1 s2.0 S0165188908000158 Main

Journal of Economic Dynamics & Control 32 (2008) 31923217
A patent race in a real options setting: Investment

strategy, valuation, CAPM beta, and return volatility
Rujing Meng
School of Economics and Finance, University of Hong Kong, 908 K.K.L., Pokfulam Road, Hong Kong
Received 5 March 2006; accepted 14 January 2008
Available online 5 February 2008
Abstract
In this paper, we study nancial properties of R&D intensive rms through a continuous-time real-
options patent-race model. Numerical analysis in this study shows that intense competition drives a
rm to invest more aggressively, which then pushes up its cost of capital and return volatility while
introducing negative return correlation with its competitor. Furthermore, we nd that a rms
position in competition has important impacts on its nancial properties. For instance, a rms cost
of capital is a non-monotonic function of its relative position in the race. In addition, the relationship
between cash ow uncertainty and investment can be negative when a rm is far ahead or far behind,
or positive when rms are close in the race.
r 2008 Elsevier B.V. All rights reserved.
JEL Classication: C73; G12; G31; O31; O32
Keywords: Strategic real options; Patent race; R&D investments; Stochastic game
1. Introduction
This paper uses a continuous-time real-options methodology to develop a duopoly
patent-race model, which is applicable for examining nancial properties of R&D intensive
rms such as rms in the business of creating new software, inventing NANO or WI-FI
technology, or innovating new drugs. Many of these rms are early stage, private, venture
ARTICLE IN PRESS
www.elsevier.com/locate/jedc
0165-1889/$ - see front matter r 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.jedc.2008.01.001
Tel.: +852 2859 1048; fax: +852 2548 1152.

E-mail address: meng@hku.hk
capital backed startups, with limited available nancial data.
1
Therefore, investors,
nancial analysts, and other market participants often have difculty analyzing their
nancial properties. For instance, how do these rms make investment decisions? How are
they valued? How are their cost of capital and the pattern of their return volatility and
return correlation with competitors determined? This paper aims to answer these questions
and develop a model to serve as guidance for future empirical work.
R&D intensive rms face three primary types of uncertainty during the innovation
process, namely, cash ow, technological feasibility, and competitive uncertainties. As a
rm typically does not receive any cash inows until the creation/innovation of a new
product/technology is completed (like a biotechnological project), there is cash ow
uncertainty. If a rm is uncertain about its technological ability or feasibility of its project
(like a new software product), then it faces technological uncertainty. R&D intensive rms
typically operate in highly competitive environments. A patent, like any intellectual
property protection, is granted to the rst inventor. So, rms face competitive uncertainty
resulting from their changing relative position in a patent race. As a result, each rm has to
strategically interact with its competitors when making investment decisions.
We build these three uncertainties into our duopoly model in which two rms compete
to invent a new technology. Each of the two rms chooses an investment rate at which it
develops the new technology. The winner of the race is awarded a patent, from which point
it receives a sequence of cash ows, which we value by using the Capital Asset Pricing
Model (CAPM). The loser of the race receives nothing. A duopoly competition for a
patent is modeled as a stochastic game in which there are three publicly observable state
variables: the value of the patent and the expected cost-to-completion of each of the two
rms. Each state variable is governed by its own source of risk. The value of the patent is
the present value of the cash ows received upon winning, which follow a geometric
Brownian motion subject to both systematic and idiosyncratic risks. The overall
uncertainty underlying the patent value is referred to as cash ow uncertainty, which
gives the rms real options to delay their investment and wait for more information about
the protability of the patent. The position of each rm in the race is described by a state
variable called the expected cost-to-completion, which measures the expected amount of
money a rm needs to succeed in developing the new technology and winning the patent
race (the follow-on capital). Each rms expected cost-to-completion follows a diffusion
process governed by its own idiosyncratic technological risk, which is a special case of the
technology described by Pindyck (1993).
2
Given a rms individual technological
uncertainty, it has an incentive to learn more about the difculty of the project and to
invest in order to resolve this uncertainty (learning by doing). A rms investments may
lead to technological improvements as well as setbacks. Faced with competitive
uncertainty, a rm receives either preemptive threats to invest or incentives to withhold
investments depending on the relative position of this rm to its competitor. In an
environment with multiple sources of uncertainty, when making strategic investment
decisions, a rm has to strike a balance among the real options to wait, the incentives to
invest or withhold investments, and the preemptive threats to invest.
ARTICLE IN PRESS
1
For startups backed by venture capitals, VentureXpert provided by Venture Economics and VentureSource
provided by VentureOne are good data sources.
2
The general process of the expected cost-to-completion proposed by Pindyck (1993) includes technological risk
as well as input cost risk. We ignore the input cost risk for simplicity.
R. Meng / Journal of Economic Dynamics & Control 32 (2008) 31923217 3193
We employ the full information Markov Perfect equilibrium in our model. Each rm
chooses an investment rate constrained by its money burn rate to maximize its rm
value given the other rms investment rate. Because there is no closed-form solution to
our continuous-time model, we solve it by using numerical approximations. Using a lattice
method, we develop a discrete-time implementation on a 201 61 61 grid with 600
monthly decisions, upon which we examine the investment strategy, valuation, CAPM
beta, return volatility, and return correlation of the two rms in the race. We report these
results for various relative positions of the rms throughout the race.
Through numerical analyses, we nd that competition drives the rms to over-invest
relative to a case of joint monopoly, which then pushes up the rms CAPM beta and
return volatility, and introduces negative return correlation between the two competitors.
With regard to a rms CAPM beta, and thus its cost of capital, our numerical analyses
depict that it depends on the rms relative position to its competitor in a non-monotonic
way. A rms beta is closer to the beta of the underlying patent when it gets closer to
winning the race because its investment option becomes deep-in-the-money and the
leverage induced by investment expenditures declines. The rms beta becomes higher
when its position is closer to its competitor, because intense competition drives the two
rms more eager to invest which increases their leverage. The rms beta becomes low
again when its investment option becomes deep-out-of-the-money as it lags far behind and
stops investing.
The numerical analyses also show that annual return volatility of R&D intensive rms
can be in excess of 100%, which is quite high compared to the range of 2040% per annum
for a typical rm.
3
Further decomposition in our numerical example reveals that this high
level of return volatility is largely attributable to the rms idiosyncratic technological
risks. These results are consistent with recent empirical ndings on venture capital-backed
startups by Cochrane (2005).
Another interesting result is that competition introduces negative return correlation.
When both rms are not investing, their returns are perfectly positively correlated by the
common factor of patent value. When one of the rms or both are investing aggressively as
competition intensies, a progress of one rm means a setback for the competitor. As a
result, this rm enjoys positive returns whereas the competitor suffers negative returns.
The return correlation between the two rms thus turns negative.
In addition, we conduct a comparative-static analysis of such parameters as cash ow
uncertainty, technological uncertainty, and dividend yield to study how the leader and the
follower react differently to the change in these key underlying parameters. One interesting
nding is that the relationship between cash ow uncertainty and investment can be
either negative or positive; it is negative when a rm is far ahead or far behind and positive
when rms are close in the race. This nding contributes to the latest literature on
strategic real options by showing the impact of a rms position in competition on the
relationship between uncertainty and investment.
4
In all, our sensitivity analysis shows
ARTICLE IN PRESS
3
See Hull (2000, p. 241).
4
Kulatilaka and Perotti (1998), Kogan (2001), and Kulatilaka and Lin (2007) show that the relationship
between uncertainty and investment is ambiguous when there are strategic interactions between rms. Earlier
literature nds this relationship purely positive or purely negative. Hartman (1972) and Abel (1983, 1984, 1985)
nd a positive relationship between uncertainty and investment by predicting that the marginal revenue product
of capital is convex in some random variable representing a certain aspect of the rms environment. Pindyck
(1988), Bertola (1988), and Caballero and Pindyck (1996), however, nd a negative relationship by predicting a
that a rms relative position to its competitor is an important control variable for
studying the investment strategies of R&D rms, which has been ignored in previous
empirical studies. We provide a discussion of empirical proxy for this control variable for
future studies.
This paper unies two lines of research, the literature of patent race in economics and
that of investment under uncertainty using standard real-options techniques, and
contributes to the latest literature on strategic real options by showing how a rms
position in competition affects its nancial properties. Along the literature of patent race,
technological competition has been investigated rst through stationary games under
uncertainty by Loury (1979), Dasgupta and Stiglitz (1980), and Lee and Wilde (1980), and
then through dynamic games under uncertainty but without explicit strategic interactions
pioneered by Reinganum (1981, 1982), or through dynamic games with strategic
interactions but without uncertainty pioneered by Fudenberg et al. (1983) and Harris
and Vickers (1985). Finally, strategic interactions and technological uncertainties are
combined within a dynamic structure, as in Judd (2003), Grossman and Shapiro (1987),
and Harris and Vickers (1987). Models in this line of research all lack the cash ow
uncertainty since the patent value is set as a constant, in which case, a real-options
methodology has not been applied. The value of the option to delay investments embedded
in cash ow uncertainty has been ignored. As a result, the investment strategies derived in
those papers lack the delay feature. On the other hand, the traditional real-options
literature, represented by McDonald and Siegel (1986), Pindyck (1988), and Dixit (1989),
highlights cash ow uncertainty.
5
Therefore, the real-options premium of delaying
investments, which has been ignored in the literature on the patent race, is the main nding
of this category of research. However, those papers have typically ignored strategic
interactions between rms and regard rms as either monopolists or entities in a perfectly
competitive market.
6
Our continuous-time model combines real options with strategic interactions, which
contributes to a recently growing body of literature on strategic real options represented by
Kulatilaka and Perotti (1998), Grenadier (2002), and Weeds (2002).
7
Most existing models
assume risk-neutrality precluding the analysis of CAPM beta and return volatility of rms
evolved in competition. In addition, technological uncertainty has also often been
ignored. Childs and Triantis (1999) examine dynamic R&D investment policy and
valuation for a rm with multiple R&D projects, which can be run in parallel or in
sequential to each other. They provide a thorough analysis of the interactions across
different projects. In our model, both rms are assumed to have a single project. Novy-
Marx (2007) investigates optimal investment decisions of heterogeneous rms in a
ARTICLE IN PRESS
(footnote continued)
concave marginal revenue product of capital in models with irreversible investments. Craine (1988) also points to
a negative relationship in a version of the Capital Asset Pricing Model. The empirical work that tests this
relationship is limited. One exception is Leahy and Whited (1996) which measures uncertainty by the volatility
over stock returns and favors the theory with irreversible investments and nds that investment is negatively
correlated with uncertainty through Tobins q. Leahy and Whited do not control for a rms relative position to
its competitors. Further empirical research is worth conducting.
5
See Dixit and Pindyck (1994) and Trigeorgis (1996) for reviews on this subject.
6
Schwartz and Moon (2000), Schwartz (2004), and Berk et al. (2004) consider the possibility that an
exogenously catastrophic event may put an end to the rm.
7
See Grenadier (2000) for a survey on this literature.
competitive, uncertain environment and shows that the strategic equilibrium real-options
premia are signicant. Grenadier (1999), Lambrecht (2000), and Lambrecht and Perraudin
(2003) study option exercise games under incomplete information. In our model, we
assume that the two rms have complete information for simplicity.
Our continuous-time model is also related to parallel works by Miltersen and Schwartz
(2004) and Garlappi (2004). Miltersen and Schwartz (2004) studies R&D spending
with competitive effects and multiple sources of uncertainty from a welfare perspective.
Firms compete in both innovation and production stages, while the innovation stage is
emphasized in our paper. Miltersen and Schwartz illustrate an example in which rms
tie at the start of competition. In comparison, we provide results for various relative
positions of the rms throughout the race, which allows us to analyze how relative
positions of rms in competition affects their nancial properties. Garlappi (2004)
studies the dynamics of rms risk premia in a duopoly patent race. In his model,
rms jump over a series of hurdles before winning the patent race. He provides analytical
solutions to a two-stage game and numerical analysis for a ve-stage game. Our
numerical approximation is a 61-stage game. This more general scenario allows us to
analyze rms costs of capital when a rm is well ahead or behind. We nd a rms CAPM
beta is a non-monotonic function of its position relative to its competitor in contrast to
monotonically increasing function found in Garlappis model. We also introduce
technological setbacks, which is omitted in his model. This enables us to study what
happens to the cost of capital when a rm realizes more time and more money than
previously expected have to be spent to complete the project. We nd that a rms setbacks
increase its own cost of capital and decrease its competitors cost of capital when
competition is erce.
The paper is organized as follows. Section 2 describes the model set-up. Section 3
describes the numerical implementation. Section 4 discusses the equilibrium investment
strategy, valuation, CAPM beta, return volatility, and return correlation of rms involved
in a patent race. Section 5 provides a comparative-static analysis on parameters of
cash ow uncertainty, dividend yield of the patent value, and technological uncertainty.
Section 6 concludes with a discussion of further research.
2. The model
In our model, we create a game between two all equity-nanced rms, indexed by
A and B, pursuing a single project.
8
At time t 0, the two rms enter a race against each
other to complete an invention. The rst rm to complete an invention (hereafter, the
winner of the patent race) will be awarded a patent. During the course of the patent
race, each rm makes continuous-time sequential investment decisions until a winner
emerges. A duopoly competition is modeled as a game in which there are three publicly
observable state variables: the value of the patent, and the expected cost-to-completion of
each of the two rms. The value of the patent is the present value of a series of cash ows
received upon winning the race. Each rms expected cost-to-completion is the expected
investment costs to complete the invention. These state variables are described in details as
follows.
ARTICLE IN PRESS
8
The model and solution can be easily extended to an oligopolistic case involving a nite number of rms.
2.1. Patent value and cash ow uncertainty
A rm receives no positive cash ows until it becomes the winner of the patent race, i.e.,
the rst to complete the project and receive a patent. If the winning rm obtains the patent
at time t, it immediately begins to receive an innitely-lived stochastic stream of cash ows,
denoted D(t), valued using the CAPM. The value of the winning rm is the value of this
stream of cash ows. The losing rm receives nothing and as such has no value. The
stochastic process D(t) is assumed to follow the geometric Brownian motion. Let m denotes
drift, b denotes CAPM beta, s
m
denotes market volatility, s
I
denotes idiosyncratic
volatility, s denotes dividend volatility, and white noise processes dz
m
, dz
I
, and dw denote
systematic, idiosyncratic, and total nancial risks, respectively. Then we can formulate
cash ows, D(t), as follows
9
:
dD
D
mdt bs
m
dz
m
s
I
dz
I
mdt sdw. (1)
Here, dividend volatility, s
b
2
s
2
m
s
2
I
_
, is referred to as cash ow uncertainty. It
contains both systematic and idiosyncratic components. Let l
m
denotes the risk premium
on the market portfolio, r
f
denotes the risk-free rate, and P
m
(t) denotes the value of a share
in the market portfolio with dividends reinvested. Then the stochastic process P
m
(t) is
assumed to follow the geometric Brownian motion:
dP
m
P
m
r
f
l
m
dt s
m
dz
m
. (2)
Using Gordons growth formula, the present value of the stream of the patents cash ows,
denoted by x(t), is given by x(t) (D(t)/r
f
+bl
m
m)). The patent value x(t) is one of the
state variables affecting the value of both rms. From Eq. (1) and Gordons growth
formula, we can derive the stochastic process of the patent value x(t), which also follows
the geometric Brownian motion with the same drift and diffusion parameters as in the
process of cash ows:
dx
x
mdt bs
m
dz
m
s
I
dz
I
mdt sdw. (3)
Let d denotes dividend yield, then the denominator in Gordons Growth Formula is
given by
d
Dt
xt
r
f
bl
m
m. (4)
Intuitively, the dividend yield d represents the rate at which the present value of the patent
decays through time as a result of no patent being granted; it represents the opportunity
cost of deferring victory. Intuitively, if the dividend yield d is smaller (holding the patent
value x(t) constant), then the option to delay is more valuable and, accordingly, the two
rms will have an incentive to pursue the patent less aggressively. We thus assume d40 to
avoid innite patent value and to eliminate the scenario whereby rms wait forever for a
more favorable patent value.
ARTICLE IN PRESS
9
The white noise process of total nancial risk dw is dened as dw bs
m
=s dz
m
s
I
=s dz
I
.
2.2. Investment rates and technological uncertainty
The position of the rms in the patent race is represented by two state variables, k
A
(t)
and k
B
(t), denoting the two rms expected cost-to-completion, respectively. We assume
both k
A
(t)40 and k
B
(t)40 at the start of the race. Let I
A
(t) and I
B
(t) denote the two rms
respective investment rates. A positive investment rate I
A
(t) tends to reduce k
A
(t), and
similarly a positive investment rate I
B
(t) tends to reduce k
B
(t). A rm wins the patent race
when its expected cost-to-completion falls to zero before the other rm does.
A special case of the random technology as described by Pindyck (1993) is used. The
expected cost-to-completion k
A
(t) and k
B
(t) are assumed to follow a diffusion process of
the following form, where y
A
and y
B
are cost technological volatility parameters, which are
referred to technological uncertainty of rms A and B, respectively:
10
dk
A
I
A
dt y
A
I
A
k
A
1=2
dz
A
, (5)
dk
B
I
B
dt y
B
I
B
k
B
1=2
dz
B
. (6)
Here, dz
A
and dz
B
denote idiosyncratic white noise processes, distributed independently
from the nancial risks dz
m
and dz
I
. Our CAPM assumption implies that any risk
associated with the expected cost-to-completion is diversiable, i.e., it is priced in a risk-
neutral manner. This specication is numerically tractable and has several attractive
features as follows.
11
First, the drift terms I
A
dt and I
B
dt indicate that a dollar of
investment reduces the expected cost-to-completion by a dollar. Second, the diffusion
terms y
A
(I
A
k
A
)
1/2
dz
A
and y
B
(I
B
k
B
)
1/2
dz
B
indicate that the expected cost-to-completion
changes randomly, with many small unexpected improvements (reductions in k
A
(t) and
k
B
(t)) as well as many small unexpected setbacks (increases in k
A
(t) and k
B
(t)).
Technological setbacks mean that a rm realizes that more money must be spent or
more time is needed to complete the project than previously believed.
12
Third, there is a
learning by doing feature which is two-fold. First, a rms expected cost-to-completion
does not change if the rm does not invest, i.e., dk
i
0 if I
i
0, i A, B. Second,
investing not only tends to improve the rms chances of victory by reducing the expected
cost-to-completion but also tends to reduce uncertainty about how much investment will
eventually be required to drive k
A
(t) or k
B
(t) to zero. Let I
A
max
and I
B
max
denote the
ARTICLE IN PRESS
10
There are several ways to model technological uncertainty. Loury (1979), Dasgupta and Stiglitz (1980), and
Lee and Wilde (1980) model technological uncertainty by using a Poisson process with a constant arrival
probability, which implies that the success probability is independent of current and past R&D efforts. Our
specication implies that the success probability depends on both a rms current R&D effort through the
investment rate and past R&D efforts through the level of its expected cost-to-completion. Doraszelski (2003)
considers knowledge accumulation and models the success probability as an addictive function of current and past
R&D efforts. Lukach et al. (2007) model the random outcome of the costs of R&D by introducing a simple two-
stage R&D process with an uncertain outcome in the rst stage.
11
This specication is very similar to those used in Schwartz and Moon (2000), Schwartz (2004), and Miltersen
and Schwartz (2004). The assumption of an exogenous expected cost may cause potential dynamic inconsistency
in the sense of rational expectation. We haveensured, however, that the algorithm converges and the results are
robust to various discrete approximations and a wide spectrum of parameter values. So, any potential dynamic
inconsistencies should not affect the results of the paper.
12
This can be justied by business news. For example, the 64-bit Intel super chip Itanium was initially to be
released in 1997. However, a long series of delays pushed the chips release date all the way into the rst half of
2001.
respective xed maximum investment rates of rms A and B, reecting their research
bandwidths that is the maximum amount of capital a rm can effectively deploy per unit of
time. We assume rms have the same magnitude of research bandwidth, i.e.,
I
A
max
I
B
max
I
max
.
13
Firms A and B choose I
A
(t) and I
B
(t), respectively, from the
closed-interval [0, I
max
].
As the race progresses, a rm could become the follower, whose expected cost-to-
completion is higher than that of the other rm (k
i
4k
j
; i A, B; j B, A), or the leader,
whose expected cost-to-completion is lower than that of the other rm (k
i
ok
j
; i A, B; j B,
A). The rms could also tie in the race (k
A
k
B
), which is referred to as head-to-head.
2.3. Firm value
The state of competition between the two rms is described by the three variables: the
patent value x(t), the expected cost-to-completion of rm A, k
A
(t), and the expected cost-
to-completion of rm B, k
B
(t). We assume that both rms observe all the three state
variables at all times. Firm value of each rm is a function of the three state variables. Let
F
A
(x, k
A
, k
B
) and F
B
(x, k
A
, k
B
) denote the respective rm value of rms A and B.
2.4. Equilibrium
We employ the solution concept of Markov Perfect Equilibrium (see Maskin and Tirole,
1988).
14
For a given date t and for every state (x, k
A
, k
B
), there is a sub-game starting from
that state. At each sub-game, the information set of either rm is a set of realizations of the
state variables x, k
A
, and k
B
at time t. There are countless paths leading to this set of
realizations. However, this set of realizations is the only payoff-relevant information at
time T. The action set available to each rm includes any investment rate level in the range
of [0, I
max
]. In equilibrium, a Markovian strategy maps the information set to the action set
for each rm.
Each rms investment strategy is a function of the three state variables, I
A
(x, k
A
, k
B
) and
I
B
(x,k
A
,k
B
). In the sub-game (x, k
A
, k
B
), let the investment strategy prole {I
AN
(x, k
A
, k
B
),
I
BN
(x, k
A
, k
B
)} represent a Nash Equilibrium in which a rm maximizes its rm value by
optimally choosing an investment rate given the other rms optimal investment strategy.
Since the mathematical derivation is symmetric for the two rms, we show only the derivation
of rm As strategy. Given rm Bs optimal investment strategy I
BN
(x, k
A
, k
B
) at sub-game
(x, K
A
, k
B
), I
A
(x, K
A
, k
B
) I
AN
(x, k
A
, k
B
) solves rm As value function:
F
A
x; k
A
; k
B
Max
I
A
20;I
max
fI
A
dt E
fe
r
f
dt
F
A
x dx; k
A
dk
A
; k
B
dk
B
gg,
s:t:
dx m
xdt sxdw
;
dk
A
I
A
dt y
A
I
A
k
A
1=2
dz
A
;
dk
B
I
BN
dt y
B
I
BN
k
B
1=2
dz
B
;
8t
_
_
ARTICLE IN PRESS
13
Kyle and Meng (2005) relax this assumption and study how a rm with larger research bandwidth makes
strategic acquisition and investment in a rm with smaller bandwidth, subject to transactions costs.
14
The Markovian strategies imply that rms cannot engage in history-dependent punishment strategies, which
would support a more collusive equilibrium. It tends to support an equilibrium in which there is more rent
dissipation than under alternative equilibrium concepts where rms behave more co-operatively.
The expectation operator notation E
*
{?} refers to the expectation with respect to the risk-
neutral probabilities. The process dw
*
is the white noise process of the total nancial risk
under the risk-neutral probability measure. Its relationship with the counterparty under the
true probability measure is dw
dw bl
m
=s dt according to the CameronMartin
Girsanov theorem. The risk-neutral growth rate on the patents cash ows m
*
is obtained by
deating the true growth rate m by the risk premium on the patents cash ows bl
m
, so that
Gordons growth formula gives the same value for winning the patent in the CAPM model as
in a risk-neutral model, i.e.
xt
Dt
r
f
bl
m
m

Dt
r
f
m
. (7)
Note that the dividend yield d remains the same and is given by
d r
f
bl
m
m r
f
m
. (8)
The intuition behind this result is that the investors can hedge the market risk associated with
investments in the two rms by trading the market portfolio; this leaves them with
idiosyncratic risk, with respect to which they are risk-neutral. Therefore, the value of the rm
can be evaluated without knowing investors risk preference or the true growth rate on the
patents cash ows. This point is evident from the derivation of Firm As Bellman equation as
detailed in Appendix A.
Intuitively, the value of a rm is the present value of its risk-adjusted cash ows,
discounted at the risk-free rate. The positive cash ows are the cash ows on the patent,
received if and when the rm wins. If a rm loses the race, the rm receives zero. The
negative cash ows are the expected cost-to-completion, which we can think of as new
equity being injected into the rms. A rms expected cost-to-completion k
i
can be
interpreted as the expected amount of follow-on capital that it needs to be successful.
The maximum investment rate I
max
can be interpreted as the maximum burn rate. As the
race progresses, a rms value may increase because the new capital invested in the rm
allows it to improve its chances of winning the patent race. Even under the risk-neutral
probabilities, the value of a rm increases faster than the risk-free rate because new capital
(like negative dividends) is being injected into the rm when it makes further investments
in the technology it is developing.
In the continuous-time framework, the Bellman Equation of rm A given rm Bs
optimal investment strategy I
BN
is
Max
I
A
2 0;I
max

1
2
s
2
x
2
F
A
xx
r
f
dxF
A
x
_ _
I
A 1
2
y
A
2
k
A
F
A
k
A
k
A
F
A
k
A
1
_ _ _ _
I
BN 1
2
y
B
2
k
B
F
A
k
B
k
B
F
A
k
B
_ _
r
f
F
A
_
0 (9)
where rm As value with a subscript denotes partial derivative of rm As value with
respect to the variable in the subscript. For instance, F
A
x
and F
A
xx
denote the respective
rst-order and second-order partial derivatives of rm As value F
A
with respective to the
patent value x. The derivation of rm As Bellman Equation is provided in Appendix A.
As indicated by Eq. (9), the Bellman Equation is nicely grouped with brackets into four
parts as shown above. The rst part corresponds to cash ow uncertainty, the second part
corresponds to rm As own technological uncertainty, and the third part corresponds to
the competitor rm Bs technological uncertainty. Even though the two rms
ARTICLE IN PRESS
technological uncertainties are independent, each rm must consider its competitors
technological uncertainty in order to make a strategic decision on investments. The fourth
part represents opportunity costs.
Because of the square root term in the Eqs. (5) and (6) of the two rms expected cost-to-
completion and an assumption of no adjustment costs on capital for numerical tractability,
the Bellman Equation of each rm is linear in its investment strategy. The linearity of the
Bellman equation implies that in the continuous-time limit, the investment strategy of rm
A depends on the coefcient of I
A
in the Bellman equation of rm A, and, similarly, the
investment strategy of rm B depends on the coefcient of I
B
in the Bellman equation of
rm B. If the coefcient is positive, then the rms dominant strategy is to invest at the
maximum rate I
Max
. Conversely, if the coefcient is negative, then the rms dominant
strategy is to choose a zero investment rate. If the coefcient is zero, the rm is indifferent
as to its investment rate. In the continuous-time limit, a zero coefcient is a zero-
probability event. Hence, a rm invests at the maximum rate or does not invest at all. We
assume that the rms make the binary choice in the numerical implementation.
In the numerical implementation, each rm plays out its dominant strategy if it has
one.
15
Otherwise, there are multiple equilibria or equilibra with mixed strategies. When
there are multiple equilibria in which only one rm should invest and both rms are
indifferent as to who invests, we let the rm with more to gain from the investment invest.
When there are multiple equilibria in which both rms are indifferent between the option
to invest or mothball, we assume that both rms would invest. Our results remain robust
when these assumptions are relaxed. When the equilibrium involves mixed strategies, we
assume that the rms play mixed strategies although this is not reected in the above
notation. Except for a small region of mixed strategies, the strategies of the rms can be
described by four sets of points, i.e., both rms invest, only rm A invests, only rm B
invests, and neither rm invests.
To complete a description of the equilibrium, it is also necessary to discuss the scenario
in which two rms become separate divisions of a joint monopoly where the research
bandwidths of the two rms remain uncombined. Let F
JM
(x, k
A
, k
B
) denote the value of
the joint monopoly, the Bellman equation has the joint monopoly maximize joint value by
choosing both investment rates:
Max
I
A
20;I
max
;I
B
20;I
max
1
2
s
2
x
2
F
JM
xx
r
f
dxF
JM
x
_ _
I
A 1
2
y
A
2
_ _
k
A
F
JM
k
A
k
A
F
JM
k
A
1
_
I
B 1
2
y
B
2
k
B
F
JM
k
B
k
B
F
JM
k
B
1
_ _
r
f
F
JM
_
0.
3. Numerical implementation
There is no a closed-form solution for this game, so we solve the model numerically by
using both a brute force lattice method and an explicit nite difference method. The
properties of the model are robust to both methods. Here we only provide a detailed
discussion of the lattice method and related properties for brevity.
ARTICLE IN PRESS
15
The scenario that both rms have dominant strategies happens at more than 99% of the decision nodes in the
numerical implementation.
The assumed values of the exogenous parameters in the base model are as follows:
d 3%, r
f
5%, b 1.00, s
m
20%, s 30%, I
max
10.00, y
A
y
B
0.60. These
values imply m
*
r
f
d 2%, s
I
s
2
b
2
s
2
m
1=2
22%.
16
Time is discretized with
decisions made at N+1 dates 0 t
0
, t
1
, y, t
N
T, so that Dt t
n+1
t
n
1 month. The
horizon T is chosen as 50 years.
17
Thus, each rm has up to N 600 monthly decisions in
the numerical example.
For the purpose of numerical implementation, we also discretize the space of the three
state variables. Instead of discretizing three diffusion state variables themselves, we
discretize three transformed variables: ln(x(t)), (k
A
(t))
1/2
, and (k
B
(t))
1/2
. Note that the
dynamics of the transformed variables are given by Itos lemma as follows:
dlnx r
f
d
1
2
s
2
_ _
dt sdw
, (10)
dk
A
1=2

1
2
I
A
1
1
4
y
A
2
_ _
k
A
1=2
dt
1
2
y
A
I
A
1=2
dz
A
, (11)
dk
B
1=2

1
2
I
B
1
1
4
y
B
2
_ _
k
B
1=2
dt
1
2
y
B
I
B
1=2
dz
B
. (12)
The chosen transformations have the property that the instantaneous variance in the
transformed variables is constant. For the value of the patent x(t), we choose 201 points,
equally spaced in the transformed variable ln(x), such that the smallest point corresponds
to x 1 and the largest point corresponds to x 10,000. For each of the expected cost-to-
completion variables k
A
and k
B
, we choose 61 points, equally spaced in the transformed
variables (k
A
)
1/2
and (k
B
)
1/2
, such that the smallest value corresponds to k
A
k
B
0 and
the largest value corresponds to k
A
k
B
225.
The game is solved by using the brute force lattice method of backward induction
from the terminal date t
N
50 years, corresponding to N 600. We use an Euler
approximation for all three transformed state variables. We approximate the normally
distributed transition probabilities in the Euler approximation by mapping the true
outcome to the closest transformed outcome in the space of transformed state variables
within two standard deviations (rounded up) of the expected outcome. When an expected
cost-to-completion variable hits its minimum value, we assume that the respective rm
wins the patent at that point. If both rms win the patent at the same time, we split the
patent value equally between the two rms. When the state variables otherwise hit their
maximum or minimum values, we do not assume that they are absorbed; random
uctuations are allowed to push them back into the interior of the space of state variables
at a later date.
Since the strategy space is three-dimensional, the equilibrium would have been depicted
in four-dimensional graphs. It is too complex. So, we choose a patent value of x 100
(e.g., $100 million), the midpoint of the grid of the transformed variable ln(x) and illustrate
investment strategies and nancial characteristics of the rms in contour plots in the next
ARTICLE IN PRESS
16
The results are robust to a wide spectrum of parameter values.
17
Regarding how long it takes for the patent race to be over, that is entirely endogenous in the model depending
on the rms investment decisions and realizations of the three random state variables. Our continuous-time
model allows the rms to compete forever if no one wins the race. In the numerical implementation, a nite
horizon T is chosen long enough to approximate the potential innite horizon in the continuous-time model as
one of the boundary conditions. If neither rm has won by time T, then both rms get zero and the patent race is
forced to terminate in the numerical approximation. The results in the paper are robust to different lengths of
horizon and different number of decision nodes.
section. A contour plot in this paper displays isolines/level curves of a certain nancial
property with respective to the plane of the two rms expected cost-to-completion given
the patent value x 100. So, the contour plots illustrate the nancial properties for
various relative positions of the two rms in the race.
4. Financial properties of equilibrium
This section examines nancial properties of the equilibrium revealed in the numerical
example in terms of investment strategy, valuation, CAPM beta, return volatility, and
return correlation of the two rms.
4.1. Over-investment and rent dissipation
Competition makes the rms more eager to invest and as a result they may over-invest in
the patent race relative to the scenario of joint monopoly. The over-investment induced by
competition erodes the aggregate value of the rms.
4.1.1. Over-investment
The result of over-investment is evident from comparison of Panels A and B in Fig. 1.
Panel A shows investment behaviors of the two rms in the benchmark case of joint
monopoly, which implements a collusive outcome. Panel B shows the investment behaviors
under the patent race. In the gures, the value of the patent is xed at x 100, the horizontal
axis is the transformed expected cost-to-completion of rm A, C
A
k
A
_
and the vertical
axis is the transformed expected cost-to-completion of rm B, C
B
k
B
_
. The 451 line is
called equal-cost-to-completion line where the rms are head-to-head in a tied race. The
area with | is where rm A invests and rm B does not; the area with is where rm B
invests and rm A does not; the area with + is where both rms invest; and the un-shaded
white area is where neither rm invests. In the case of joint monopoly, as shown in Panel A,
the rm which invests is typically the rm with the lower expected cost-to-completion. The
delay feature of a real option is also well illustrated in Panel A; the point where each rms
expected cost-to-completion is equal to the value of the patent C
A
C
B
x
p

100
p

10 is well inside the area where neither rm invests. The real option value of waiting induces
the joint monopolist to delay investment even when a nave NPV rule would suggest that the
project has a positive NPV. In comparison, in Panel B, there is an extensive area with +
around the equal-cost-to-completion line (451 line) where both rms invest inefciently
from the perspective of a joint monopolist. Furthermore, the northeast boundary of this
inefcient investment area in Panel B reaches a point where each rms expected cost-to-
completion is approximately equal to the value of the patent C
A
C
B
x
p

100
p
10
indicating that competition dissipates the real option value of waiting.
To illustrate the impact of investment decisions on the rms nancial characteristics, we
will plot the contour of the two rms investment regions as dotted-curves in Figs. 26
when discussing valuation, CAPM beta, return volatility, and return correlation.
4.1.2. Rent dissipation
To examine the degree of value dissipation, we plot level curves for the percentage value
dissipation as a function of the two rms expected costs-to-completion in Fig. 2. The
ARTICLE IN PRESS
ARTICLE IN PRESS
Fig. 1. Investment strategies of rms A and B under joint monopoly (Panel A) and under patent race (Panel B)
given the patent value x(t) 100. The horizontal axis is the square root of rm As expected cost-to-completion.
The vertical axis is the square root of rm Bs expected cost-to-completion. The area with | is where rm A
invests and rm B does not; the area with is where rm B invests and rm A does not; the area with + is
where both rms invest; and the un-shaded white area is where neither rm invests.
Fig. 2. Level curves for the percentage value dissipation given the patent value x(t) 100. Percentage value
dissipation is measured as the percentage difference between the combined values of the two rms under patent
race and joint monopoly. The horizontal axis is the square root of rm As expected cost-to-completion. The
vertical axis is the square root of rm Bs expected cost-to-completion. The dotted-curves are the contour plot of
rms investment regions under patent race from Panel B of Fig. 1.
ARTICLE IN PRESS
Fig. 3. Level curves for CAPM beta of rm A under patent race given the patent value x(t) 100. The horizontal
axis is the square root of rm As expected cost-to-completion. The vertical axis is the square root of rm Bs
expected cost-to-completion. The dotted-curves are the contour plot of the rms investment regions under patent
race from Panel B of Fig. 1.
Fig. 4. Level curves for portfolio CAPM beta under joint monopoly (Panel A) and under patent race (Panel B)
given the patent value x(t) 100. The horizontal axis is the square root of rm As expected cost-to-completion.
The vertical axis is the square root of rm Bs expected cost-to-completion. The dotted-curves in Panel A are the
contour plot of the rms investment regions under joint monopoly from Panel A of Fig. 1 and the dotted-curves
in Panel B are the contour plot of the rms investment regions under patent race from Panel B of Fig. 1.
ARTICLE IN PRESS
Fig. 5. Level curves for annualized return volatility (Panel A) and fraction of return volatility caused by its
idiosyncratic risks (Panel B) of rm A under patent race given the patent value x(t) 100. The horizontal axis is
the square root of rm As expected cost-to-completion. The vertical axis is the square root of rm Bs expected
cost-to-completion. The dotted-curves are the contour plot of the rms investment regions under patent race
from Panel B of Fig. 1.
Fig. 6. Level curves for return correlation between the rms A and B under patent race given patent value
x(t) 100. The horizontal axis is the square root of rm As expected cost-to-completion. The vertical axis is the
square root of rm Bs expected cost-to-completion. The dotted-curves are the contour plot of the rms
investment regions under patent race from Panel B of Fig. 1.
expected rent dissipation is measured as the percentage difference between the combined
values of the two rms in the duopoly patent race and in the joint monopoly. The graph
shows that rent dissipation is greater than 2% around the equal cost-to-completion line. In
this region, it is not efcient from the perspective of a joint monopolist for the rm with
higher expected cost-to-completion to invest, but the rm nevertheless invests because it is
afraid to lose the patent. The maximum amount of rent dissipation can be in excess of
35%, which is economically signicant.
4.1.3. Implications of over-investment and value dissipation
The results of over-investment and value dissipation have implications for the venture
capital industry. A venture capitalist can invest in several start-up rms called portfolio
rms, over which it typically has substantial control. Therefore, portfolio rms are in some
ways like rms in a joint monopoly. This model implies that a venture capitalist should
encourage its portfolio rms to coordinate more. In addition, if a venture capitalists
portfolio includes the leader, it is valuable to also include the follower into the portfolio
which cannot only potentially improve efciency by reducing over-investment and value
dissipation, but can also earn a good hedge plan since the returns of the leader and the
follower could be negatively correlated as analyzed later in this paper.
4.2. CAPM beta, return volatility, and return correlation
In this paper, return refers to the return on rm value over interval dt, that is, the
percentage change in the rm value over interval dt, taking into account any dividends
distributed over that interval, i.e., a rms investments can be regarded as negative
dividends. Since the rms are all-equity nanced, the return on rm value therefore equals
the return on equity. The derivation of formulas of CAPM beta, return volatility and
return covariance of the two rms is in Appendix B.
4.2.1. Properties of CAPM beta in a duopoly competition
In the continuous-time limit, a rms CAPM beta is the product of the patent beta b and
the elasticity of the rms value with respect to the patent value as denoted by e
A
for rm A
and e
B
for rm B. Take rm A as an example:
b
A
b
A
b
x
F
A
F
A
x
_ _
b
x
F
A
_ _
D
A
. (13)
As each rms value can be regarded as the value of an investment option on the patent,
the elasticity can be further decomposed into the product of a ratio of patent value to rm
value x/F
A
and the option delta D
A
. If we x the patent beta to be b 1.00 for simplicity
purposes, then the magnitude of rm beta will depend on that of the value ratio and the
option delta.
Fig. 3 shows level curves for CAPM beta of rm A in the patent race. The dotted-curves
are the contour plot of the rms investment regions under patent race from Panel B of
Fig. 1. Fig. 3 makes it clear that a rms beta is a complicated non-monotonic function of
its relative position in the race. When rm A is close to victory (in the points close to the
vertical axis), it is about to win and receive the patent and thus the investment option is
deep-in-the-money. Both the ratio of patent value to rm value and option delta are
consequently close to 1.00. As a result, rm As beta is close to the patent beta that is set to
ARTICLE IN PRESS
be 1.00. Fig. 3 shows that rm As beta is always slightly above 1.00 because of the real
options value to delay.
Firm As beta increases when it drops back from the point of victory (points move east in
Fig. 3) or competition with its competitor becomes intense (in the common investment area in
Fig. 3). The beta can be in excess of 2.00 as read from the gure. When rm A drops back
from the point of victory, its expected cost-to-completion increases, which in turn decreases
its rm value. In addition, the intensive competition drives both rms more eager to invest as
patent value rises, which in turn dissipates rm values. As a result, the ratio of patent value to
rm value becomes larger and the delta becomes smaller. Firm As beta is higher or lower
depending on the relative effect of the value ratio and the delta as shown by the above
Eq. (13) given the patent beta is set to be 1.00. Our numerical results in Fig. 3 suggest that the
effect of the value ratio dominates the effect of the delta and, hence, rm As beta increases
when it experiences setbacks after having been ahead. Similarly, rm As beta increases as
rm B makes improvements (points move south in the common investment area in Fig. 3).
As rm A continues to experience setbacks and nally stops investing (in the region to
the right of the common investment area in Fig. 3), its rm beta declines with its expected
cost-to-completion. When rm A lags far behind, the ratio of patent value to rm value
increases dramatically as rm As value drops signicantly. Meanwhile, the investment
option becomes deep-out-of-the-money, and thus the delta approaches zero. The size of a
rms beta is dependent again upon the dominance of either the value ratio or the delta
effect. Intuitively speaking, for example, when rm A is far behind in the race, a small
increase in patent value would not be enough to encourage it to restart investing. Firm A
would wait for rm B to experience a major setback such that it can take advantage of the
situation to reenter the game. Under this condition, rm As value is insensitive to the
uctuations in patent value and more sensitive to its competitors technological risk.
Therefore, the delta effect dominates and thus rm As beta is small when far behind.
We can also explain the intuition for the non-monotonic relationship between a rms beta
and its position in the race by looking at the leverage induced by the investment
expenditures.
18
A rms beta is closer to the beta of the underlying patent when the rm gets
closer to winning the race because its expected investment costs to complete the project
declines. The rms beta becomes higher as the patent race gets more intense, because both
rms are investing more aggressively which increases their leverage. The rms beta becomes
low once the rm falls well behind in the race and nally stops investing.
Fig. 4 plots the beta of the portfolio comprising rms A and B, which is calculated as the
value weighted betas of the two rms. Panel A is under joint monopoly and panel B is
under patent race. Just an eyeball comparison reveals that competition pushes up the rms
betas as a result of the increased leverage caused by over-investments.
4.2.2. Properties of return volatility in a duopoly competition
A rms return volatility over interval dt can be decomposed into cash ow uncertainty and
the idiosyncratic technological uncertainties of rms A and B. Take rm A as an example:
Vol
dF
A
I
A
dt
F
A
_ _

sx
F
A
x
F
A
_ _
2
I
A
k
A
y
A
F
A
k
A
F
A
_ _
2
I
B
k
B
y
B
F
A
k
B
F
A
_ _
2
_

dt
p
. (14)
ARTICLE IN PRESS
18
We would like to thank an anonymous referee for suggesting this interpretation.
Panel A of Fig. 5 plots level curves for the annualized return volatility of rm A under
patent race and Panel B plots level curves for the fraction of rm As return volatility
caused by idiosyncratic risks that is dened as the sum of the second and third components
of rm As return volatility formula as in the above Eq. (14). As shown in Panel A of Fig. 5,
rm As return volatility can be in excess of 100% when involved in a patent race. This
high level of return volatility is mainly attributable to idiosyncratic risks which account
for 90% of the return volatility as shown in Panel B of Fig. 5. Intuitively, competition
drives the rms to over-invest. The over-investments then speed up the resolution of the
rms technological uncertainty, which further more frequently alters the rms relative
position in the race. Accordingly, there appear more changes in the value of the rms.
This result is consistent with the empirical ndings on venture capital-backed
startups documented by Cochrane (2005). Cochrane (2005) studies the risk and return of
venture capital and nds that the arithmetic returns of venture capital, even after correction
of selection bias, are very volatile, and the standard deviation is about 100%.
When rm A is close to victory (the vertical axis), it is almost certain for rm A to win.
Therefore, its idiosyncratic risk becomes negligible and thus its return volatility is mainly
attributable to the cash ow uncertainty. As shown in Panel A of Fig. 5, rm As return
volatility is equal to the parameter value set for cash ow uncertainty s 30% when rm
A is close to the vertical axis. When rm A is far behind in the race (the southeast corner of
Panel B in Fig. 5), rm A stops investing unless rm B experiences signicant setbacks.
Firm Bs idiosyncratic technological risk accounts for about 90% of rm As return
volatility in that region.
4.2.3. Properties of return correlation in a duopoly competition
A rms return covariance with its competitor over interval dt can also be decomposed
into cash ow uncertainty and the idiosyncratic technological uncertainty of rms A and B
as follows:
Cov
dF
A
I
A
dt
F
A
;
dF
B
I
B
dt
F
B
_ _
sx
2
F
A
x
F
A
F
B
x
F
B
_
I
A
k
A
y
A
2
F
A
k
A
F
A
F
B
k
A
F
B
I
B
k
B
y
B
2
F
A
k
B
F
A
F
B
k
B
F
B
_
dt.
The return correlation of the two rms can be derived as the scaled return covariance
by the product of return volatilities of the two rms. Fig. 6 provides level curves for
return correlation. As in previous graphs, the dotted-curves are the contour plot of
both rms investment regions under patent race from Panel B of Fig. 1. In the region
where neither rm invests, the returns of the two rms are in principle perfectly correlated
by the common factor of patent value. The correlation shown in the graph is 1.00 or
very close to 1.00 due to discretized approximation. While the resolution of uncertainty
about the patent value makes the rms returns positively correlated, the resolution
of uncertainty about the expected cost-to-completion introduces negative correlation.
As rms resolve technological uncertainty through investments, a progress of one
rm (reductions in the rms expected cost-to-completion) means a setback for
the competitor. As a result, this rm enjoys positive returns whereas the competitor
suffers negative returns. The return correlation can be 0.7 in the region where both rms
invest.
ARTICLE IN PRESS
5. Comparative-static analysis
In this section, we analyze how cash ow uncertainty, dividend yield, and technological
uncertainty affect the rms investment decisions.
19
5.1. Cash ow uncertainty effect (s effect)
In the base model, the cash ow uncertainty parameter s is set to s 30%. By
increasing cash ow uncertainty from s 30% to 50%, we nd the following results:
On investment strategy: The effect of an increase in cash ow uncertainty on investment
strategies of a rm depends on the rms position in the patent race. As cash ow
uncertainty increases, a rm delays making investments when it is either far ahead or
behind in the race; it is eager to invest when the two rms are close in the race.
Empirical implication: The relationship between cash ow uncertainty and investment
therefore depends on a rms position: it is negative when a rm is far ahead or behind,
and it is positive when the rms are close in the race.
Given cash ow uncertainty, it is optimal for rms to delay making irreversible
investments and wait for more information about the protability of the new technology.
This is why the traditional real-options literature has concluded that the relationship
between uncertainty and investment is negative. The latest literature on strategic real
options shows that the relationship is ambiguous when there are strategic interactions
between rms (see, for example, Kulatilaka and Perotti, 1998; Kogan, 2001; Kulatilaka
and Lin, 2007). The results in this paper are consistent with the literature on strategic real
options, and contribute to this literature by showing how a rms position in competition
changes the relationship between uncertainty and investment. The intuition of our results
is as follows. An increase in cash ow uncertainty leads to a higher value of the option to
delay as well as a higher upside potential on patent value. In an environment with strategic
interactions between competing rms, any delay in investment may result in losing the
higher patent value to the competitor. Balancing the pros and cons, a rm may nd it
optimal to delay making investments when it is either well ahead or far behind in the race,
but eager to invest when competition is intense. In sum, we nd that the relationship
between cash ow uncertainty and investment depends on a rms position in the race.
5.2. Dividend yield effect (d effect)
In the base model, the dividend yield d is set to d 30%. By increasing dividend yield
from d 3% to 5%, we nd the following results.
On investment strategy: The effect of an increase in dividend yield on investment
strategies of a rm depends again on the rms position in the race. As dividend yield
increases, a rm is eager to invest when it is ahead, but delays making investments when
it is behind or close to the other rm.
ARTICLE IN PRESS
19
Graphs of the comparative-static analysis are omitted for brevity. They are available upon request.
Intuitively, the dividend yield d represents the rate at which the present value of the
patent decays through time as a result of no patent being granted (as no victor has
emerged); in other words, it represents the opportunity cost of deferring victory. One might
expect that a rm always has incentives to invest when d increases. This is the case when
the project can be completed immediately, as in the model of McDonald and Siegel (1986).
However, when a project cannot be completed immediately, the result is not at all straight
forward due to conicting factors as explained by Majd and Pindyck (1987). In that case,
the winning rm can only obtain the patent upon completion. Holding the risk free rate,
the market risk premium and the patent beta constant, the true growth rate m and the risk-
neutral growth rate m
*
decrease as dividend yield increases, which is evident from Eq. (8).
Therefore, as dividend yield increases, the value of the patent at completion may be lower,
discouraging rms from investing. When a rm is close to victory, it is eager to invest more
to save dividends. However, when a rm is close to or lags behind its competitor, it will
withhold investments expecting that the patent value will be reduced.
5.3. Technological uncertainty effect (y effect)
In the base model, the rms technological volatility parameters y
A
and y
B
are set to
y
A
y
B
0.6. Here we set rm Bs technological volatility parameter to y
B
0.6 and
change rm As technological volatility parameter to y
A
0.9 or y
A
0.3. We nd the
following results:
On investment strategy: An increase in the followers technological uncertainty will
encourage both rms to invest. An increase in the leaders technological uncertainty,
however, will encourage the leader to invest but discourage the follower from investing.
In other words, an increase in a rms technological uncertainty encourages the rm to
invest regardless of its position in the race. In contrast, an increase in the competitors
technological uncertainty encourages a rm to invest if it is leading but discourages a
rm from investing if it lagging behind.
Empirical implication: The relationship between the aggregate investments of two rms
involved in a patent race and a rms idiosyncratic technological risk depends on its
position in the race. The aggregate investments are increasing in the followers
idiosyncratic technological risk.
One of the interesting features of our specication of the expected cost-to-completion is
that investments of a rm can result in technological improvements as well as technological
setbacks. From the followers perspective, an increase in its own technological uncertainty
gives it a chance to catch up, so it invests aggressively. In contrast, an increase in
technological uncertainty of the leader has two opposite effects: (1) the leader may advance
further, which discourages the follower from investing, (2) the leader may suffer setbacks,
which gives the follower a chance to catch up, and thus encourages the follower to invest.
Since investments are irreversible, it is optimal for the follower to withhold investments
pending more information on the leaders move in order to save investment costs. From
the leaders perspective, an increase in its own technological uncertainty gives it an
incentive to invest so that it can learn more information about its investment costs. An
increase in the followers technological uncertainty may help the follower catch up, and
therefore gives the leader an incentive to invest so as to maintain its leading position. In
ARTICLE IN PRESS
sum, the relationship between aggregate investments of both rms and a rms
idiosyncratic technological risk depends on the position of this rm in the race.
5.4. Implication of comparative-static analysis
The above sensitivity analysis reinforces the analysis of nancial properties of the
equilibrium in Section 4 on the following point.
The relative position of a rm to its competitor in a race is an important control variable
when we investigate a rms investment behavior and other nancial properties. We can
use the degree of maturity or superiority of a rms technology as a proxy for its position in
competition.
Regarding private start-ups, venture capitalists typically do stage-nancing to monitor
start-ups. There are nancing round A, round B, round C y Future empirical studies could
proxy a start-ups position by its current nancing stage. For public rms, it is harder to
proxy a rms position; however, there should be some specic industrial characteristic,
such as a ladder of technology milestone that can be used as a criterion to rank technologies
within a sector. In the disk drive industry, for example, a higher density indicates a more
superior drive (as in Lerner, 1997). If the objective is to reach a certain level of density, it is
reasonable to assume that the rm with the higher level of density needs less money or time
to reach the targeted level. Furthermore, in the biotechnology industry, there are standard
stages of clinical trials to indicate the degree of maturity of a certain technology. A rm at
an FDA approval stage could be reasonably assumed to spend less money or time to receive
an approval than a rm who is still at clinical trial stage I.
6. Conclusion
In this paper, we develop a continuous-time real-options patent-race model to analyze
nancial properties of R&D intensive rms. The numerical analysis provided in this study
shows that competition causes over-investment and value dissipation relative to the case of
joint monopoly, which then pushes up CAPM beta and return volatility, and introduces
negative return correlation between the competing rms. These results imply that venture
capitalists including competing rms in the portfolio cannot only improve efciency by
mitigating over-investment and value dissipation, but can also reduce portfolio risk via
hedging as the returns of competing rms could appear negative correlation. In addition,
we nd that a rms position in competition is an important factor in the determination of
its nancial properties. In particular, the CAPM beta, and thus the cost of capital of an
R&D rm, is a non-monotonic function of its relative position to its competitor. The cost
of capital is higher when a rm intensively competes with its competitor than when well
ahead or well behind. We also nd that, as another example, the relationship between cash
ow uncertainty and investment can be negative when a rm is far ahead or far behind, or
positive when rms are close in the race. So, the relative position of competing rms in a
race should be taken as an important control variable in future empirical studies of
nancial properties of R&D intensive rms.
To extend the model, it could be interesting to add explicit learning process about prot
ows of the patent and investment opportunities for a rm and its competitor. It may be
also worthwhile to study a rms investment behavior and other nancial properties in a
competitive environment with the presence of technological spillovers. While competition
ARTICLE IN PRESS
motivates a rm to invest, technological spillovers may motivate a rm to withhold
investments so as to obtain a free ride on investment efforts of its rival. We leave these
extensions to future research.
Acknowledgments
This paper is a modied version of Chapter 1 of my Ph.D. Dissertation at Duke
University. I would like to give special thanks to my supervisor Albert S. Pete Kyle for
his continuous advice and insightful comments. I also thank my other committee members,
Itay Goldstein, Leslie M. Marx, Curtis R. Taylor, S. Viswanathan, for their advice and
support. Thanks also to Michael Bradley, Carl Chiarella (the editor), Lorenzo Garlappi,
John Graham, Campbell R. Harvey, Tracy Lewis, Vojislav Maksimovic, Richmond
Mathews, Hui Ou-yang, Alexander Triantis, three anonymous referees, and seminar
participants at Duke University, the Board of Federal Reserve, the University of Hong
Kong, Tulane University, 2005 WFA, 2005 WEA and 2005 CICF. All errors are mine. I
gratefully acknowledge the nancial support from Hong Kong Research Grants Council
(Grant no. HKU 7670/05H).
Appendix A. The derivation of the bellman equation of rm A
To evaluate the rms, we assume that the systematic risk associated with the patents
cash ows can be hedged by trading the market portfolio. With this assumption, we can
price the rms without knowing investors risk preference or the true growth rate m of the
patents cash ows. Construct a hedge portfolio: buy rm A, which is worth F
A
, and sell
short n
A
units of the market portfolio. The value of a share in the market portfolio with
dividends reinvested is P
m
. The hedging demand n
A
will be chosen so that the systematic
risk of the hedge portfolio is eliminated. The value of the hedge portfolio is
Q F
A
n
A
P
m
. The total return on the hedge portfolio over interval dt is
dF
A
n
A
dP
m
I
A
dt. By applying Itos lemma and assuming that the white noise processes
dz
A
and dz
B
are mutually independent, and are both independent of the white noise
process of total nancial process dw:
dF
A
F
A
x
dx F
A
k
A
dk
A
F
A
k
B dk
B
1
2
F
A
xx
dx
2
1
2
F
A
k
A
k
A
dk
A
1
2
F
A
k
B
k
Bdk
B
2
F
A
x
mxdt sxdw
F
A
k
A
I
A
dt y
A
I
A
k
A
1=2
dz
A
F
A
k
B I
B
dt y
B
I
B
k
B
1=2
dz
B

1
2
F
A
xx
s
2
x
2
dt
1
2
F
A
k
A
k
A
y
A
2
Ik
A
dt
1
2
F
A
k
B
k
By
B
2
I
B
k
B
dt F
A
x
mx I
A
F
A
k
A
I
B
F
A
k
B
1
2
F
A
xx
s
2
x
2
1
2
F
A
k
A
k
A
_
y
A
2
I
A
k
A
1
2
F
A
k
B
k
By
B
2
I
B
k
B
_
dt sxF
A
x
dw y
A
I
A
k
A
1=2
F
A
k
A
dz
A
y
B
I
B
k
B
1=2
F
A
k
B dz
B
.
where rm As value with a subscript denotes partial derivative of rm As value with
respect to the variable in the subscript. For instance, F
A
x
and F
A
xx
denote the respective
rst-order and second-order partial derivatives of rm As value F
A
with respective to the
patent value x.
Note that we can decompose dw into systematic and idiosyncratic components
dw b(s
m
/s)dz
m
+(s
I
+s)dz
I
. Hence, the total return on the hedge portfolio over interval
ARTICLE IN PRESS
dt is
dF
A
n
A
dP
m
I
A
dt F
A
x
mx I
A
F
A
k
A
I
B
F
A
k
B
1
2
F
A
xx
_
s
2
x
2
1
2
F
A
k
A
k
A
y
A
2
I
A
k
A
1
2
F
A
k
B
k
By
B
2
I
B
k
B
n
A
r
f
l
m
P
m
I
A
_
dt bs
m
xF
A
x
n
A
s
m
P
m
_ _
dz
m
s
I
xF
A
x
dz
I
y
A
I
A
k
A
1=2
F
A
k
A
dz
A
y
B
I
B
k
B
1=2
F
A
k
B dz
B
.
To eliminate the systematic risk, that is the term with dz
m
, we set the hedging demand as
n
A
bxF
A
x
P
m
. In doing so, the only risks associated with the hedge portfolio are
diversiable, and hence the expected rate of return on the portfolio must equal to the risk-
free rate r
f
, that is,
Max
I
A
20;I
max
fEdF
A
n
A
dP
m
I
A
dtg r
f
F
A
n
A
P
m
dt.
After substituting n
A
bxF
A
x
=P
m
into this equation and re-grouping, the Bellman
equation of rm A is:
Max
I
A
20;I
max
1
2
s
2
x
2
F
A
xx
r
f
dxF
A
x
_ _
I
A 1
2
y
A
2
k
A
F
A
k
A
k
A
F
A
k
A
1
_ _ _ _
I
B 1
2
y
B
2
k
B
F
A
k
B
k
B
F
A
k
B
_ _
r
f
F
A
_
0.
Appendix B. The derivation of CAPM beta, return volatility, and return covariance
By Itos lemma, the rate of return on rm i, i A, B, is:
dF
i
I
i
dt
F
i
mx
F
i
x
F
i
I
A
F
i
k
A
F
i
I
B
F
i
k
B
F
i

1
2
s
2
x
2
F
i
xx
F
i
_

1
2
y
A
2
I
A
k
A
F
i
k
A
k
A
F
i

1
2
y
B
2
I
B
k
B
F
i
k
B
k
B
F
i
I
i
F
i
_
dt sx
F
i
x
F
i
dw y
A
I
A
k
A
1=2
F
i
k
A
t
F
i
dz
A
y
B
I
B
k
B
1=2
F
i
k
B
F
i
dz
B
.
Hence, the expected rate of return on rm i, i A, B, is:
E
dF
i
I
i
dt
F
i
_ _
mx
F
i
x
F
i
I
A
F
i
k
A
F
i
I
B
F
i
k
B
F
i

1
2
s
2
x
2
F
i
xx
F
i
_

1
2
y
A
2
I
A
k
A
F
i
k
A
k
A
F
i
1
2
y
B
2
I
B
k
B
F
i
k
B
k
B
F
i

I
i
F
i
_
dt.
(1) CAPM beta of rm i, i A, B:
The rate of return on the underlying patent is
dx dxdt
x

mxdt sxdw dxdt
x
m d dt sdw.
Hence, the expected rate on return of the underlying patent is
Edx dxdt=x m d dt.
We assume the market return, R
m
dP
m
=P
m
r
f
l
m
dt s
m
dz
m
, is normally
distributed with mean (r
f
+l
m
)dt and variance s
2
m
dt.
ARTICLE IN PRESS
Therefore, the beta of the underlying patent by CAPM is,
b
Covdx dxdt=x; R
m
VarR
m
Edx dxdt=x Edx dxdt=xR

m
ER
m
s
2
m
dt

Esdws
m
dz
m
s
2
m
dt
.
Hence, the beta of rm i, i A, B, by CAPM is:
b
i
CovdF
i
I
i
dt=F
i
; R
m
VarR
m

EdF
i
I
i
dt=F
i
EdF
i
I
i
dt=F
i
R
m
ER
m
s
2
m
dt
EsxF
i
x
=F
i
dws
m
dz
m
s
2
m
dt

Esdws
m
dz
m
s
2
m
dt
x
F
i
x
F
i
_ _
b
i
,
where
i
xF
i
x
=F
i
is the elasticity of rm is value with respect to the patent value,
i A, B.
(2) By assuming that the white noise processes dz
A
and dz
B
are mutually independent,
and are both independent of the white noise process of total nancial process dw, the
return volatility of rm i over interval dt, i A, B, is:
Vol
dF
i
I
i
dt
F
i
_ _

E
dF
i
I
i
dt
F
i
E
dF
i
I
i
dt
F
i
_ _ _ _
2
_ _
sx
F
i
x
F
i
_ _
2
I
A
k
A
y
A
F
i
k
A
F
i
_ _
2
I
B
k
B
y
B
F
i
k
B
F
i
_ _
2
dt
p
; i A; B.
(3) By assuming that the white noise processes dz
A
and dz
B
are mutually independent,
and are both independent of the white noise process of total nancial process dw, the
covariance of the two rms returns over interval dt:
Cov
dF
A
I
A
dt
F
A
;
dF
B
I
B
dt
F
B
_ _
E
dF
A
I
A
dt
F
A
E
dF
A
I
A
dt
F
A
_ _ _ _ _
dF
B
I
B
dt
F
B
E
dF
B
I
B
dt
F
B
_ _ _ __
sx
2
F
A
x
F
A
F
B
x
F
B
I
A
k
A
y
A
2
F
A
k
A
F
A
F
B
k
A
F
B
_
I
B
k
B
y
B
2
F
A
k
B
F
A
F
B
k
B
F
B
_
dt.
References
Abel, A.B., 1983. Optimal investment under uncertainty. American Economic Review 73, 228233.
Abel, A.B., 1984. The effects of uncertainty on investment and the expected long-run capital stock. Journal of
Economic Dynamics and Control 7, 3954.
Abel, A.B., 1985. A stochastic model of investment, marginal q and the market value of the rm. International
Economic Review 26, 305322.
Berk, J., Green, R.C., Naik, V., 2004. Valuation and return dynamics of new ventures. The Review of Financial
Studies 17, 135.
ARTICLE IN PRESS
Bertola, G., 1988. Adjustment costs and dynamic factor demands: investment and employment under uncertainty.
Ph.D. Dissertation, MIT (Chapter 2).
Caballero, R., Pindyck, R., 1996. Uncertainty, investment, and industry evolution. International Economic
Review 37, 641662.
Childs, P., Triantis, A., 1999. Dynamic R&D investment policies. Management Science 45, 13591377.
Cochrane, J., 2005. The risk and return of venture capital. Journal of Financial Economics 75, 352.
Craine, R., 1988. Risky business: the allocation of capital. Journal of Monetary Economics 23, 201218.
Dasgupta, P., Stiglitz, J., 1980. Uncertainty, industrial structure and the speed of R&D. Bell Journal of
Economics 11, 128.
Dixit, A.K., 1989. Entry and exit decisions under uncertainty. Journal of Political Economy 97, 620638.
Dixit, A.K., Pindyck, R.S., 1994. Investment Under Uncertainty. Princeton University Press, Princeton, NJ.
Doraszelski, U., 2003. An R&D race with knowledge accumulation. The Rand Journal of Economics 34, 2042.
Fudenberg, D., Gilbert, R., Stiglitz, J., Tirole, J., 1983. Preemption, leapfrogging and competition in patent races.
European Economic Review 22, 331.
Garlappi, L., 2004. Risk premia and preemption in R&D ventures. Journal of Financial and Quantitative
Analysis 39, 843872.
Grenadier, S.R., 1999. Information revelation through option exercise. Review of Financial Studies 12, 95130.
Grenadier, S.R., 2000. Game Choices: The Intersection of Real Options and Game Theory. Risk Books.
Grenadier, S.R., 2002. Option exercise games: an application to the equilibrium investment strategy of rms.
Review of Financial Studies 15, 691721.
Grossman, G.M., Shapiro, C., 1987. Dynamic R&D competition. The Economic Journal 97, 372387.
Harris, C., Vickers, J., 1985. Perfect equilibrium in a model of a race. Review of Economic Studies 52, 193209.
Harris, C., Vickers, J., 1987. Racing with uncertainty. Review of Economic Studies 54, 121.
Hartman, R., 1972. The effects of price and cost uncertainty on investment. Journal of Economic Theory 5,
258266.
Hull, J., 2000. Options, Futures, & Other Derivatives, fourth ed. Prentice-Hall.
Judd, K., 2003. Closed-loop equilibrium in a multi-stage innovation race. Economic Theory 21, 673695.
Kogan, L., 2001. An equilibrium model of irreversible investment. Journal of Financial Economics 62, 201245.
Kulatilaka, N., Lin, L., 2007. Strategic options and rm value. Managerial Finance 33, 893903.
Kulatilaka, N., Perotti, E., 1998. Strategic growth options. Management Science 44, 10211031.
Kyle, A.S., Meng, R., 2005. Strategic acquisitions and investments in a duopoly patent race under uncertainty.
Duke University Working Paper.
Lambrecht, B., 2000. Strategic sequential investments and sleeping patents. In: Brennan, M., Trigeorgis, L. (Eds.),
Project Flexibility, Agency, and Competition: New Developments in the Theory and Application of Real
Options. Oxford University Press, Oxford.
Lambrecht, B., Perraudin, W.R., 2003. Real options and preemption under incomplete information. Journal of
Economic Dynamics and Control 27, 619643.
Leahy, J.V., Whited, T.M., 1996. The effect of uncertainty on investment: some stylized facts. Journal of Money,
Credit and Banking 28, 6483.
Lee, T., Wilde, L., 1980. Market structure and innovation: a reformulation. Quarterly Journal of Economics 94,
429436.
Lerner, J., 1997. An empirical exploration of a technology race. Rand Journal of Economics 28, 228247.
Loury, G., 1979. Market structure and innovation. Quarterly Journal of Economics 93, 395410.
Lukach, R., Kort, P.M., Plasmans, J., 2007. Optimal R&D investment strategies under the threat of new
technology entry. International Journal of Industrial Organization 25, 103109.
Majd, S., Pindyck, R., 1987. Time to build, option value, and investment decisions. Journal of Financial
Economics 18, 728.
Maskin, E., Tirole, J., 1988. A theory of dynamic oligopoly I: overview and quantity competition and a theory of
dynamic oligopoly II: price competition, kinked demand curves, and Edgeworth cycles. Econometrica 56,
549599.
McDonald, R., Siegel, D., 1986. The value of waiting to invest. Quarterly Journal of Economics 101, 707727.
Miltersen, K., Schwartz, E., 2004. R&D investments with competitive interactions. Review of Finance 8, 355401.
Novy-Marx, R., 2007. An equilibrium model of investment under uncertainty. Review of Financial Studies 20,
14611502.
Pindyck, R.S., 1988. Irreversible investment, capacity choice, and the value of the rm. American Economic
Review 79, 969985.
ARTICLE IN PRESS
Pindyck, R.S., 1993. Investments of uncertain cost. Journal of Financial Economics 34, 5376.
Reinganum, J., 1981. Dynamic games of innovations. Journal of Economic Theory 25, 2141.
Reinganum, J., 1982. A dynamic game of R&D: patent protection and competitive behavior. Econometrica 50,
671688.
Schwartz, E., 2004. Patents and R&D as real options. Economic Notes 33, 2354.
Schwartz, E., Moon, M., 2000. Evaluating research and development investments. In: Brennan, M., Trigeorgis, L.
(Eds.), Project Flexibility, Agency, and Competition: New Developments in the Theory and Application of
Real Options. Oxford University Press, Oxford.
Trigeorgis, L., 1996. Real Options: Managerial Flexibility and Strategy in Resource Allocation. MIT Press,
Cambridge, MA.
Weeds, H., 2002. Strategic delay in a real options model of R&D competition. Review of Economic Studies 69,
729747.
ARTICLE IN PRESS

1 s2.0 S0165188908000158 Main

Загружено:

Сведения о документе

Исходное описание:

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

1 s2.0 S0165188908000158 Main

Загружено:

Авторское право:

Доступные форматы

Journal of Economic Dynamics & Control 32 (2008) 31923217

A patent race in a real options setting: Investment

Tel.: +852 2859 1048; fax: +852 2548 1152.

Edx dxdt=x Edx dxdt=xR

Вам также может понравиться