Professional Documents
Culture Documents
2 Real Options in Theory and Practice
2 Real Options in Theory and Practice
2.1 Introduction
In this chapter, the theory and practice of real options will be introduced. Although their application belongs to the eld of Corporate Finance, the valuation methods origined in the option pricing theory for nancial securities.
In 1977, Stewart C. Myers introduced real options as a new area of nancial
research with the publication of his famous article1 on this subject. He also
coined the term real option and, as a result, is often referred to as the father
of real options theory.
The introduction to the concept of real options is the topic of Section 2.2.
Sections 2.3, 2.4, and 2.5 are devoted to Financial Engineering. The language
of Financial Engineering and hence that of real options is stochastic calculus, which will be used throughout this book when quantifying real options.
Section 2.4 introduces the basics of this theory in order to provide an understanding of the mathematical language used to describe real options valuation
and stochastic term structure models. The term diusion process will take center stage in the following chapters. However, since the meaning of diusion
processes can be best explained by applying methods of classical probability
theory, a mathematical detour will be taken in Section 2.3 about classical
probability theory for diusion processes. Section 2.5 is devoted to numerical simulation methods of continuous-time stochastic processes as needed for
the computer simulation program. The available literature on the quantitative description of real options models is the topic of Section 2.6. This section
comprises four subsections: The rst section (2.6.1) deals with decision-tree
analysis. Subsequently, Section 2.6.2 introduces the idea of contingent-claims
analysis and Section 2.6.3 categorizes the various real options valuation methods, distinguishing between analytical and numerical methods. The exibility
of an investment project due to interest rate uncertainty is the topic of Section
2.6.4. The nal section (2.7) gives a summary of Chapter 2.
M. Schulmerich, Real Options Valuation: The Importance of Interest Rate Modelling in Theory
and Practice, DOI 10.1007/978-3-642-12662-8_2, Springer-Verlag Berlin Heidelberg 2010
21
22
3
4
5
6
7
8
9
23
and Kulatilaka approached this area in their book Real Options10 , published
in 1999. Copeland and Antikarov oered an applied, less quantitative, approach to real options as well11 .
Real options practitioners and academics have also contributed to the diusion of the real options concept. A book published by Hommel, Scholich, and
Vollrath12 in 2001 on the real options approach in Corporate Finance practice
and theory is one of the most successful books in Germany on this subject.
Another book on real options was published in 2003 by Hommel, Scholich,
and Baecker13. Both books contain articles from various authors, from the
corporate sector as well as from the academic world.
Although all these books are a valid introduction to real options for a broad
audience (and this list is by no means complete), the main valuation tool in
capital budgeting is still the DCF method. Until recently, capital budgeting
problems, i.e. the valuation of investment projects, were almost exclusively
done via the DCF method. This method calculates the projects net present
value NPV given a deterministic cash ow structure of the project and a
known discount factor k as
NPV
N
i=0
Cti
.
(1 + k)ti
See
See
See
See
See
24
25
5. Platform investments:
These investments create valuable follow-on contingent investment opportunities. This is particularly important in R&D, the classic platform
investment. The value of such an investment results from products it releases for further development that may in turn lead to marketable products. Traditional tools greatly undervalue these investments, whereas the
real options approach is ideally suited to value such an investment.
6. Learning investments:
Learning investments are investments that are made to obtain information that is otherwise unavailable. A typical learning investment is the
exploration work undertaken on a site prior to exploiting it.
Based on this list of dierent investment types, a categorization was developed
and is currently widely used in Finance. This list is based on Trigeorgis [133],
pages 1-14. Trigeorgis groups real options into the following classes:
Time-to-build option:
This is an option that allows to stop a step-by-step investment within a
project if market conditions turn unfavorable. Such an option is particularly valuable in R&D.
Option to switch:
This option comprises the possibility to react upon changed market conditions by changing the input and output factors via input shifts and/or
output shifts. This is the classical real option.
26
Growth option:
Growth options are strategic options. They are particularly revelant for
projects which are not advantageous in themselves but may generate lucrative opportunities in the future. This type of option can especially be
found in R&D17 . In the pharmaceutical industry, e.g., it takes more than
ten years for a product to develop from the original idea to the nal product with a success probability of only a few percent18 . However, during the
course of a project the original investment may generate various other applications which are protable and generate a positive NPV for the whole
investment.
For a detailed analysis of R&D and the valuation approach with multi-dimensional
models on American compound options in that area see Pojezny [107].
See Schwartz & Moon [122], pages 87-89.
See Pritsch & Weber [108].
See Pritsch & Weber [108], page 38.
See Pritsch & Weber [108], page 31.
Figure 2.2 is based on the articles Klammer & Walker [76] (for 1955 and 1984),
Klammer [75] (for 1959, 1964, and 1970), Gitman & Forrester [47] (for 1970 and
1977) as well as Reichert, Moore & Byler [110] (for 1992).
27
!
28
of exibility but are not fully aware of the academic theory of real options27 .
Moreover, only about 50% of the managers interviewed in this study saw the
necessity to quantitatively implement a real options approach for their capital
budgeting. Vollraths main results are summarized28 in the following:
29
This book intends to bridge the gap between the theoretical and the practical
importance of real options by providing various real options valuation tools
that can readily be used in practice29 . Moreover, by applying a stochastic
term structure instead of a constant risk-free interest rate, the book intends
to improve these numerical methods in order to model reality more accurately.
30
This has been pointed out as an important point of critique and as an area of
future research by Trigeorgis, see Trigeorgis [133], page 375, Section 12.3 - Future Research Directions: Developing generic options-based user-friendly software
packages with simulation capabilities that can handle multiple real options as a
practical aid to corporate planners. This is a major purpose of this book which is
realized with the computer simulation program that was developed for all analyses
in Chapter 5 based on the models presented in Chapters 3 and 4.
See Karlin & Taylor [73], page 149.
30
=:
(x, t),
(2.1)
1
E([h Xt ]2 | Xt = x)
h
=:
2 (x, t)
(2.2)
lim
h0
lim
h0
for each l < x < r and t 0, where l and r are the left and right interval
points of the state space I and h Xt := Xt+h Xt . The functions (x, t)
and 2 (x, t) are the innitesimal parameters of the diusion process; (x, t)
is called innitesimal expectation, expected innitesimal change or (innitesimal) drift; 2 (x, t) is called innitesimal variance or diusion parameter.
Functions and 2 are called innitesimal expectation and innitesimal variance, respectively, since from (2.1) and (2.2) it is known31 :
E(h Xt | Xt = x) = (x, t) h + o1 (h),
V ar(h Xt | Xt = x) = 2 (x, t) h + o2 (h),
o1 (h)
= 0,
h0
h
o2 (h)
= 0.
lim
h0
h
lim
31
x
2()
d d, l < x < r,
exp
S(x) :=
2 ()
c
2()
d , l < < r,
S () = exp
2 ()
0
2
d,
() s()
2
l < x < r,
:=
M (b) M (a),
2
,
2 (x)s(x)
l < x < r,
the function m : I IR is dened, which is called speed density of the diusion process. This density is a Radon-Nikodym density.
The functions S and s are unique only up to an ane transformation with
a monotonically increasing transformation function in the case of S and
up to a positive multiplicative factor in the case of s. For a given S, the
33
34
The term exible constant refers to the mathematical idea that the choice of c
and does not matter. However, if certain values for c and 0 are chosen, these
values cannot be changed any more.
It is common in literature to use the M for the speed measure and for the speed
function. This does not create confusion since both functions operate on dierent
sets.
32
[u q1 ] [q2 v]
,
q2 q1
q1 < u v < q2
[q2 u] [v q1 ]
,
q2 q1
q1 < v u < q2
(2.4)
(2.3)
=
(x,x+)
=
(x,x)
(2.4)
x
([x + ] x)(y [x ])
m(y) dy +
[x + ] [x ]
x+
(x [x ])([x + ] y)
m(y) dy
[x + ] [x ]
x
x+
1
=
(y x + ) m(y) dy +
(x + y) m(y) dy .
2
x
33
Therefore:
1
E(Txx+ |X0 = x)
2
x+
x
1
1 1
(y
x
+
)
m(y)
dy
+
(x
+
y)
m(y)
dy
2 2
2
x
x
1
1
(x 1 () x + ) m(x 1 ()) (x [x ])
2 2
1
(x
+
(x
+
()))
m(x
+
())
([x
+
]
x)
,
+
2
2
2
1 () and 2 () [0, 1] appropriate with
0 (i = 1, 2)
i
0
1
{(1 1 ()) m(x 1 ()) + (1 2 ()) m(x + 2 ())}
=
2
m(x), (m is smooth, see 2.3.2 and denition of m).
0
This means that the speed density m(x) is the speed with which the process
(Xt )t0 moves forward if it is in state x I.
:=
m(x) dx
l
be the total mass of the speed density m on I. If |m| < then there exists a
stationary distribution of the process (Xt )t0 which has a Lebesgue measure
with density
m(x)
(x) =
, x I.
|m|
Since m and |m| are unique up to the same multiplicative constant (see 2.3.3),
is uniquely dened. The proof can be found in Rogers & Williams [113], page
303, theorem 54.5. According to Schulmerich [117], Section 1.16, the restriction in natural scale on the diusion process (Xt )t0 above can be omitted.
This is the main result of 2.3. In the following an example is given of a special
diusion process, the one-dimensional Brownian Motion.
34
2x
= exp 2 , where 0 := 0 I,
2x
2
1 exp 2
, where c := 0,
=
2
1
2x
exp
1
, where c := 0,
=
2
2
2x
= 2 exp
.
Case II with = 0:
s(x) = 1, where 0 := 0 I,
S(x) = x, where c := 0,
2
M (x) = 2 x, where c := 0,
2
m(x) = 2 .
35
the transformation formula in Karlin & Taylor [73], pages 173-175, it holds
for the Geometric Brownian Motion:
2
2
z
and
Z
Z (z) = +
(z) = 2 z 2 , z IR+ .
2
(t, )
f (t, ) is B F -measurable, t 0, , i.e.:
function f is progressively measurable.
(b):
(c):
S, if
(n)
j 2n , if
tj
:=
T, if
(n)
A function n : IR+
0 IR from V is called elementary function, if it
has the form
36
n (t, )
jIN0
j+1
t 0, .
(n)
j+1
(n)
T
for each as
T
n (t, ) dBt ()
:=
jIN0
E |f n |2 dt 0.
n
T
T
T
f (t, ) dBt ()
:=
lim
This limit exists as an element of the set L2 (, P ). The Ito integral is welldened, i.e., independent from the choice of the sequence of elementary functions (n )nIN .
2.4.2 Properties of an Ito Integral
The following rules hold for an Ito integral:
2
T
T 2
f dt f V(S, T )
1. E f dBt = E
S
(Ito isometry).
U
f dBt =
(a) :
S
T
f dBt +
f dBt
U
almost sure.
T
(b) :
(cf + g) dBt = c
S
(c) :
(d) :
T
T
S
T
T
f dBt +
37
g dBt
f dBt = 0.
f dBt
is a martingale.
T S
T
f dBt is FT -measurable.
(e) :
S
3. For each f V(0, T ) there exists an in t smooth version of the Ito integral
t
f dBs ,
0 t T,
t
P Xt = f dBs = 1 for each 0 t T.
0
In the following only in t smooth Ito integrals will be considered. The proof
of these properties can be found in ksendal [103], pages 26-30.
38
Analogous to the description in 2.4.1, the Ito integral can also be dened
for g W. Yet this will not be done here. For details see ksendal [103],
pages 31-32, whose denition will be used in the following. All of the
properties in 2.4.2 are fulllled for each g W as well (see Karatzas &
Shreve [72], Chapter 3.2).
2. It is also possible to dene the Ito integral from 2.4.1 not only with respect to the Standard Brownian Motion (Bt )t0 as done in the original
construction from Ito, but also, under suitable conditions, with respect
to each stochastic process (Mt )t0 that is a smooth, quadratic integrable
martingale. This broader denition of the stochastic integral can be found
in Karatzas & Shreve [72], pages 128-139.
t
b(s, )dBs (), and t 0. (2.5)
a(s, )ds +
0
Here, it is assumed that b W holds and that for a the following equation
holds true:
t
P
|a(s, )| ds < t 0 = 1.
Moreover, let a be adapted to (Ht )t0 where (Ht )t0 is a family of increasing
-elds from W. The dierential notation for (2.5) is:
dXt () = a(t, ) dt + b(t, ) dBt (),
t 0,
or in short notation:
dXt
a dt + b dBt ,
t 0.
(2.6)
39
(ii)
(2.7)
has a unique and in t smooth solution (Xt )t0 that is an element of V(0, T ).
More precisely, the process (Xt )t0 is referred to as a strong solution since
(Bt )t0 is given and the solution process is Ft -adapted, t 0. A solution is
called a weak solution if the Standard Brownian Motion is not given but a
pair ((Bt )t0 , (Xt )t0 ) is searched for that fullls (2.7).
A strong solution (Xt )t0 of (2.7) possesses the strong Markov property and
fullls
E |Xt |2
<
t 0.
If and do not depend on t, i.e., if the SDE
dXt
t 0, X0 = Z,
x, y IR.
SDE (2.7) is the type of an SDE that describes most of the term structure
models in Chapter 3, e.g., the Vasicek model, the Cox-Ingersoll-Ross (CIR)
model, the Ho-Lee model, and the Hull-White one-factor model.
The goal for the remainder of this section is to give explicit solutions for
some specic SDEs. These formulas will be then further used in Chapter 3 to
describe various term structure models.
40
The opposite direction does generally not hold true as the Feller-McKean
process demonstrates (see Rogers & Williams [113], page 271). This process
is a diusion process according to version 1 but not according to version 2.
Therefore, there exist more general diusion processes (version 1) than diusion processes which are solutions of an SDE (version 2).
In the following, a process is called a diusion process if it complies with
the more general denition 2.3.1 (1st version of a diusion process). This is
common in literature.
41
(2.8)
t
t
Xt = (t) X0 + 1 (u) [a(u) C(u)c(u)] du + 1 (u) c(u) dBu ,
0
where
t
(t) = exp
1
A(u) C 2 (u) du +
2
t
0
C(u) dBu ,
t 0.
t
! 1
"2
(u)(u) du < t 0
E
0
this yields:
t
1 (u)a(u) du ,
st
t 0.
2
(u)(u) du (t),
s, t 0.
36
The theorem holds also true for a multi-dimensional linear SDE, see Schulmerich
[117], Sections 2.16, 2.19, and 2.20. In this case, A and C are real matrices and a
and c are real vectors.
is a so-called fundamental solution in the one-dimensional case. For a multidimensional stochastic process is a fundamental matrix. For more details see
Schulmerich [117], pages 22-26. The proof of the solution is done with the Ito
formula and is explained in detail in G
oing [48], pages 60-62.
42
(2.9)
I = (l, r) denotes the state space of the process (see Section 2.3). Such a
process is called a mean reverting or mean reversion process; parameter is
called mean reversion force and is called mean reversion level . These names
are based on the notion that is the force that pulls the process back to
its mean
. The model is called a mean reversion model.
According to this denition only a constant mean reversion level is allowed.
However, this can be extended to a time-dependent mean reversion level
or even a mean reversion level that contains a stochastic process, e.g., an
Ornstein-Uhlenbeck process37 . This process will be introduced in the following.
2.4.11 Example: Ornstein-Uhlenbeck Process
The Ornstein-Uhlenbeck process is a one-dimensional Ito diusion that is
given as the solution of the following linear SDE in the stronger sense:
dXt = Xt dt + dBt , t 0, X0 = x0 IR constant, IR+ , IR+ .
This process mean reverts around zero38 . According to 2.4.9 it holds
(t) = et , t 0,
and
Xt
x0 e
+ e
t
es dBs ,
t 0.
0
37
38
An example is the Hull-White two-factor model which includes a drift term with
a stochastic part which is modelled via an Ornstein-Uhlenbeck process.
This property follows directly from 2.4.9 (i) for t . See also Schulmerich
[117], page 30.
43
44
N1
(0,1)
N2
:=
:=
%
%
2 ln(U1 ) cos(2U2 ),
2 ln(U1 ) sin(2U2 )
(2.10)
(0,1)
:= 1 N1
N1
+ 1
(2 ,22 )
and N2
(0,1)
:= 2 N2
+ 2
(2.11)
= N (1 , 12 )
N1
and
(2 ,22 ) d
N2
= N (2 , 22 ).
B (1) = N (0, 12 ),
and
(2.13)
Y0
:= x0 ,
(1)
Yi+1 := Yi + (i , Yi ) s + (i , Yi ) Bi ,
i = 0, . . . , N 1,
(2.14)
45
(1) d
with = 0 and 1 = s .
2.5.3 Milstein Scheme
As opposed to the Euler scheme, the Milstein scheme uses one more term of
the Ito-Taylor formula. In the following, a partial derivative with respect to
the local component (=second component) is denoted with a stroke next to
N
the variable. The sequence (Yi )N
i=0 (Yi )i=0 can now iteratively be dened
via:
Y0
:= x0 ,
(1)
Yi+1 := Yi + (i , Yi ) s + (i , Yi ) Bi
(2.15)
'
&
(1) 2
1
+ 2 (i , Yi ) (i , Yi ) (Bi ) s ,
i = 0, . . . , N 1,
(1) d
(2)
= N (0, s ) and Bi
=
eratively be dened, see (2.16). Here, Bi
N (0, 13 1.5 ) for i = 0, . . . , N 1. In the computer simulation program realiza(1)
(2)
tions of Bi and Bi are calculated via the method presented in (2.12),
which
according to (2.13) yields the appropriate distributions with = 0 and
1 = s . (Yi )N
i=0 is the Taylor 1.5 approximation of the continuous-time
process (Xt )0tT in the observation period [0, T ]. (2.16) is the Taylor 1.5
scheme. The denomination of the simulation method (2.16) in the book as
Taylor 1.5 scheme stems from the following introduction of the term strong
convergence of order with IR+ which will be used for comparing the
three simulation methods introduced above with each other.
46
Y0
Yi+1
(1)
:= Yi + (i , Yi ) s + (i , Yi ) Bi
'
&
(1)
1
2
+ 2 (i , Yi ) (i , Yi ) (Bi ) s
(2)
+ (i , Yi )(i , Yi ) Bi
#
$
1
1
2
2
+ 2 (i , Yi ) (i , Yi ) + 2 (i , Yi ) (i , Yi ) s
#
$
+ (i , Yi ) (i , Yi ) + 21 2 (i , Yi ) (i , Yi )
&
'
(1)
(2)
Bi s Bi
1
+ 2 (i , Yi ) ((i , Yi ) (i , Yi )
'
&
(1)
(1)
1
2
2
[B
B
]
,
+ [ (i , Yi )]
s
i
i
i = 0, . . . , N 1.
:= x0 ,
(2.16)
See Kloeden & Platen [77], page 323, page 345, and page 351.
2.6 Evolution of the Real Options Theory and Models in the Literature
47
See
See
See
See
48
projects. In the words of Trigeorgis44: The basic inadequacy of the NPV approach and other DCF approaches to capital budgeting is that they ignore,
or cannot properly capture, managements exibility to adapt and revise later
decisions (i.e., review its implicit operating strategy). The traditional NPV
approach, in particular, makes implicit assumptions concerning an expected
scenario of cash ows and presumes managements commitment to a certain
operating strategy.
One of the results of this criticism was the development of the decision-tree
analysis by Hertz and Magee45 in 1964. Interestingly enough, the idea of the
decision-tree analysis was published by these authors in the Harvard Business
Review, which is more focused on application and is not a classical academic
journal. In the following the basic idea of decision-tree analysis will be explained and elaborated to show its advantanges and disadvantages.
2.6 Evolution of the Real Options Theory and Models in the Literature
49
The calculations of the values in the tree are done by discounting with the
cost of capital k, which is assumed to be 6%, and taking into consideration
the investment cost and the management decision at point B:
Point b1 :
Point b2 :
Point B :
Point A :
100 000 =
32
Applying DTA yields a negative NPV of 32, which means that the platform
investment would not be undertaken: stop R&D.
50
X(cH
2 ) = 24 000
X(cM
1 ) = 10 000
X(cM
2 ) = 20 000
X(cL
1) =
X(cL
2 ) = 18 000
8 000
Management expects the following NPVs (with respect to the end of year 3,
thus index 3) for the cash ows of the years 4 - 10:
N P V3 (Cash ow of years 4 - 10, cH
1 )
145 000
70 000
7 000
200 000
90 000
14 000
This also gives the following cash ows at the end of year 3 for year 3 as
E(cji ) = cji N P V3 (cji ) :
E(Cash ow in year 3, cH
1 ) = 180 000 145 000
35 000
E(Cash ow in year 3, cM
1 ) = 100 000 70 000
30 000
3 000
E(Cash ow in year 3, cH
2 ) = 270 000 200 000
70 000
E(Cash ow in year 3, cM
2 ) = 140 000 90 000
50 000
E(Cash ow in year 3, cL
1)
E(Cash ow in year 3, cL
2)
46
= 10 000 7 000
2.6 Evolution of the Real Options Theory and Models in the Literature
51
At the end of year 3 management will only sell the plant if N P V3 is smaller
than the salvage value. For both plant types this is only the case for a low
market when plant type I sells for 8 000 and plant type II sells for 18 000.
Taking this salvage value into consideration at the end of year 3, gives the
following situation (compared with the right side of Figure 2.3):
Replace 10 000 at cL
1 with 3 000 + max(7 000, 8 000) = 11 000
Replace 20 000 at cL
2 with 6 000 + max(14 000, 18 000) = 12 000
The remaining calculations are the same as in example 1. They produce the
decision tree of Figure 2.4 in the case of an option to abandon for a salvage
value in this example of a platform investment in R&D. It nally yields an
NPV of 1 043, so that the investment would be undertaken. This NPV is called
a strategic NPV since it includes a real option. The NPV of 32 previously
calculated in the absence of a real option is called static NPV. The value of
the real option can therefore be calculated as:
Value of the option to abandon = option premium
= strategic NPV static NPV
= 1 043 (32)
= 1 075.
In summary, DTA is well suited to price some types of real options. It can price
sequential investment decisions in which management decisions are made at
discrete points in the future and uncertainty is resolved at discrete points in
time as well. DTA is able to handle this embedded exibility but the practical
application has serious limitations: the number of discrete points in time can
get large, thus, creating an extremely complex tree. Trigeorgis refers to this as
decision-bush analysis47 instead of decision-tree analysis. Realistic corporate
budgeting situations cannot be properly handled that way. Second, the applied
discount rate poses a problem since it is usually assumed to be constant. So,
when uncertainty gets resolved at decision nodes, the DTA method does not
use a changed discount factor. DTA cannot therefore reect this change in the
riskiness of the project expressed in the discount rate.
47
52
Fig. 2.4. Example of a platform investment in R&D with option to abandon within
the framework of decision-tree analysis.
2.6 Evolution of the Real Options Theory and Models in the Literature
53
The characteristic of a real option is that it alters the risk structure of the
project. In example 2 of the previous Section 2.6.1, the option to abandon
after year 3 gets rid of negative cash ows in a low-case market. This should
result in a new risk-adjusted interest rate as discount factor below 6%. Since,
however, the discount factor stays the same, the true NPV is understated.
Therefore, the true NPV is above the calculated 1 043.
Trigeorgis characterizes DTA and CCA in a very precise way49 : DTA can
actually be seen as an advanced version of DCF or NPV - one that correctly
computes unconditional expected cash ows by properly taking account of their
conditional probabilities given each state of nature. As such, DTA is correct in
principle and is particularly useful for analyzing complex sequential investment
decisions. Its main shortcoming is the problem of determining the appropriate
discount rate to be used in working back through the decision tree. [ . . . ] The
fundamental problem with traditional approaches to capital budgeting lies in
the valuation of investment opportunities whose claims are not symmetrical or
proportional. The asymmetry resulting from operating exibility options and
other strategic aspects of various projects can nevertheless be properly analyzed by thinking of discretionary investment opportunities as options on real
assets (or as real options) through the technique of contingent-claims analysis.
The CCA method builds upon the DTA method and eliminates the weak
point of constant interest rate. With DTA the basic idea is to calculate with
real probabilities and a constant, risk-adjusted interest rate as the discount
rate. With CCA the basic idea is to transform the real probabilities into riskadjusted probabilities such that the algorithm can use a constant, risk-free
interest rate that is independent of the projects risk structure. To explain the
idea of CCA some denitions are necessary:
V
54
As the starting point the current values for V and S are assumed to be
V = 100 and S = 20. S is the price of the so-called twin security that exhibits
the same risk prole as the project with value V and can be traded on the
capital markets. Furthermore, the real probability q is assumed to be 0.5. This
gives a development of S and V with u = 1.8 and d = 0.6 as shown in Figure
2.5.
Fig. 2.5. Development of the total value V of a project and of the price S of a twin
security.
The goal is to replicate the pay-o structure of the project after one year
through the purchase of n shares of the twin security and through issuing
of a 1-year Zero bond with nominal value B and risk-free return rf . This is
illustrated in Figure 2.6. The issue of the Zero is equivalent to getting a loan
of value B for the interest rate rf . Therefore, a positive B indicates a bond
issue.
By choosing the replicating portfolio approach E = nS B (i.e., E is replicated via n shares of stocks and the issue of a Zero worth nominal B) equations
2.6 Evolution of the Real Options Theory and Models in the Literature
55
(a) and (b) in Figure 2.6 have to be solved simultaneously to arrive at n and
B. The solutions are summarized in Figure 2.7.
When dealing with a project that starts immediately, it is obvious that V =
E, Vu = Eu , and Vd = Ed have to hold since the value of the project has to be
exactly the same as the value the shareholder gets out of the project. Putting
the given data into the formulas, with a risk-free rate of rf = 8%, yields the
following result:
(1 + rf ) d
ud
pVu + (1 p)Vd
1 + rf
Vu Sd Vd Su
(Su Sd )(1 + rf )
Vu Vd
Su Sd
=
=
(1 + 0.08) 0.6
1.8 0.6
0.4 180 + (1 0.4) 60
1 + 0.08
0.4
100
180 12 60 36
(36 12)(1 + 0.08)
180 60
36 12
qSu + (1 q)Sd
0.5 36 + (1 0.5) 12
1 =
1 = 20%.
S
20
This gives:
V DT A =
In this case, therefore, the same value of V is yielded by DTA and CCA. This
will change when the project contains a real option. Only the CCA with its
risk-free interest rate gives the correct value for projects with real options,
unlike the DTA with its risk-adjusted interest rate. This will be illustrated
in the example of an option to defer by calculating the values for V and
E if the project starts immediately or one year from today. For the NPV
calculation the development of V, S, and E over two years with Suu = u2 S,
Sud = udS = Sdu , and Sdd = d2 S needs to be determined. Sud = Sdu leads to
a recombining tree, i.e., the branches of the tree merge in the next stage and
each additional stage creates one new node.
56
Fig. 2.8. Example of an option to defer the project start by one year within the
framework of contingent-claims analysis.
An immediate start of the project has so far been assumed with an investment
cost of I0 = 104. If the start of the project is in one year from today, an
investment cost of I1 = (1 + rf )I0 = 112.32 can be assumed50 . The valuation
of the real option is explained in Figure 2.8. The shareholder values in Figure
2.8 can be calculated as follows:
Eu = max(Vu I1 , 0)
= max(180 112.32, 0)
= 67.68
Ed = max(Vd I1 , 0)
= max(60 112.32, 0)
= 0
Thus, in the case of a 1-year deferred project start, dierent values for V and
E at t = 0 are calculated: V = 100 but E = 25.07. In this situation the NPV
as the equity value to the shareholder is 25.07. The question that arises here is:
What is an option worth that allows the start of the project to be deferred by
one year? It has to be the dierence of both NPVs, i.e., 25.07 (4) = 29.07.
This means that management would be willing to pay up to 29.07 to have the
option to defer the project start by exactly one year.
Applying the DTA method yields a dierent value for E and, therefore, for
the NPV:
N P V DT A = E DT A =
50
2.6 Evolution of the Real Options Theory and Models in the Literature
57
So the question is which NPV is the correct one? This question can be easily answered by applying the theory of the replicating portfolio, see Figure 2.7.
n
Eu Ed
Su Sd
67.68 0
36 12
2.82
shares of stocks and the issue of a 1-year Zero with nominal value
B
Eu Sd Ed Su
(Su Sd ) (1 + rf )
67.68 12 0 36
(36 12) (1 + 0.08)
31.33
are needed to replicate the equity value in the up- and down-cases for a
risk-free interest rate of rf = 8%. The cost of a replicating portfolio is
2.82 20 31.33 = 25.07, so that E = 25.07 is the correct answer. DTA,
therefore, overestimates the equity value. The reason for this is that the DTA
approach uses a constant interest rate, which is not appropriate if real options
are present.
In summary, asymmetrical cash ows change the risk structure of the project
and, in so doing, the discount rate - something the DTA does not consider.
The aw of the DCF is nicely described in Trigeorgis [133], page 152: The
presence of exibility embedded in future decision nodes, however, changes the
payo structure and the risk characteristics of an actively managed asset in
a way that invalidates the use of a constant discount rate. Unfortunatelly,
classic DTA is in no better position than DCF techniques to provide any recommendation concerning the appropriate discount rate.
This error is corrected in the CCA method since CCA transforms the real
probabilities into risk-adjusted probabilities and the risk-adjusted interest
rates into risk-free interest rates to correctly mirror the asymmetry within
a real option. In the words of Trigeorgis51: Traditional DTA is on the right
track, but although mathematically elegant it is economically awed because of
the discount rate problem. An options approach can remedy these problems.
Therefore, only a CCA approach can capture the exibility that is inherent
in investment projects. Flexibility, according to Trigeorgis [133], page 151, is
nothing more than the collection of options associated with an investment opportunity, nancial or real. However, the general question of the justication
of an option based approach with replicating portfolios and its limits has to be
answered. A good justication for this approach is given in Trigeorgis [133],
pages 124-129. While his full justication will not be restated here, a summary
of Trigeorgis most important points is needed52 : Can the standard technique
of valuing options on the basis of a no-arbitrage equilibrium, using portfolios
51
52
58
Table 2.1. Comparison between a real option on a project and a call option.
Call option on stock
current value on stock
exercise price
time to expiration
stock value uncertainty
risk-free interest rate
53
54
55
56
57
See
See
See
See
See
2.6 Evolution of the Real Options Theory and Models in the Literature
59
60
is the instantaneous expected return on the project and its instantaneous standard deviation. Paddock, Siegel, and Smith use a similar
process given via the SDE
dVt = ( D) Vt dt + Vt dBt , t 0, IR+ , IR+ ,
with D as the payout rate for the valuation of o-shore petroleum
leases. Majd and Pindyck use the same process for the market value
of the completed project in valuing the option to defer irreversible
construction in a project where a series of outlays must be made sequentially but construction has an upper speed boundary.
b) Option to shut down or abandon (McDonald & Siegel [91], 1985; Myers
& Majd [101], 1990): McDonald and Siegel assume that the diusion
process (Pt )t0 for the output price is given by the SDE
dPt = Pt dt + Pt dBt , t 0, IR+ , IR+ ,
(2.17)
i.e., the same process as in McDonald & Siegel [92]. Myers and Majd
use a similar process, given by
dPt = ( D) Pt dt + Pt dBt , t 0, IR+ , IR+ , (2.18)
where D represents the instantaneous cash payout (calculated as cash
ow divided by the project value).
c) Option to switch (Margrabe [87], 1978; Stulz [127], 1982): Margrabe
analyzes the value of an option to exchange one non-dividend paying
risky asset for another where the prices V and S of the risky assets
are modelled via the same processes as in (2.17), however with different coecients for the V process and the S process. Stulz analyzes
European options on the minimum or maximum of two risky assets,
assuming the same two processes as in (2.17) and (2.18).
d) (Simple) Compound options (Geske [44], 1979): Geske calculates the
value of an option on a stock where the stock can be seen as a European call option on the value of the rms assets. Trigeorgis stresses
this application for real options as to value a compound growth option
where earlier investments have to be undertaken as prerequisites59 .
The value of the underlying is assumed to follow the same process as
in (2.17).
e) Compound options to switch (Carr [27], 1988; Carr [28], 1995): Carr
uses the same processes to describe the value of two risky assets to
analyze European compound (or sequential) exchange options as in
(2.17), however with dierent coecients for the two processes.
59
2.6 Evolution of the Real Options Theory and Models in the Literature
61
In addition to these articles several books were published in the past presenting specic real options models. 17 articles on real options by various
authors were compiled in Brennan & Trigeorgis [21], published in 2000.
The models presented are mathematically very complex and specic. The
analytical analysis is often accompanied by empirical valuation results for
the situation analyzed. Important to notice is that all articles assume a
constant risk-free rate apart from Miltersen [96]. Another (less quantitative) work on real options was published in 1995 by Trigeorgis60. This
book also contains 17 separate articles on real options, all assuming a
constant risk-free rate.
Two main disadvantages are inherent in all analytical solutions. First, the
capital budgeting problem needs to be nice in order to be analytically
trackable. This means that for a capital budgeting problem one must be
in the position to write down the describing partial dierential equation
with the underlying stochastic process. In practice this is almost never
the case. Second, even if a solution to the capital budgeting problem can
be found for each single real option, this does not say anything about the
value of the total option package. The valuation for complex options, i.e.,
option packages with many dierent real options and real option types61 ,
need to take option interactions into consideration, something almost always impossible to handle in an analytical approach.
2. Numerical methods. One criticism of analytical approaches is that
they are not suited to valuing complex real options. The ability to value
such complex options has been enhanced through various numerical techniques62 . These can be divided into methods that approximate the partial
dierential equation and methods that approximate the underlying stochastic process63 :
a) Approximation of the partial dierential equation: This includes numerical integration methods as well as the explicit and implicit nite dierence methods.
b) Approximation of the underlying stochastic process: This includes Monte Carlo simulation and lattice approaches like the CoxRoss-Rubinstein binomial tree method and the Trigeorgis log-transformed binomial tree method.
In the following the basic characteristics of both numerical approximation
methods will be described:
60
61
62
63
See
See
See
See
Trigeorgis [132].
Section 2.2.
Trigeorgis [133], pages 20-21.
Trigeorgis [133], page 21.
62
64
65
66
Consistency, also called accuracy, refers to the idea that the discrete-time process
used for calculation has the same mean and variance for every time-step size as
the underlying continuous process. Stability, or numerical stability, means that the
approximation error in the computations will be dampened rather than amplied.
Eciency refers to the number of operations, i.e., the computational time needed
for a given approximation accuracy. For more information see Trigeorgis [133],
page 308. These terms are addressed in Chapter 4 where the various methods are
introduced in detail.
See Trigeorgis [133], page 305.
See Trigeorgis [133], page 306.
2.6 Evolution of the Real Options Theory and Models in the Literature
63
Analytic approximation:
Johnson [71], 1983; Geske & Johnson [45], 1984; MacMillan [84],
1986; Blomeyer [10], 1986; Barone-Adesi & Whaley [4], 1987; Ho,
Stapleton & Subrahmanyam [58], 1997.
St dt + St dBt ,
t 0,
(2.19)
67
68
64
:=
Si + Si s + Si Bi ,
i = 0, 1, . . . , N,
where Bi = N (0, s ). This is the so-called Euler scheme as presented in 2.5.2. Having simulated such a path, the further steps depend on the type of option that needs to be priced. In the following a
European call option with strike price X will be valued. To do this the
nal value ST of the path is needed and has to be compared with the
strike price of the option. Let Pj := max(ST X, 0) be the price of
a European call at maturity date T where index j indicates that this
is the price for the j th simulated path. This simulation of the paths
has to be done many times with each simulation independent of each
other. Let A denote the number of simulations (e.g., A := 10 000).
Each j gives a dierent value Pj . The mean
B
:=
A
Pj
(2.20)
j=1
then gives the value of the option at time T , and B erf T gives the
current discounted value of the option with rf as the current risk-free
interest rate. An interesting application of path simulations and option pricing can be found in Lieskovsky, Onkey, Schulmerich, Teng &
Wee [81] for pricing and hedging Asian options on copper.
As already mentioned, the latest publications also show how Monte
Carlo simulation can be used to price American type options. If the
underlying does not pay any dividends, the exercise of an American
call option prior to expiration is never optimal69 . However, if the underlying pays dividends, it can be optimal to exercise the call option
prior to maturity. For an American put option it can be optimal to
exercise early even in the absence of dividend payments70 .
69
70
2.6 Evolution of the Real Options Theory and Models in the Literature
65
71
72
73
74
See
See
See
See
Lattice methods:
Cox, Ross & Rubinstein [36], 1979; Boyle [14], 1988; Hull & White
[63], 1988; Trigeorgis [131], 1991.
66
A nal word has to be addressed to the valuation of multiple options. Interactions among multiple options are the reason why option value additivity can
usually not be assumed when valuing multiple options. Valuing each option
of such an option package separately and then adding up the values does not
take into consideration the fact that options usually interact. Consider, for
example, an option to abandon and an option to switch. The value of the option to switch is zero if the option to abandon has been exercised. Therefore,
the value of the option to switch depends on whether the project is still valid
(i.e., option to abandon has not been exercised and is still active) or not. In
such a case the values of each single option cannot simply be added together;
rather, both options have to be valued together. This gives a multiple option
value less than the sum of the single option values, since the sum is less the
more interactions there are among the single options. Trigeorgis elaborates on
the aspect that the incremental value of an additional option, in the presence
of another option, is generally less than its value in isolation (in Chapter 7 of
Trigeorgis [133]). An example of value additivity is an option package comprising an option to expand and an option to contract, each one year after the
project was started. The underlying value, the present value of the project,
can be split into exercise ranges for each of the options. These ranges do not
overlap in the case of an option to expand or an option to abandon, which is
obvious.
2.6 Evolution of the Real Options Theory and Models in the Literature
67
78
79
80
81
82
83
84
85
Ingersoll and Ross used the Cox-Ingersoll-Ross model to describe the term structure of interest rates whereby the parameters and are omitted, i.e., the shortrate process has a zero-expected change.
See Trigeorgis [133].
See Trigeorgis [133], page 199.
See Sandmann [115].
See Ho, Stapleton & Subrahmanyam [58].
See Miltersen [96].
See Alvarez & Koskela [1].
See Miltersen & Schwartz [97].
See Miltersen [96], page 196.
68
The recent article of Alvarez and Koskela builds86 on Ingersoll & Ross [69]
and analyzes irreversible investments under interest rate variability. According to Alvarez and Koskela87 , the current extensive literature on irreversible
investment decisions usually makes the assumption of constant interest rate.
As Alvarez and Koskela further state88 : In these studies dealing with the impact of irreversibility in a variety of problems and dierent types of frameworks the constancy of the discount rates has usually been one of the most
predominant assumptions. The basic motivation of this argument is that interest rates are typically more stable (and consequently, less signicant) than
revenue dynamics. [. . .] If the exercise of such investment opportunities takes
a long time, the assumed constancy of the interest rate is questionable. This
observation raises several questions: Does interest rate variability matter and,
if so, in what way and how much?
Consequently, their studies are motivated by the importance of interest rate
variability89 : It is known from empirical research that interest rates uctuate
a lot over time and that in the long run these follow a more general mean reverting process (for an up-to-date theoretical and empirical survey in the eld
see e.g. Bj
ork [6] and Cochrane [32]). Since variability may be deterministic and/or stochastic, we immediately observe that interest rate variability in
general can be important from the point of view of exercising real investment
opportunities.
The article of Alvarez and Koskela is very thorough and focuses on the mathematical aspects but does not provide detailed real world examples or historical
backtesting. In this book, the focus is on numerical real options pricing with
interest rate uncertainty. Although articles about option pricing in general
are plentiful in the literature, thorough numerical analyses using simulations
and historical backtesting of complex real options situations with stochastic
interest rates are rare at best. Therefore, the goal of this book is to ll this
gap by providing in-depth insight into how a stochastically modelled interest
rate inuences the real options value in various real options cases common in
practice and to derive rules for application in Corporate Finance.
86
87
88
89
See Alvarez & Koskela [1], page 2: [. . .] we generalize the important ndings by Ingersoll and Ross both by allowing for stochastic interest rate of a mean reverting
type and by exploring the impact of combined interest rate and revenue variability on the value and the optimal exercise policy of irreversible real investment
opportunities.
See the abstract in Alvarez & Koskela [1].
See Alvarez & Koskela [1], page 1.
See Alvarez & Koskela [1], page 2.
2.7 Summary
69
2.7 Summary
Chapter 2 introduced the idea of real options: qualitatively, quantitatively,
and with respect to the historical development of this research eld. Various
real options approaches have been categorized and explained in terms of their
development over time and use in nancial practice. While Section 2.2, Basics
of Real Options, introduced the concept of real options in general, Sections
2.3, 2.4, and 2.5 introduced the necessary nancial mathematics needed in the
following chapters. Section 2.3, Diusion Processes in Classical Probability
Theory, presented the idea of the diusion process that is necessary to model
real options and the term structure of interest rates. Section 2.4, Introduction to Ito Calculus, introduced Ito calculus in its basic form, especially with
a special focus on diusion processes. Finally, Section 2.5, Discretization of
Continuous-Time Stochastic Processes, provided the necessary algorithms to
numerically calculate a realized discrete path of a continuous-time stochastic
process in a computer simulation program. The Evolution of the Real Options
Theory and Models in the Literature was the topic of Section 2.6, a section
divided into four subsections: Decision-Tree Analysis in 2.6.1, ContingentClaims Analysis in 2.6.2, Categorization of Real Options Valuation Methods
in 2.6.3, and Flexibility due to Interest Rate Uncertainty in 2.6.4.
In the remaining chapters of this book the focus is on real options pricing
under interest rate uncertainty, which was introduced in 2.6.4. The central
thrust will be to thoroughly analyze complex real options situations with
non-constant (especially, stochastically modelled) risk-free interest rates using numerical simulations and historical backtesting. The primary valuation
tools will be methods used for valuing nancial options but modied to value
real options problems (i.e. the investment opportunities) since the similarities
between these two are manifold. To quote Dixit and Pindyck90 : Opportunities are options - rights but no obligations to take some action in the future.
Capital investments, then, are essentially about options.
The stochastic interest rate movement in these models will be described via
term structure models. Each term structure model is specied by a diusion
process that will be introduced in the following chapter. The real options
valuation methods will be explained in Chapter 4 and numerically analyzed
in Chapter 5.
90
http://www.springer.com/978-3-642-12661-1