European Journal of Operational Research 271 (2018) 720–732

Contents lists available at ScienceDirect

European Journal of Operational Research

journal homepage: www.elsevier.com/locate/ejor

Innovative Applications of O.R.

Valuing multistage investment projects in the pharmaceutical industry

Luiz E. Brandão a,∗, Gláucia Fernandes a, James S. Dyer b
a
Pontifical Catholic University of Rio de Janeiro, IAG Business School, Rua Marques de São Vicente 225, Gávea, Rio de Janeiro, 2451-900 RJ, Brazil
b
The University of Texas at Austin, McCombs School of Business, 2110 Speedway, Austin, TX 78712, United States

a r t i c l e i n f o a b s t r a c t

Article history: Multistage investment projects are subject to several sources of uncertainty, but also present significant
Received 14 September 2017 embedded flexibilities both during and after the development process. An important characteristic of
Accepted 21 May 2018
this class of projects is that as the firm incurs a cost and invests, it learns both about the difficulty of
Available online 29 May 2018
developing and implementing the project and also about market conditions. We develop a real options
Keywords: model where the firm continuously updates its prospects of timely completion and future market con-
OR in research and development ditions. This information can then be used to optimally decide whether further investment is warranted
Real options or not, given the expected future revenues of the whole venture. We apply this model to a Research and
Multistage projects Development (R&D) project in the pharmaceutical industry and value this investment opportunity as a
Valuation compound contingent claim where the underlying asset is the value of the completed project, and find
that the opportunities to abandon and to expand into related markets and applications once the original
product is proven successful have a significant impact on the value of the project as a whole. This article
differs from previous work in the field by capturing the potential upsides that may occur during product
development through a novel quality model, and their effect on the expansion opportunities during the
market phase. We use a simulation approach for the solution and show that for complex real options
models, this method can be an effective way to value such projects while providing adequate precision
compared to other models in the literature.
© 2018 Elsevier B.V. All rights reserved.

1. Introduction Another characteristic of these projects is that by investing, the

Most capital budgeting problems involve analyzing the trade-
offs between a fixed and certain capital investment and an uncer-
tain stream of future cash flows. On the other hand, a large class
of problems, such as the valuation of startup firms, complex in-
dustrial and construction ventures, and R&D projects such as soft-
ware, drugs and technology initiatives, include complications not
captured by this simple model. In addition to the uncertainty over
the final payoffs of the venture, for this class of problems there is
also considerable uncertainty over the cost and timing of the to-
tal investment required and over the quality and performance of
the final product. The design of an advanced microprocessor chip,
the development of a new aircraft, a startup firm or a new drug,
for example, all involve investing an uncertain amount of capital
and time in order to obtain and to bring to the market a product
whose performance characteristics or final quality is uncertain.
∗
Corresponding author.
E-mail addresses: brandao@iag.puc-rio.br (L.E.
glaucia.fernandes@iag.puc-rio.br (G. Fernandes), j.dyer@mccombs.utexas.edu (J.S.
Dyer).
firm learns about the difficulty of designing and building a new
product or of performing research on a new drug, and updates
its prospects of a successful development, timely completion and
the quality of the final product as it progresses during each devel-
opment phase. This new information also allows the firm to op-
timally determine whether to abandon or to continue investing in
the project at any time. This gives these projects the characteristics
of a contingent claim over the value of the completed project, and
we analyze the problem from a real options approach. According
to Lo Nigro, Morreale, and Enea (2014), Real Options Analysis is ac-
knowledged as a powerful tool to evaluate uncertain projects that
have an intrinsic flexibility. The valuation of multi-staged pharma-
ceutical R&D ventures, for example, can be interpreted as a chain
of nested real options (Cassimon, De Backer, Engelen, Van Wouwe,
& Yordanov, 2011), where there is uncertainty of both the true in-
vestment costs and the future cash flows the product will generate,
as well as managerial flexibility to optimally abandon the develop-
ment (Managi, Zhang, & Horie, 2016).
These projects are subject to several different types of stochas-
tic cost and demand uncertainties. We model the investment cost
Brandão), as a diffusion process with a negative drift equivalent to the
instantaneous rate of investment, in an approach that follows

https://doi.org/10.1016/j.ejor.2018.05.044
0377-2217/© 2018 Elsevier B.V. All rights reserved.
 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732
Pindyck (1993), where the firm starts out with an exogenously de-
fined expected cost to completion and updates this expected cost
as new information becomes available. There is also uncertainty
concerning the changes in the competitive and market environ-
ment, both during and after product introduction, that may render
the project worthless, such as a preemptive move by a competitor
or a technological breakthrough that makes the product obsolete.
This is modeled as an exogenous Poisson death process.
We assume the firm has the option to abandon the project con-
tinuously at any time during the development process, even before
a particular stage is completed. If successfully completed, through
additional investment the project provides the firm an opportu-
nity to optimally expand the scope of the project to new market
segments that were not originally contemplated, or to derivative
products, with the possible extension of patent life. This may in-
volve new uses for the original product or making improvements
or modifications that would allow it to be marketed for different
uses. As an example, a low power consumption version of a suc-
cessful cell phone model may expand the original market for this
product to include other mobile appliances, or a drug that is tar-
geted to the adult patients may be altered to also be used by chil-
dren. Obviously, the prerequisite for a profitable expansion is that
the development of the original product is successfully completed
and that the project has an adequate performance.
Prior to the beginning of the project, the firm specifies a set of
performance characteristics the final product is expected to have.
This can be the clock speed of a new microprocessor chip design,
the operating range of a new aircraft, the maximum sustainable
output of a power plant or the effectiveness of a new drug. As
the firm invests in the project it also learns about any deviations
from the expected performance and updates this information as
the project progresses through each stage. We assume that the fi-
nal product performance is correlated with the deviations from the
expected cost and time to completion in each stage. Also, we as-
sume that the project is subject to a finite economic life due to
technical obsolescence, increased competition or patent expiration.
We adopt a discrete simulation model to obtain a solution for
the value of this more complex multistage investment problem. We
first model the more general case of an n-stage investment project
subject to several sources of uncertainties and then analyze the
case of a three stage drug R&D project in the pharmaceutical in-
dustry. We obtain a discrete approximate solution to this applica-
tion and show the comparative statics.
This paper is organized as follows. In the next section we dis-
cuss previous work in this field that is related to our work. In
Section 3 we present our basic model, the notation and the val-
uation equations. In Section 4 we apply the model to value a R&D
project in the pharmaceutical industry and show analytical results.
In Section 5 we draw our conclusions.
2. Related work
The valuation of R&D projects as a contingent claim on the
value of the completed project has been subject to much interest
in the literature. With the increasing complexities related to ana-
lyzing uncertainty, R&D decision making has become more difficult
for firms seeking to enhance their competitive position and drive
their sustainable profitable growth (Song, Di Benedetto, & Na-
son, 2007). Therefore, real options models can help R&D managers
evaluate and determine optimal investment decisions to maximize
market payoff under different demand structures (Wang, Wang, &
Wu, 2015) and to quantify managerial flexibilities typical of these
projects (Martín-Barrera, Zamora-Ramírez, & González-González,
2016).
Majd and Pindyck (1987) analyzed a multistage investment
problem where the value of the project upon completion is un-
721
certain. The problem of cost uncertainty in an irreversible invest-
ment was first modeled as a real options problem by Pindyck
(1993), where he addressed the issue of an uncertain investment
cost subject to both market and private risks and where the value
of the completed project is known with certainty. By undertak-
ing R&D activities and incurring costs, the firm not only produces
an R&D output but also learns about the difficulty of the research
project and gets better information about the expected remaining
time and costs to completion. Based on this information, the firm
can optimally choose to continue its R&D efforts or abandon the
project altogether. The project is valued as a single period contin-
gent claim on a fixed asset value and a closed form solution is ob-
tained.
Schwartz and Moon (2001) extended this valuation model to
include a market uncertainty for the project revenues and the pos-
sibility of catastrophic events in a multi period setting to allow for
different investment rates for each stage, where the firm has the
option to abandon its R&D efforts at any time. Since no closed
solution exists, they solve by numerical methods. Miltersen and
Schwartz (2004) further extended the model to include the effects
of market competition in a duopolistic market setting.
According to Li, Watada, and Zhang (2015) and Lin et al. (2007)
the future movement of the markets is influenced by various un-
certain factors, then an investment project can be divided into
multiple stages that can effectively reduce the risks of the invest-
ment project, i.e., firms may have options to invest in multi-stage
R&D with an abandonment option (Dobbelaere, Luttens, & Peters,
2015). Hsu and Schwartz (2008) examined the response of a phar-
maceutical firm involved in a two stage R&D process of a new drug
to different types of government incentives. They model this as a
discrete two period project where the firm has the option to aban-
don development only at the end of each stage. The novel aspect
of this work is that it incorporates uncertainty in the quality of
the R&D output that is endogenously specified and which in turn
affects the efficiency and the market size for the product, and then
use simulation to value both the option to abandon and the project
itself. Lee, Rhee, and Cheng (2013) adopt a stochastic model to
analyze the impact of quality uncertainty in the context of sup-
ply chain coordination. Alexander, Mo, and Stent (2012) argue in
favor of using Arithmetic Brownian Motion (ABM) stochastic pro-
cess as an alternative to the common Geometric Brownian Motion
(GBM) assumption for modeling the underlying process, especially
for projects that are treated as part of an investment portfolio of
the firm
Technological uncertainties are also a factor when market com-
petitors are considered. Leung and Kwok (2016) used the game
theoretical real options model to analyze the characterization of
strategic equilibria associated with an asymmetric R&D race to de-
velop a new innovative product between two firms, an incumbent
and an entrant firm, under market and technological uncertain-
ties. The authors modeled the random arrival of the discovery of
the patent as a Poisson jump process and showed that preemptive
equilibrium does not occur in the R&D race due to the presence
of dominant second mover advantage. The sequential equilibrium
would occur if the sunk cost asymmetry is significant. Pennings
and Sereno (2011) also evaluated a pharmaceutical R&D venture
where the technical uncertainty is modeled as a Poisson process
and the firm has a series of compound options to abandon the de-
velopment, and show that project uncertainties increase the option
value. Morreale, Robba, Lo Nigro, and Roma (2017) examine the op-
timal timing problem from the point of view of a pharmaceutical
firm who must decide when to offer a licensing deal to a biotech
firm under an option game theoretical framework. Loch and Bode-
Greuel (2001) show that growth options play a relevant role in the
valuation of pharmaceutical products. Martzoukos (2003) demon-
strated the valuation of real options under incomplete information
722 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732

in the presence of endogenously generated actions of control (R&D)
and exogenous information arrival.
Cassimon et al. (2011) model a pharmaceutical project as a
compound option, using both the Geske (1979) approach and the
generalized n-fold compound option model of Cassimon, Engelen,
Thomassen, and Van Wouwe (2004), and arrive at a closed form
solution. Discrete success–failure probabilities are used to reflect
technical failure in each stage of the project, with a single volatil-
ity for each of the product development and market phases. Our
model differs from their research both in the formulation and so-
lution procedure. We use the Longstaff and Schwartz (2001) model
to price the options, a Poisson model to capture the probability
of exogenous catastrophic events, such as market or patent pre-
emption by a competitor or financial distress of the firm, and a
quality model to reflect the endogenous technical uncertainties,
rather than discrete success–failure probabilities. We also incorpo-
rate the option to expand during the market phase, allow for dis-
tinct volatilities in all stages of development and for the fact that
the duration of the cash flows in the market phase depends on
how long the development of the project took.
Our paper is close to Hsu and Schwartz (2008) in the sense
that it adopts a real options approach to value the project, and
we adopt their notation in order to facilitate a comparison. Their
simulation model though, only allows the option to abandon to be
exercised at the end of each stage, and thus requires the firm to
complete any stage already initiated regardless of the cost that will
be incurred by doing so. Even if it becomes clear to the firm that
further investment is not warranted, the firm is assumed to be in-
capable of interrupting the investment and abandon its research
efforts until this stage is completed. Since there is no possibility of
exercise while a stage is in progress, any learning that occurs be-
fore a particular stage is completed has no effect on that stage, as
any costs already incurred are sunk costs, and thus, irrelevant for
the investment decision of the firm. The only benefit of the learn-
ing that does occur is through the correlation between the actual
cost of the current stage and the expected initial cost of the sub-
sequent one. This is an essential feature of their model, without
which there would be no optionality involved in the problem, and
which allows the project, once initiated, to have a negative value.
Our model differs from Hsu and Schwartz (2008) in that it has
no such limitations on the timing of the exercise of the Ameri-
can option. We allow the firm to decide on the optimal operat-
ing strategy and, if necessary, abandon the project even before a
particular stage is completed, and thus avoid committing resources
to a project that will offer a negative expected Net Present Value
(NPV). Due to this, there will be a positive probability that any
particular stage will not be completed due to the early exercise of
the option to abandon. We also model the project cash flows and
the endogenous performance variable differently. Besides being af-
fected by market uncertainties, in our model the cash flows have
an additional dimension as they are also a function of the uncer-
tain time required to complete the project. Quality, on the other
hand, is defined as a function of a stochastic performance variable
decision of whether to abandon the project during any of the de-
velopment stages and may alter the optimal investment strategy
of the firm, and is, to the best of our knowledge, along with the
option for within stage abandonment, a novel addition to the liter-
ature on the valuation of R&D projects.
3. The model
Consider a multistage investment project where the first R&D
stage (1) ends at time t = τ 1 , the second stage (2) ends at time
t = τ 2 and the last R&D stage (n) ends at time t = τ n . If success-
fully completed, the project provides the firm with a continuous
stream of stochastic cash flows C˜(t ) during the subsequent market
phase, where t ∈ [τ n ,τ m ]. The project is subject to uncertainties
concerning the total cost of the investment required to complete
each stage, the final quality of the finished product, the level of the
cash flow streams and the risk of catastrophic failure that would
instantaneously render the project worthless. All these uncertain-
ties are assumed to be project and firm specific idiosyncratic risks,
uncorrelated to the market, and the firm and its shareholders are
assumed to be adequately diversified. Accordingly, we assign no
risk premium to these risks and discount them at the risk free
rate r. The stochastic cash flows the firm receives once all stages
are completed and the product is marketed is the only source of
systematic risk of the project, and thus commands a risk adjusted
discount rate μ > r. Fig. 1 shows the stages of the general model.
The initial expected costs to complete each stage are E0 [K˜i ],
i = 1,2…n, and it is assumed that each stage has a fixed rate of
investment Ii . Prior to the beginning of the project, the firm also
specifies the expected performance of the final product E0 [P˜(τn )],
which is one of the input factors for the quality model Ǫ, as we
explain subsequently.
Without uncertainty, the actual investment costs in each stage
are equal to their expected costs and there will also be no devia-
tions from the expected quality level of the final product. The so-
lution is straightforward since the project presents no managerial
flexibility other than the decision on whether to commit to the
project or not at the outset, and the value V (τn ) of the expected
cash flows the firm will receive upon completion of the project at
t = τ n is
 τm

V ( τn ) = E C˜(t )e−μ(t−τn ) dt (1)
τn
Considering that Ki = E0 [K˜i ], i = 1,2…n, the value F(V,Ki ,Q) of the
project is


F (V, Ki , Q ) = max V (τn )e−μτn

n−1
  τi+1
−r τi −r (t−τi )
− e Ii+1 e dt |Ki , Q ,0 (2)
i=0
τi

n
Ki
where Q is the quality model, τ0 = 0, τ1 = and τn =
that is negatively correlated with the deviations from the expected K1
I1 Ii
cost to completion of each stage, whereas Hsu and Schwartz resort i=1
to random draws from a Beta distribution bounded between 0 and The first term is the value of the completed project cash flows
1. Our quality model (Q) allows for values greater than 1, given that discounted to the current time t = τ 0 , while the remaining terms
significant positive externalities can occur during product develop- represent the present value of the total investment costs of each
ment such as the discovery of additional and unforeseen applica- stage of an n-stage project. Integrating the costs over all the devel-
tions. opment stages we arrive at
Our work also differs from these previous efforts in the sense


that we also analyze and incorporate the effects of an option to F (V, Ki , Q ) = max V (τn )e−μτn
expand the original project during the production and marketing
phase once development of the new product is successfully com- 
I j+1  −rτ j+1 
n−1

pleted. This expansion opportunity, which is analyzed as a com-
+ e − e−rτ j |Ki , Q , 0 (3)
pound American option, has important consequences on the prior r
j=0
 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732
Product Development Stages
Stage 1 Stage 2
τ1 τ2-τ1
0 τ1 τ2 τn-1
Firm continuously decides whether to continue investing
based on updated expectations of remaining costs to
completion, of final product performance and of value of
market stage
Decision to invest
in the project
Fig. 1. Stages and decisions of the general model.
The optimal rule when no uncertainty exists is to invest when-
ever F(V, Ki , Q) > 0, which is simply the traditional discounted cash
flow investment criteria. Since there are no uncertainties to be re-
solved in time, no learning occurs and whatever decision the firm
makes at the onset (to invest or not to invest) remains optimal
throughout the full life of the project. Without any new informa-
tion forthcoming, the firm has no reason to change its decision at
any future time t > τ 0 .
Under uncertainty, the firm can decide not only whether to
invest in the project or not, but, once the project is underway,
whether it should continue its investment efforts until all stages
are successfully completed. We now present the model in more
detail.
3.1. The investment cost model
The total cost of the investment and the time required to com-
plete each stage are uncertain. We follow closely the model de-
scribed by Pindyck (1993), where the expected cost of each stage i
is E0 [K˜i ] and there is a constant investment rate Ii in each stage.
We assume that the expected costs in each stage follow a ran-
dom walk with a negative drift term that reflects the instantaneous
(negative) investment rate of this cost stage, and that this diffusion
process is given by:
dKi = −Ii dt + σi dzi τi−1 < t < τi (4)
where dzi is the standard Wiener process that governs the invest-
ment cost process for stage i.
As the firm invests, the expected cost to completion tends to
decrease, but is also subject to random shocks which are assumed
to be the consequence of purely idiosyncratic (private) risks which
can be totally diversified away. Depending on direction and inten-
sity of these shocks, the firm may take a longer or a shorter time
to complete a particular stage than expected, which will cause the
actual investment cost to be respectively greater or less than ex-
pected.
We assume that the initial values of the expected costs to com-
pletion for the stages are correlated across consecutive stages. This
implies that the expected cost to completion of the next stage is
affected by the learning that occurs in the immediately preceding
stage, and undue delays that may occur in a development stage
negatively affect the initial expected costs to completion of the fol-
lowing stage. Likewise, if a stage is developing faster and at a lower
cost than expected, then the expected cost to completion of the
723
Market Stage
Stage n Option to Expand
τn-τn-1 τm-τn
τn τE τm
Firm continously decides whether to
expand based on current market
performance of product and cost of
additional investment
Decision to launch End of economic
product in the market life of product
following stage will also decrease, even though this next stage has
not yet begun. The information about the actual cost and timeli-
ness of a particular stage i is conveyed by the random component
σ i dzi of the cost diffusion process. In addition to the process de-
scribed by Eq. (4), for any two consecutive stages i and i + 1 we
will also have in stage i the process described by Eq. (5):
dKi+1 = −Ii+1 dt + σi+1 dzi+1 τi−1 < t < τi (5)
where ρ dt = E[dzi dzi+1 ] is the instantaneous correlation between
dzi and dzi+ 1 . The expected cost to completion of stage i + 1 will be
updated during the time period τi−1 < t < τi prior to the beginning
of this stage through the correlation between the two processes,
even though stage i + 1 has not yet begun.
3.2. Learning
The firm may abandon the project at any point in time, includ-
ing during any particular stage up to the beginning of the market
phase. At each instant it assesses the expected value of the project
given the actual costs incurred up to then, the expected final qual-
ity level conditional on the deviation from the expected times to
completion and the value of the future expected cash flow that will
be generated by the project. The probability that the project may
suffer a sudden and catastrophic ending is also taken into account
and factored into the final value of the project. A negative value
indicates that the expected future revenues will be insufficient to
cover the realized and expected costs and the firm will be better
off if it abandons the project at that point. If this value is positive,
then the firm continues on until the next instantaneous exercise
time when it again repeats this analysis with updated information.
Once all stages are successfully completed and the product is
ready to go to market at t = τ n , the firm has one final opportunity
to abandon the venture prior to the beginning of the market phase.
This option to abandon will only be exercised if the expected value
of the future project revenues net of any capital expenditure in-
vestment required for the production of the final goods is nega-
tive, since all prior costs incurred to develop the project are now
sunk costs. The level of the capital investment at t = τ n will depend
on the nature of the R&D venture, ranging from low for intellec-
tual property goods such as software and pharmaceutical drugs, to
high for technologically complex products such as aircraft and mi-
croprocessors.
As the product is brought to market, the firm immediately be-
gins to analyze possible enhancements into derivative products and
724 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732
product extensions in order to capture additional market segments
that may not have been targeted initially. The value of this ex-
pansion opportunity is a function of the actual performance of the
current product in the market, which must be sufficiently large to
warrant the additional investment required for the expansion. We
assume that the investment cost required to implement the expan-
sion is a fraction of the total costs originally incurred to develop
and manufacture the product and involves little or no uncertainty.
3.3. Quality model
During the design phase and prior to the beginning of the
project, the firm identifies a set of technical specifications, which
we define as performance, that the completed product is expected
to achieve. Once the project is initiated and investment begins, the
firm obtains information that allows it to revise these expectations
to adjust them to the facts that are being uncovered by the re-
search, development or construction that is taking place. Extend-
ing the model to include additional measures of performance is
straightforward.
We now define our quality model in more detail. The firm ex-
ogenously determines the time zero expected performance of the
final product, conditional on the evolution of the investment stages
of the project, which may be actual construction phases or R&D
stages. As more information becomes available and the firm learns
about the ease or difficulty of completing each stage, the firm up-
dates this expectation. In all, the performance is affected by the
learning that occurs in each stage, and if there is no deviation
from the expected costs, then no learning occurs and the initial
performance level assessed remains unchanged. This implies that
this stochastic variable has no drift, although random events may
affect the performance of the final product.
The time zero expectation of the final performance of the prod-
uct is E0 [P˜(τn )] where P˜(τn ) is the (uncertain) performance of the
final product at the end of all the R&D stages. During each stage
i the firm assesses its expectation of the performance of the final
product based on the learning that occurs in that stage. Although
this assessment is done continuously, we assume that the firm
only updates its expectation of the performance Eτi [P˜(τn )] at the
final product Eτi [P˜(τn )] will then be a function of the initial ex-
pected performance level Eτi−1 [P˜(τn )] of the previous stage and the
correlation of S˜(t ) with the deviations from the expected invest-
ment costs E0 [K˜i ], and the only source of drift from the expected
value of S˜(t ) will come from the correlation with the deviations
in cost. As information flows and learning occurs, the performance
expectation is updated at the end of each stage and Eτi [P˜(τn )] be-
comes the revised expectation for the performance of the final
product. The process is repeated for each stage until all devel-
opment stages are completed and the final product performance
P (τn ) is determined.
If actual R&D costs are the same as the expected costs for a
particular stage, then no learning has occurred, S(t) remains at its
initial value S(0) = 0 and the expected performance level remains
constant, since eS(t ) = 1 and Eτi [P˜(τn )] = 1 − (1 − Eτi−1 [P˜(τn )] ) =
Eτ [P˜(τn )]. If product development is progressing at a fast pace
i−1
with no major hurdles in a particular stage, the actual time and
costs of this stage will be less than expected and the cost deviation
will be negative. S(t) will then increase and become greater than
zero, and the expected final performance will be revised upwards.
For very high values of S(t) we have eS(t ) → ∞ and Eτi [P˜(τn )] =
1 − (1 − Eτi−1 [P˜(τn )] )∞ = 1.
On the other hand, if the R&D stage takes longer than expected
and costs increase beyond the expected costs for this stage, the de-
viation will be positive and S(t) will decrease in value and become
negative. This will cause a similar decrease in the expected per-
formance of the product, indicating that unforeseen problems are
hampering the R&D efforts and that product performance will suf-
fer. For large cost overruns the corresponding decrease in S(t) may
result in large negative values of S(t) and we will have eS(t ) → 0
and Eτi [P˜(τn )] = 1 − (1 − Eτi−1 [P˜(τn )] )0 = 0.
Under this process, the stochastic performance of the final
product evolves endogenously, spanning all stages of the project.
The value function of the project then incorporates this uncer-
tainty, in addition to the uncertainty in the investment cost, risk of
catastrophic failure and market demand for the product by means
of the quality model Q, which we define in Eq. (8)
 γ
P ( τn )
end of each stage, after all the information for that particular stage Q= 
(8)
becomes available. This updated expectation is then the starting E0 P˜(τn )
point for the performance assessment during the next stage. This
process ends at t = τ n when the actual value P (τn ) = Eτn [P˜(τn )] is where P (τn )
is the ratio between the observed and the expected
E0 [P˜(τn )]
finally determined.
final product performance and γ ≥ 1 is the sensitivity of the ini-
We model the product performance P˜(τn ) as a function of a
tial market cash flows to deviations in the performance of the fi-
diffusion process S˜(t ), 0 ≤ t ≤ τi − τi−1 , fluctuating stochastically
nal product. The quality model leverages the results of the pro-
within the interval [0,1]. We define the stochastic variable S(t) as a
cess. For γ = 1, it is simply the ratio between the observed and the
driftless random walk over (-∞, + ∞) as shown in Eq. (6), which is
expected quality of the product. For higher values of γ , Q will in-
negatively correlated with the deviations from the expected invest-
crease if the product performance turned out to be better than ini-
ment costs associated with each stage i as defined by Eq. (4). This
tially expected, and at the limit, for very high values of γ , Q → ∞.
is in keeping with the assumption that the expected performance
On the other hand, if the final product performance turns out to be
must remain unchanged if no learning occurs during the develop-
worse than expected, then Q → 0 for high values of γ . The sensi-
ment stages.
tivity parameter γ can be used to calibrate the model to reflect
dS(t ) = σS (t )dzS (6) the expected impact of the product performance on the value of
the project.
where ρS dt = E[dzS dzi ], i = 1, 2, ...n, ρS ≤ 0 is the instanta-
neous correlation between dzs and the Wiener process dzi from
Eq. (4), and S(0) = 0. We assume that the expected performance 3.4. Risk of catastrophic failure
at the end of stage i Eτi [P˜(τn )] is a function of S˜(t ) as specified in
Eq. (7) where t = τi − τi−1 and which assures that Eτi [P˜(τn )] re- At any time, including in the market phase, the project may be
mains within the range of [0,1]. rendered worthless due to a preemptive move by a competitor, loss

  
 of the technical capability required to undertake the necessary re-
Eτi P˜(τn ) = 1 − exp eS(t ) log 1 − Eτi−1 P˜(τn ) (7)
search such as the loss of strategic personnel, or changes in reg-
We take the deviations between the actual and expected costs ulations that may require the firm to withdraw the product from
in each stage as a proxy for the learning that occurs during each the market. If such an event occurs, for whatever reason, the result
R&D stage. At the end of stage i, the expected performance of the is that the value of the project for the firm goes instantly to zero.
 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732 725

We model this uncertainty with a Poisson process, where at the volatility σ of the stock, and the risk adjusted discount rate for
any time t the probability of a catastrophic event occurring in the the project market cash flows is also μ.
next small interval of time dt is λdt. Suppose a project producing If we consider only the net revenues associated with the orig-
a cash flow of F in perpetuity is subject to this uncertainty. Then, inal product assuming that the project offers no managerial flex-
the probability that it will continue for the short time period dt ibility during the market phase, the value of the project at any
∞
is 1-λdt. The value of the project is V (F ) = 0 E[C] e−μt dt, where time t during the market phase can be determined by simply tak-
n n λ
E[F ] = (1 − λdt ) F , and (1 − λdt ) = e . Substituting in the value
− t ing expectations and discounting at the risk adjusted discount rate
∞
function, we arrive at V = 0 e−(μ+λ )t F dt . Therefore the impact μ as shown in Eq. (10), where τ n < t < τ m . This is also the stan-
of the probability of a catastrophic event that suddenly ends the dard value that is obtained from the traditional discounted cash
project is equivalent to an increase in the discount rate by an flow method.
amount λ. If this value is different for each of the development  τm

stages, we then represent this uncertainty as λi , where i represents V (C, t ) = Et C (τ ) e−μ(τ −t ) dτ |C (τn ), Q (10)
t
the ith stage.

3.6. Option to expand

3.5. Market phase

Once a project is successfully completed, the firm engages in

Once all the R&D stages are successfully completed at t = τ n and
the market phase where it begins to manufacture the product and
the product is approved for release, the firm has the final option to
receive the net revenues associated with its sales. If additional ap-
bring the product to market or to abandon the project altogether.
plications for the product were discovered during the R&D phase
Although the firm could possibly defer the decision to launch the
as signaled by the quality model, the firm has the opportunity to
product in the market at this point, technological obsolescence,
expand the scope of its original market. The firm may also choose
loss of patent life and the risk of having a competitor preempt the
to continue research on the product beyond the R&D stages in or-
market by launching a similar product increases the downside ef-
der to develop derivative products and/or additional markets seg-
fect of this decision and we assume the firm will go straight to
ments. This presents the firm with an opportunity to expand the
market.
original market with tailored versions of the product and to de-
Bringing the product to market entitles the firm to an uncer-
velop extensions of the basic product.
tain cash flow stream for the duration of the market phase whose
Usually the firm knows well in advance of the actual develop-
value at t = τ n is C˜(τn )Q, where Q is the final product quality as
ment of the product both the expected benefits and the capital
defined in (8). At this point, except for cases where significant cap-
costs involved and also whether such an opportunity will be avail-
ital expenditures are required to begin the manufacture of the final
able or not for any particular product, and thus we assume that if
product, all investment costs already been irreversibly expended,
exercised, the project value is increased by a factor of κ at a cost
so only the expectation of a negative net cash flow stream would
of ψ .
lead the firm to abandon the project.
The decision of whether to exercise the option to expand dur-
The initial level of these cash flows C (τ0 , E0 [Q] ) at the start of
ing the market phase is modeled as an American type option and
the project (t = τ0 ) is exogenously defined and reflects the firm’s
the value of this opportunity will be a function of the cash flow
expectation of the commercial success of the developed product
stream generated by the original product, the benefits of the ex-
and final product quality. These cash flows are subject to system-
pansion, the volatility of the project value, the risk free rate, the
atic risk over market uncertainties concerning pricing and demand
time to expiration and the exercise cost, which is the investment
for the final product, and are assumed to evolve according to a Ge-
required to implement and market the modified product. At each
ometric Brownian Motion diffusion process with growth rate of α
instant t, τn < t < τm , the firm assesses the expected value of the
and volatility of σC as the project progresses through the develop-
remaining project cash flow stream V (t ) and compares this to the
ment and market phases, as shown in Eq. (9). At any point in time
expected value of the project were the expansion to take place at
they represent the net cash flows that would accrue to the firm
that time, V  (t ) = κ V (t ) − ψ , net of the required capital invest-
were the project instantaneously completed at that time.
ment ψ and chooses the greatest of the two. The instantaneous
dC (t ) = αC (t )dt + σC C (t )dz τ0 < t < τm (9) value of the project with the option to expand during the mar-
ket phase is then F (t ) = max{V (t ), V  (t )}, τn < t < τm . We assume
While the firm does not receive any cash flows until the project
that the firm has a single opportunity to exercise the option during
is completed, it updates this value during the investment and de-
the market phase and does so at an uncertain time t = τ E , when
velopment stages as it observes the market and obtains more in-
V  (τE ) > V (τE ).
formation about the true potential of the product. The information
imparted by this variable also impacts the project value and the
3.7. Valuation and solution
firm’s optimal investment and operational strategies.
Given that it takes time to develop and complete the project,
As is usual in dynamic programming, we begin by assessing the
the actual level of the cash flows that will occur during the market
value function from the terminal value of the project and work our
phase is uncertain and will depend on how soon the project will
way backwards. We rely on the equivalent risk neutral valuation
be completed, among other factors. While the firm can observe the
and modify the cash flow diffusion process accordingly while dis-
level of these cash flows at any time, it only starts to receive them
counting future cash flows under the risk neutral measure at the
when, and if, the project is completed at t = τn .
risk free rate r, as is standard in the option pricing literature. The
Under market equilibrium, the risk adjusted rate of return of
modified cash flow process will be
the firm’s shares according to the CAPM is μ = r + ϕ σ ρm,S where
r is the risk free rate, ϕ is the market price of risk, σ is the volatil- dC (t ) = rC  (t )dt + σC C  (t )dz τ0 < t < τm (11)
ity of the stock returns, and ρ m,S is the correlation between the
The Bellman equation for the value of this cash flow stream is
returns of the market and that of the stock of the firm. We assume
shown in Eq. (12).
that the project is typical of all the projects of the firm and thus
the stochastic changes in the value of the project are spanned by 1 
rV (C  , t ) = C (t ) + E dV (C  , t ) (12)
the firm’s stock so that the volatility of the project is the same as dt
726 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732

Applying Ito’s Lemma to dV(C’,t) we obtain future project cash flows at t = τ n- 1 is

  τn
∂ V (C  , t ) ∂ V (C  , t )  V (τn−1 ) = E In e−(r+λn )(t−τn−1 ) dt + F (τn )e− (μ + λn )(τn
dV (C  , t ) = + (r − δ ) C (t )
∂t ∂ C  (t ) τn−1
 
1 ∂ 2V (C  , t ) 2  2 ∂ V (C  , t )  −τn−1 )C (τ0 ), Eτn−1 [Kn ], Eτn−1 [Qτn ] (17)
+ σ C (t ) dt + σ C (t )dz
2 ∂ C  (t ) 2 C
∂ C  (t ) C and given that the firm will abandon the project if Vτn−1
is negative, the value of the project at t = τ n- 1 is F (τn−1 ) =
Substituting this in Eq. (12), we arrive at the value Eq. (13) that
max{V (τn−1 ), 0}. As the firm progresses in this development stage
the project value must satisfy.
it updates its expectation of the cost to completion while amounts
1 ∂ 2V (C  , t ) 2  2 ∂ V (C  , t )  ∂ V (C  , t ) already invested become sunk costs, and this new information is
σ C (t ) + ( r − δ ) C (t ) +
2 ∂ C  (t )2 C
∂ C  (t ) ∂t then used to optimally exercise the option to abandon at any time
during this stage. As previously mentioned, we assume the firm
−rV (C  , t ) + C  (t ) = 0 (13)
only updates its expectation of the quality of the final product at
The project value is monotonically decreasing due to the cash the end of each stage.
flows that accrue to the firm and their shareholders during the We progress steadily back in time continuously through all the
market phase, and thus the boundary condition at t = τ m is development stages in the same fashion, adopting the different
V (C  , τm ) = 0, which states that the terminal value is zero. The values for the variables and parameters for each of these stages.
solution to this partial differential equation is shown in Eq. (14), For example, the value of the project during the first development
which is the risk neutral version of Eq. (10). stage will be
   τ1 
τm 
V (C  , t ) = Et C  (τ ) e−r (τ −t ) dτ |C (τn ), Q (τn ) (14) V (t ) = E I1 e−(r+λ1 )(τ −t ) dτ + F (τ1 )e−(μ+λ1 )(τ1 −t ) C (τ0 ), E0 [K1 ],
t
t

We now include both the option to expand the project and the ..., E0 [Kn ], E0 [Qτn ] (18)
probability of sudden catastrophic failure of the project, which is
represented by the Poisson death parameter λm . For simplicity of where τ0 < t < τ1 . The value of the project at time t is then
notation we drop the subscript C’ of V(C’, t). If it is optimal to ex- F (t ) = max{V (t ), 0}, and we proceed in this manner until time
pand during the market phase we assume this will occur at time t = τ 0 when
t = τ E , and the remaining expected future cash flows would in-
crease by a factor of κ while the project would suffer an imme- F (0 ) = max {V (0 ), 0} (19)
diate and instantaneous investment cost of ψ . The value V  (τE ) of
Fig. 2 presents the project valuation model where the value of
the project with this expansion and also considering the possibility
the project is shown at discrete times at the beginning of each de-
of catastrophic failure would then be
velopment and market phases of the project.
  τm

V  (τ E) = E κ C  (t ) e−(r+λm )(t−τE ) dt |C (τn ), Q (τn ) −ψ (15) 4. Application: R&D in the pharmaceutical industry
τE

With this in mind, we can now determine the value of the
project at time t = τ n , prior to the beginning of the market com-
mercialization of the product. All the development stages have
been successfully completed and all investment expenses are now
sunk costs. If we ignore any additional capital expenditures, the
firm will proceed to market if the expected value of the market
phase is positive. The value of the future cash flows at this point
is then:
 τE
 sive and risky investment. The PhRMA Association, which repre-
V (τn )=E C  (t ) e−(r+λm )(t−τn ) dt +F (τ E )e −(r +λm )(τE −τn )
|C (τn ), Q (τn )
τn
(16)
At time t = τ n the firm gains V(τ n ) if it continues and noth-
ing if it abandons. Accordingly, the project will be abandoned
only if V(τ n ) < 0. The value of the project will then be F (τn ) =
max{V (τn ), 0}.
At any time prior to τ n , the firm must assess and deduct the
expected cost of the investment still to be incurred from the dis-
counted value of the future project cash flows, while continuously
determining whether it is optimal to abandon or to continue in-
vesting conditional on the expected quality of the final product.
While this process is continuous, the expected investment costs,
investment rates and other parameters such as cost correlation
across contiguous development stages may be different for each
development stage.
At the beginning of the last development stage n at time
t = τ n- 1 , the firm must invest an uncertain amount Kn at an in-
vestment rate In in order to complete this stage. The value of the
Drugs are a substance or combination of substances presented
as having properties for treating or preventing disease in hu-
man beings (European Union, 2014). Drugs contain molecules that
present biological activity against a targeted disease, and since not
all molecules have these properties, the task of the drug maker is
to discover which ones are effective against the disease. The devel-
opment of pharmaceutical drugs has evolved significantly in the
last decades and has since become an increasingly capital inten-
sents the leading research-based pharmaceutical and biotechnol-
ogy companies in the United States, estimates that its member
companies invested $58.8 billion dollars in 2015 on research to de-
velop new treatment for diseases, on sales of $297 billion dollars
(PhRMA, 2016).
Before a firm can apply for approval by the Federal Drug Ad-
ministration (FDA) to take a drug to market, the firm must suc-
cessfully complete several well defined research stages and clinical
trials in order to establish its efficacy, safety, dosage and other pa-
rameters. Fig. 3 shows the required stages for full development of
a drug in the United States.
Each of these stages provides the firm with additional informa-
tion about the medical and economic feasibility of the compound
as an effective and profitable drug, and based on this information
the firm can decide whether to continue or abandon the project.
Each stage carries a significant chance of failure and ultimately
only a small fraction of the drugs that enter into pre-clinical test-
ing get to the FDA approval phase.
The R&D process begins with the discovery stage, where a new
molecule is identified, synthesized and screened as a potential
 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732 727

Stage I Stage II Stage n Market Stage

τ2 -
τ1 τn-τn-1 τM-τn
τ1

τ0 τ1 τ2 τn-1 τn τE τM
F(τE)
V(τE)
F(τn)
V(τn)
F(τn-1)
V(τn-1)
F(τ2)
V(τ2)
F(τ1)
V(τ1)
F(τ0)
V(τ0)

Fig. 2. Dynamic Programming Valuation Model.

Discovery/
Phase Phase Phase Phase
Preclinical FDA
I II III IV
Testing

Years 6.5 1,5 2 3,5 1,5

20 to 100 100 to 500 1,000 to 5,000

Test Laboratory and

File NDA at FDA

healthy patient patient
File IND at FDA

Population animal studies Additional

volunteers volunteers volunteers
post-
Review
marketing
Confirm process/
Assess safety, Evaluate testing
Determine effectiveness, approval
biological effectiveness, required
Purpose safety and monitor adverse
activity and look for by FDA
dosage reactions from
formulations side effects
long term use
5,000
Success 5 1
compounds
Rate enter trials approved
evaluated

$5M - $100M $300M - $800M $5M - $50M

Fig. 3. R&D Stages for Drug Development. (B.A. Spilker. “The drug discovery, development, and approval process”, in Medicines in Development for Women, 2004.)

drug. Most of the potential molecules are abandoned during this
stage, and only a few candidate molecules progress to preclinical
trials, where they are then analyzed for pharmacological activity
against the targeted disease and toxicity using animals. If results
of this stage indicate that the drug is safe enough to be tested on
humans, the firm then files a IND (Investigational New Drug Appli-
cation) request to inform the FDA that it intends to enter clinical
trials.
Each of the three stages that comprise the clinical trials involves
a growing number of human subjects and has the objectives of
testing the potential drug for safety, appropriate dosage, negative
side effects and of assessing its effectiveness for the intended use
in controlled groups. If all three stages of clinical trials are suc-
cessfully completed, the firm then files a NDA (New Drug Appli-
cation) where it requests FDA authorization to market the drug to
the general population, where only one in five drugs are eventually
approved.
Once the project is successfully completed and FDA approval is
secured, the drug is brought to market. This step presents the firm
with the opportunity to expand and extend the original market
for the product to include additional patient groups or markets by
making improvements and/or modifications to the existing drug.
The feasibility of this expansion will center on the cost of these
improvements and the revenue of the existing drug, which in turn
is a function of the quality and the market demand for the drug.
Once the market phase is initiated, the firm measures the level of
revenues and decides whether an expansion is warranted.
Although the United States provides a 20 year patent life for a
new drug, the average commercial life during which the firm has
exclusive marketing rights is estimated to be 13 years. This is due
to the fact that the patent application is usually filed and granted
while the drug is still under development, which reduces the ef-
fective patent duration. We assume that once the patent expires,
due to market competition continuing operation will yield no fur-
ther value creation for the firm and the project expires at the end
of its effective patent life with no terminal value.
For the purpose of applying the ideas developed in our model,
we analyze a hypothetical, but typical, drug development project
involving three development stages, followed by a market phase
equal to the duration of its remaining patent protection, where the
firm earns a stream of uncertain cash flows and has the opportu-
nity to expand its operations.
728 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732

Table 1 value is the average of these values. The difference in the number
Parameters for development stage.
of simulations between the steps is to facilitate the computational
Development stage Units Stage I Stage II Stage III process. For computational purposes we divide each calendar year
Expected investment cost Ki MUSD∗ 30 50 90 into nine discrete and equal time periods where each of these is
Investment rate Ii MUSD∗ 15 25 30 a potential exercise period. While the time step n and the pre-
Cost volatility σi % 10 10 15 cision parameters chosen provide a level of discretization that is
Cost correlation ρ − 0.30 0.30 adequate for the valuation purpose of this application, it may be
∗
Million USD. coarse for graphic displays and the approximation errors appear as
jagged lines in some of the figures. Greater detail can be obtained
4.1. Analytical results by increasing the time step parameter n and increasing the num-
ber of iterations, at the cost of increased computational processing
The industry data and parameters adopted are assumed to be times.
typical of the industry and are shown in Tables 1 and 2 (Berk, In order to generate simulation paths for the project uncertain-
Green, & Naik, 2004; DiMasi, Grabowski, & Hansen, 2016; DiMasi, ties, the processes for the continuous expected cost and for the
Hansen, Grabowski, & Lasagna, 1995; Gu, 2016; Hansen & Grunow, future cash flow are discretized, respectively, as follows:
2015; Hsu & Schwartz, 2008; Struck, 1996). The use of different Kt+
t = Kt − Ii
t + σi (Ii Kt )1/2 (
t )1/2 εi , i = 1, 2, 3 (20)
input data can be easily accommodated by the model.
We assume that the R&D project is subject to the following un-
certainties: the total cost of the investment required to complete
t = Ct − exp (r − 0.5σc2 )
t + σc (
t ) ε (21)
each stage and the time to complete each stage; the risk of catas-
trophic failure; the final quality of the finished product and the
level of the future cash flow streams. We assume that these uncer-
tainties are uncorrelated to the market and we discount them at
the risk free rate r. The stochastic cash flow is the only source of
systematic risk of the project, and thus commands a risk adjusted
discount rate μ > r.
We consider the opportunity to abandon, during the R&D phase,
and to expand the project during the market phase. With the op-
tion to abandon, the firm can opt out of the project if it learns that
development is lagging, and thus is able to avoid a potential loss.
This has the net effect of increasing the project value whenever
the option is exercised, which is more likely to occur when the
initial cash flow level is close to the threshold level. The option to
expand the project, on the other hand, has the power to increase
the project revenues after the product has been successful during
the development phase. The less time taken for the product to be
launched on the market, the greater the opportunities to earn prof-
its with the expansion option.
The solution is calculated using the least squares method (LSM)
of Longstaff and Schwartz (2001) for valuating American options.
Longstaff and Schwartz (2001, pg. 131 and Table 4) show that the
difference between their method and the pricing of American op-
tions using finite differences is very small relative to the level of
the option values, and well within the bid-ask spread bounds of
financial options. We use polynomials as a set of basis functions
of the state variables to implement the algorithm. We verified our
computational model by simplifying its structure to match the sin-
gle R&D phase model analyzed by Schwartz (2004), and we ob-
tained very similar results.
We assume the project has a total duration of 20 years, which
represents the duration of its patent life, and which will be di-
vided between its R&D and marketing phases. Thus, the greater
the time for R&D development, the less time the firm will enjoy
the benefits of the cash flows from the marketing phase, and we
assume that if τ n > 10 years, the project is immediately abandoned
as the remaining time would be insufficient to recoup the invest-
ment. Otherwise, we simulate 10,0 0 0 total R&D cost scenarios and
use LSM to calculate the option value of abandoning and expand-
ing the project for each cost scenario.
In this case, by the end of each R&D phase, we run 10,0 0 0 itera-
tions for the simulation of the market phase for each cost scenario,
bringing the total number of iterations to 100 million, and again
use LSM to calculate the option value of the expansion option dur-
ing the market phase. Then, these values are discounted to present
value and distributed over the R&D phase, when a new LSM is cal-
culated to find the value of the abandonment option. The option
 1/2
Ct+
From these stochastic processes, we first determine the dura-
tion of stages 1, 2, 3 (τ 1 , τ 2 , τ 3 ) using Eq. (20), and then calculate
the cost of each stage as the investment rate times the stage du-
ration. The total duration of the R&D phase is the sum of the du-
ration of each stage (τ n= τ 1 + τ 2 + τ 3 ), while the total cost is the
sum of the present value of all costs.
Next, we divide the market simulation in two steps. First, we
consider that the market phase is not correlated with the prod-
uct development (R&D) phase, so this process is modeled using
Eq. (21) with c0= 35. At the end of the R&D phase we calculate the
product quality and multiply this value by the last value of the
previous simulation to find the initial value (c0 ) of the second sim-
ulation of the market phase. In both simulations we consider the
same parameters, except for the initial value. Finally, the present
value of the market phase is calculated and compared with the to-
tal cost. Fig. 4 shows the result of one of the paths generated by
the random simulation for the expected remaining cost to comple-
tion and for the project cash flows.
For this particular path the investment took 10 years to com-
plete, at which time the remaining cost to completion is zero. At
the moment the investment came to the end, the cash flow was
worth approximately $3.4 million. During the project life the cash
flow has gone through ups and downs and reached a peak of ap-
proximately $6.5 million between the sixteenth and seventeenth
year. Although this figure only represents one of the simulated
paths, the result is an illustration of how the state variables may
behave in the simulation.
Under traditional discounted cash flow analysis (DCF), where no
consideration is given to project options, investment cost correla-
tions or product quality, the value of the project will grow linearly
as a function of the initial level of the cash flow stream once the
minimum cash flow investment threshold level is overcome. This
threshold represents the minimum cash flow level at which the
firm is indifferent about whether to invest or not in the project,
which in our case is $36 million. Since the initial value of the cash
flow stream is $35 million, the project will not be undertaken. The
options to abandon and expand increase the project NPV to $16.73
million and decrease the threshold level to $20 million. This low
value is due to the high percentage of project abandonment and
expansion decisions. Fig. 5 illustrates the probability distribution
of research time in each development phase, where the expected
times for phases I, II and III are respectively two, two and three
years.
The quality of the final product can enhance or reduce the
market value of the project depending on the main parameters
of the process, as shown in Table 3. Increases in expected perfor-
mance reduce both the value of the project without options and
 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732 729

Cost to Completion per stage ($millions) 160 7

140 6

120

Cash Flows ($ millions)

5
100
4
80
3
60
2
40

20 1

0 0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Time (year)

Expected Cost to Completion Project Cash Flows

Fig. 4. Simulation of state variables of the R&D project.

8
30

6
% Probability
% Probability

20
4

10
2

0
0
0,0 5,0 10,0 15,0
1,0 2,0 3,0
Phase III Research Time (years)
Phase I Research Time (years)

20 6

15
% Probability

4
% Probability

10

2
5

0 0
1,0 2,0 3,0 4,0 0,0 5,0 10,0 15,0
Phase II Research Time (years) Time to Compleon (years)
Fig. 5. Distribution of research time in each phase.
730 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732

Table 2
Parameters for market phase.

Market phase Other Parameters

Initial cash flow rate C0 35 MUSD∗ Expected performance level P 0.80

Cash flow growth rate α 8% Performance correlation ρS 0.30
Cash flow volatility σC 30% Performance volatility σS 0.40
Time for patent expiration D 13 years Sensitivity parameter γ 2.0
Annual risk adjusted rate μ 13% Annual risk free rate r 5%
Expansion cost ψ 100 MUSD∗ Poisson parameter λ 0.08
Expansion factor κ 0.50 N° of time steps per year n 7
∗
Million USD.

Table 3
Comparative statics with respect to the quality model.

Expected performance (P) NPV project value Option value Value with options

0.50 −1.222 18.696 17.474

0.60 −1.477 18.697 17.220
0.70 −1.757 18.680 16.923
0.80 −1.960 18.689 16.729
0.90 −2.196 18.701 16.505

Performance volatility (σ S ) NPV project value Option value Value with options

0.20 −2.087 18.779 16.693

0.30 −2.010 18.737 16.727
0.40 −1.960 18.689 16.729
0.50 −1.949 18.624 16.675
0.60 −1.939 18.565 16.626

Sensitivity parameter (γ ) NPV project value Option value Value with options

1.00 −2.164 18.749 16.585

1.50 −2.109 18.704 16.595
2.00 −1.960 18.689 16.729
3.00 −1.713 18.633 16.920
4.00 −1.314 18.637 17.322

Table 4
Comparative statics of model parameters.

Probability of failure (λ) NPV project value Option value Value with options

0.05 20.254 6.364 26.618

0.06 11.200 9.588 20.787
0.07 3.571 13.626 17.197
0.08 −1.960 18.689 16.729
0.09 −8.239 23.974 15.735

Risk free rate (r) NPV project value Option value Value with options

0.03 9.118 10.239 19.356

0.04 2.912 14.223 17.135
0.05 −1.960 18.689 16.729
0.06 −6.853 22.766 15.913
0.07 −10.655 26.669 16.014

the value of the options, as higher expectations are harder to meet
and the project value suffers accordingly. The volatility of the
performance variable, on the other hand, has a negligible impact
on the final value of the project due to its symmetrical nature. The
effect of the sensitivity parameter depends on whether the final
product performance turns out to be better or worse than initially
expected, as shown in Section 3. In our numerical example this
ratio is greater than 1, which accounts for the increase in project
value as the sensitivity parameter increases.
Given that the occurrence of a catastrophic event prevents the
product from being successfully developed, an increase in the
probability of failure (λ) affects both the project and the option
value, as shown in Table 4. As the project value decreases with
the increase in the probability of catastrophic failure, the option to
abandon becomes more valuable. Nonetheless, the overall project
value is still less for greater values of the failure parameter. The
project is also very sensitive to the discount rate, which is typical
of projects with long maturity times such as drug development.
Table 5 illustrates the effect on the project of changes in the
initial cash flow rate, expansion factor and patent expiration times
considering the project value without options as determined by its
NPV, with only the option to abandon and finally with both the op-
tions to abandon and expand. We can see that the cash flow rate
has a significant impact on all three values, while the expansion
factor affects only the option to expand during the market phase.
The sensitivity of the project with respect to changes in the patent
life has a significant impact on the value of the project. Extending
the duration of the patent by 10% increases the value of the project
by 50.2%, while a decrease of 10% cuts the value of the project
by half. We can also see that the opportunity to expand during
the market phase has a significant impact on the value of the
project.
For any given level of the cash flow rate, there is a critical
cost to completion above which it is not optimal to invest in the
project. For the base case cash flow rate of $36 million, this criti-
cal cost is $170 million. Fig. 6 shows the investment threshold as
 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732 731

Table 5
Comparative statics with respect to cash flow rate, expansion factor and patent expiration time.

Cash flow rate NPV project value Project with abandon option Project with abandon and expansion

25 −28.512 0.695 3.494

30 −15.44 5.463 12.198
35 −1.960 10.974 16.729
40 10.704 14.081 17.579
45 23.775 24.113 25.081

Expansion factor NPV project value Project with abandon option Project with abandon and expansion

0.1 −1.960 10.974 16.416

0.3 −1.960 10.974 16.459
0.5 −1.960 10.974 16.729
1.0 −1.960 10.974 17.201
1.5 −1.960 10.974 18.446

Patent expiration NPV project value Project with abandon option Project with abandon and expansion

18 −7.586 3.894 9.436

19 −5.976 5.457 11.317
20 −1.960 10.974 16.729
21 1.472 15.307 21.203
22 4.469 19.203 25.130

40 We show that the existence of the opportunities to abandon

the project and further expand the market revenues of the fin-
35 ished product once the development stages have been successfully
Initial Cash Flow Rate

Investment area
30
25
Non-investment area
20
15
100 120 140
Cost to Compleon (Millions)
Fig. 6. Investment threshold as function of initial cash flow rate and cost to com-
pletion.
a function of the cost to completion. Above or to the left of the
gray boundary, when cash flows are high and/or costs are low it is
optimal to invest. Below and to the right of the gray boundary it is
optimal to abandon the project.
5. Conclusions
The analysis of multistage investment problems is a typical ap-
plication of real options analysis given the characteristics of a com-
pound option problem. We modeled a R&D investment problem
with multiple sources of uncertainty and compound abandon and
expansion American options using a simulation approach.
Traditional real options models usually require simplifying as-
sumptions that limit the complexity of the model in order to
maintain tractability of the solution. In this sense, the model we
propose is more realistic and better reflects the actual decision
making process of R&D initiatives. The main contributions of this
paper are twofold. First, the timing of the exercise of the option to
abandon is not limited to the moment of completion of each stage.
In our model, if necessary, the firm can abandon the project be-
fore any particular stage is fully completed and avoid committing
to a project with negative NPV. Second, the quality model, which
is original, is correlated with the results of the R&D efforts and is
updated continuously as the project develops. In addition, at the
end of the development phase, we model an additional option for
the firm to expand the project into new markets.
completed can significantly affect the value of the project and, as
a result, the optimal investment decision. In particular, we show
that the option to expand after product development is completed
can increase the project value by an order of magnitude. Addition-
ally, the uncertainty over the quality of the finished product also
impacts the valuation problem, and is in turn affected by the diffi-
culties that arise from the product development.
The model developed in this article can assist firms in obtaining
a better valuation assessment of their R&D projects and in improv-
170
ing risk management mechanisms and decision making processes.
In the pharmaceutical industry, this model may also be of interest
of government officials responsible for establishing public policies
in developing countries for the development of new drugs targeted
at diseases that would otherwise be economically unfeasible. The
model may also be useful for applications in other industries as
discussed in the introduction.
Acknowledgements
The authors wish to thank CNPq Brazil for the support for this
article, under grant no 305422/2014-6.
References
Alexander, D. R., Mo, M., & Stent, A. F. (2012). Arithmetic Brownian motion and
real options. European Journal of Operational Research, 219(1), 114–122. https:
//doi.org/10.1016/j.ejor.2011.12.023.
Berk, J. B., Green, R. C., & Naik, V. (2004). Valuation and return dynamics of new
ventures. Review of Financial Studies, 17(1), 1–35. doi:10.1093/rfs/hhg021.
Cassimon, D., Engelen, P. J., Thomassen, L., & Van Wouwe, M. (2004). The valuation
of a NDA using a 6-fold compound option. Research Policy, 33(1), 41–51. https:
//doi.org/10.1016/S0048-7333(03)00089-1.
Cassimon, D, De Backer, M, Engelen, P. J, Van Wouwe, M, & Yordanov, V (2011).
Incorporating technical risk in compound real option models to value a phar-
maceutical R&D licensing opportunity. Research Policy, 40(9), 1200–1216. http:
//dx.doi.org/10.1016/j.respol.2011.05.020.
DiMasi, J. A., Grabowski, H. G., & Hansen, R. W. (2016). Innovation in the pharma-
ceutical industry: New estimates of R&D costs. Journal of Health Economics, 47,
20–33. https://doi.org/10.1016/j.jhealeco.2016.01.012.
DiMasi, J. A., Hansen, R. W., Grabowski, H. C., & Lasagna, L. (1995). Research and
development costs for new drugs by therapeutic category. PharmacoEconomics,
7(2), 152–169. doi:10.2165/0 0 019053-199507020-0 0 0 07.
Dobbelaere, S., Luttens, R., & Peters, B. (2015). Demand lotteries, abandonment op-
tions and the decision to start R&D and process innovation. Management Revue,
26(1), 25–51. Retrieved from http://www.jstor.org/stable/24331308.
European Union (2004). Official Journal of the European Union. Directive 2004/27/Ec
of the European Parliament and of the council of 31 March 2004. L136.
732 L.E. Brandão et al. / European Journal of Operational Research 271 (2018) 720–732
Geske, R. (1979). The valuation of compound options. Journal of Financial Economics,
7(1), 63–81. https://doi.org/10.1016/0304- 405X(79)90022- 9.
Gu, L. (2016). Product market competition, R&D investment, and stock returns. Jour-
nal of Financial Economics, 119(2), 441–455. https://doi.org/10.1016/j.jfineco.2015.
09.008.
Hansen, K. R. N., & Grunow, M. (2015). Planning operations before market launch
for balancing time-to-market and risks in pharmaceutical supply chains. Inter-
national Journal of Production Economics, 161, 129–139. https://doi.org/10.1016/j.
ijpe.2014.10.010.
Hsu, J. C., & Schwartz, E. S. (2008). A model of R&D valuation and the design
of research incentives. Insurance: Mathematics and Economics, 43(3), 350–367.
doi:10.1016/j.insmatheco.20 08.05.0 03.
Lee, C. H., Rhee, B.-D., & Cheng, T. C. E. (2013). Quality uncertainty and quality-
compensation contract for supply chain coordination. European Journal of Oper-
ational Research, 228(3), 582–591. https://doi.org/10.1016/j.ejor.2013.02.027.
Leung, C. M., & Kwok, Y. K. (2016). Real options game models of R&D competition
between asymmetric firms with spillovers. Decisions in Economics and Finance,
39(2), 259–291. doi:10.1007/s10203- 016- 0176- 2.
Li, C., Watada, J., & Zhang, H. (2015). A granularity approach to compound real op-
tion in multi-stage capital investment project. Intelligent Decision Technologies,
9(4), 331–341. doi:10.3233/idt-140228.
Lo Nigro, G., Morreale, A., & Enea, G. (2014). Open innovation: A real option to
restore value to the biopharmaceutical R&D. International Journal of Production
Economics, 149, 183–193. http://dx.doi.org/10.1016/j.ijpe.2013.02.004.
Loch, C. H., & Bode-Greuel, K. (2001). Evaluating growth options as sources of value
for pharmaceutical research projects. R&D Management, 31(2), 231–248. https:
//doi.org/10.1111/1467-9310.00212.
Longstaff, F. A., & Schwartz, E. S. (2001). Valuing American options by simulation: A
simple least-squares approach. Review of Financial studies, 14(1), 113–147. doi:10.
1093/rfs/14.1.113.
Majd, S., & Pindyck, R. S. (1987). Time to build, option value, and investment
decisions. Journal of Financial Economics, 18(1), 7–27. http://dx.doi.org/10.1016/
0304- 405X(87)90059- 6.
Managi, S., Zhang, Z., & Horie, S. (2016). A real options approach to environmental
R&D project evaluation. Environmental Economics and Policy Studies, 18(3), 359–
394. doi:10.1007/s10018-016-0147-4.
Martín-Barrera, G., Zamora-Ramírez, C., & González-González, J. M. (2016). Applica-
tion of real options valuation for analysing the impact of public R&D financing
on renewable energy projects: A company s perspective. Renewable and Sustain-
able Energy Reviews, 63, 292–301. http://dx.doi.org/10.1016/j.rser.2016.05.073.
Martzoukos, S. H. (2003). Real R&D options with endogenous and exogenous learn-
ing. In D. Paxson (Ed.), Real R & D Options (pp. 111–129). Oxford: Butter-
worth-Heinemann.
Miltersen, K. R., & Schwartz, E. S. (2004). R&D investments with competitive inter-
actions. Review of Finance, 8(3), 355–401. doi:10.1007/s10679- 004- 2543- z.
Morreale, A., Robba, S., Lo Nigro, G., & Roma, P. (2017). A real options game of al-
liance timing decisions in biopharmaceutical research and development. Euro-
pean Journal of Operational Research, 261(3), 1189–1202. https://doi.org/10.1016/
j.ejor.2017.03.025.
Pennings, E., & Sereno, L. (2011). Evaluating pharmaceutical R&D under technical
and economic uncertainty. European Journal of Operational Research, 212(2), 374–
385. http://dx.doi.org/10.1016/j.ejor.2011.01.055.
PhRMA — Pharmaceutical Research and Manufacturers of America. (2016). Pharma-
ceutical Industry Profile 2016. Washington DC http://www.phrma.org.
Pindyck, R. S. (1993). Investments of Uncertain Cost. Journal of Financial Economics,
34(1), 53–76 Doi. doi:10.1016/0304-405x(93)90040-I.
Schwartz, E. S. (2004). Patents and R&D as Real Options. Economic Notes - Review of
Banking, Finance and Monetary Economics, 33(1), 23–54. doi:10.1111/j.0391-5026.
20 04.0 0124.x.
Schwartz, E. S., & Moon, M. (2001). Rational pricing of internet companies revisited.
Financial Review, 36(4), 7–26. doi:10.1111/j.1540-6288.20 01.tb0 0 027.x.
Song, M., Di Benedetto, C. A., & Nason, R. W. (2007). Capabilities and financial per-
formance: the moderating effect of strategic type. Journal of the Academy of Mar-
keting Science, 35(1), 18–34. doi:10.10 07/s11747-0 06-0 0 05-1.
Struck, M.-M. (1996). Chances and risks of developing vaccines. Vaccine, 14(14),
1301–1302. https://doi.org/10.1016/S0264-410X(96)00111-9.
Wang, J., Wang, C.-Y., & Wu, C.-Y. (2015). A real options framework for R&D planning
in technology-based firms. Journal of Engineering and Technology Management,
35, 93–114. http://dx.doi.org/10.1016/j.jengtecman.2014.12.001.