JRO Vol 1 2011 Complete

ISSN 1799-2737 Open Access http://www.rojournal.
com
Volume 1, Issue 1, 2011

Editor: Mikael Collan

From the editor
Mikael Collan

Research Articles
Real Option Valuation of Offshore Petroleum Field Tie-ins, 1-17
Stein-Erik Fleten, Vidar Gunnerud, ystein Dahl Hem, and Alexander Svendsen

Valuing Real Options Projects with Correlated Uncertainties, 18-32
Luiz E. Brando and James S. Dyer

Estimating Changing Volatility in Cash Flow Simulation-Based Real Option Valuation with
the Regression Sum of Squares Error Method, 33-52
Tero Haahtela

Contact information
Journal www homepage is to be found at:
www.rojournal.com
Primary channel of communication with the journal is email:
rojournal@gmail.com
Postal address:
Journal of Real Options
Raivaajankatu 5 as. 3
24100, Salo, Finland

Submission information
Submissions to the Journal of Real Options are exclusively done
through the on-line submission system that can be found at the journal
homepage at www.rojournal.com.
Author guidelines with information about paper formatting with
templates, and all other information about publication in the journal is
available on the journal homepage.
All published articles go through at least double-blind peer review and
are required to adhere to a high academic standard of writing and to
present new and interesting contributions in the field.
All accepted papers are published on-line and are available for free
download.

Editorial Board
Editor-in-Chief, Mikael Collan, Lappeenranta University of
Technology, Finland

Co-Editor-in-Chief, Christer Carlsson, bo Akademi University,
Finland

Editorial Board Members:

Robert Fullr, Etvs Lornd University, Hungary
Mario Fedrizzi, University of Trento, Italy
Yuri Lawryshyn, University of Toronto, Canada
Richard de Neufville, Massachusets Institute of Technology, United
States of America
Gill Eapen, Charles River Associates, United States of America
Frode Kjaerland, University of Nordland, Norway
Markku Heikkil, bo Akademi University, Finland
Peter Majlender, Stockholm University, Sweden
Joszef Mezei, bo Akademi University, Finland
Marco Antonio Dias, Petrobras, Brazil
Farhad Hassanzadeh, Sharif University of Technology, Iran
Michael Flanagan, Manchester Metropolitan University, United
Kingdom
Makoto Goto, Hokkaido University, Japan
Lauri Frank, University of Jyvskyl, Finland
Scott Mathews, The Boeing Company, United States of America
Luiz Brando, Pontificia Universidade Catolica, Brazil
Peter Linquiti, George Washington University, Washington DC, USA
Tero Haahtela, Aalto University, Finland

Contents
Volume 1, Issue 1
From the editor
Mikael Collan

Research Articles
1-17
Real Option Valuation of Offshore Petroleum Field Tie-ins,
Stein-Erik Fleten, Vidar Gunnerud, ystein Dahl Hem, and
Alexander Svendsen

18-32
Valuing Real Options Projects with Correlated Uncertainties,
Luiz E. Brando and James S. Dyer

33-52
Estimating Changing Volatility in Cash Flow Simulation-Based Real
Option Valuation with the Regression Sum of Squares Error Method,
Tero Haahtela

From the Editor
Mikael Collan
Lappeenranta University of Technology, Business School, PO BOX 20, 53850 Lappeenranta,
Finland
mikael.collan@lut.fi

Welcome to the first issue of the Journal of Real Options! It has already been a
long time since research on real options merits a dedicated journal and now the time
has come. In the modern world of academia it is challenging to start a new journal. I
have still decided to go against the grain and have set up the Journal of Real Options
to be a channel of academic publication for research on real options. These efforts are
backed up by the board that composes of many international top names in real options
research today.
In this inaugural issue we have three academic research papers:

In the first paper Stein-Erik Fleten, Vidar Gunnerud, ystein Dahl Hem, and
Alexander Svendsen examine the valuation of real options related to offshore
petroleum production. The problem is to look at the analysis of investing in adding a
smaller tie-in oil field into an already existing larger field.

In the second paper Luiz E. Brando and James S. Dyer present a
straightforward way of implementing real option valuation using standard decision
tree tools, while taking the correlation between private and market risks into
consideration. Considering the correlated risk brings an additional level of complexity
into the modeling. The paper demonstrates how the correlation can be addressed in a
practical way.

In the third paper Tero Haahtela presents a volatility estimation method for
simulation based real option valuation under changing volatility that can also handle
negative asset price. The paper discusses how uncertainty decreases as the future is
revealed and shows how the method presented is a good example of rolling
valuation with respect to new more accurate information.

It is my sincere hope that this journal will become a place where researchers on
real options will be able to share their research results and their thoughts about the
application and theoretical issues of real options. And remember, the Journal of Real
Options welcomes also short case reports, discussion papers, method notes, state-of-
the-art articles, and book reviews in addition to high quality academic research
articles.

Real Option Valuation of Offshore Petroleum Field
Tie-ins
Stein-Erik Fleten
1,
, Vidar Gunnerud
2
, ystein Dahl Hem
1
, and Alexander Svendsen
1
1
Norwegian University of Technology, Department of Industrial Economics and Technology,
Trondheim 7491, Norway
2
Norwegian University of Technology, Department of Engineering Cybernetics, 7491
Trondheim, Norway
Abstract. We value two real options related to offshore petroleum production.
We consider expansion of an offshore oil eld by tying in a satellite eld, and
the option of early decommissioning. Even if the satellite eld is not protable to
develop at current oil prices, the option to tie in such satellites can have a signi-
cant value if the oil price increases. Early decommissioning does not have much
value for reasonable cost assumptions. Two sources of uncertainty are consid-
ered: oil price risk and production uncertainty. The option valuation is based on
the Least-Squares Monte Carlo algorithm.
Keywords: Investment uncertainty, satellite elds, petroleum development, oil elds,
energy commodities
1 Introduction
We explore the exibility related to investment timing in offshore oil exploration and
production. Offshore oil production can require large investments in infrastructure, off-
shore and onshore facilities and well-drilling costs. These costs are to a large degree
sunk once the investment has been made. Since 2000, oil prices have been increasingly
volatile, thereby creating uncertainty about whether marginal projects can deliver a suf-
cient return on the investment.
During the nancial turmoil in 2008/2009 the development of several smaller elds
on the Norwegian Continental Shelf were postponed due to uncertainty related to whether
they could deliver a sufcient return, among others the satellite eld Alpha connected
to the Sleipner eld. This should make the problem of optimal investment timing in-
teresting for practitioners assessing investment opportunities and both government and
researchers forecasting the future level of investment in petroleum production.
The most critical decisions in a petroleum production project with regards to prof-
itability is when and if the eld should be developed and the largest part of the in-
vestment is made. Depending on the eld and the technology used to produce it, the
Corresponding author, Stein-Erik.Fleten@iot.ntnu.no, Phone: +47 73 59 12 96, Fax: +47 73

59 10 45
Journal of Real Options 1 (2011) 1-17
ISSN 1799-2737 Open Access: http://www.rojournal.com 1
operator might also have choices available after the eld has started to produce. Often
there are smaller reservoirs surrounding the main eld. Generally too small to warrant
an independent production unit, these reservoirs can be developed via the production
unit at the main eld. The operator also has to decide when to abandon the eld and
decommission the production unit by taking into account future production as well as
equipment lifetime and operating cost. Once the eld is abandoned, restarting produc-
tion is in most cases not realistic.
Real options valuation (ROV) has been applied to petroleum projects for a long time
as they have many attributes that make them suitable for this valuation. These projects
often involve a large initial investment, the output is a risky and easily traded commod-
ity, and management have many choices available related to timing, production tech-
nology and size. Siegel, Smith and Paddock [27] assess investing in offshore petroleum
leases. Cortazar and Schwartz [7] use a Monte Carlo model to nd the optimal timing
of investing in a eld with a set production rate that declines exponentially and with
varying, but known, operating costs. With this predetermined production rate, the value
of the eld becomes a function of the oil price, which is modeled as a two-factor model
where the spot price follows a geometric Brownian motion and the convenience yield
follows a mean-reverting process. Smith and McCardle [20] consider the timing of in-
vestment, the option to abandon and to vary the production rate by drilling additional
wells. Both prices and production rates are modeled as stochastic processes, where the
price follows a geometric Brownian motion. Ekern [12] uses a ROV model to value
the development of satellite elds and adding incremental capacity using a binomial
lattice model. He nds that satellite elds that are currently unprotable can have an
option value. Lund [18] considers an offshore eld development by using a case from
the North Sea eld Heidrun. The model used is a dynamic programming model, and
take into account the uncertainty regarding both reservoir size and well rates in addi-
tion to the oil price. The paper models the price as a geometric Brownian motion, and
use a binomial valuation model to nd the optimal size of the production rig and in-
vestment timing. Armstrong et al. [2] uses information from production logging and a
copula-based Bayesian updating scheme for real options valuation of oil projects. Dias
et al. [11] use Monte Carlo simulations together with non-linear optimization to nd an
optimal development strategy for oil elds when considering three mutually exclusive
alternatives. Chorn and Shokor [6] combine dynamic programming and real options
valuation to value investment opportunities related to petroleum exploration. Dias [10]
provides a more thorough review of ROV related to petroleum exploration and produc-
tion.
The contribution of our paper is that we consider the decision to add a known
(smaller) tie-in eld to an existing one, taking into account possible abandonment, price
risk and technical risk. We use a real options approach where the valuation and optimal
exercise is found using the least-squares Monte Carlo (LSM) algorithm presented by
Longstaff and Schwartz [17]. We do not consider the problem of initial investment in
the main eld, as this problem has been considered both in petroleum production and
other industries before, see e.g. Kort et al. [15]
3
. Furthermore, the value of deferring an
investment is generally low in petroleum production [18]. Instead, we focus on deci-
sions being made as the eld is in production, where we model expanding production
by tying in surrounding satellite elds as well as the option to abandon the eld early
and selling the production equipment.
In Section 2 we present the data used in this work, and discuss its properties. Further,
in Section 3 we study the option to expand the production with a tie-in eld as well as
abandoning the eld early. We then apply these models in a case study of two such real
options. Finally, in Section 4 we conclude and offer suggestions for further work.
2 Data
To estimate the long term behavior of the oil price and to nd a suitable time-series
model, we have used the real price of crude oil denominated in 2008 USD from Reuters
EcoWin [24]. The series can be seen in Fig. 1 and consist of the US average price in
the years from 1861 to 1944, then Arabian Light posted at Ras Tanura from 1945 to
1985, and Brent spot since 1985 to today. It has 148 annual observations going back to
1860. We could have used more high-frequency data for the latter years, but these are
not available prior to 1946 for monthly data and 1977 for daily data. To avoid mixing
the different series, only the annual observations have been considered.
To obtain risk-neutral growth and the oil lease rate we use forward prices at time t
expiring at time T, F
t,T
, from Wall Street Journal [29] for Light Crude Oil, as seen in
Fig. 2. The series have contracts for each month till December 2014 and semiannual
contracts expiring as late as December 2017. We nd an estimated risk-neutral long
term growth of 2.70%. The longest duration for the forward contracts used, T, is eight
years, but we assume that the growth indicated by these, ln
F
0,T
S
0
is a good estimate for
the growth in our twenty-year period. We use the expected growth together with an
estimate for the risk-free rate, r
f
to nd the oil lease rate, , by using (1). This puts the
oil lease rate at 1.6%:
= r
f
1
T
ln
F
0,T
S
0
(1)
For a market-based estimate for the volatility, we have used implied volatility from
options quoted at ICE [14] for Brent oil options. The series have options expiring at 17
different dates with several options set to expire at each date. The longest time to expiry
is 3 years. We have used an average value over all strike prices available to nd a mean
implied volatility for each date. The implied volatility is falling with longer expiration
time, implying that the 3-year forecast might not be valid for the long-term real options.
Even so, we use the implied volatility for the 3-year option as the oil price long term
volatility, with an implied volatility of 29.5%. This is quite a lot higher than the historic
3
They study what inuences the choice of developing the whole project at once versus devel-
oping it in gradual steps.
average for the last 148 years, but close to the volatility in the last 40 years of 28.8%.
It is also higher than the 20% that Pindyck [21] found when estimating volatility from
historical data. Costa Lima and Suslick [8] refer to Pindyck [21] and also argue that the
volatility has been stable around 20%. We use the implied volatility as an estimate for
the long term volatility. One reason for the difference between the market view and the
conclusions of Costa Lima and Suslick [8] and Pindyck [21] could be the increase in
oil price volatility in the last years.
To estimate the USD-denominated risk free rate, we have used 20-year US Treasury
bonds from [24] as an estimator for the risk free rate. The risk free rate is estimated to
be 4.3%.
Fig. 1: Real price adjusted Brent spot price, USD 2008
Fig. 2: Light crude oil forward prices with increasing time to maturity. Observation date
20090911
3 Real Option Valuation
3.1 Flexibilities in Petroleum Production
In this section we consider two cases where the operator has exibility, and develop
valuation models for this exibility. We include both input (resource) uncertainty and
output price uncertainty, as in Bobtcheff and Villeneuve [3]. Unlike their analysis, we
ignore capacity choice issues.
The Value of Including Satellite Fields We assume that the search and exploration
phase has been completed; see e.g. Martinelli et al. [19] for a bayesian network analysis
of which location to drill a prospect well. In many situations, the operator knows of a
smaller and nearby eld that can be produced through the main production platform.
These smaller elds will often have higher per-barrel costs due to economies of scale
and are more interesting to consider in a real option model than ordinary elds since
they are not necessarily economical to develop. Typically, such elds will not be large
enough to warrant an independent platform, but it can be protable to tie the elds to ex-
isting platforms. Tying in a small eld will increase the produceable reserves connected
to the platform, but will require an investment. The deterministic NPV of tying in such
a satellite eld can be calculated by using the reservoir model presented in Sect. 3.2
and valuing the incremental production from the satellite, given the capacity constraints
and the time of connection. Given that the increased costs by adding the satellite are
xed, the value of extra production will vary only with the price of oil and the time of
connection. If the satellite eld is connected before the production declines, then it will
not increase the production from the platform until the main eld is off its plateau, since
the plateau is given by the platforms maximum production rate. Further, if the satellite
is connected near the end of the platforms life time, much of the extra elds reserves
will be left in the ground unless one extends the lifetime of the platform, which might
not be possible depending on the availability of infrastructure etc. Developing a satellite
eld can require a large initial investment, and it is assumed that any extra operational
costs are included in the investment cost. Since these are modeled as deterministic cash
ows, the NPV of the future costs are simply added to the investment. Thus, the value
of being able to include a satellite eld takes the form of a call option to acquire the
extra production by paying the investment cost.
The increase in production is the difference between the line and the dotted line in
Fig. 3. We can calculate the net present value of increased production when connecting
the tie-in at time t by (2):
NPV
S,t
= S
T
j=t
Prod
j
e
j
I (2)
S represents the price of oil, Prod
i
the extra production from the satellite in period
j, the convenience yield and I the present value of the investment and operational
costs.
The Timing Dimension of Including a Tie-in Field The process of valuing a project
with a xed end date is different than for an ordinary stock. Even with uncertainty in
the output, one is certain that the tie-in will be worthless at the time the main platform
is decommissioned. In the case study we have used a production prole from Robinson
[25] to calibrate the model of Lund [18] in order to get a representative production
prole.
The Value of Early Shut Down Some offshore production units can be moved if the
value of the remaining production is low, and the production unit is not near the end of
its life. This can be the case if the true eld reserves are lower than estimated. To model
this, we have used the same price and reservoir model as in the expansion case, but now
it is the whole project value that is relevant. Thus, the value of ending the production
prematurely can be calculated by using (3):
NPV
S,t
= K
t
S
T
i=t
(Prod
i
e
i
C
i
e
ri
) (3)
This states that the value of decommissioning the eld early is the income from
selling the production unit, K
t
, less the future expected prot, stated as remaining pro-
duction less the operational costs, C
i
. It is assumed that it is possible to sell the unit
either to another project or another company for a positive price. We have assumed that
the unit depreciates linearly and that the income from a sale follows this value, and that
it has a planned lifetime equal to the the elds lifetime. The strike will take the form:
K
t
= K
0
T t
T
(4)
The Effect of Uncertain Production In Sect. 3.2, we model the production uncertainty
as a mean-reverting process. Unlike a Brownian motion, the expected value of a mean-
reverting process at time t is dependent on both its current value and its equilibrium
value.
E() = +(
0
)e
t
, (5)
where represents the production level, the mean index level, and the speed of
mean-reversion.
Finding a Suitable Model for the Oil Price One of the most signicant factors in
valuing a potential oil eld is the price of oil. Like the price of other tradeable items
the oil price is governed by supply and demand. The theoretical ideal model would take
into account all the factors that affect supply and demand and produce a forecast of
the oil price based on this information [21]. Several such models have been developed,
among others the Hubbert model of supply [13] and the LOPEC model [23]. These
model the price development by looking at the underlying factors that drive supply and
to some extent demand. There are two major obstacles for implementing such a model
for generating long term forecasts. First, identifying all of the factors affecting the oil
price is in itself a difcult task. Second, producing good forecasts for all of these factors
might be just as difcult as producing a forecast for the oil price. A time series model
thus seems like an attractive alternative model formulation. Pindyck [21] nds that the
oil price can be modeled both as a mean-reverting model and a geometric Brownian
motion. Postali and Picchetti [22] shows that a geometric Brownian motion is a good
approximation for the oil price movements in the long run. We model the oil price as a
geometric Brownian motion because we are interested in the long-term behavior. The
geometric Brownian motion is described by (6).
dP
P
= dT +dZ (6)
3.2 Reservoir Model
The eld production prole is useful when valuing real options, since it provides infor-
mation on volume and time of production. A realistic model of reservoir performance
is challenging to create and to calculate, because of the need to model many parameters
in a 3D-setting with many non-linear relations. In this work, a simple zero-dimensional
model of Wallace et al. [30] is used. This models the reservoir as a tank with a uniform
uid and with uniform properties in the whole reservoir. Thus, it does not account for
differences in permeability in different areas or local differences in pressure caused by
the well ow as the areas surrounding the producing wells empties. It is, however, a
simple model that has great computational advantages compared to a more complex
reservoir model, and it does reect the form of reservoir production proles of several
types of petroleum elds [18].
Table 1: Reservoir parameters
P
w,0
- Initial reservoir pressure
P
w,t
- Reservoir pressure at time t
P
min
- Abandonment pressure
R
0
- Initial reservoir volume
R
t
- Reservoir volume at time t
q
r,t
- Maximum reservoir depletion rate at time t
q
w
- Maximum well rate
q
max
- Maximum capacity, or plateau production
q
rampup,t
- Maximum production during eld development
N
t
- Number of wells producing at time t
The reservoir pressure follows the following relation:
P
w,t
= P
w,0
R
0
R
t
R
0
(P
w,0
P
min
) (7)
The reservoir pressure provides the maximum well ow, which decays exponen-
tially with time with continuous production if there are no other constraints on the well
ow. The maximum well rate is based on the capacity of the wells installed.
q
r,t
= N
t
q
w
t
P
w,t
P
min
P
w,0
P
min
(8)
Together, (7) and (8) becomes the simple equation
q
r,t
= N
t
q
w
R
t
t
R
0
(9)
This is the maximum production from the eld, given that there is no water injec-
tion or other types of pressure maintenance performed. It is rarely optimal to construct
the production unit so that it can produce at the maximum rate q
r,t
, because of high
investment costs. When the eld has a maximum processing capacity that is lower than
the eld maximum production, the production prole will have a at region where the
production is equal to the capacity maximum. This level is called the plateau produc-
tion. The optimal plateau level is mainly a function of investment cost, production and
required rate of return, since it is a trade off between investment cost and the ability to
get the oil quickly out of the ground. There might also be technical reasons to limit the
capacity. We have included a ramp-up period of three years, which is similar to the case
found in Robinson [25]. During this ramp-up period we have assumed that the produc-
tion grows linearly to capacity maximum over the three year period. The background
for such a ramp-up period is among other topics well drilling. It will not be possible
to drill all wells at the same time, and connecting the streams to the platform will also
require some time. The actual production thus becomes the minimum of q
r,t
, q
max
and
q
rampup,t
.
Production Prole with a Tie-in Field To model the increase in production by a tie-in
satellite eld, the new reserves, R
new
are added to the initial reserves. This increases
both the initial reserves, R
0
and the reserves at the connection time, R
t
. The effect of
this increase is dependent on when the new eld is built. If the satellite is connected
before the eld goes into decline, then the plateau production will be maintained longer
as seen in Fig. 3a.
Uncertainty in Production Production volumes are often uncertain as wells can pro-
duce more or less than planned. Lund [18] models this by a changing well capacity. The
well capacity is modeled as a simple stochastic function, where the well can either have
a high or a low well rate. The probability of one of the wells changing regime from a
high rate to a low or opposite is 0.1 per period of 6 months. Each well capacity will be
highly random, but with a large number of wells the process resemble a mean-reverting
stochastic process. The variance of the eld production will be very dependent on the
number of wells connected to the eld. McCardle and Smith [20] take a different ap-
proach by modeling the decline rate as a geometric Brownian motion. This might be
appropriate when the eld is in decline, but it does not take into account the effect of
the production capacity limit and it does not clarify which fundamental property that
(a) t=5 (b) t=10
Fig. 3: Production proles with tie-in at t=5 and t=10
varies. We consider changing well rates as the main source of uncertainty, as in the the
switching model in Lund [18]. We do not model each well individually, however, in-
stead we consider the whole eld production by assuming a number of wells. This is
implemented as a production factor for the whole eld,
t
, as a mean-reverting process.
We believe that this aggregate production factor is more versatile than the model of
Lund [18], as operators can create historic production factors from current and previous
elds and easily take into account other risk factors like technology development or
unscheduled maintenance. The production factor follows:
t
=
tT
+(
tT
)dT +dZ (10)
where
t
is the well production factor at time t, and , and are mean rever-
sion parameters from the regression. The parameter values can be seen in Table 2. The
parameter values are found by Monte Carlo simulations from the model used by Lund
[18], and regressing the simulation results to nd a mean-reverting model.
Table 2: Production factor mean reversion parameters
Parameter Value
0.665
0.218
0.050
3.3 Valuation Framework
There are mainly two ways of calculating the present value of future cash ows. One
solution is using risk-adjusted rates of return and real expected growth rates. The other
is risk-neutral pricing.
The rate of return used in the valuation of real options have a signicant inuence
on both optimal exercise policy and option value. Especially with long term valuations,
like many real options, a slight change in the rate of return can make a substantial
difference due to the compounding effect. Using an appropriate discount rate is thus
important to obtain correct results.
Another procedure of obtaining a valuation is to price the cash ows using other
securities with similar risk proles that are traded in the market. By replacing the real
price growth with the risk-neutral price growth obtained from traded forward-contracts,
one can use the risk-free rate to obtain the value of the project and connected options.
This treats risk in a consistent manner compared to the market, avoiding biases that can
occur otherwise (Laughton [16]). This is commonly called risk-neutral valuation. Since
all parameters are estimated from nancial markets, which are assumed to be efcient,
this leads to an accurate valuation of the project.
Using risk-adjusted rates has the advantage of being familiar to decision-makers
in most rms today, and is perhaps the most intuitive of the two approaches. We do
however choose to use risk-neutral pricing, since this ties the valuation of the risky
cash ows directly to observed prices of this risk. The risk-neutral method is also the
most common approach when valuing options. One issue with using risk-neutral pric-
ing is that the risk-neutral method can underestimate capital costs when risk of default
is present (Almeida and Philippon [1]). This can lead to inaccurate valuations when the
cost of distress is high. This was the case during the nancial crisis in 2008/2009, when
the risk-free rates went down but the cost of capital for rms increased. Thus, the risk
neutral valuation would advice rms to invest more in a time where rms capital costs
increased, which is clearly the wrong advice. However, in more stable conditions the
distortions related to the risk of default should be low, specially when considering large
petroleum companies.
3.4 Case Study
For valuing nite-maturity American call options one must use numerical methods.
Common approaches include lattice methods, a la Cox et al. [9], or nite difference
methods, see Brennan and Schwartz [4]. However, these methods are cumbersome
when there are multiple and possibly heterogeneous sources of uncertainty. In such sit-
uations, approaches based on Monte Carlo simulation come to the fore; see [5,28,17].
Input Data In this section, we use the model developed in previous sections to value
two real options connected to an offshore oil project with the Least Squares Monte
Carlo algorithm developed in Longstaff and Schwartz [17]. First and second degree
monomials of the forward price of the underlying asset as presented in (2) and (3) are
used as regressors in the LSM calculation. We use risk neutral pricing.
Table 3: Financial parameters
S
0
- Current oil price - USD 60
r
f
- Risk free rate of return - 4.3%
- Lease rate - 1.6%
- Annual volatility oil price - 29.5%
K
E
- Expansion option strike/Total cost tie-in eld - MUSD 600
K
A
- Early decommissioning option strike/Initial production unit sales price - MUSD 500
Table 4: Reservoir parameters
R
0
- Initial reservoir reserves - 300 MMbbl
R
tiein
- Initial tie-in reserves - 15 MMbbl
q
w
- Maximum well production - 66 MMbbl/yr
q
p
- Platform production capacity - 33.17 MMbbl/yr
T - Field life time - 20 Years
I
Tot
- Total Investments - MUSD 2,228
T
Rampup
- Production Ramp-up time - 3 Years
Expansion Option The option to invest in a tie-in eld takes the form of a call option,
as discussed in Sect. 3.1. To acquire this option the operator might have to invest in extra
deck-space or other forms of extra capacity today, denoted C
tiein
. This will be the cost
of obtaining the real option, and should not be confused with K
E
which is the investment
needed when the tie-in is connected. Using the input data in the previous section and
taking the price growth into account, we nd that the maximum static NPV is obtained
at T = 8 which is the last year of plateau production. However, after deducting invest-
ment costs the NPV is MUSD 176 at the optimal investment time discounted back to
t = 0. In a deterministic setting, it does not pay off to produce the satellite and based
on this the operator should not invest in excess capacity in order to have the opportunity.
When we add price uncertainty the answer changes. By valuing the investment op-
portunity as an American call option on the incremental production, the option to invest
is estimated to be worth MUSD 150. This implies that if the investment needed today,
C
tiein
, is less than MUSD 150, the operator should invest in order to have the option.
This helps explain why operators frequently invest in extra capacity, since having the
opportunity of producing nearby satellite elds creates valuable real options.
Adding further uncertainty by introducing uncertainty in production, the option
value is still in the same range as before with an option value of MUSD 161. The lower
contribution is not surprising, as the variation in production is lower compared to price
variation and the production follows a mean-reverting process rather than a Brownian
motion.
Sensitivity Analysis As we can see from Fig. 4a, the option value increase with in-
creasing initial oil price. Unlike a static NPV calculation the option value increases
nonlinearly with low initial oil prices, but the growth becomes linear at higher prices.
Table 5: Monte Carlo parameters
N - Number of realizations - 100 000
M - Number of time points - 100
This is natural, as the tie-in is almost certain to be developed at high prices, and the
extra value from the option is low. In this case, the option value is almost equal to a
static NPV. However, unlike the static NPV the option value is never negative. Because
the operator has the choice but not the obligation to develop the tie-in, it will never be
developed if it has a negative NPV.
Another important variable is oil price volatility, and the option sensitivity to this
variable can be seen in Fig. 4b. The option does not have any signicant value for
volatilities below 5% per year, and this conrms the conclusion that the project would
not have positive NPV in a static valuation method. That the value of a project should
increase with larger volatility is contrary to common intuition. The crucial difference
between real option valuation and a discounted cash ow approach is that the project
owner has the option to not exercise the option. Thus the owner is protected from the
case where the price falls, since the satellite eld will not be developed in this case.
High volatility increases the value because it increases the probability of a very high
payoff, without increasing the probability of a large loss. However, higher volatility
will increase the optimal exercise price and delay the investment time as seen in Fig.
5b. This is because one needs to have a price high above the break-even price to be
certain that the price will not drop to a level where the project has a negative NPV when
the volatility is high. Also, we observe that the volatility has less effect on the option
value than the initial oil price.
(a) Initial oil price (b) Oil price volatility
Fig. 4: Expansion option value sensitivities
Another important output from a ROV is the optimal oil trigger price that triggers
the investment. For the option to develop the satellite eld, the development of the
trigger price can be seen in Fig. 5. The trigger price is dened as the smallest price that
triggers investment in the LSM-algorithm. As expected, the trigger price increases with
increasing volatility and with decreasing satellite size.
(a) Volatility (b) Satellite size
Fig. 5: Trigger price sensitivities
Early Decommissioning Option The opportunity of decommissioning the eld pre-
maturely could be a response to lower production volume than expected, or very low oil
prices. The operational costs of an oil project are often low compared to the investment
cost, and the value of being able to prematurely abandon the eld is believed to be low.
When disregarding uncertainty in reservoir reserves, making price risk the only
source of uncertainty, the option value is MUSD 4.4. Adding uncertainty in the reser-
voir reserves, we obtain an option value of MUSD 4.5. We conclude that the option of
abandoning the eld prematurely is not very valuable, and that the exibility related
to being able to sell the production unit can be disregarded when choosing production
technology.
Sensitivity Analysis Since the decommissioning option is similar to a put option, we
expect the option value to decrease with rising oil prices. This is also the case, as can be
seen in Fig. 6a. Unlike a regular put, the option is worth more than the strike price as the
oil price approaches zero. This is because as the project is abandoned the operator also
avoids the operating costs. The option value of abandoning is high when the oil price is
low, but since the project as a whole will have a negative NPV it will not be built in the
rst place. Also, we have assumed that the value of the production unit is deterministic.
A more realistic assumption would be that the sales price is positively correlated with
the oil price, as few new projects will be initiated if the price is low. This will further
reduce the value of early decommissioning. For initial prices close to todays price the
option value is negligible compared to the investment. The option value is sensitive to
the price volatility, as seen in Fig. 6b. If the price volatility should continue to increase
in the future, decommissioning options could become valuable.
(a) Initial oil price (b) Oil price volatility
Fig. 6: Abandonment option value sensitivities
When considering the trigger prices, we nd that the oil price will have to fall below
40 USD per barrel if early decommissioning is to be considered. Compared to historical
oil prices this is not an unrealistic situation. Early exercise is however most likely at the
end of the production units lifetime when the expected sales price is low. We also note
that the oil price volatility does not have a large impact on the exercise trigger price.
Fig. 7: Abandonment option trigger price
4 Conclusion
In this paper we study the exibility related to investment timing in offshore oil explo-
ration and production. The oil price is the main source of risk that inuence the value of
real options related to the project. It is shown that the option to abandon by moving the
production unit is not signicant compared to the cost of developing the eld. The op-
tion to expand the production by adding new elds adds value and the value of making
initial investments in order to be able to connect such satellite elds in the future can
be large even when the current NPV from the satellite elds are negative. As expected,
both options increase in value when faced with increased volatility.
For further work, exploring if other price models, e.g. the two-factor model pre-
sented by Schwartz and Smith [26], leads to different option valuations would be an
interesting extension. Another extension related to the option value framework would
be to introduce a stochastic process governing when and if a tie-in eld is found. This
would be more general than our assumption that the operator knows from the start if
there is a nearby eld.
Acknowledgements
We would like to thank Afzal Siddiqui for comments, and Marta Dueas Diez of Rep-
sol YPF for her advice related to the issues in petroleum production. We acknowledge
the Centre for Sustainable Energy Studies at the Norwegian University of Science and
Technology (NTNU), and are grateful for support from the Center for Integrated Oper-
ations in the Petroleum Industry at NTNU, and from the Research Council of Norway
through project 199908. All errors are solely our responsibility.
References
1. Almeida, H., Philippon, T.: The risk-adjusted cost of nancial distress. Journal of Finance
62(6), 25572586 (2007)
2. Armstrong, M., Galli, A., Bailey, W., Cout, B.: Incorporating technical uncertainty in real
option valuation of oil projects. Journal of Petroleum Science and Engineering 44(1-2), 67
82 (2004)
3. Bobtcheff, C., Villeneuve, S.: Technology choice under several uncertainty sources. Euro-
pean Journal of Operational Research 206(3), 586600 (2010)
4. Brennan, M.J., Schwartz, E.S.: Finite difference methods and jump processes arising in the
pricing of contingent claims: A synthesis. Journal of Financial and Quantitative Analysis
13(3), 461474 (1978)
5. Carriere, J.F.: Valuation of the early-exercise price for options using simulations and non-
parametric regression. Insurance: Mathematics and Economics 19(1), 1930 (1996)
6. Chorn, L., Shokhor, S.: Real options for risk management in petroleum development invest-
ments. Energy Economics 28(4), 489 505 (2006)
7. Cortazar, G., Schwartz, E.S.: Monte Carlo evaluation model of an undeveloped oil eld.
Journal of Energy Finance & Development 3(1), 7384 (1998)
8. Costa Lima, G.A., Suslick, S.B.: Estimation of volatility of selected oil production projects.
Journal of Petroleum Science and Engineering 54(3-4), 129 139 (2006)
9. Cox, J.C., Ross, S.A., Rubinstein, M.: Option pricing: A simplied approach. Journal of
Financial Economics 7(3), 229263 (1979)
10. Dias, M.A.G.: Valuation of exploration and production assets: An overview of real options
models. Journal of Petroleum Science and Engineering 44(1-2), 93 114 (2004)
11. Dias, M.A.G., Lazo, J.G.L., Pacheco, M.A.C., Vellasco, M.M.B.R.: Real option decision
rules for oil eld development under market uncertainty using genetic algorithms and Monte
Carlo simulation. 7th Annual Real Options Conference, Washington DC (2003)
12. Ekern, S.: A option pricing approach to evaluating petroleum projects. Energy Economics
10(2), 9199 (1988)
13. Hubbert, M.K.: Nuclear energy and the fossile fuels. Tech. rep., American Petroleum Insti-
tute Drilling and Production Practice, Spring Meeting, San Antonio, Texas (March 1956)
14. ICE: Daily volumes for ICE Brent crude options. Retrived 20091111 (2009),
https://www.theice.com/marketdata/reports/
15. Kort, P.M., Murto, P., Pawlina, G.: Uncertainty and stepwise investment. European Journal
of Operational Research 202(1), 196203 (2010)
16. Laughton, D., Guerrero, R., Lessard, D.R.: Real asset valuation: A back-to-basics approach.
Journal of Applied Corporate Finance 20(2), 46 65 (2008)
17. Longstaff, F.A., Schwartz, E.S.: Valuing American options by simulation: a simple least-
squares approach. Rev. Financ. Stud. 14(1), 113147 (2001)
18. Lund, M.W.: Real options in offshore oil eld development projects. 3rd Annual Real Op-
tions Conference, Leiden (1999)
19. Martinelli, G., Eidsvik, J., Hauge, R., Frland, M.D.: Bayesian networks for prospect analy-
sis in the North Sea. AAPG bulletin 95(8), 14231442 (2011)
20. McCardle, K.F., Smith, J.E.: Valuing oil properties: Integrating option pricing and decision
analysis approaches. Operations Research 46(2), 198 217 (1998)
21. Pindyck, R.S.: The long-run evolution of energy prices. The Energy Journal 20(2) (1999)
22. Postali, F.A.S., Picchetti, P.: Geometric brownian motion and structural breaks in oil prices:
A quantitative analysis. Energy Economics 28(4), 506 522 (2006)
23. Rehrl, T., Friedrich, R.: Modelling long-term oil price and extraction with a Hubbert ap-
proach: The LOPEC model. Energy Policy 34(15), 2413 2428 (2006)
24. Reuters EcoWin: Reuters EcoWin database. Retrieved 20090902 (2009)
25. Robinson, R.: The economic impact of early production planning (EPP) on offshore fron-
tier developments. In: Offshore Mediterranean Conference and Exhibition in Ravenna, Italy
(March 2009)
26. Schwartz, E.S., Smith, J.E.: The short-term variations and long-term dynamics in commodity
prices. Management Science 46(7), 893911 (July 2000)
27. Siegel, D.R., Smith, J.L., Paddock, J.L.: Valuing offshore oil with option pricing models.
Midland Corporate Finance Journal 5, 2230 (1988)
28. Tsitsiklis, J.N., Van Roy, B.: Regression methods for pricing complex American-style op-
tions. IEEE Transactions on Neural Networks 12(4), 694703 (2001)
29. Wall Street Journal: Markets data center (2009), http://online.wsj.com/, retrieved 200909
11
30. Wallace, S.W., Helgesen, C., Nystad, A.N.: Generating production proles for an-oil eld.
Mathematical Modelling 8, 681 686 (1987)
Valuing Real Options Projects with Correlated
Uncertainties
Luiz E. Brando
1
and James S. Dyer
2
1
IAG Business School, Pontifcia Universidade Catlica do Rio de Janeiro, Rio de Janeiro, RJ,
22451-900, Brazil
brandao@iag.puc-rio.br
2
McCombs School of Business, University of Texas at Austin, Austin, Texas 78712, United
States
jim.dyer@mccombs.utexas.edu
Abstract. Contingent claims are traditionally priced through the use of
replicating portfolios, or equivalently, risk neutral valuation. When markets are
incomplete, as occurs with many projects and is often the case with claims on
real assets when firms are subject to private, project specific risks, these risks
cannot be hedged and a replicating portfolio cannot be construed. We proposed
a modified approach that enhances the methodology originally developed by
Copeland and Antikarov (2003) and provides a practical method for pricing
project where management flexibility and correlated risks are present, using the
concept of partially complete markets of Smith and Nau (1995).
Keywords: Real Options, Correlated Uncertainties, Decision Analysis
1 Introduction
It is widely recognized that discounted cash flow methods do not adequately value
contingent claims, such as options on financial or real assets. The solution to the
problem of valuing financial options was pioneered by Black & Scholes [1] and
Merton [2] and this approach has been further extended to the valuation of
investments in real assets that present managerial flexibility in an approach known as
the real options methodology.
Contingent claims are traditionally priced through the use of a market asset or
portfolio of marketed assets that replicate the payoffs of the claim in all states and
times. Since this replicating portfolio offers the same payoffs and risks as the claim,
arbitrage considerations imply that their prices must also be the same. While this
analysis is straightforward in the case of complete markets, the markets for real assets
are usually incomplete as the number of marketed assets is insufficient to set up the
replicating portfolio.
Market risks are due to uncertainties that are market correlated and can be fully
hedged by trading in securities. An example is the risk derived from the uncertainty
over future oil prices in an oil exploration and development project. For projects
where a replicating portfolio can be constructed and valued, the market is complete
and all project risks can be hedged by shorting this portfolio. On the other hand, even
though the basic project uncertainty may be market related, there may not be an active
traded market for this particular product or commodity, or there may be project or
firm specific uncertainties, such as the uncertainty over the size of the oil reservoir or
discrete lumpy events that may have a one time effect on the project and are not
correlated with the market. In these cases, the project cannot be hedged with market
securities and markets are incomplete for projects that bear these types of risks.
When the output of the project is a traded commodity, a standard solution in the
real options literature [3-4] is to use this asset to hedge the projects risks and construct
the replicating portfolio. Another approach is to treat the project without options as a
traded asset, where its present value is assumed to be its true market value, and to use
the project to create an underlying portfolio to value options associated with the
project [5]. Both approaches address the problem by making assumptions that
transform an incomplete market setting into a complete one.
The incomplete market problem can only be addressed directly if we are willing to
place restrictive assumptions on the investors or managers utility functions. Smith
and Nau [6] introduce the concept of a partially complete market where the market is
complete for market risks, and private events convey no information about future
market events. This implies that if
m
i
e and
1
p
t
e

are the vectors of all possible market
and private states at t and t-1 respectively, then
m
i
e and
1
p
t
e

are independent. Under
this framework, the market component of the project cash flows is valued using the
traditional complete market setting, and the private component may be priced
assuming risk neutrality if the investors are sufficiently diversified, or by using a
utility function that reflects the investors subjective beliefs and preferences
otherwise.
There are, however, projects where the distinction between market and private
risks is either not so clear, or not a meaningful concept, such as when these two
uncertainties are correlated in some way. An example of this is the uncertain change
in oil drilling rig rates, which cannot be directly hedged in any existing markets, but
nonetheless, are loosely correlated with oil prices.
In this paper we demonstrate how the correlation between market and private risks
can be addressed within the framework of a contingent claims valuation when private
risks are conditioned to market risk. We solve this problem with a discrete time model
based with risk neutral probabilities, and provide a practical computational solution
for this approach based on the use of binomial decision trees. This approach is similar
to the work of Wang and Dyer [7], who develop a copula based approach for
modeling dependent multivariate uncertainties, but differs from the Smith and Nau
approach in the sense that Smith and Nau do not explicitly consider the problem of
correlation between the market and private uncertainties.
The remainder of the paper is organized as follows: Section 2 introduces the
concept decision tree analysis and risk neutral probabilities. Section 3 presents an
enhanced approach to project valuation with correlated private uncertainty. In Section
4 we apply the model to solve a sample problem and in Section 5 we conclude with a
summary and discussion of further research issues regarding model formulation and
solution procedures.
2 Decision Tree Analysis for Real Option Valuation
Real options derive their value from the managerial flexibility present in a project,
which allows firms to affect the uncertain future cash flows of a project in a way that
enhances expected returns or reduces expected losses. The flexibility to delay
investment in a project, for example, may be viewed as a call option on the project
where the required investment is the exercise price. Other typical project flexibilities
are the option to switch inputs or outputs or otherwise expand, abandon, suspend,
contract or resume operations in response to future uncertainties. Due to the option-
like characteristics of management flexibility, discounted cash flow methods cannot
be used to capture this value and one must resort to option pricing or decision analysis
methods.
Managerial flexibility can be modeled with decision tree analysis (DTA) by
incorporating the decision instances that allow the manager to maximize the value of
the project conditioned on the information available at that point in time, after several
uncertainties may have been resolved. A nave approach to valuing projects with real
options would be simply to include decision nodes corresponding to project options
into a decision tree model of the project uncertainties, and to solve the problem using
the same risk-adjusted discount rate appropriate for the project without options.
Unfortunately, this nave approach is incorrect because the optimization that occurs at
the decision nodes changes the expected future cash flows, and thus, the risk
characteristics of the project. As a consequence, the standard deviation of the project
cash flows with flexibility is not the same as that of the project without flexibility, and
the risk-adjusted discount rate initially determined for the project without options will
not be the same for the project with real options. However, real option problems can
be solved by DTA with the use of risk neutral probabilities, which implies that we can
discount the project cash flows at the risk free rate of return and make any necessary
adjustments for risk in the probabilities of each state of nature.

Fig. 1. The Project with Objective Probabilities and a Risk-adjusted Discount Rate
Up state
.50
59.1
Down state
.50
-19.1
Chance
Accept 20
Reject 0
Decision
20
Net Payoff
59.1 = 120/1.1 - 50
Net Payoff
-19.1 = 34/1.1 - 50
We illustrate this concept with an example. Suppose there is a two state project
with equal chances of cash flows of $120 or $34 one year from now that has a risk-
adjusted discount rate of 10%, and which can be implemented at a cost of $50. The
expected present value of the project is $70 = [0.5($120) + 0.5($34)]/1.10 and the
NPV is $20, as shown in Fig. 1., where the square represents a decision, the circle
represents an uncertainty and the triangles represent path endpoints.
Suppose now that the decision to commit to the project can be deferred until next
year, after the true state of nature is revealed, and that the risk free rate is 8%. The
original discount rate of 10% cannot be used because the risk of the project has now
changed due to the option to defer the investment decision. On the other hand, a set
of risk neutral probabilities for the original project can be determined and used to
value the project with the deferral option, since the expected cash flows for both
problems are the same ($120 and $34).
While the correct risk-adjusted discount rate of a project with options is difficult to
determine due to the effect these options have on the project risk, the risk free rate of
return can be readily observed in the market. By switching from objective
probabilities to risk neutral probabilities, the project NPV with options can then be
estimated even without knowing the correct risk-adjusted discount rate. In this
example this can be done by solving for the risk neutral probability p
r
in
( ) $70 ($120) (1 )($34) / 1.05
r r
p p = +
and we obtain p
r
= 0.4593.
The project with the option to defer has net payoffs of $120-$50=$70 in the up
state and zero in the down state as illustrated in Fig. 2. , as there will be no investment
if it is known beforehand that the down state will prevail. The net present value of the
project with the option to defer is $30.6 = [0.459($66.7) + 0.541($0)] / 1.05, up from
$20, which implies that the value of the option to defer is $10.62.

Fig. 2. Project with Risk Neutral Probabilities and Risk Free Discount Rate
The DTA model is based on the idea proposed by Copeland and Antikarov [5],
which requires two key assumptions: MAD (Marketed Asset Disclaimer), where the
present value of the project assumed to be the best estimate of its market value, and
that variations in the project returns follow a random walk. We refer the reader to
Accept 66.7
Reject 0
Decision
Up state
.459
66.7
Accept -15.2
Reject 0
Down state
.541
0
Chance
30.6
Net Payoff
(120 - 50)/1.05 = 66.7
Net Payoff
(34 - 50)/1.05 = -15.2
both Copeland and Antikarov [5] and Brando and Dyer [8] for a more thorough
discussion of these ideas.
While both these assumptions are also subject to a number of caveats, we will
adopt this point of view for the purpose of this discussion. Let V
i
be the value of a non
dividend paying project at time period i and V
i+1
/V
i
be its return over the time period
between i and i+1. Under the random walk assumption, the logarithm of the return
) / ln(
1 i i
V V
+
is normally distributed, and we define v and o
2
as the mean and variance
of this distribution. When the time period length is small, this stochastic model can be
expressed as an Arithmetic Brownian Motion (ABM) random walk process in the
form dz dt V d o v + = ln where dz dt c = is the standard Wiener process, (0,1) N c ~ .
Accordingly, changes in V
i
will be lognormally distributed, and can be modeled as a
Geometric Brownian Motion (GBM) stochastic process in the form
Vdz Vdt dV o + = where
2
2 v o = + . The random walk assumption implies that
any number of uncertainties in the model of the project can be combined into one
single representative uncertainty, the uncertainty associated with the stochastic
process of the project value V, and the parameters of this process can be obtained
from a Monte Carlo simulation of the project cash flows.
The value of the underlying project at time i is determined by simply discounting
the expected cash flows
| | { }
, = 1, 2, ...,
i
E C i m
at the risk-adjusted discount rate ,
such that
| |
( )
( )
m
u t i
i
i
V E C t e dt

=
}
. If the project pays dividends, then its value will
decrease in each period by the amount of dividends that is paid out. We assume these
dividends are equal to the cash flows in each period. The distribution of V
i
can be
fully defined by the mean and standard deviation of the project returns. Assuming that
markets are efficient, purchasing the project at its present value guarantees a zero
NPV and the expected return of the project will be exactly the same as its risk-
adjusted discount rate. In this sense, the mean return is exogenously defined and is
usually set equal to the firms WACC. The volatility o of the returns can be
determined from a Monte Carlo simulation of the stochastic process
ln d V vdt dz o = + where
( )
1 0
ln v V V =
%
% and . )
~
( v v = E The value of the project can
be modeled in time as a GBM stochastic process by means of a discrete recombinant
binomial lattice according to the model of Cox, Ross and Rubinstein (CCR) [9]. The
pre-dividend value of the project in each period and state is given by
, 0
i j j
i j
V V u d
=
,
where
t
u e
o A
= and
t
d e
o A
= are the parameters governing the size of the up and
down movements in the lattice. The objective probability of an up movement
occurring is
.t
e d
p
u d
, where i = period (i = 0, 1, 2, ..., m) and j = state (j = 0, 1, 2,

..., i). The continuous time stochastic process associated with this dividend-paying
project is
( )
t
dV Vdt Vdz o o = +
, where
t
is the instantaneous dividend distribution
rate at time t. Under uncertainty, the pre-dividend value of the project V
ij
in period i,
state j, is given by the recursive equation
1
, 0
1
(1 )
i
i j j
i j k
k
V V u d o
=
=
[
where the
probability P
i,j
that the value V
ij
will occur is
,
(1 )
i j j
i j
i
P p p
j

| |
=
|
\ .
.
The value of the project can be modeled in time as a GBM stochastic process by
means of a discrete recombinant binomial lattice according to the CRR model. We
choose to model the options as function of the project cash flows rather than on the
value, even though both approaches are equivalent. These cash flows, which we will
call pseudo cash flows (C
i,j
), will themselves be a function of the expected cash flows
of the project C
i
(i = 1, 2, ..., m), of and of the parameters u and d of the binomial
mode and can be shown to be defined as:

1 1
1,
1
1, ,
1
1, 1 ,
1
1
(1 )
2, 3,..., 0,1, 2,...
(1 )
j j
j
i
i j i j
i
i
i j i j
i
C
C u d i
C
C C u
C
i m j i
C
C C d
C
+
+
+
+ +
= =
`
+
)
=

+

= =
`
=
+
)
(1)
Since we are using risk neutral probabilities
rt
r
e d
p
u d
, these cash flows are

discounted at the risk free rate to arrive at the present value of the project at time
t = 0.
3 Correlated uncertainties
Situations where risks are correlated bring an additional level of complexity to the
valuation problem. The consolidation of all the market uncertainties that affect the
project resulted in the lognormal diffusion process for the project value V, which we
defined in the previous steps. We now consider the case where an additional market
or private uncertainty is conditional on the consolidated market uncertainties by
means of a correlation factor and derive a binomial approximation following the
CRR model, but now with conditional probabilities.
Let V and P be the value of the project and of the additional uncertainty at time t,
V
and
P
the respective drift rates,
V
and
P
the volatilities of each process and
and
V P
dz dz
the standard Wiener processes respectively for V and P. The diffusion
process for these risks is then given by:

where
, (0,1)
V V V V V
P P P P P i
dV Vdt Vdz dz dt
dP Pdt Pdz dz dt N
o c
o c c
= + =
= + =

with correlation
| |
, ,
.
V P
V P dz dz V P
E dz dz dt = =

Since we will be using risk neutral probabilities to obtain a solution, we must
substitute the drift rate for the risk free rate r = r(t), and the process becomes
V V
dV rVdt Vdz o = +
and
P P
dP rPdt Pdz o = +
. In the binomial mode each asset price
can go up or down at each step, so that we will have four possible states after one time
period, each with probability p
i
, i = 1,2,3,4, as shown in Fig. 3.

V, P
u
V
V, u
P
P
p
1
= p(u
V
,u
P
)
p
2
= p(u
V
,d
P
)
p
3
= p(d
V
,u
P
)
p
4
= p(d
V
,d
P
)
Probabilities
d
V
V, d
P
P
u
V
V, d
P
P
d
V
V, u
P
P

Fig. 3. States and Probabilities
The size of the up and down movements (u
V
, u
P
, d
V
, d
P
) and the value of the
probabilities p
i
must be such that the discrete probability distribution converges to the
bivariate lognormal distribution when the time period tends to zero. We achieve this
by equating the mean, variance and correlation of the binomial and continuous time
models, as suggested by Boyle, Evnine and Gibbs [10]. In analogy with CRR, we
make u
j
d
j
=1, and
j
t
j
u e
o A
=
, j = V,P. The derivation of the formulas is provided in
the appendix. The conditional probabilities for the uncertainties are:
( | ) (( , ) , ), 1, 2, 3, 4
( )
k
i j
j
p
P P V j u d i u d k
P V
= = = =
where
( )
r t
u
e d
P V
u d
A

1
( | )
4 ( )
V P
V P
u u
u
v v
t
P P V
P V
o o
| |
+ + A +
|
\ .
=

1
( | )
4 ( )
V P
V P
u d
d
v v
t
P P V
P V
o o
| |
+ A
|
\ .
=

(2)

1
( | )
4 ( )
V P
V P
d u
u
v v
t
P P V
P V
o o
| |
A
|
\ .
=

1
( | )
4 ( )
V P
V P
d d
d
v v
t
P P V
P V
o o
| |
+ A +
|
\ .
=

4 Example
We illustrate this approach to the evaluation of real options with a simple four-
period project. Due to limitations in the software used to model the problem, the
decision tree representation is essentially a binary tree augmented by decision nodes,
and it is not recombining like a binary lattice. This results in a large tree due to the
unnecessary duplication of nodes, but provides a visual interface and a convenient and
flexible modeling tool. We consider initially the case of a project subject to a market
uncertainty and next analyze the effects of a private uncertainty on the project value.
4.1 Project subject to market uncertainty
Consider a firm that has just developed a new product and is deciding whether to
invest in the manufacturing and marketing of this product. Due to the very
competitive nature of this market, the product life is expected to be no more than four
years. The spreadsheet with the expected value of the future cash flows and the
present value of the project at time zero is shown in Table 1. The risk-adjusted
discount rate is assumed to be 10% and the risk free rate is 5%.

0 1 2 3 4
Revenue 1000 1080 1166 1260
Variable Cost (400) (432) (467) (504)
Fixed Cost (240) (240) (240) (240)
Depreciation (300) (300) (300) (300)
EBIT 60 108 160 216
Tax Rate (50%) (30) (54) (80) (108)
Depreciation 300 300 300 300
Investment (1,200)
Cash Flow (1,200) 330 354 380 408

PV
0
= 1,157 WACC = 10%
Invest = (1,200)
NPV

= (43)
Table 1 Project Expected Cash Flows
The present value of $1,157 is assumed to be the best estimate of the market value
of the project and is our base case value. Since the required investment is $1,200, the
project has a negative NPV, which indicates that it should not be implemented.
We assume that the project is subject to a single source of market uncertainty, the
future value of its revenue stream; although other sources of market uncertainties
could be easily incorporated into the model by adding additional uncertainty
distributions to the simulation. Suppose the future project revenues R follow a GBM
diffusion process with a mean
R
= 7.70% (equivalent to a discrete annual growth of
8.0%) and volatility
R
= 30%. Using these parameters, a Monte Carlo simulation of
the project cash flows may be used to compute the standard deviation of
( )
1 0
ln / v V V =
%
% , and to obtain an estimate of the project volatility = 24.8%. Finally,
we assume that the project rate of return is normally distributed, so the project value
will have a lognormal distribution at any point in time that may be approximated by a
binomial lattice.
Modeling the binomial approximation requires that we determine the values of u,
d, and the risk neutral probability p
r
, according to the formulas defined previously.
The pseudo cash flows of the project are computed using equations shown in (1), and
the value of the project is determined applying the usual procedures of dynamic
programming implemented in a binomial tree, and discounting the expected cash
flows at the risk free rate of return. Risk neutral probabilities are used to arrive at the
project expected value, which is the same as the one calculated with the spreadsheet.
Note that the values for o, , r and the project expected cash flows C
i
can be entered
as parameters in a decision tree model, and all the necessary formulae can be
incorporated into the tree structure. In effect, tree building can be greatly simplified
by developing a standard template for a binary tree for any given number of time
periods.
Once the projects stochastic parameters are determined and the decision tree is
structured, the project options can be added with ease. Suppose the project can be
abandoned in years two and three for a constant terminal value of $350, and that there
exists an option to expand the project by 30% in year 2 at a cost of $100. Given the
binary tree representation, these options can be evaluated by simply inserting the
appropriate decision nodes in the time period that models the existing managerial
flexibility in each year.
The decision tree model is shown in Fig. 4. The project value, computed using the
same risk neutral probabilities, increases to $1,280, and the expansion option will be
exercised in all states of year 2, except one, while the abandon option will continue to
be exercised only in year 3, as can be seen by the lines in bold. Additional options and
time periods can be added in a straightforward manner.
4.2 Project subject to correlated uncertainties
Traditional financial theory seeks to obtain market values for assets, and assumes
that firms and/or their shareholders are sufficiently diversified so that they become
risk neutral in relation to private risks since these risks can be eliminated by an
adequate diversification strategy. In this case, the private, firm specific risks must be
computed at their expected values and discounted at the risk free rate in order to
estimate the market value of a project. On the other hand, for small family and owner
operated businesses, for employees that hold large stock investment in the firms they
work for or managers that have significant amount of stock options, this may not be
the case.


Fig. 4. Decision Tree with Option to Expand and Abandon
To solve this problem we consider the partially complete market concept of Smith
and Nau [6] and decompose the project cash flows into their market and private
components and value each one separately: option pricing in complete markets for the
market risk and use a utility function that reflects the firms risk preferences to value
the private risk, if we consider an undiversified investor. In our example, the private
risk is due to the fact that the projects cash flows are also affected by the efficiency
of the plant, which in turn is correlated with the firms manufacturing technology.
Since the firms technology cannot be hedged in the market, this results in a private
risk for which we have no way of determining an appropriate discount rate unless we
make restrictive assumptions about the firms utility function. We will also assume
that the efficiency of the plant is positively correlated with the project value, since a
T4
Continue [2044]
Abandon
302.3
[1688]
Dec 3
High
674.5 .538
[2044]
T4
Continue [1523]
Abandon
302.3
[1424]
Dec 3
Low
410.8 .462
[1523]
T3
Expand
-90.7
[1803]
T3
Continue [1642]
Abandon
317.5
[1119]
Dec 2
High
435.8 .538
[1803]
T4
Continue [1352]
Abandon
302.3
[1254]
Dec 3
High
410.8 .538
[1352]
T4
Continue [1035]
Abandon
302.3
[1093]
Dec 3
Low
250.1 .462
[1093]
T3
Expand
-90.7
[1233]
T3
Continue [1196]
Abandon
317.5
[949]
Dec 2
Low
265.4 .462
[1233]
T2
High
366.1 .538
[1540]
T4
Continue [1209]
Abandon
302.3
[1111]
Dec 3
High
410.8 .538
[1209]
T4
Continue [891.9]
Abandon
302.3
[950.1]
Dec 3
Low
250.1 .462
[950.1]
T3
Expand
-90.7
[1090]
T3
Continue [1053]
Abandon
317.5
[805.8]
Dec 2
High
265.4 .538
[1090]
T3
Expand
-90.7
[801.2]
T4
Continue [764.8]
Abandon
302.3
[879.3]
Dec 3
High
192.4 .538
[879.3]
T4
Continue [616.1]
Abandon
302.3
[804.1]
Dec 3
Low
117.2 .462
[804.1]
T3
Continue [844.6]
Abandon
317.5
[702]
Dec 2
Low
161.6 .462
[844.6]
T2
Low
223 .462
[976.4]
T1
[1280]
greater cash flow stream from higher volumes and prices would generate
manufacturing economies of scale and more investment in manufacturing technology.
We assume that the plant operates with a mean efficiency of 80 %, volatility of
0.10 and correlation of = 0.20. The conditional probabilities for the private risk are
determined from equations (2) and the private uncertainty is added as a chance node
in each period. Fig. 5. shows the model for the base case. If the investor is fully
diversified, the we would expect the value of the project to be smaller, since the mean
efficiency is less than 100%, and accordingly, solving the decision tree provides a
value of $1,058 for the base case, as compared to $1,157.

Fig. 5. Private Uncertainty: Base Case
The project options can be determined by inserting the appropriate decision nodes
in each time period in the same manner as before. The decision tree model is shown in
Fig. 6. , and the solution to this tree provides the value of $1,181.

Fig. 6. Model with correlated private risk
On the other hand, if the firm or its investors are not sufficiently diversified and
this investment represents a significant portion of their wealth, they may be risk
averse toward private risks. In this case, assuming an exponential utility function
( )
x RT
u x e
= and a risk tolerance of $200, this level of risk aversion leads to the
High
T2*Priv2/(1+r)^2
Low
T2*Priv2/(1+r)^2
a
High
Low
Priv2
High
T1*Priv1/(1+r)
Low
T1*Priv1/(1+r)
T2
High
Low
Priv1 T1
High
T4*Priv4/(1+r)^4
Low
T4*Priv4/(1+r)^4
High
Low
Priv4
High
T3*Priv3/(1+r)^3
Low
T3*Priv3/(1+r)^3
T4
High
Low
Priv3
a
T3
Expand
-Invest/(1+r)^2
a
Continue
a
Abandon
Abn_Value/(1+r)^2
High
T2*Priv2/(1+r)^2
Low
T2*Priv2/(1+r)^2
Dec 2
High
Low
Priv2
High
T1*Priv1/(1+r)
Low
T1*Priv1/(1+r)
T2
High
Low
Priv1 T1
High
T4*Priv4/(1+r)^4
Low
T4*Priv4/(1+r)^4
High
Low
Priv4
Continue
T4
Abandon
Abn_Value/(1+r)^3
High
T3*Priv3/(1+r)^3
Low
T3*Priv3/(1+r)^3
Dec 3
High
Low
Priv3
a
T3
project value decreasing to $1,134. The breakeven point for the risk tolerance factor is
$134, below which the optimal project strategy changes. For a further discussion on
the use of utility functions for investment valuation see Kasanen & Trigeorgis [11].
5 Conclusions and Recommendations
The method proposed represents a simple and straightforward way of
implementing real option valuation techniques using standard decision tree tools
easily available in the market. By modeling the correlated uncertainty explicitly,
private or otherwise, into the problem, a more accurate estimate of the project value
can be obtained that takes into account the possibility the different natures of these
uncertainties and their correlation.
Even for a simple model such as this one, the decision tree becomes large very
quickly. In most practical problems the complexity of the decision tree will be such
that full visualization will be impossible; however, even large problems with literally
millions of endpoints for the tree can be solved using this approach. Additional
computational efficiencies can be obtained by using specially coded algorithms,
although at the cost of having to forgo the simple user interface that decision tree
programs offer and the advantage of visual modeling and a logical representation.
Suggested extensions include the implementation of recombining lattice capability in
current decision tree generating software to cut down on processing time. While a n
period recombining binary lattice has a total of n(n+1)/2 nodes, a similar binary tree
has 2
n+1
-1 nodes, which becomes a significant difference for large values of n. On
the other hand, the extension of this model to projects with non-constant volatility can
be easily implemented, whereas the effect of changes in volatility cannot be modeled
with a recombining lattice.
Perhaps the primary caveat regarding this methodology for the evaluation of
projects with real options relates to the assumptions underlying the Copeland and
Antikarov [5] approach itself, since the use of decision trees is simply a
computational enhancement of their concepts. The use of the Market Asset
Disclaimer as the basis for creating a complete market for an asset that is not traded
may lead to significant errors, since the valuation is based on assumptions regarding
the project value that cannot be tested in the market place. For example, the
appropriate choice of the project discount rate for the project without options is left to
the discretion of the analyst, and the use of WACC may not be appropriate for all
projects. Therefore, it is important to realize that this thorny issue is not resolved by
this methodology.

6 References
1. Black, F. and M. Scholes, The Pricing of Options and Corporate Liabilities.
The Journal of Political Economy, 1973. 81(3): p. 637-654.
2. Merton, R.C., Theory of Rational Option Pricing. Bell Journal of Economics
and Management Science, 1973(4): p. 141-183.
3. Dixit, A.K. and R.S. Pindyck, Investment under Uncertainty. 1994, Princeton:
Princeton University Press. 476.
4. Trigeorgis, L., Real options, Managerial Flexibility and Strategy in Resources
Allocation. 1996, Cambridge, Massachussets: MIT Press.
5. Copeland, T. and V. Antikarov, Real Options: A Practitioners Guide. 2003,
Texere, New York. 368.
6. Smith, J.E. and R.F. Nau, Valuing Risky Projects: Option Pricing Theory and
Decision Analysis. Management Science, 1995. 41(5): p. 795-816.
7. Wang, T. and J.S. Dyer, A copula based approach for modeling Dependence in
Decision Trees. Forthcoming in Operations Research, 2011.
8. Brandao, L. and J.S. Dyer, Decision analysis and real options: A discrete time
approach to real option valuation. Annals of Operations Research, 2005.
135(1): p. 21-39.
9. Cox, J.C., S.A. Ross, and M. Rubinstein, Option pricing: A simplified
approach. Journal of Financial Economics, 1979. 7(3): p. 229-263.
10. Boyle, P., J. Evnine, and S. Gibbs, Numerical evaluation of multivariate
contingent claims. Review of Financial Studies, 1989. 2(2): p. 241-250.
11. Kasanen, E., & Trigeorgis, L. (1994). A market utility approach to investment
valuation. European Journal of Operational Research, 74(2), 294-309.

7 Appendix: Conditional probabilities for a bivariate
distribution
To simplify the analysis, we convert to the natural logarithm of the variables, so
that x
i
= ln(S
i
). It can then be shown through an Ito process that the log of a GBM is
an ABM of the form:
ln
2
ln , (0,1)
2
V
V V V
P
P P P i i i
dV dx r dt dz
dP dx r dt dz where dz dt N
o
o
o
o c c
| |
= = +
|
\ .
| |
= = + =
|
\ .

Discretizing the time steps we have:
V V V V
P P P P
x v t z
x v t z
o
o
A = A + A
A = A + A
where
2
2
i
i
v r
o
=

The mean, variance and correlation of the continuous time process are:

| | | |
V V V V V
E x E v t z v t o A = A + A = A

| | | |
P P P P P
E x E v t z v t o A = A + A = A

| | | |
2 2
V V V
Var x E x E x ( A = A A

But since
| | ( )
2
2
0
V V
E x v t A = A =
, we remain with
| |
2 2
V V V
Var x E x t o ( A = A = A

| |
2
P P
Var x t o A = A

and finally,
| |
V P V P V P
E x x E z z o o ( A A = A A

, but
| |
V P
t E z z A = A A
, so we have
| |
V P V P
E x x t o o A A = A

The discrete binomial distribution yields:
| |
1 2 3 4
( ) ( )
V V V V V
E x p x p x p x p x A = A + A + A + A

| |
1 2 3 4
( ) ( )
V V V
E x p p x p p x A = + A + A

| |
1 2 3 4
( ) ( )
P P P P P
E x p x p x p x p x A = A + A + A + A

| |
1 2 3 4
( ) ( )
P P P
E x p p x p p x A = + A + A

( ) ( )
2 2
2 2 2
1 2 3 4 V V V V V
E x E p x p x p x p x
(
( A = A + A + A + A

2 2
V V
E x x ( A = A

2 2
P P
E x x ( A = A

| |
1 2 3 4 V P V P V P V P V P
E x x p x x p x x p x x p x x A A = A A A A A A + A A

| |
1 2 3 4
( )
V P V P
E x x p p p p x x A A = + A A

We equate the first and second moments of the discrete distribution to the
continuous distribution. Equating also the correlations and making the sum of the
probabilities add to one, we arrive at a system of six equations with six unknowns.

1 2 3 4
( ) ( )
V V V
v t p p x p p x A = + A + A
(1.1)
1 2 3 4
( ) ( )
P P P
v t p p x p p x A = + A + A
(1.2)
2 2
V V
t x o A = A
(1.3)
2 2
P P
t x o A = A (1.4)
1 2 3 4
( )
V P V P
t p p p p x x o o A = + A A
(1.5)
1 2 3 4
1 p p p p + + + =
(1.6)
Solution:
V V P P
x t and x t o o A = A A = A
1
1
1
4
V P
V P
v v
p t
o o
| | | |
= + + A +
| |
|
\ . \ .
(1.7)
2
1
1
4
V P
V P
v v
p t
o o
| | | |
= + A
| |
|
\ . \ .
(1.8)
3
1
1
4
V P
V P
v v
p t
o o
| | | |
= A
| |
|
\ . \ .
(1.9)
4
1
1
4
V P
V P
v v
p t
o o
| | | |
= + A +
| |
|
\ . \ .
(1.10)
The conditional probabilities can be obtained from the joint probabilities of the
market and private uncertainties.
( )
( | ) (( , ) , )
( )
i j
i j
j
P P V
P P V j u d i u d
P V
= = =

where
( ) ( , ) , 1, 2, 3, 4
i j k
P P V p j u d i u d k = = = =

Estimating Changing Volatility in Cash Flow
Simulation-Based Real Option Valuation with the
Regression Sum of Squares Error Method
Tero Haahtela
Aalto University, School of Science and Technology
P.O. Box 15500, FI-00076 Aalto, Finland
tero.haahtela@aalto.fi
Abstract. This paper presents a volatility estimation method for cash flow
simulation based real option valuation with changing volatility. During cash
flow simulation, the present values of future cash flows and their corresponding
cash flow state variable values are recorded for all time periods. Then, for each
time period, regression analysis is used to relate the present value to the cash
flow state variables of the same time period. Each regression equation provides
an estimate of the expected present value as a function conditioned on the
resolution of all uncertainties up to that time. Then, basic regression statistics of
Pearsons correlation R
2
and the sum of squares error (or standard error) for
each equation provides all the information required for estimating how the
standard deviation and volatility of the stochastic process change over time. The
method requires only one pass of simulation runs, allows negative underlying
asset values, and is easy for a practitioner to apply.

Keywords: Real options, cash flow simulation, volatility estimation

1 Introduction
Real options analysis is a framework for valuing managerial flexibility under
uncertainty. It has adopted advanced methods from financial derivatives valuation and
made valuation of projects with several sequential and parallel decision alternatives
more accessible. Difficulty in volatility estimation has been one of the reasons for the
slow acceptance of the new valuation framework [1, 2]. Volatility is probably the
most difficult input parameter to estimate in real options analysis [1], and this is also
the case with financial options. However, volatility estimation in the case of financial
options is easier because of past observable historical data and future price
information. With real options, especially if related to R&D, such information is not
necessarily available [3, 4]. Therefore, volatility estimation has to be based on some
other method.
One alternative is to use Monte Carlo simulation for the gross present value and
volatility estimation [1, 2, 5-15]. In this approach, forecast data for future cash values
with probabilities is converted into an estimated underlying asset value and volatility.
ISSN 1799-2737 Open Access http://www.rojournal.com 33
The cash flow model is usually a gross present value (i.e. a conventional net present
value of the completed project's expected operating revenues less investments) where
uncertainties related to parameters are presented as objective or subjective
distributions with different correlations, time series and other constraints. After the
simulation, the mean and the standard deviation of the rate of return, i.e. volatility, are
calculated.
Monte Carlo simulation on cash flow calculation consolidates a high-dimensional
stochastic process of several correlated variables into a low-dimension process. The
most common assumption is the univariate geometric Brownian motion (gBm)
process. Firstly, the Monte Carlo simulation technique is applied to develop a
probability distribution for the rate of return. Then, volatility parameter (or other
parameters for stochastic processes other than gBm) of the underlying asset can be
estimated with alternative approaches, of which the most common is presumably
calculating the standard deviation of the simulated probability distribution for the rate
of return.
Several authors have suggested different approaches for applying Monte Carlo
simulation on cash flow calculation to estimate the volatility. The existing cash flow
simulation-based volatility estimation methods are the logarithmic present value
approach of Copeland and Antikarov [6] and Herath and Park [7], the conditional
logarithmic present value approach of Brando, Dyer and Hahn [8, 9], the two-level
simulation and the least-squares regression methods of Godinho [10], and the
generalized risk-neutral volatility estimation over different time periods [2, 13-15].
All these methods are based on the same basic idea. The Monte Carlo simulation
technique is applied to develop a probability distribution for the rate of return. Then,
the volatility parameter of the underlying asset is estimated by calculating the
standard deviation of the rate of return.
The previously mentioned methods have different strengths and qualities. Different
aspects related to them are theoretical correctness, computational efficiency,
capability to handle negative underlying asset values, capability to estimate changing
volatilities, ability to separate ambiguity from volatility, and ease of use from the
intuitive aspects of understanding the logic behind the volatility estimation as well as
ease of use in terms of available software to do the estimation. Luckily, not all of
these aspects are relevant in all practical valuation cases. Therefore, usability of the
methods depends heavily on the case in which they are applied.
The approach presented in this paper can be considered a mixture of the previously
mentioned methods. The purpose is to sustain and combine many of their good
qualities while keeping the procedure as straightforward as possible for a practitioner.
In the method presented in this paper, the present value of the cash flows and the cash
flow state variable values are recorded for each time period during cash flow
simulation runs. Then a regression analysis is run for relating the PV for each year to
the corresponding cash flow state variables. Each regression equation provides an
estimate of the expected present value as a function conditioned on the resolution of
all uncertainties up to that time. Then, the basic regression statistics of Pearsons
correlation R
2
and sum of squares error (or standard error) for each equation provide
all the information required for estimating how the standard deviation and volatility
change over time. The method is computationally efficient as it requires only a single
pass of simulation runs, it is easy to understand and use, and it can handle changing
volatility and negative underlying asset values.
After introductory Section 1, Section 2 discusses in detail earlier cash flow
simulation based methods, because the procedure presented in this paper is based
heavily on their ideas and results. Section 3 presents the intuition behind the
suggested method, the idea of the proportion of unsolved and solved uncertainty
during the projects timespan. The main contribution of this paper is in Section 4
which describes the regression sum of squares error and R
2
based method for
estimating changing volatility in cash flow simulation-based real options valuation.
Section 5 illustrates the use of this method in a case example. Finally, Section 6
concludes the paper.
2 Cash Flow Simulation Based Volatility Estimation Methods
2.1 Logarithmic Present Value Approach of Copeland and Antikarov
Copeland & Antikarov [6] (henceforth C&A) presented the first detailed
explanation of the use of Monte Carlo simulation for volatility estimation based on
cash flows. This logarithmic present value approach in terms of Mun [1, 16] is based
on several assumptions. Firstly, it relies on the marketed asset disclaimer and
Samuelsons proof [17] that the correctly estimated rate of return of any asset follows
random walk regardless of the pattern of the cash flows. Secondly, the approach is
based on the idea that an investment with real options should be valued as if it were a
traded asset in markets even though it would not be publicly listed. Thirdly, the
present value of the cash flows of the project without flexibility is the best unbiased
estimate of the market value of the project were it a traded asset. This is called the
marketed asset disclaimer assumption. Therefore, the simulation of cash flows should
provide a reliable estimate of the investments volatility.
The idea of C&As logarithmic present value approach is similarly to most other
consolidated volatility approaches analogous to stock price simulations where the
theoretical stock price is the sum of all future dividend cash payments, and with real
options, these cash payments are the free cash flows. The sum of free cash flows
present value PV
0
at time zero is the current stock price (asset value), and at time one,
the stock price PV
1
in the future. As the stock price at time zero is known while the
future stock price is uncertain, only the uncertain future stock price is simulated. As a
result, the C&A approach uses the Monte Carlo simulation on a projects present
value to develop a hypothetical distribution of one period returns. On each simulation
trial run, the value of the future cash flows is estimated at two time periods, one for
the present time and another for the first time period. The cash flows are discounted
and summed to the time 0 and 1, and the following logarithmic ratio is calculated
according to Equation (1):

z = ln[
Pv
1
+PCP
1
Pv
0
(1)
where PV
1
means present value at time t = 1, FCF
1
means free cash flow at time 1,
and PV
0
projects present value at the beginning of the project at time t = 0. Then,
volatility is calculated as the standard deviation of z.
Modifications to this method include duplicating the cash flows and simulating
only the numerator cash flows while keeping the denominator value constant. This
reduces measurement risks of auto-correlated cash flows and negative cash flows [1].
Whereas simulating logarithmic cash flows gives a distribution of volatilities, and
therefore also a distribution of different real options values, this alternative gives a
single-point estimate.
Although the fundamental idea of C&As volatility estimation approach is correct,
it has one significant technical deficiency, as noted by Smith [18] and Godinho [10].
The method would be an appropriate volatility estimate if the PV
1
were period 1s
expected NPV of subsequent cash flows and this volatility would reflect the
resolution of a single years uncertainty and its impact on expectations for future cash
flows. However, in C&As solution, this PV
1
is the NPV of a particular realization of
future cash flows that is generated in the simulation, and therefore the calculated
standard deviation is the outcome of all future. Therefore, the approach overestimates
the volatility [10, 18].
2.2 Logarithmic Present Value Approach of Herath and Park
Herath and Parks [7] volatility estimation is very similar to C&As and is based on
the same Equation (1). However, whereas in C&A only the numerator is simulated
and the denominator is kept constant, Herath and Park [6] apply simulation to both the
numerator and denominator with different independent random variables: both
PV
0
and PV
1
are independent random variables. Therefore, a different set of random
number sequences has to be generated when calculating PV
0
and PV
1
. However,
this alternative has the same over-estimation problem as the original C&A, and by
having a non-constant denominator, the approach actually worsens the situation and
over estimation of the volatility.
2.3 Conditional Volatility Estimation of Brando, Dyer and Hahn
Other authors have resolved the original problem of C&As approach. The
conditional volatility estimation of Brando, Dyer, and Hahn [8, 9], based on
comments of Smith [19], and similarly to Godinho [10], suggest an alternative where
the simulation model is changed so that only the first years cash flow FCF
1
is
stochastic, and the following cash flows FCF
2
,,FCF
n
are specified as expected
values conditional on the outcomes of FCF
1.
Thus, the only variability captured in PV
1

is due to the uncertainty resolved up to that point. The method works well if the
conditional future values are straightforward to calculate or estimate. Then, the
standard deviation of the following Equation (2) is used to estimate the volatility of
the rate of return:

z = ln [
PCP
1
+Pv
1
(L
1
(PCP
2
),,L
1
(PCP
n
)|PCP
1
)
Pv
0
(2)

The deficiency with the method is that it may be hard to compute the expected
future values given the values simulated in earlier periods. This is true especially for
both auto- and cross-correlated input variables in cash flow simulations.
2.4 Two-level Simulation of Godinho
The two-level simulation of Godinho [10] is also based on the idea of
conditionality in expected cash flows given stochastic C
1
. In comparison with the
conditional volatility estimation, it works also in situations where conditional
outcomes given C
1
cannot be calculated analytically. Firstly, the simulation is done
for the project behavior in the first year. Secondly, project behavior given the first
year information is simulated for the rest of the project life cycle. Thirdly, average
cash flows after the first year are used to calculate PV
1
, which is then used to
calculate a sample of z. Finally, the volatility of z (standard deviation) is calculated.
The method provides correct estimation of volatility, but it suffers mostly from the
required computation time. This is because the calculation is iterative, meaning that
after each first year simulation, a large number of second-stage simulations is
required. Therefore, the total number of simulations is the product of first and second
stage simulation runs. In practice, whereas other methods compute the volatility
within a few seconds even with larger models, this procedure requires at least several
minutes of computation time with present computers and algorithms.
2.5 Least Squares Regression Method of Godinho
Inspired by Longstaff and Schwartz [19]
1
, Godinho [10] presents the least squares
regression method for volatility estimation. This procedure consists of two
simulations. In the first simulation, the behavior of the project value and the first year
information is simulated. Then, PV
1
is explained with linear regression with first year
information as follows according to Equation (3):

PI
1
= o
0
+[
1
X
1
+[
2
X
2
++[
n
X
3
(3)

Then, in the second simulation round, volatility is calculated as the standard
deviation of z:

1
Who got their idea from Carriere [20]
z = ln [
Pv
1
Pv
0
(4)
Often a good and straightforward approximation is to use the first years free cash
flow FCF
1
as the explaining variable with intercept term. Then, in the second
simulation round, only the first year cash flow is simulated, and the estimation model
is used to calculate the expected value of PV
1
for calculating the sample z. Similarly
to conditional volatility estimation [9] and two-level simulation [10], this method also
estimates volatility correctly. This least-squares volatility estimation approach is
computationally fast with only two simulation rounds required and it is also much
easier to apply in practice than conditional volatility estimation.
2.6 Generalized Risk-neutral Valuation Approach
Another straightforward volatility estimation approach is based directly on the
assumptions and qualities of the geometric Brownian motion and its lognormal
underlying asset value distribution. Smith [18] suggests that correct parameterization
for the mean value and volatility could be found by changing the volatility until the
underlying asset mean and standard deviation match the simulated cash flow and its
standard deviation. However, if common gBm assumptions hold, this can actually be
solved analytically [2, 13-15]. Given that PV
0
is known, and it is possible to simulate
future cash flows, distribution of the cash flows in future can be simulated. As well as
discounting all the cash flows to the present value, they can also be undiscounted to
their future value. Because the present value of cash flows is known (PV
0
), as well as
their simulated undiscounted future value with standard deviation, it is possible to
find the volatility parameter analytically without any unnecessary additional
computations and simulations. It is known that in a risk-neutral world, asset value
increases with time according to Equation (5), and that the standard deviation of the
process increases according to Equation (6).

S
t
= S
0
c
t
(5)

stJ(S
t
) = S
0
c
t
c
c
2
t
- 1 = S
0
c
t
c
Z c
i
2
t
i
-1 (6)

Therefore, if the standard deviations of the underlying asset process at certain time
points are known, it is possible to compute the average volatility for each time period.
Starting from the beginning of the process, each
i
can be calculated according to
Equation (7) as

o
=
_
In__
std(S)
i
S
0
c
t
]
2
+1_- (c
i
2
t
i
)
i-1
i=0
t
i
(7)
The information required is the length of time period T for volatility estimation,
value of asset S
0
at the beginning, interest rate , and standard deviation of the asset
value std(S) at the end of the volatility estimation period. Interest rate does not
change the volatility estimation as long as the same interest rate is used both in cash
flow simulation and when computing the volatility. Therefore, even if the risky
expected return is used in volatility estimation, the option valuation still follows risk-
neutral pricing with risk-free interest rate used.
2.7 Least Squares Regression Method with the Risk-neutral Approach of
Haahtela
Haahtela [14] further extends the idea of Godinho [10] and properties of
generalized risk-neutral valuation for estimating changing volatilities for different
time periods
2
. The approach uses ordinary least squares regression for estimating PV
t

with cash flow simulation state variables as explaining input parameters. However,
instead of the common approach of directly estimating volatility as the standard
deviation of the rate of return as z = ln(S
t+1
/S
t
), PV regression or response surface
estimators are used directly for estimating both the underlying asset value and its
standard deviations at different time points. Figure 1 illustrates this. Based on this
information, volatilities are estimated for different time periods. After cash flow
simulation and constructing the regression equations, the actual procedure for the
standard deviation calculation is presented as follows:

Calculate the estimated expected value for PI
t
from the cash flow parameters X

i,k

generated in the simulation trial run using the regression equation.
Calculate the differences between values predicted by the regression estimator and
the values actually observed from the realized simulations.
a) Square the differences,
b) add them together,
c) divide them by the number of observation, and
d) take the square root.

After that, volatility for different time periods is calculated period by period
according to Equation (7). While the method provides sound volatility estimation, it is
not very usable or comfortable in practice. If there are changes in any cash flow
simulation parameters, the whole volatility estimation procedure has to be repeated
again for all the time periods. The advantage of the approach is that it also allows
negative underlying asset values and use of the displaced diffusion process instead of
the commonly assumed geometric Brownian motion.

2
Volatility estimation is not the main topic of the paper, it is only presented as an example of
that it is possible to estimate changing volatilities with regression equations based on
simulated cash flows

Fig. 1. Response surface equations are used to estimate underlying asset value for different
time periods. This information can be used to estimate volatility according to the generalized
risk-neutral valuation approach assuming geometric Brownian motion.
3 Intuition Behind the Regression Sum of Squares Error
Method: Proportion of Solved and Unsolved Uncertainty
Before going to the actual regression solution procedure, it is good to illustrate the
idea behind the calculation. Usually, uncertainty reveals itself over time and we have
a better understanding and knowledge about the situation and the investment viability
of a project during its later stage. Also, proportional uncertainty is likely to decline
over time, especially in the case of R&D and learning. Therefore, volatility is often
time changing and declining. Finally, all the uncertainty has been revealed and there
is no uncertainty and volatility left anymore.


Fig. 2. Unsolved uncertainty and solved uncertainty during investment. Usually uncertainty
resolves faster in the early stages of a project.
Figure 2 illustrates this situation, and the same is demonstrated in the forth-coming
example. In the beginning, we estimate the uncertainty of the investment project (in
terms of variance) to be 12 000. During each time period of the process, more and
more of the uncertainty has been revealed. At the end, all the uncertainty has been
revealed. Now, if we were able to construct good forward-looking value estimators
PV
1
PV
n
for each time period, we could estimate the overall uncertainty solved (and
unsolved) for each time point, and this information could then be used to calculate the
volatility for each time period according to Equation (7).
For simplicity, we assume in this example use of arithmetic Brownian motion, and
the interest rate is assumed to be zero. Arithmetic Brownian motion is normally
distributed with mean and variance
2
. The sum of normally distributed variables is

J(p, o
2
) =J(p
1
, o
1
2
) ++J(p
, o
2
) = J(p
1
++p
, o
1
2
+ +o
2
) (8)

The standard deviation of the arithmetic Brownian motion is:

stJ(S
t
) = S
0
c
1
o
2
t
(9)

The example has only four annual cash flows, each with an expected value of 100.
These cash flows are all normally distributed with an expected value of 100 and the
standard deviation for time periods 14 are 80, 60, 40, and 20. Therefore, based on
the properties of normal distribution, the sum of these cash flows is 400 and the
variance is 12 000, which is the same as our present value of cash flows because the
interest rate is assumed to be zero.

Table 1: Example calculation of how uncertainty reveals itself over time.
Cash flows, variance and resolving of uncertainty during time periods
Time period 1 2 3 4
Cash flow N(100; 80
2
) N(100; 60
2
) N(100; 40
2
) N(100; 20
2
)

Variance (variation, uncertainty) of
the time period
6 400 3 600 1 600 400

Solved (explained) uncertainty
after time period
6 400 10 000 11 600 12 000

Unsolved (unexplained) uncertainty
after time period
5 600 2 000 400 0

Proportion of total variance of 12 000

0.533 0.300 0.133 0.033
Cumulative proportion of solved
uncertainty
0.533 0.833 0.967 1.000
Proportion of unsolved uncertainty 0.467 0.167 0.033 0.000
Volatility of the time period
20%
(80/400)
15%
(60/400)
10%
(40/400)
5%
(20/400)

If we consider this cash flow calculation period by period, we can recognize how
the variance uncertainty reveals over time. Table (1) shows this. After we know the
realization of the first cash flow, N(100; 80
2
), actually 6 400 (or 53,3 %) of the overall
variance and uncertainty of 12 000 has been resolved, leaving 5 600 of the uncertainty
unsolved. During the second time period, 3 600 of the total variance is solved, and
thus the cumulative portion of solved uncertainty is 10 000, leaving 2 000 left as
unsolved uncertainty. After the third time period, 1 600 more is solved, and in the
final stage, the remaining 400 is resolved. As a result, the amount of solved
uncertainty increases over time until all the uncertainty is solved, while the amount of
unsolved uncertainty diminishes. This is precisely the same as demonstrated in Figure
2.
When the variance and standard deviation changes between time periods are
known, this information can be used for the volatility estimation. Starting from the
last time period 4, and using Equation (9) with the given parameters Std(S
4
) = 20, S
0

= 400, r = 0 %, and t
4
= 1, we get
4
= 0.05. This means that standard deviation is 5 %
of the expected value. This is also quite self-evident, given that we already knew that
standard deviation (20) is 5% of the expected value (400). Then, having this
information available, we can step backwards into time period 3 and knowing also
Std(S
3
), calculate the volatility
3
. This is 0.10, which is also consistent with standard
deviation of 40 being 10% of the mean value 400. With the same backward logic,
respectively, we can calculate
2
= 0.15 and
1
= 0.20.
4 Regression-Based Volatility Estimation
As mentioned in the example of the previous section, if we are able to construct
good forward-looking value estimators PV
1
PV
n
with information of their expected
values and standard deviations for each time period, we can estimate the overall
uncertainty solved (and unsolved) for each time point, and this information can then
be used for volatility estimation. An alternative way to do this is to use ordinary least
squares regression equations.
In statistics and econometrics, ordinary least squares (OLS), or linear least squares,
is a method for estimating unknown parameters in a linear regression model. The
method is a standard approach to the approximate solution of overdetermined
systems, i.e. sets of equations in which there are more equations than unknowns. The
most important application is in data fitting. The OLS method minimizes the sum of
squared distances, also called residuals or squares of errors, between the observed
responses in the dataset, and the responses predicted by the linear approximation
model. The resulting estimator can be expressed by a simple formula, especially in the
case of a single regressor on the right-hand side.
Regression analysis includes any techniques for modeling and analyzing several
variables, when the focus is on the relationship between a dependent variable and one
or more independent variables. Regression analysis helps us understand how the
typical value of the dependent variable changes when any one of the independent
variables is varied, while the other independent variables are held fixed. Most
commonly, regression analysis estimates the conditional expectation of the dependent
variable given the independent variables that is, the average value of the dependent
variable when the independent variables are held fixed. In all cases, the estimation
target is a function of the independent variables called the regression function. In
regression analysis, it is also of interest to characterize the variation of the dependent
variable around the regression function, which can be described by a probability
distribution.
Regression divides the sum of squares (SS) in Y, total variation, into two parts: the
sum of squares predicted due regression (SSDR) and the sum of squares error (SSE),
so that SS = SSDR + SSE. SSDR is the sum of the squared deviations of the predicted
scores from the mean predicted, and the SSE is the sum of the squared errors of
prediction. The SSDR therefore describes the variation that the regression model can
explain and the SSE is the unexplained variation. The square root of the SSE is the
standard error.
A common measure of the regression model is a coefficient of determination, R
2
. It
is the proportion (percentage) of the sum of squares explained due to regression
divided by the total sum of squares, i.e. R
2
= SSDR/SS. R
2
is the same as Pearsons
correlation measure between the predicted and observed values. As a result, if we
know R
2
and standard error, we can also calculate the explained variation, the square
root of SSDR, according to:

ExploincJ :oriotion = _[
Std.Lo
2
1-R
2
-StJ. Error
2
(10)

As a result, we can use regression equations to estimate the expected value and
standard deviation of PV, time period by time period, by linking the explaining cash
flow variables to the explanatory estimators of PV
t
. These explaining variables can be
any parameters in the cash flow calculation. Often sufficient explanatory variables for
estimating PV
t
are free cash flows of the given and the previous years
3
, i.e.
FCF
0,
,FCF
t
. As a result, we can calculate PV
0
according to Equation (11). This
increases for the following time periods with the risk-free interest rate according to
Equation (12).

PI
0
=
CP
t
(1+)
t
1
t=1
(11)

PI
1
= PI
0
c
1
(12)

Then, we can construct the regression estimators as suggested for each time period
according to the following equations:

PI
1
= o
1
+[
1,1
FCF
1
(13)
PI
2
= o
2
+[
1,2
FCF
1
+[
2,2
FCF
2
(14)
:
PI
t
= o
t
+[
1,t
FCF
1
+[
2,t
FCF
2
++[
t,t
FCF
t
(15)

Now we can use Monte Carlo simulation on our example cash flow model. Before
the cash flow is simulation run, we set the free cash flows of the time periods (FCF
t
)
and the present values of project (PV
t
) as output parameters. These parameter values
are then saved during the simulation run so they can be used in regression calculations
after the simulation. Then we use regression Equations (13 - 15) for forecasting the
expected values and standard errors of PVs.
There are many different alternatives to conduct the simple OLS regression
analysis. Even the basic tools provided with the common spreadsheet programs (e.g.
MS Excel or OpenOffice Calc) are sufficient for this purpose
4
. As an example of this,
the following Table 2 is taken from the Excel summary output regression statistics
report for the PVs of the previous example. As we can see from the results in Table
2, the numbers of R Square are very close to those of Cumulative proportion of solved
uncertainty in Table 1.

3
Haahtela [14] has more details about constructing good regression estimators.
4
In practice, majority of the potential users would rather use other software (e.g. SPSS, SAS,
R, Statistix, Matlab) for the data analysis, because more advanced software is required
anyway for many other common tasks related to data analysis and other financial analysis.
Table 2: Regression statistics of the example.
SUMMARYOUTPUT
RegressionStatistics year1 year2 year3 year4
MultipleR 0.730 0.912 0.983 1.000
RSquare 0.532 0.831 0.966 1.000
AdjustedRSquare 0.532 0.831 0.966 1.000
StandardError 74.48 44.78 20.00 0.000
Observations 5000 5000 5000 5000

The only information needed of the summary output regression statistics is the
correlation coefficient R Square and Standard Error used in the calculation of the
explained variation according to Equation (10). These results are presented in Table 3.
The explained variation is indeed the standard deviation of the underlying asset
process at the particular time point, i.e. how much of the uncertainty has been
revealed until that time point from the earlier time points. With this information, we
can calculate the volatilities for the different time periods according to Equation (7).
Also SSDR, the square of the explained variation, and SSE, the square of standard
error, are presented in the table to compare the simulated results with the calculated
results presented earlier in Table 1.
Table 3: Explained variation and the volatilities for different time periods.
Explainedvariation 79.47 99.30 107.09 (108.9)*
SSDR 6315 9860 11467 (11860)*
SSE 5547 2005 400 0
Volatility 19.87% 14.89% 10.02% 4.96%
*observeddirectlyfromthesimulationresults

As we can see from the numbers in Table 3, results of the SSDR are very similar to
the results of Solved (explained) uncertainty, and the results of SSE are close to the
numbers of Unsolved uncertainty in Table 1. If more simulation runs are used, the
regression based results converge gradually to the accurate results of Table 1. The
results thus also show that the logic and intuition behind the procedure is in line with
the idea of resolving uncertainty during the project timespan presented in Section 3.
In the following section the use of the regression based volatility method is
illustrated with an actual case example. We cant directly calculate comparative
results such as those presented in Table 1 because of the serial- and cross-correlations
among the cash flow input parameters; therefore, modeling the forthcoming cash flow
parameters conditional on earlier realizations of several input parameters is
impossible in practice. The suggested regression-based method is not sensitive to that.
It can also be adjusted for stochastic interest rate. Another advantage of the approach
is that it can be used with different stochastic processes (e.g. with arithmetic,
geometric, or displaced diffusion process).
5 Case Example
The following example illustrates the use of the approach. The input parameters
(unit sales price, sales quantity, and variable unit costs) are presented as normal
distributions, and they also have correlations with each other. For example, if the
sales price is low in 2011, it is also likely to be lower in the years ahead. Also fixed
costs have some normal variation. These uncertain input parameters of cash flow
calculation are presented in italic in Table 4. As such, this is a typical simplified cash
flow calculation that has several partly correlated parameters and some uncertainty.
Table 4: Cash flow calculation of the case example.
Year 2011 2012 2013 2014 2015 2016 2017
Unit sales price 2.00 1.80 1.62 1.46 1.31 1.18
Sales quantity 25 000 37 500 37 500 28 125 21 094 15 820
Variable unit costs 1.70 1.36 1.16 0.98 0.88 0.80

Cash flow calculation

Revenue 50 000 67 500 60 750 41 006 27 679 18 683
Variable costs -42 500 -51 000 -43 350 -27 636 -18 654 -12 591
Fixed costs 5 000 5 000 5 000 5 000 5 000 5 000
Depreciation -833 -833 -833 -833 -833 -833
EBIT 11 667 20 667 21 567 17 538 13 192 10 259
Taxes -4 667 -8 267 -8 627 -7 015 -5 277 -4 104
Depreciation 833 833 833 833 833 833
Cash flows (CFt) 7 833 13 233 13 773 11 356 8 748 6 988
Discounted CF's (11%) 7 057 10 740 10 071 7 480 5 192 3 736
Present value

44 276

PVt(Project) PV0 PV1 PV2 PV3 PV4 PV5 PV6
(rf = 5.0 %) 44 276 46 546 48 933 51 442 54 079 56 852 59 767

The ordinary cash flow calculation model requires one additional row that
describes how the expected present value of the discounted cash flows PV
0
increases
over time according to the risk-free interest rate. Before the cash flow simulation, the
free cash flows of the time periods (FCF
t
) and the present values of the project (PV
t
)
are set as output parameters. These parameter values are saved during the simulation
run so they can be used later in regression calculations.
After the simulation, we can investigate the overall uncertainty of the project.
Figure 3 shows the simulated underlying asset terminal value distribution discounted
to the present value. We know the expected underlying asset value in a risk-neutral
world increases according to the risk-free interest rate, but we do not know how the
uncertainty changes over the time periods. Based on the assumption of generalized
risk-neutral valuation and assuming the stochastic process to follow geometric
Brownian motion
5
, we can calculate according to Equation (7) the annualized
volatility to be 18.4 %
6
. However, we want to know how the uncertainty and thus the
volatility evolves over time; therefore, we use the regression-based volatility
estimation presented earlier.

Fig. 3. Probability distribution of the case example present value PV
0
. Basic statistics of mean
and standard deviation are also presented as the parameterization for the shifted lognormal
distribution.
Then we use the regression approach similarly to the approach presented in
previous Section 4. The regression estimator Equations (13-15) were used with each
PV
t
as an explanatory variable, and the free cash flows of the corresponding and
previous years as the explaining variables for each case.
After that, the regression analysis is done, and we get the following regression
analysis summary output results presented in Table 5. Based on the R Square
(proportion of uncertainty solved) and Standard Error (uncertainty left), we calculate
the explained variation for each time period according to Equation (10). After that,
volatility for each time period is calculated according to Equation (7). The method
and calculation of the results can be confirmed by using the approach of Haahtela [14]
for volatility estimation. The results given by both methods are exactly the same when
calculated using the same simulated data set. However, using the approach presented
in this paper is much easier and faster for a practitioner to apply.

5
Later, this assumption of gBm is also omitted and displaced diffusion process is assumed
instead.
6
Given S
0
= 44 321, t = 6, Std(S)
0
= 20 981, and r
f
= 5%.
Table 5: Summary output of the regression analysis and calculation of explained variation,
volatility and shifted volatility.
SUMMARYOUTPUT
RegressionStatistics 2011 2012 2013 2014
MultipleR 0.794 0.943 0.981 0.994
RSquare 0.630 0.890 0.961 0.988
AdjustedRSquare 0.630 0.890 0.961 0.988
StandardError 13419 7698 4788 2858
Observations 10000 10000 10000 10000

Explainedvariation 17506 21875 23895 25506
Volatility
i
36.34% 22.32% 11.47% 7.39%
Shifted volatility
dd,i

16.89% 10.75% 5.59% 3.62%

However, as the distribution of the present value (Figure 3) shows, the shape of it
is between a normal and a lognormal distribution, and there are also negative
underlying asset values. Therefore, the common assumption of the geometric
Brownian motion does not hold. One viable alternative is to use the displaced (also
called shifted) diffusion process suggested by Haahtela [15]
7
that has the evolution of
the underlying asset S out to a time horizon T given by equation

S
1
= S
0,0
c
(-Vc
d
2
)1+c
d
1z)
+0
0
c
1
witb z J(u, 1) (16)

where S
0
is the displaced mean parameter,
d
is volatility of the displaced diffusion
and is shifting parameter. The corresponding underlying asset value distribution is a
displaced (shifted) lognormal distribution. This process is similar to the displaced
diffusion process of Rubinstein [21] and D-binomial process of Camara and Chung
[22]. The displaced diffusion process of Equation (16) with different values of S

(mean) and (shift) is capable of modeling stochastic processes that are between
multiplicative (lognormal) and additive (arithmetic) processes, and it also allows
negative underlying asset values.
6 Discussion and Conclusions
This paper presented a practical volatility estimation method for cash flow
simulation based real option valuation cases with changing volatility. During cash
flow simulation runs, the present value of the cash flows and the corresponding cash

7
This diffusion process is used also in Haahtela [11], but [15] discusses the topic more
thoroughly.
flow state variable values are recorded for each time period. Then a regression
analysis is run for relating the PV for each year to the cash flow state variables. Each
regression equation provides an estimate of the expected present value as a function
conditioned on the resolution of all uncertainties up to that time. Then, basic
regression statistics of Pearsons correlation R
2
and sum of squares error (or standard
error) for each equation provide all the information required for estimating how the
standard deviation of the stochastic process and the volatility change over time.
The method is computationally very efficient as it requires only one pass of
simulation runs regardless of the number of time periods. Straightforward calculation
of required regression statistics and their availability in any statistical software and
even in spreadsheet programs make this approach very easy for a practitioner to
apply. Also, the intuitive logic behind the procedure and the capability to handle
negative underlying asset values can be regarded as strengths of the method. As such,
the approach is sufficiently robust for a cash flow simulation-based volatility
estimation.
The method presented in this paper is already quite good in practical terms. It takes
into account the stochastic time-dependence of volatility and the level of the
underlying assets price. The latter of these means the model is able to take into
account the common characteristic of the underlying asset to have lower volatility
with higher underlying asset values. The next logical step for improving the accuracy
of volatility estimation is to use a complete state-space dependent volatility structure.
This is very close to the logic of applying an implied volatility term structure method
where simulated cash flows determine the underlying asset price state-space.
However, this approach requires more computation and forward-looking estimation
and is not that easy to implement in practice in comparison with the method this paper
suggested.
One alternative between the two previously mentioned alternatives (time-varying
volatility for the displaced diffusion process and the implied volatility term structure
approach) is to model the change in volatility as a function of several underlying
parameters. One alternative to determining these functions would be to use statistical
software on simulated data to investigate what kind of response surface functions
could be feasible. However, finding a good function (or a set of functions for different
time periods) for describing state-dependent volatility may be somewhat difficult. On
the other hand, this approach allows direct linking of the cash flow calculation
parameters to the parameters describing the underlying asset value process. This may
be a useful feature for sensitivity analysis purposes if we want to investigate how
changes in cash flow calculation parameters affect the project value with real options.
However, similarly to the complete state-space implied volatility approach, this
approach is not a significant improvement over the approach suggested in this paper.
This is mostly because cash flow simulation follows the law of large numbers.
According to the central limit theorem, the sum of a sufficiently large number of
independent random variables of fairly finite mean and variance, is approximately
normally distributed. If the process is multiplicative, i.e. the random variables are
multiplied, the result is lognormally distributed. Typical cash flow calculation has
both of these properties i.e. they are sums of several cash flow variables, with some of
these variables having multiplicative correlated properties over the time periods.
Therefore, when we sum up a large number of cash flows over multiple simulation
runs, the terminal value distribution is often very close to the shape of the displaced
lognormal distribution. Therefore, the method presented in this paper, especially when
applied to the displaced diffusion process, is a sufficiently accurate volatility
estimation method, given the realities of accuracy of any consolidating cash flow
simulation-based volatility estimation method
8
.
Another question is how much we are interested in knowing the volatility changes
over time. The consolidated cash flow simulation-based volatility measure is not an
observable market variable that could be used for hedging purposes. As such, the
volatility parameter may not even be needed in valuation. If the approach based on the
simulated implied value distribution approach is used with an implied binomial
lattice, calculating the volatility parameter for each node does not provide significant
advantages over the approach of directly estimating the risk-neutral implied
probabilities for the transitions
9
. For example, Wang and Dyer [23] use this approach
without calculating the volatility parameter for each node. Another advantage of this
approach is that it offers a very flexible distribution assumption for project values.
The shortcoming of this approach is that it is more complex to apply in comparison
with this papers method. On the other hand, such approach is better for such cases
where the underlying asset is clearly tractable, which is the case for many projects
related to natural resource investments. If this assumption does not hold, the approach
presented here gives similar results with less effort.
However, even if the procedure presented here may be technically feasible, it still
has many non-computational drawbacks common to similar cash flow simulation-
based methods. Firstly, we are consolidating a high-dimensional simulated process
into a very low-dimensional process. As also discussed in Brando et al. [8] and
Smith [18], the applicability of such approaches is usually case dependent. Secondly,
the high-dimensional process is based on a simulated cash flow, not on a real
observable process. As such, it may have highly subjective estimates, especially when
estimating the correlations between different cash flow variables. Thirdly, the whole
valuation may appear as a black box for the decision-maker after the consolidation,
because it is harder to monitor and understand how different parameters and their
changes affect the consolidated process. This may significantly reduce the capability
to exercise the real options optimally. The use of the sensitivity analysis and linking
the primary cash flow calculation parameters to the option valuation parameters with
regression equations may mitigate this problem [24].
There are several computational issues that could be engaged to technically
improve the approach presented here. One suggested alternative for future research is
to take into account the level of the underlying asset value in more detail than

8
This does not mean that the reality would follow this kind of pattern. It is just a consequence
of how sampling based methods work.
9
If we know the risk-neutral transition probabilities between the nodes and assume some
stochastic process (most often geometric Brownian motion), we can easily calculate the
implied volatilities for each node.

provided by the displaced diffusion process. Also, it would be of high practical
relevance if the link between the single cash flow calculation parameters and the
stochastic process properties of mean and standard deviation could be better modeled.
However, from both the academic and practitioners viewpoint it, would be far more
interesting and relevant to investigate, based on several real life cases from different
industries, how and when the consolidated approach-based methods are of practical
use in comparison with some other valuation methods.

References
1. Mun, J.: Real Options Analysis Course: Business Cases and Software Applications.
John Wiley & Sons, New Jersey, USA (2003)
2. Han, H.: Estimating project volatility and developing decision support system in real
options analysis. Dissertation, Auburn University, Alabama, Dec 17, 2007 (2007)
3. Newton, D., Pearson, A.: Application of Option Pricing Theory to R&D. R&D
Management. Vol. 24, No. 1, pp. 8389 (1994)
4. Lint, O., Pennings, E.: A Business Shift Approach to R&D Option Valuation. In:
Trigeorgis, L. (Ed.) Real Options and Business Strategy. Risk Books, pp. 117132
(1999)
5. Trigeorgis, L.: Real Options: Managerial Flexibility and Strategy in Resource
Allocation. MIT Press (1996)
6. Copeland, T., Antikarov, V.: Real Options: A Practitioners Guide. Texere. New
York (2001)
7. Herath, H., Park, C.: Multi-stage Capital Investment Opportunities as Compound
Real Options. The Engineering Economist. Vol. 47, No.1, pp. 127 (2002)
8. Brando, L., Dyer, J., Hahn, W.: Using Binomial Decision Trees to Solve Real-
Option Valuation Problems. Decision Analysis. Vol. 2, No. 2, pp. 6988 (2005)
9. Brando, L., Dyer, J., Hahn, W.: Response to Comments on Brando et al. (2005).
Decision Analysis. Vol. 2, No. 2, pp. 103109 (2005)
10. Godinho, P.: Monte Carlo Estimation of Project Volatility for Real Options Analysis.
Journal of Applied Finance. Vol. 16, No. 1, pp. 1530 (2006)
11. Haahtela, T.: Extended Binomial Tree Valuation when the Underlying Asset
Distribution is Shifted Lognormal with Higher Moments. 10th Annual International
Conference on Real Options, 14-17 June, New York, USA. (2006)
12. Haahtela, T.: The Effect of Cross-Correlated Input Variables on Real Options
Valuation. 30th Annual Congress of the European Accounting Association, Lisbon,
Portugal, 25.-27. April 2007. (2007)
13. Haahtela, T.: Separating Ambiguity and Volatility in Cash Flow Simulation Based
Volatility Estimation. Available at SSRN: http://ssrn.com/abstract=968226 (2007)
14. Haahtela, T.: Recombining Trinomial Tree for Real Option Valuation with Changing
Volatility. 14th Annual International Conference on Real Options, 16-19 June 2010,
Rome, Italy. Available at SSRN: http://ssrn.com/abstract=1932411. (2010)
15. Haahtela, T.: Displaced Diffusion Binomial Tree for Real Option Valuation.
November 23, 2010. Available at SSRN: http://ssrn.com/abstract=1932408 (2010)
16. Mun, J.: Real Options Analysis: Tools and Techniques for Valuing Investments and
Decisions. John Wiley & Sons, New Jersey, USA (2006)
17. Samuelsson, P.: Proof That Properly Anticipated Prices Fluctuate Randomly.
Industrial Management Review. Spring 1965, pp. 4149 (1965)
18. Smith, J.: Alternative Approaches for Solving Real-Options Problems (Comments on
Brando et al. 2005). Decision Analysis. Vol. 2, No. 2, pp. 89102 (2005)
19. Longstaff, F., Schwartz, E.: Valuing American Options by Simulation: A Simple
Least-squares Approach. Review of Financial Studies. Vol. 14, No. 1, pp. 113147
(2001)
20. Carriere, J.: Valuation of Early-exercise Price of Options Using Simulations and
Nonparametric Regression. Insurance: Mathematics and Economics. Vol. 19, pp.
1930 (1996)
21. Rubinstein, M.: Displaced Diffusion Option Pricing. Journal of Finance. Vol. 38,
No.1, pp. 213217 (1983)
22. Camara, A., Chung, S-L.: Option Pricing for the Transformed-Binomial Class,
Journal of Futures Markets, Vol. 26, No. 8, pp. 759-787. (2006)
23. Wang, T., Dyer, S.: Valuing Multifactor Real Options Using an Implied Binomial
Tree. Decision Analysis. Vol. 7, No. 2, pp. 185195 (2010)
24. Haahtela, T.: Sensitivity Analysis for Cash Flow Simulation Based Real Option
Valuation. 15th Annual International Conference on Real Options. 15-18 June 2011,
Turku, Finland. Available at SSRN: http://ssrn.com/abstract=1864909. (2011)

JRO
JOURNAL OF REAL OPTIONS
ISSN 1799-2937
OPEN ACCESS: http://www.rojournal.com

JRO Vol 1 2011 Complete

Diunggah oleh

Informasi Dokumen

Hak Cipta

Format Tersedia

Bagikan dokumen Ini

Bagikan atau Tanam Dokumen

Opsi Berbagi

Apakah menurut Anda dokumen ini bermanfaat?

Apakah konten ini tidak pantas?

Hak Cipta:

Format Tersedia

JRO Vol 1 2011 Complete

Diunggah oleh

Hak Cipta:

Format Tersedia

ISSN 1799-2737 Open Access http://www.rojournal.

Volume 1, Issue 1, 2011

Corresponding author, Stein-Erik.Fleten@iot.ntnu.no, Phone: +47 73 59 12 96, Fax: +47 73

, where i = period (i = 0, 1, 2, ..., m) and j = state (j = 0, 1, 2,

, these cash flows are

from the cash flow parameters X

Anda mungkin juga menyukai