Ann Oper Res (2011) 186:465–490

DOI 10.1007/s10479-011-0856-9

A decision-making tool for project investments based

on real options: the case of wind power generation

J.I. Muñoz · J. Contreras · J. Caamaño · P.F. Correia

Published online: 10 May 2011

© Springer Science+Business Media, LLC 2011

Abstract This paper presents how to apply a decision-making tool based on real options to
assess the investment in a wind energy plant.
The work shows six case studies where the main model’s parameters are analyzed. The
uncertainty coming from wind regimes is simulated by using Weibull distributions and the
volatility of market prices is obtained from the mean reverting process of the Ornstein-
Uhlenbeck type, also known as Geometric Mean Reversion (GMR). From these and other
values, such as investment and maintenance costs, the Net Present Value (NPV) curve, made
up of different values of NPV in different periods of the investment is calculated, as well as
its average volatility.
Having the key parameters of the model, a real options valuation method is applied. The
volatility, strength of reversion and long-term trend of the NPV curve reflecting different
periods are inserted into a trinomial investment option valuation tree. From this, it is possi-
ble to calculate the probabilities of investing right now (exercise), deferring the investment
(wait), or not investing at all (abandon).
This powerful decision-making tool allows wind energy investors to decide whether to
invest in many different scenarios.

Keywords Real options · Risk management · Monte Carlo · Wind energy

J.I. Muñoz () · J. Contreras

E.T.S. de Ingenieros Industriales, University of Castilla–La Mancha, Avda. Camilo José Cela s/n.,
13071 Ciudad Real, Spain
e-mail: joseignacio.munoz@uclm.es
url: www.ulcm.es

J. Caamaño
E.T.S. de Ingenieros Industriales, University of País Vasco, Alameda de Urquijo, s/n., 48013 Bilbao,
Spain

P.F. Correia
Instituto Superior Técnico, Technical University of Lisboa, Avda. Rovisco Pais, 1049-001 Lisbon,
Portugal
466 Ann Oper Res (2011) 186:465–490

1 Introduction

World population and energy demand are rapidly increasing. According to current rates, in
less than 30 years there will not be enough fuel-based energy production. Due to the high
variability of fossil commodities prices, the need to reduce dependence on conventional
energy sources and to protect the environment has fomented the use of clean sources of
energy.
Renewable energy sources have been supported and subsidized by different countries.
Among them, wind energy seems to be the most successful in terms of market penetration.
Increasing generation capacity in liberalized power markets requires the producers to ac-
count for future long-term uncertainties. In particular, this is true for wind producers since
the inherent intermittency of its energy production is due to wind regimes. Due to the consid-
erable lifetime of these plants and the mentioned uncertainties, traditional discounted cash
flow methods used to obtain the NPV of the investment are unsuitable for these projects.
To take into account uncertainty, real options models can be used to develop flexible
investment strategies. Depending on the present value in a certain time, the decision maker
can decide to execute, wait, or abandon the construction project of a power plant.
The impact of renewable energy is traditionally measured from the societal perspective
and focuses on social welfare and other overall system benefits (El-Khattam et al. 2004).
A complementary approach is to assume that the builder of a new wind plant wants to
maximize its own profits and has to decide whether or not to build and when. This requires
careful evaluation of the financial aspects involved (Khatib 2003) and adequate analytical
decision tools are needed.
Real options analysis is widely used as a tool to help decision making in many fields
(Dixit and Pindyck 1994; Trigeorgis 1996). It is successfully applied both in operation and
planning of power systems:
• to perform a generation asset valuation; and
• to find the optimal timing for new generation investments under uncertainty.
Regarding the first application, the operation of a power plant can be studied from a
generation asset perspective, where fuel and electricity prices are modeled using mean re-
verting processes (Tseng and Barz 2002; Tseng 2000) considering seasonal patterns (Lucía
and Schwartz 2002) as any other commodities (Schwartz 1997). Financial option theory is
used to value the plant using the concept of Value at Risk. Sophisticated models add opera-
tional constraints, as in Deng and Oren (2003), or define multi-stage stochastic models of the
operational decisions, with two-factor lattices describing fuel and electricity prices, such as
in Tseng and Lin (2007). Another multi-stage investment under market uncertainty model,
where the authors use a stochastic process to estimate the power plant’s present value, is
shown in Correia et al. (2008).
Secondly, the problem of investment is addressed using the real options approach. In
Fleten and Näsäkkälä (2003), spark spread and emission costs are used to value a gas-fired
plant. The same authors study the upgrade of a base-load plant in Näsäkkälä and Fleten
(2005) into a peak-load plant. The evaluation of an investment for two inter-related plant
projects is described in Wang and Min (2006), where a quadrinomial tree represents the mar-
ket value of the units. In Fleten et al. (2007), a real options method to evaluate investments in
renewable generation, including wind generation, makes extensive use of price volatility to
decide upon investment timing. In addition, the NPV break even price is determined under
price uncertainty.
Ann Oper Res (2011) 186:465–490 467

The remaining sections of the paper are organized as follows, Sect. 2 presents the model
used and describes in detail the main steps: wind speed simulation, electricity price mod-
eling and parameter estimation, NPV curve calculation and the piecewise estimation of its
parameters and, finally, how to get the real option value and its associated probabilities,
as well as developing the risk profile characterization of the model. Section 3 provides six
illustrative examples and several results, and Sect. 4 presents the conclusions.

2 Proposed investment model

The following flowchart in Fig. 1 shows the structure of the model.

Briefly, the main steps that are necessary to obtain the final value of the real option related
to the investment in each scenario are:
(1) Wind speed estimation using scale and shape parameters of a Weibull distribution. These
parameters depend on the location of the power plant. The model generates hourly wind
speeds.
(2) Calculation of the production curve of the wind turbine using the hourly wind speed
data transformed into monthly data.
(3) Calculation of the electricity prices to remunerate the wind producer using a Geometric
Mean Reversion (GMR) model that takes into account parameters (volatility, strength
of reversion and trend) obtained from the Spanish market’s historical data. It is assumed
that wind generation is paid at the spot prices resulting from the day-ahead electricity
market, but any incentive scheme can be easily added to the model.
(4) Addition of the investment cost estimation from national standards, the stochastic cash
flows and NPVs are obtained.
(5) Estimation of the NPV curves by shifting the initial time for investment. Monte Carlo
simulation is used in order to obtain enough data points to achieve high accuracy in the
results. The NPV curves represent the evolution of the NPVs that would result if the
investment were done at different times.
(6) A real options methodology is applied from these NPV curves and their GMR parame-
ters:
(a) a trinomial investment decision tree is built, where the values of the parameters are
only valid for specific time intervals, since a unique GMR model only holds for
prices (not NPVs) during the entire lifetime of the project;
(b) to evaluate the attractiveness of the investment, an American option of the project
(that allows investing before the expiration of the option takes place) is formulated,
as well as the probabilities of three possible alternatives: execute, wait, or abandon.
The main steps of the overall model that obtain the final value of the real option of the
investment are described in the following sections.

2.1 Wind speed simulation and electricity price model

The probability function that most closely resembles the wind speed regime is the Weibull
distribution function as a result of the orthogonal composition of two correlated Gaussian
functions. The Weibull density function is given by:
 
β x β−1 ( x )β
f (x, α, β) = e α , (1)
α α
468 Ann Oper Res (2011) 186:465–490

Fig. 1 Model’s flowchart

Ann Oper Res (2011) 186:465–490 469

Fig. 2 Mean reversion parameters

where α and β are the scale and shape factors, respectively. Parameter α refers to the average
speed and the β parameter indicates the degree of dispersion of the samples.
Due to the inherent uncertainty of the parameters involved in the estimation of the NPV
of a project, it is advisable to reproduce its behavior by means of a stochastic process. One
of the most commonly used models is the mean reverting process of the Ornstein-Uhlenbeck
type (Schwartz 1997), also known as GMR (Metcalf and Hassett 1995):
 
dx/x = λ φ − ln(x) dt + σ dz, (2)

where the price x(t) follows a stochastic process, λ corresponds to the speed of adjustment
of the reversion, φ is the average long-term return, σ is the standard deviation of the process,
and z defines a Wiener process, as shown in Fig. 2.
The impact of these five parameters (α, β, σ, λ, φ) is the key point of the model’s para-
meterization. All of them are analyzed in detail in Sect. 4.
If the reader wants to go deeper into how to calculate wind regimes and productions,
please refer to Muñoz et al. (2009).

2.2 NPV curve calculation

In this work it is assumed that the income resulting from the energy produced is the product
of two stochastic processes: the energy produced and the price of the electricity sold in the
day-ahead market, represented by a GMR model, as in (2). The price is estimated for a
period of more than one year, since the NPV calculation requires long periods. The model
could also be enriched by using long-term contract values, but this is outside the scope of
this paper.
The yearly cash flows resulting from the income model, subtracting the investment, main-
tenance costs, depreciation and interests, and applying corporate taxes, can be used to es-
timate the NPV of the wind investment. This is the standard calculation of the value of an
investment used in economics and finance textbooks. However, a more interesting approach
is to study the evolution of the NPV for long periods of time. The NPV curve represents
the construction of the trajectory of the resulting NPVs when the investment takes place at
successive times, updating the investment costs and cash flow values to the corresponding
reference periods. In this paper, the NPV curve is calculated on a monthly basis and the an-
nual value corresponds to the average value of the 12 months. Figure 3 presents an example
of how to obtain the different NPVs, and Fig. 4 shows the resulting NPV curve.

2.3 NPV curve parameter estimation

It is assumed that the model’s NPV curves follow GMR trajectories. To facilitate the cal-
culation of the value of the parameters of the GMR process, a transformation y = ln(x) is
470 Ann Oper Res (2011) 186:465–490

Fig. 3 NPVs calculation for a period of 20 years

Fig. 4 Example of the

construction of an NPV curve for
a 20-year period

needed in order to apply Itō’s lemma:

dy = λ[ − y]dt + σ dz, (3)

where
σ2
=φ− . (4)
2λ
This is equivalent to the Ornstein-Uhlenbeck process whose mean and variance values are
given by the expressions:

E[y(t)] ≡ μy (t) = y(t0 )e−λ(t−t0 ) + [1 − e−λ(t−t0 ) ], (5)

σ2
Var[y(t)] ≡ σy2 (t) = [1 − e−2λ(t−t0 ) ]. (6)
2λ
Using the transformed values, and knowing that x = ey , the mean and variance of the origi-
nal function are obtained:
σy2 (t)
E[x(t)] ≡ μx (t) = e[μy (t)+ 2 ] , (7)
[2μy (t)+σy2 (t)]
 σy2 (t)

Var[x(t)] ≡ σx2 (t) = e e −1 . (8)
Ann Oper Res (2011) 186:465–490 471

Next, the transformed function in (7) is discretized taking discrete time steps δt as follows:

yk = yk−1 e−λδt + (1 − e−λδt ) + εk , (9)

where the uncorrelated independent residual error term, εk , has a standard distribution of the
form:
 
σ2
ε ≈ N 0, (1 − e−2λδt ) , (10)
2λ
or, equivalently,
σ2
με = 0 and (1 − e−2λδt ).
σε2 = (11)
2λ
Taking differences of adjacent elements of the process in (9), the following expression is
obtained:
yk − yk−1 = (e−λδt − 1)yk−1 + (1 − e−λδt ) + εk . (12)
This expression can be assimilated to a first-order autoregressive process AR(1) defined as:

yk − yk−1 = β0 + β1 yk−1 + εk . (13)

Equalizing (12) and (13):

β0 = (1 − e−λδt ), (14)

−λδt
β1 = e − 1. (15)

The Cholesky decomposition expressed as a product of matrixes is used to estimate the

values of β0 and β1 :

β̂
β̂ = 0 = (A · A)−1 · A · Z. (16)
β̂1
The matrixes used in (16) are composed of the transformed (ln) values of the GMR process:
⎡ 1 y ⎤ ⎡ y −y ⎤
2 2 1
⎢ 1 y2 ⎥ ⎢ y 3 − y2 ⎥
⎢ ⎥ ⎢ ⎥
A = ⎢... ...⎥; Z=⎢ ... ⎥. (17)
⎣ ⎦ ⎣ ⎦
... ... ...
1 yn yn − yn−1
Clearing λ in (15), the expression for the first estimated parameter is attained:

ln(1 + β̂1 )
λ̂ = − . (18)
δt
On the other hand, from (14), the estimated value of  is:

β̂0
ˆ =−
 , (19)
β̂1
and from (4), the estimated value of φ is:

σ̂ 2 β̂0
φ̂ = − . (20)
2λ̂ β̂1
472 Ann Oper Res (2011) 186:465–490

To calculate the volatility it is necessary to have residual terms resulting from the difference
between the values of the matrix Z and the ones in the estimated matrix Ẑ:

Ẑ = β̂ · Y = β̂0 + β̂1 · Y (21)

where
⎡ y ⎤
1
⎢ y2 ⎥
⎢ ⎥
Y = ⎢ ... ⎥ and ε = Ẑ − Z. (22)
⎣ ⎦
...
yn−1

The standard deviation of the residual terms, σ̂ε , is calculated and, from (11), the expression
of the estimated variance is attained:

2λ̂σ̂ε2
σ̂ 2 = . (23)
1 − e−2λ̂δt

From (23) and (18), the value of the estimated volatility is obtained:

2 ln(1 + β̂1 )
σ̂ = σ̂ε . (24)
δt[(1 + β̂1 )2 − 1]

In this work, we do not assume that the NPV curve follows a GMR process with constant
parameters, because the stochastic parameters of the NPV curve change with time without
following the exact price pattern shown in (2). This is due to the fact that the NPV comes
from the combination of stochastic processes with different characteristics. As mentioned,
the values of the NPV parameters change, in particular φ and λ, and the σ parameter remains
approximately constant showing a slightly increasing trend. To solve this problem, the es-
timation of the parameters is done piecewise. In this way, the matrixes are created with the
information of each of the processes in a piecewise fashion. Therefore, the parameters φ and
λ are optimally adjusted to the actual trajectory.
Thus, the matrixes calculated in (17) are created with the information of each of the
pieces so that the parameters, φ and λ, are optimally adjusted to the actual trajectory, as
shown in (25).

⎡ 1 y2 ⎤ ⎡ 1 y n ⎤
( tn )tn−1
⎢... ... ⎥ ⎢... ... ⎥
⎢ ⎥ ⎢ ⎥
A1 = ⎢ . . . . . . ⎥ , . . . , An = ⎢ . . . ... ⎥;
⎣ ⎦ ⎣ ⎦
... ... ... ...
1 y( tn )t1 1 yn
⎡ n
y2 − y1 ⎤ ⎡y n ⎤ (25)
( tn )tn−1 − y( tnn )tn−1 −1
⎢ ... ⎥ ⎢ ... ⎥
⎢ ⎥ ⎢ ⎥
Z1 = ⎢ ... ⎥ , . . . , Z1 = ⎢ ... ⎥.
⎣ ⎦ ⎣ ⎦
... ...
y( tn )t1 − y( tn )t1 −1 yn − y( tn )tn −1
n n n
Ann Oper Res (2011) 186:465–490 473

Fig. 5 Monte Carlo simulation and piecewise calculation of the NPV curve parameters

Thus, the parameter triads are estimated for each block as follows:
⎡ λ̂ σ̂1 φ̂1 ⎤
1
⎢ λ̂2 σ̂2 φ̂2 ⎥
⎢ ⎥
⎢... ... ...⎥. (26)
⎣ ⎦
... ... ...
λ̂n σ̂n φ̂n

Figure 5 shows the piecewise estimation of the parameters of the NPV curves. In this
case, 12 pieces or blocks have been used. The green curves represent each of the NPV curves
coming from Monte Carlo simulation, the blue one represents the average of the curves, and
the black one represents the initial investment.

2.4 Calculation of the real option value and the probabilities

To calculate the value of an option from a tree (binomial, trinomial, etc.) that is already built
for a specific process, it is necessary to subtract the project investment cost from the NPV
of the project once the last period is reached.
Once the investment costs are discounted from the NPV in the last period, the real op-
tion tree is built backwards up to the initial node. Mathematically this can be modeled by
dynamic programming.
The possibility to exercise an American option is available at any time at each node,
unlike the European option, that only depends on the values of previous nodes and their
associated probabilities. The value of the option is the maximum of three possible choices:
invest or exercise now, wait, or abandon.
474 Ann Oper Res (2011) 186:465–490

Fig. 6 Graphical depiction of the alternative probabilities: to exercise, to wait, or to abandon

To calculate the value of an option from a tree (binomial, trinomial, etc.) that is already
built for a specific process, it is necessary to subtract the project investment cost from the
NPV of the project once the last period is reached.
Figure 6 shows graphically the three different alternatives: exercise the option, wait, and
abandon. It also depicts the probability distribution of each alternative throughout the life of
the real option. The probability to exercise the option is marked in red, the waiting option
in green, and the abandoning option in black. The visual check corresponding to the sum of
the three probabilities is marked in blue.
In parallel to the construction of the real options decision tree there must be an estimation
of the probability of each possible investment situation. Due to the probability distribution
of the nodes, the central zone of the tree has the highest weight in the estimation.
A detailed mathematical formulation of the construction of the trinomial trees can be
found in Muñoz et al. (2009).

2.5 Risk profile characterization

Valuation is usually carried out in a risk-neutral world, by risk-adjusting the process of the
asset value. In Cox et al. (1979, 1985) it has been shown that this adjustment requires a
Ann Oper Res (2011) 186:465–490 475

reduction of the process drift by the risk premium. The reason behind this procedure is
that, by doing this, we avoid discounting expected cash flows at risk-adjusted rates (and
ultimately using preferences on risk premiums). Instead, in a risk-neutral world, cash flows
are discounted at the risk-free rate (Trigeorgis 1995).
Since the drift for the real process is a = λ[φ − ln(x)], denoting the expected capital gain,
and in order to risk-adjust it, we just have to reduce it by the risk premium, η − τ , where
η denotes the total expected (or required) rate of return and τ denotes the risk-free rate of
return.
Therefore, the risk-adjusted drift is given by ar = a − (η − τ ) and the risk-adjusted
process is represented as:

dxr
= λ[φr − ln(xr )]dt + σ dz, (27)
xr

where
 
η−τ
φr = φ − . (28)
λ
Note that risk-neutrality valuation assumes that the position in the plant value and the project
remains riskless by constant re-balancing. By doing this, we implicitly assume that the li-
cence to build the plant is a tradable derivative (Hull 2005).

3 Case studies

This section shows six case studies where the investment model described is applied. A sen-
sitivity analysis is applied with respect to the key parameters of the model. The base case
assumes a wind farm of 10 MW, composed of five generators of 2 MW each. Table 1 shows
the base parameters.
Table 2 shows the range of the parameters for the cases developed in this work. Due to
space limitations, only six cases from Table 2: 1, 6, 7, 8, 9, and 10, are shown in this paper.

Case study 1: Parameterization of the project financing percentage

This case study analyzes the effect produced in the model by changing the external financing
percentage of the required investment. The base value of the financing parameter is 80% in
order to mimic what is already in use in Spain, see Royal Decree 661/2007 (2007). The pa-
rameter values oscillate between 100 and 0% of external financing, applying a 10% increase
for each of the 11 scenarios analyzed. To solve this case study there are four steps to follow:
Step 1: Construction of the NPV curves: Fig. 7 shows the evolution of the NPV curves for
each of the scenarios for a period of 20 years. It can be observed that the curves that represent
the evolution of the shareholders’ equity increase linearly. In the NPV curve of scenario 1
the percentage of external financing is 100%, so there is no equity capital involved, but at
the other extreme, scenario 11, the percentage of leverage is 0%, where the equity needed is
equal to the total investment cost.
Step 2: Estimation of the GMR parameters: from the different NPV curves the parame-
ters of their piecewise GMR associated parameters are obtained. According to the speed of
change of the NPV curves, the optimal number of blocks whose parameters remain constant
is found. See Fig. 5.
476 Ann Oper Res (2011) 186:465–490

Table 1 Model parameters

Type Parameters Value

General NPV calculation period 20 years

General Estimation period for the cash flows 40 years
General NPV discount rate 7%
General Monte Carlo simulations 200
Investment Installed kW price €1000
Investment Price increase (annual) 7%
Costs O&M cost €15/MWh
Wind Shape factor 1.8
Wind Scale factor 5.3
Production Maximum power 10 MW
Production Min., control, and max. rotor speed 4, 11, 25 m/s
Electricity price Initial price €60/MWh
Electricity price Annual volatility 0.005
Electricity price Strength of reversion 0.005
Electricity price Long-term trend 16.95
Electricity price Annual increase of trend 7%
Risk Estimated expected yield 10%
Risk Risk-free asset return 5%

Table 2 Case studies parameters range

Case Parameters Range (step)

1 Project financing 100 to 0% (10)

2 Investment cost (100% of equity) €750 to 1250/kW (50)
3 Investment cost (50% of equity) €750 to 1250/kW (50)
4 Annual investment cost rise 5 to 10% (0.5)
5 Risk aversion 0 to 10% (1)
6 Spot price volatility 0 to 0.01 (0.001)
7 Spot price trend (annual increase) 5 to 10% (0.5)
8 Spot prices’ strength of reversion 0.025 to 0.075 (0.005)
9 Wind speed’s shape parameter 1.5 to 2.5 (0.1)
10 Wind speed’s scale parameter 5 to 6 (0.1)

Step 3: Construction of the trinomial trees for the real options (see Fig. 6). Once tested
that the parameters fit with the original data the different values of the American option to
invest are calculated. Figure 8 shows their values for a specific NPV curve: wait, execute or
abandon. In the figure, the first scenarios, in which the percentage of external financing is
high, the “invest now” alternative is present throughout the entire lifetime (20 years). When
the external financing decreases, there is more need of equity and the “wait” alternative has
more weight. Finally, the “abandon” alternative only shows up in scenarios with more equity
at the end of the lifetime.
Ann Oper Res (2011) 186:465–490 477

Fig. 7 NPV curves for each scenario

Fig. 8 Probabilities of the alternatives for each scenario

478 Ann Oper Res (2011) 186:465–490

Fig. 9 Real option values to invest for each scenario

Fig. 10 Initial NPV vs. equity

Step 4: Calculation of the option to invest: as a result of the above calculations, the value
of the real option is attained. As observed in Fig. 9, more equity implies that the “invest
now” alternative loses value, up to scenarios where the value of the option is zero; therefore,
an investment is not advised.
These results can be easily compared to the ones resulting from a more traditional analy-
sis. In particular, a classic investment valuation method would estimate the value of the NPV
at the beginning of the lifetime (month 0, year 0) and depending on its value would decide
to invest or not. Figure 10 shows how these NPV values correspond to the values of NPV
Ann Oper Res (2011) 186:465–490 479

Fig. 11 NPV curves for each scenario

curves at the beginning of the lifetime. For this case, it is observed that in the initial scenar-
ios, the NPV is above the equity and, the difference is diminishing until it becomes negative
in scenario 8.
As a conclusion for this case, the “invest now” alternative is present from 100% to 40%
of the project financing, i.e., the first 7 scenarios.

Case study 2: Parameterization of the volatility of the electricity prices

This case illustrates the effect on the investment of the changes in the volatility of the prices.
Due to the procedure followed in the model, the volatility of the prices has a slight effect
due to the filtering effect of the Monte Carlo trials done to estimate the average value for
each scenario.
In this case, the base parameter of the volatility is 0.005, ranging from 0 to 0.01 in 11
scenarios. Figure 11 shows that, as the volatility increases, the variability of the NPV curve
also increases. NPV curves are depicted in color surface and equity (not financed project
investment) values in gray.
The trinomial tree produces the different alternatives of the American option to invest, as
shown in Fig. 12.
Finally, the value of the option to invest is depicted in Fig. 13. It is clear that the average
value of the option does not change, but it is not stable.
480 Ann Oper Res (2011) 186:465–490

Fig. 12 Probabilities of the alternatives for each scenario

Fig. 13 Real option values to invest for each scenario

Ann Oper Res (2011) 186:465–490 481

Fig. 14 NPV curves for each scenario

Case study 3: Parameterization of the annual increase of the long-term trend of the
electricity prices

As shown in Tables 1 and 2, the trend of the mean reversion of the NPV curves is annually
increased by a percentage. This case depicts the impact of these annual variations on the
model.
Since this parameter greatly modifies the energy average price year-by-year, its overall
effect is very important in the model’s behavior. See Fig. 14.
Figure 15 depicts the investment surfaces, whose shape is very similar to the previous
case. This one and the percentage of the project’s financing parameter are the most relevant
parameters in the sensitivity analysis study (see Fig. 26). Note that up to scenario 3 there is
no probability to execute the option.
Here, the range of values for the real option is the biggest in all the cases developed.
Therefore, the price to execute the option is the highest, reaching more than €10 M in the
last scenarios. See Fig. 16.

Case study 4: Parameterization of the strength of reversion of the electricity prices

In this case the effect of the changes in the strength of reversion of the prices is analyzed. The
higher the strength of reversion, the higher the price paid for the option. As shown in Fig. 2,
the higher the strength, the faster the NPV curves reach the long-term value, increasing the
project value. See Fig. 17.
Similarly, the differences between the probabilities of the alternatives in each scenario,
as shown in Fig. 18, change depending on the parameter value. In the first four scenarios
with low λ values, the alternative to exercise the option has zero or a very low probability.
482 Ann Oper Res (2011) 186:465–490

Fig. 15 Probabilities of the alternatives for each scenario

Fig. 16 Real option values to invest for each scenario

Ann Oper Res (2011) 186:465–490 483

Fig. 17 NPV curves for each scenario

Fig. 18 Probabilities of the alternatives for each scenario

484 Ann Oper Res (2011) 186:465–490

Fig. 19 Real option values to invest for each scenario

Real option values reflect how sensitive the model is to changes on this parameter. See
Fig. 19.

Case study 5: Parameterization of the Weibull’s shape (beta) parameter

The shape parameter in a Weibull distribution indicates the level of wind speed variability
around the average.
Wind turbines have generation limits, i.e. they can not be operated above a certain maxi-
mum speed, if the variability of the wind speed is high the turbines have to be automatically
stopped.
The higher this parameter is, the higher the volatility of the wind speed, and the lower
the production rates, showing why the NPV curves go down. See Fig. 20.
This is why the alternative of executing the option has lower probabilities in the last
scenarios, as shown in Fig. 21.
As shown in Fig. 22, very high wind speed volatilities make the option to execute the
project not worthy.

Case study 6: Parameterization of the Weibull’s scale (alpha) parameter

Weibull’s scale parameter is related to the average wind regimes. The higher the average
wind speed, the more generation and the higher revenues for the wind farm. See Fig. 23.
This is also reflected in the probabilities of the different alternatives. See Fig. 24.
The range of the option’s values for this case is quite widespread, as can be observed in
Fig. 25.
Ann Oper Res (2011) 186:465–490 485

Fig. 20 NPV curves for each scenario

Fig. 21 Probabilities of the alternatives for each scenario

486 Ann Oper Res (2011) 186:465–490

Fig. 22 Real option values to invest for each scenario

Fig. 23 NPV curves for each scenario

Ann Oper Res (2011) 186:465–490 487

Fig. 24 Probabilities of the alternatives for each scenario

Fig. 25 Real option values to invest for each scenario

488 Ann Oper Res (2011) 186:465–490

Table 3 Case studies results

Case Scenarios Max Min Ave. St. dev.

1 2 3 4 5 6 7 8 9 10 11

1 9.15 8.02 6.21 4.85 3.92 2.08 0.35 0 0 0 0 9.15 0.00 3.14 3.48
2 0.005 0 0 0 0 0 0 0 0 0 0 0.00 0.00 0.00 0.00
3 5.16 4.81 4 3.79 3.02 2.63 1.51 0.97 0.12 0.06 0 5.16 0.00 2.37 1.94
4 2.03 2.05 2.07 2.13 2.18 2.22 2.32 2.41 2.79 3.12 3.54 3.54 2.03 2.44 0.50
5 0.69 0.97 1.24 1.43 1.5 1.71 1.81 1.94 2.08 2.21 2.33 2.33 0.69 1.63 0.52
6 2.33 2.25 2.32 2.19 2.51 2.48 2.11 2.6 1.99 2.81 2.03 2.81 1.99 2.33 0.25
7 0 0 0.24 1.03 2.58 3.18 4.57 6.01 8.02 9.07 10.12 10.12 0.00 4.07 3.76
8 0 0 0 0 1.07 2.31 3.97 4.42 6.05 6.79 8.24 8.24 0.00 2.99 3.07
9 3.64 3.31 2.5 2.02 1.52 1.48 0.63 0.34 0 0 0 3.64 0.00 1.40 1.34
10 0.08 0.51 0.95 1.32 2.3 2.78 3.81 4.09 5.08 6.49 6.89 6.89 0.08 3.12 2.36

Fig. 26 Sensitivity analysis (tornado diagram)

Case studies results

Table 3 shows the results obtained from the six case studies, including the minimum, aver-
age, maximum and standard deviations of the value of the option each one of them.
According to this data, Fig. 26 depicts a tornado diagram with the variation ranges of the
real option values as a function of the range of variation of the parameters.
Related to the results of the cases and the sensitivity analysis, the following conclusions
can be extracted:
(1) It is necessary to have a minimum of 40% project financing. Below this value, the project
lacks interest, i.e., profitability.
(2) As long as the minimum project financing requirement is fulfilled, the project will be
possible if the price of the kW installed is less than €1100.
(3) Due to the integration that takes place in the calculation of the NPV curves, the annual
investment cost rise has a small effect for parameter values close to the base value, 7%.
Ann Oper Res (2011) 186:465–490 489

From this value onwards the increments are clearly reflected on the result reaching the
option value of €3.5 M.
(4) The variation in the option price is almost linear with respect to the risk aversion. For
high risk-aversion values (risk-free parameter between 0 and 3%) the “abandon” alter-
native is shifted to the middle of the option lifetime (10 years). For low risk-aversion
values (risk-free parameter between 7 and 10%) the “abandon” alternative does not take
place until the end of the option lifetime.
(5) Regarding the spot price, its volatility does not affect the results significantly, due to the
aforementioned integration effect. However, an increase in the spot price trend or in the
strength of reversion has considerable effects; the option value can reach values above
€8 M. On the contrary, for increases in the spot price trend below 6% or strengths of
reversion below 0.04, the project is not profitable due to the sensitivity of the model to
these two parameters.
(6) Finally, referring to the wind speed, the shape and scale parameters have opposite ef-
fects. When the shape parameter increases, the dispersion of the wind speed decreases,
and the value of the option decreases up to zero for parameter values higher than 2.2. If
the scale parameter increases, the profitability of the project increases. In spite of this,
for parameter values below the base value, 5.3, the profitability is significantly reduced
due to the extreme sensitivity of the project to this parameter.

4 Conclusions

The presented model obtains the price of the right to construct a wind farm that an investor is
willing to pay for. This right, but not the obligation, is represented by a real options approach
as a decision-making tool to evaluate investments on wind energy plants based on two main
concepts:
(i) a stochastic approach to calculate the project NPV and its associated curve and parame-
ters; and
(ii) a real options model built upon a trinomial tree that evaluates numerically the probabili-
ties of the alternatives of executing or investing now, waiting or abandoning the project.
This trinomial lattice is calculated piecewise in order to make the model more precise.
We propose the NPV curve concept, which is made up from the successive points corre-
sponding to the NPV values at each period (month). This idea is introduced and compared
with the traditional NPV method.
As regards to risk valuation, the model adapts the results as a function of the risk-aversion
profile of the investor instead of producing a specific risk associated with the investment.
This model permits an estimation of the best time to execute the investment within the
project’s lifetime seeking the maximum profit and also an estimation of the probability that
a specific future scenario takes place.
In order to demonstrate the usefulness of the method, six case studies are developed and
analyzed having a variable range of the most important model’s parameters.

References

Ault, G. W., McDonald, J. R., & Burt, G. M. (2003). Strategic analysis framework for evaluating distributed
generation and utility strategies. IEE Proceedings. Generation, Transmission and Distribution, 150,
475–481.
490 Ann Oper Res (2011) 186:465–490

Correia, P. F., Carvalho, P. M. S., Ferreira, L. A. F. M., Guedes, J., & Sousa, J. (2008). Power plant multistage
investment under market uncertainty. IEE Proceedings. Generation, Transmission and Distribution, 2,
149–157.
Cox, J. C., Ingersoll, J. E., & Ross, S. A. (1979). Duration and determination of basis risk. Journal of Business,
52, 51–61.
Cox, J. C., Ingersoll, J. E., & Ross, S. A. (1985). An intertemporal general equilibrium model of asset prices.
Econometrica, 53, 363–384.
Deng, S.-J., & Oren, S. S. (2003). Incorporating operational characteristics and start-up costs in option-based
valuation of power generation capacity. Probability in the Engineering and Informational Sciences,
17(2), 155–182.
Dixit, A., & Pindyck, R. S. (1994). Investment under uncertainty. Princeton: Princeton University Press.
El-Khattam, W., Bhattacharya, K., Hegazy, Y., & Salama, M. M. A. (2004). Optimal investment planning for
distributed generation in a competitive electricity market. IEEE Transactions on Power Systems, 19(3),
1674–1684.
Fleten, S.-E., & Näsäkkälä, E. (2003). Gas-fired power plants: Investment timing, operating flexibility and
abandonment. In Proceedings of the 7th annual international conference on real options, 10–12 July
2003, Washington, DC.
Fleten, S.-E., Maribu, K. M., & Wangensteen, I. (2007). Optimal investment strategies in decentralized re-
newable power generation under uncertainty. Energy, 32, 803–815.
Hull, J. C. (2005). Prentice-Hall finance series. Options, futures, and other derivatives (6th ed.). New York:
Prentice Hall. ISBN:0-1314-9908-4.
Khatib, H. (2003). IEE power series: Vol. 23. Financial and economic evaluation of projects in the electricity
supply industry. Stevenage: The Institution of Engineering and Technology.
Lucía, J. J., & Schwartz, E. S. (2002). Electricity prices and power derivatives: evidence from the Nordic
power exchange. Review of Derivatives Research, 5, 5–50.
Metcalf, G. E., & Hassett, K. A. (1995). Investment under alternative return assumptions: comparing random
walks and mean reversion. Journal of Economic Dynamics & Control, 8, 1471–1488.
Muñoz, J. I., Contreras, J., Caamaño, J., & Correia, P. F. (2009). Risk assessment of wind power generation
project investments based on real options. In Proceedings of IEEE PowerTech 2009, June 28–July 2,
2009, Bucharest (pp. 1–8). ISBN:978-1-4244-2234-0.
Näsäkkälä, E., & Fleten, S.-E. (2005). Flexibility and technology choice in gas fired power plant investments.
Review of Financial Economics, 14(3–4), 371–393. Special Issue on Real Options.
Royal Decree 661/2007 (2007). Regulating the activity of energy production in a special regime, 25th of May
2007. http://www.boe.es/boe/dias/2007/05/26/pdfs/A22846-22886.pdf (in Spanish).
Schwartz, E. S. (1997). The stochastic behaviour of commodity prices: implications for valuation and hedg-
ing. Journal of Finance, 52(3), 923–973.
Trigeorgis, L. (1995). Real options in capital investment: models, strategies and applications. New York:
Praeger. ISBN:978-0-2759-4616-6.
Trigeorgis, L. (1996). Real options: managerial flexibility and strategy in resource allocation. Cambridge:
MIT Press.
Tseng, C.-L. (2000). Exercising real unit operational options under price uncertainty. In Proceedings of the
2000 IEEE PES winter meeting, 23–27 January, Singapore (pp. 436–440).
Tseng, C.-L., & Barz, G. (2002). Short-term generation asset valuation: a real options approach. Operations
Research, 50(2), 297–310.
Tseng, C.-L., & Lin, K. Y. (2007). A framework using two-factor price lattices for generation asset valuation.
Operations Research, 55(2), 234–251.
Wang, C.-H., & Min, K. J. (2006). Electric power generation planning for interrelated projects: a real options
approach. IEEE Transactions on Engineering Management, 53(2), 312–322.
Copyright of Annals of Operations Research is the property of Springer Science & Business Media B.V. and its
content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email articles for individual use.