 g m e k 2
2

Adding the results of two segments together, the mean or first order moment m1 is

165

x = m1 =

x f ( x)dx =

2
2
a1

{erf (l2 ) erf (l1 )} + g0 e l 1 g1e l 2

2

2
2
( r )2
(
a

+ 2
{erf (k2 ) erf (k1 )} + g1e k 1 g m e k 2 + gm (e e
2

wout

(I.18)
)2

B. Variance
~

The variance of x , or the second order central moment, is given by (I.6).

Conveniently, from (I.8) we can use
(I.19)

2 = c 2 = m2 m12

Therefore we need m2 first. Setting r=2 in (I.10),

k2

k1

2 y ( ya + b) 2e y dy =
2 2

k2

(2a 2 2 y 3 + 4ab y 2 + 2b 2 y )e y dy

k1

= 2a I 3 + 4ab I 2 + 2b I1
2

(I.20)

= a 2 2 ([ y 2 e y ]kk2 + 2 I1 ) + 2ab ([ ye y ]kk2 + I 0 ) + 2b2 I1

1

= (a 2 2 k 12 + a 2 2 + 2ab k1 + b 2 )e k 1 ( a 2 2 k 22 + a 2 2 + 2ab k2 + b 2 )e k 2 + 2ab I 0

2

= {g 12 + (a ) 2 }e k 1 {g 2m + ( a ) 2 }e k 2 + ab {erf (k2 ) erf (k1 )}

Adding the results of two segments together, then m2:

2

m2 = {g 12 + (a2 ) 2 }e k 1 {g 2m + ( a2 ) 2 }e k 2 + a2b2 {erf ( k2 ) erf ( k1 )}

2

+ {g 02 + (a1 ) 2 }e l 1 {g 12 + ( a1 ) 2 }e l 2 + a1b1 {erf (l2 ) erf (l1 )}

+ g m2 (e

( w r )2

( wout )2

(I.21)

The formula of variance follows when m1 and m2 are known.

C. Skewness and Kurtosis

It becomes clear that any higher order moments or central moments can be built
for better characterization of the PDF. Also eventually any orders of cumulant can be
calculated by (I.9) and those cumulants could readily be used in the Gram-Charlier series.
Without showing the steps, the third and fourth cumulants are stated as follow.
Third cumulant:
3 = c3

(I.22)

= m3 3m2 m1 + 2m13

where

166

2
3

m3 = {g 13 + a22 2 ( g1 + b2 )}e k 1
2
2
a 2 2
3

{g 3m + a22 2 ( g m + b2 )}e k 2 + 3a2 ( 2 + b22 )

{erf (k2 ) erf (k1 )}
2
2
2
2
3

+{g 30 + a12 2 ( g 0 + b1 )}e l 1

2
2
a2 2

{g 13 + a12 2 ( g1 + b1 )}e l 2 + 3a1 ( 1

+ b12 )
{erf (l2 ) erf (l1 )}
2
2
2

+ g m3 (e

wr

)2

wout

)2

(I.23)

Fourth cumulant:
4 = c4 3c22

(I.24)

= (m4 4m3 m1 + 6m2 m12 3m14 ) 3(m2 m12 ) 2

where
2

m4 = {g14 + 2(a2 ) 2 ( g12 + b2 g1 + a22 2 + b22 )}e k 1

2

{g m4 + 2(a2 )2 ( g m2 + b2 g m + a22 2 + b22 )}e k 2 + (3a23b2 3 + 2a2 b23 ) {erf (k2 ) erf (k1 )}
2

(I.25)

+{g04 + 2( a1 ) 2 ( g02 + b1 g0 + a12 2 + b12 )}e l 1

2
2

{g14 + 2(a1 )2 ( g12 + b1 g1 + a12 2 + b12 )}e l + (3a13b1 3 + 2a1b13 ) {erf (l2 ) erf (l1 )}

+ g m4 (e

( w r )2

( wout ) 2

By definition, we can also find skewness and kurtosis excess respectively as:
1 =
2 =

c3

(I.26)

3
c4

167

(I.27)

II. M&M Propositions I and II with Corporate Taxes

A. Corollary 1
Given
WACC =

E
D
RE +
RD (1 tC )
VL
VL

RE = RU + ( RU RD )

and

D
(1 tC ) ,
E

(II.1)
(II.2)

substitute the second equation into the first one:

WACC =
=

E
D
D
[ RU + ( RU RD ) (1 tC )] +
RD (1 tC )
VL
E
VL

E
D
D
RU + ( RU RD ) (1 tC ) +
RD (1 tC )
VL
VL
VL

E
D
=
RU + RU
(1 tC )
VL
VL
= RU (1

(II.3)

D
tC )
VL

Hence WACC is expressed in terms of RU.

B. Corollary 2
By definition of WACC, discounting OCF by WACC should yield VL:
OCF
OCF
=
WACC R (1 D t )
U
C
VL
=

VU
D
(1 tC )
VL

VLVU
(VL DtC )

VLVU
VU

= VL

C. Corollary 3

168

(II.4)

Here I start from multiplying equity E by RE,

D
(1 tC )]
E
= ERU + ( RU RD ) D(1 tC )
E RE = E[ RU + ( RU RD )

= ( E + D) RU RU DtC RD D(1 tC )
= VL RU RU DtC RD D(1 tC )
= (VU + tC D) RU RU DtC RD D(1 tC )
= VU RU RD D(1 tC )
= OCF iD (1 tC )

169

(II.5)

III. Solution of an Ordinary Second Order Non-homogenous

Differential Equation
The differential equation of interest is

1 2 2 2V
V
S 2 +(r )S rV + = 0,
2
S
S

(III.1)

1 2 2 2V
V
S 2 +(r )S rV = 0 .
2
S
S

(III.2)

in which its homogenous part is

Since (III.2) is linear in the dependent variable V and its derivatives, its general solution
is a linear combination of two independent solutions. Their functional forms depend on
boundary condition of the differential equation. Since V(0) = 0, try AP which satisfies
(III.2) by substitution provided that is a root of the following quadratic equation:
1 2
( 1) + (r ) r = 0
2

(III.3)

The two roots of are

1 =

(r ) 1 2 2r
1 (r )
[
] + 2 >1

+
2
2
2
2

(III.4)

2=

(r ) 1 2 2r
1 (r )
[
] + 2 <0

2
2
2
2

(III.5)

Therefore the solution to (III.2) is

V(S) = A1S1 + A2S2

(III.6)

where A1 and A2 are constants to be determined. The complete solution to (III.1) is the
sum of a general solution, which is (III.6), and a particular solution to be found by
substitution. Suppose (S) takes the form of:
(S ) = R(S ) C

(III.7)

where R(S) is in general a function of S and C is a constant. With a few trials, the
following particular solution is spotted.

V(S) =

R(S) C

170

(III.8)

By substitution, (III.7) and (III.8) simultaneously satisfy (III.1), implying the choice of
particular solution is correct. Hence the complete solution of the second order nonhomogenous differential equation is

V(S) = A1[R(S)]1 + A2[R(S)]2 +

171

R(S) C

(III.9)

IV. Solving Ordinary Second Order Homogenous Differential

Equation with Boundary Conditions
For a second order homogenous differential equation

1 2 2
S F"(S) + (r )SF '(S) rF(S) = 0 ,
2

(IV.1)

its solution is
F ( S ) = B1S 1

(IV.2)

where
1 =

(r ) 1 2 2r
1 (r )
[
] + 2

+
2
2
2
2

(IV.3)

and subject to boundary conditions:

F (0) = 0

(IV.4)

F ( S *) = V ( S *) I

(IV.5)

F '( S *) = V '( S *)

(IV.6)

Based on our own definition of (S) for (III.1), the corresponding solution form of
V(S) from (III.8) is restated as follow.

V(S) =

R(S) C

(IV.7)

Then (IV.5) and (IV.6) can be rewritten respectively as

B1[ R( S *)] 1 =

R( S *) C
I

1B1[ R( S *)] 11 =

(IV.8)
(IV.9)

Solving coefficient B1 and the so-called optimal threshold R(S*) from the above two
equations yield:
B1 =

( 1 1) 11

(C / r + I ) 11( 1) 1

R( S *) =

1 (C / r + I )
1 1

172

(IV.10)
(IV.11)

V. Matching Mean and Variance of a Bivariate Binomial

Lattice with Geometric Brownian Motions
Given two correlated geometric Brownian motions processes:
~

(V.1)

d S 1 = 1 S 1 dt + 1 S 1 d Z 1
~

(V.2)

d S 2 = 2 S 2 dt + 2 S 2 d Z 2

Consider a multiplicative increment to S in the form of

~

S i (t + t ) = w i
e
S i(t )

(V.3)

~
ai with probability p
wi =
ai with probability 1 p

(V.4)

where

and i=1,2. After taking logarithm on (V.3) by xi = ln(Si), it becomes

~

wi = ln S i (t + t ) ln S i (t ) = x i

(V.5)

Define the discretization by four possible outcomes in combinations of x1 and

x2 as in Table V.1:
Probability of

Value changed of the underlying variable given

event

that event has occurred

Variable 1

Variable 2

puu

x1

x2

pud

x1

-x2

Pdu

-x1

x2

Pdd

-x1

-x2

Table V.1 Discretization outcomes of two correlated GBMs

Then the means and variances of the correlated GBM processes and the bivariate
binomial lattice are matched in the logarithm of Si so that the increment xi becomes
additive. Also, apply the Ito lemma on the multiplicative model (V1) and (V.2) to obtain

173

d x i = v idt + id z i
vi = i

i2

(V.6)
(V.7)

Matching means of the increments xi with the two drift rates of (V.6) yield
~

E ( x1) = ( p uu + p ud ) x1 + ( p du + p dd )( x1) = v1t

~

E ( x 2) = ( p uu + p du )x 2 + ( p ud + p dd )( x 2) = v 2t

(V.8)
(V.9)

For matching variances, first consider the identity Var(x) = E(x2) E(x)2 and then neglect
second order term in t, so that
~

E( x1) = ( puu + pud + pdu + pdd)(x1)2 = 12t + v12t 2 12t

~

E( x 2) = ( p uu + p ud + p du + p dd )( x 2) 2 = 22t + v 22t 2 22t

(V.10)
(V.11)

Since the probabilities sum to one, (V.10) and (V.11) are simplified to
(V.12)

x i = i t

Up to now there are three equations in terms of four probabilities. One more equation is
needed in order to solve all the four probabilities. By accounting the covariance of x1
and x2, we have
~

E( x1 x2) = ( puu pud pdu + p dd )(x1x 2)

= 1 2t + v1v2t 2 1 2t

(V.13)

All together a system of four equations in probabilities is obtained as follow:

p uu + p ud p du p dd =

v1 t

p uu p ud + p du p dd =

1
v 2 t

p uu p ud p du + p dd =
p uu + p ud + p du + p dd = 1

The system of equations can be solved by inverting the matrix.

174

(V.14)

1
p uu = [1 +
4
1
p ud = [1 +
4
1
p du = [1 +
4
1
p dd = [1 +
4

t (

v1

v2

v2

1 2

t (

v1

1 2

t (
t (

v1

) + ]
) ]

v2

v2

1 2
v1

1 2

(V.15)
) ]
) + ]

Note that (V.15) can be expressed in terms of i and i by the substitution of vi using
(V.7). Finally from (V.4), if we let u i = e a i and di=1/ui, (V.12) implies
u i = e i

(V.16)

Therefore, ui and di together with all the four probabilities completely specify the
bivariate binomial model. For risk-neutral valuation of option, the drift rates i are further
replaced by the risk-free rate r.

175

VI. Moment/Cumulant and Gram-Charlier series

~

Consider a random variable x which is continuous in a range, its probability

distribution or cumulative distribution function (CDF) is given by
~

P ( x x ) = F ( x)

(VI.1)

F(x) is everywhere continuous and monotonically non-decreasing. Assume it has a

derivative so that

dF ( x)
= F '( x) = f ( x)
dx

(VI.2)

f(x) is called the probability density function (PDF). The rth moment (about zero) is given
by

mr = xr f ( x)dx

(VI.3)

The rth central moment is given by

cr = ( x m1 )r f ( x)dx

(VI.4)

In particular, c1 = 0 . And by definition m0 = c0 = 1 . By binomial theorem, any rth moment

can be expressed in terms of the rth and lower order central moments:
r
r
mr = cr j m1j
j =0 j

(VI.5)

r
cr = mr j (m1 ) j
j =0 j

(VI.6)

Or vice versa:
r

The moments are a set but not the only set of constants that characterizes a
distribution. If we consider the characteristic function of the density function,

( ) = ei x f ( x)dx

(VI.7)

where i = 1 , and use Taylor series to expand the exponential function about zero, then

(i )
mr
r =0 r !

( ) =

176

(VI.8)

One can also consider another set of constants called cumulant, r , as defined as the
coefficients in the following identity in :
r

(i )
(i )
exp(
mr
r) =
r!
r!
r =1
r =0

(VI.9)

By definition, r is the coefficient of the series

(i )
in the logarithmic characteristic
r!

function ln ( ) . I readily write down the formulae of the first four cumulants in terms of
moments and central moments as follow.

1 = m1
2 = c2
3 = c3

(VI.10)

4 = c4 3c22
The density function of a standardized normal random variable

f ( z) =

1
1
exp( z 2) ,
2
2

(VI.11)

when being differentiated successively, yields the Hermite polynomials Hk(z) by the
identity:

dk
(1)
f ( z) H k ( z) f ( z)
dz k
k

(VI.12)

The first five polynomials are

H0 ( z) = 1
H1 ( z ) = z
H2 ( z) = z 2 1

(VI.13)

H 3 ( z ) = z 3z
H4 ( z) = z 4 6z 2 + 3
Now assume that a density function can be expanded formally in a series of derivatives of
f(z). One can arrive to the so-called Gram-Charlier series of Type A:

1
exp( z 2)
2
2 [1 + 3 H ( z ) + 4 H ( z ) + 5 H ( z ) + 6 + 10 3 H ( z ) + ...] (VI.14)
f ( z) =
3
4
5
6
3!
4!
5!
6!
2

177

The idea is to approximate a density function by finite number of terms. In what follows I
use up to the fourth cumulant 4 to construct the series, which reconciles to a reference
mentioned in the discussion of Kendalls Advanced Theory of Statistics.
An important application of Gram-Charlier series is we can obtain approximation
~

to the density function of the sum of n random variables. Let X n = x1 + x2 + ... + xn , observe
~

that if and only if all xs are independent, we can write the PDF of X n as multiplication of
all individual PDFs, i.e.

f ( X n ) = f1 ( x1 ) f 2 ( x2 )... f n ( xn )

(VI.15)

Then by the definition of logarithmic characteristic function, cumulants of independent

~

variables are additive. In other words, each rth cumulant of X n is the sum of the
~

corresponding rth cumulants of x1 to xn :

n

r ( X n ) = r ( xt )

(VI.16)

t =1

The Gram-Charlier series (VI.14) up to the fourth cumulant can satisfactorily

approximate the PDF of the sum and we obtain

( x )2
)
2 2 [1 + 3 H ( x ) + 4 H ( x )] (VI.17)
3
4
3!
4!

exp(
f ( x) =

where and 2 are the sum of the means and sum of the variances of the n random
~

xt respectively, and
n

r =

r ( X n )
~

(x )

=
r

[ 2 ( X n )]

t =1

r
~

is known as the rth standardized cumulant of X n .

178

(VI.18)

8 Publications
[1] Henry M. K. Cheng, Yunhe Hou, and Felix Wu, Probabilistic wind power generation
model: Derivation and applications, NAUN International Journal of Energy, Vol. 5,
No. 2, 2011.
[2] Henry M. K. Cheng, Yunhe Hou, and Felix Wu, Real option investment model for
utility wind power project, IEEE Trans. Power Syst., submitted for publication.
[3] Henry M. K. Cheng, Stochastic optimisation of independent wind power project
valuation under feed-in tariff and its optimal capital structure, working paper.
[4] Henry M. K. Cheng, Yunhe Hou, and Felix Wu, Wind Power Investment in Thermal
System and Emissions Reduction, in Proc. 2010 IEEE PES General Meeting,
Minneapolis, 25-29 July 2010.
[5] Henry M. K. Cheng, Yunhe Hou, and Felix Wu, Probability Distribution of the
Output Power of Wind Turbine, in Proc. the 16th International Conference on
Electrical Engineering, Busan, Korea, 11-14 July 2010.

179