Anda di halaman 1dari 54

Electronic copy available at: http://ssrn.

com/abstract=1421748
Strategic Investment
and Industry Risk Dynamics
Maria Cecilia Bustamante

June 2012
Abstract
This paper characterizes how rms strategic interaction in product markets aects the
industry dynamics of investment and expected returns. Under imperfect product market
competition, the investment strategy of each rm depends on the intra-industry standard
deviation in rms market to book ratios or intra-industry value spread. The insight by
asset pricers that a rms exposure to systematic risk is signicantly related to its own
investment is incomplete in industries with high value spread, in which a rms exposure
to systematic risk is also explained by the investments of others. In the model and the
data, rms betas and excess returns correlate more positively in industries with low value
spread, low dispersion in operating mark-ups, and low concentration.
Keywords: expected returns, investment, imperfect competition, strategic interaction,
value spread, industry concentration.
JEL codes: L11, L22, G11, G12, G31.

London School of Economics, Department of Finance, Houghton Street, WC2A 2AE London, United King-
dom. E-mail: m.bustamante@lse.ac.uk. Tel.: +44(0)20 7107 5011. I am grateful to the editor, an anoymous
referee, and Rick Green for insightful comments. I also thank Kerry Back, Andres Donangelo, Andrea Einsfeldt,
Jack Favilukis, Christian Julliard, Christopher Polk, Ramman Uppal, Georgios Skoulakis, Michela Verardo, the
participants of the FMG lunchtime seminar, the UBC Summer Conference 2010, the World Econometric Society
Meetings 2010, the FIRS Conference 2011, the WFA Annual Meetings 2011 and the Texas Finance Festival 2012.
An earlier version of this paper circulated under the titled "Strategic Investment, Industry Concentration and
the Cross Section of Returns."
Electronic copy available at: http://ssrn.com/abstract=1421748
Introduction
In imperfectly competitive industries, the ability of rms to aect market prices induces them
to invest strategically. The value of each rm depends not only on its own assets in place
and investment opportunities, but also on the ability of its competitors to expand capacity
and aect market prices. As a result, under imperfect competition, the fundamental insight
in the asset pricing literature that a rms exposure to systematic risk or beta is signicantly
aected by its own investment decisions is incomplete. When rms invest strategically, a
rms exposure to systematic risk may depend signicantly on the investments of other rms
in the industry.
The production based asset pricing literature focuses on the impact of corporate invest-
ment on expected returns in perfectly competitive or perfectly monopolistic industries.
1
We
contribute in exploring the intermediate case of imperfectly competitive industries, in which
rms strategic interaction aects the dynamics of investment and risk. The study of rms
intra-industry interactions is relevant in the light of the existing empirical evidence, which
suggests that commonly studied asset pricing regularities are predominantly intra-industry.
2
Our model rationalizes existing ndings on the cross section of returns, and provides additional
testable predictions which we nd support for in our empirical section.
We motivate our study with several research questions. How does a rms relative position
in its product market inuence its investment decisions, and the conditional dynamics of its
expected returns? In which types of industries are the stylized predictions of investment based
asset pricers for monopolies or perfectly competitive industries still appropriate? How does
strategic interaction aect the intra-industry correlation of rms investments and their expo-
sure to systematic risk? And lastly, how do specic industry characteristics such as demand
elasticity, demand growth or demand volatility aect the industry dynamics of investment and
risk?
The core prediction of our model is that the dynamics of rms investments and expected
returns depend critically on the intra-industry standard deviation in market to book ratios,
or intra-industry value spread. Under imperfect competition, a rms market to book ratio
reects its comparative advantage to increase its market share. As a result, in industries with
low value spread, rms have more similar investment strategies, they invest at more similar
points in time, and their betas and excess returns correlate more positively. Firms betas and
excess returns may also correlate more positively in industries with low standard deviation in
1
See, for instance, Berk et al (1999), Zhang (2005) and Carlson et al (2004).
2
See, for instance, Cohen and Polk (1996), Moskowitz and Grimblatt (1999), and Cohen et al (2003).
1
mark-ups, and low concentration.
3
We obtain these predictions in partial equilibrium, real options model of dynamic oligopoly
in which heterogeneous rms compete in capacity, with costly production and irreversible
investment. We solve for the investment strategies of rms that dier in either their production
technologies, and have a single growth option to increase capacity. This represents a signicant
departure from earlier dynamic models of imperfect competition, which focus on identical rms
and hence are silent about the intra-industry cross section of investments and risk.
4
Given
that the setting is fairly complex, we spend substantial eort in deriving rms investment
strategies in equilibrium. We derive testable implications on the impact of rms strategic
interaction on expected returns by examining their betas.
We group our contributions into three dierent sets of results. Our rst set of results relates
to the predictions of the model on the dynamics of investment under imperfect competition.
In neoclassical models, the investment of each rm solely depends on its own marginal product
of capital or .
5
In contrast, we nd that under imperfect competition the investment strategy
of each rm depends on the marginal product of capital or of all rms in the same industry.
In our model, a rms reects its comparative advantage to increase its market share relative
to other rms in the industry. As a result, rms strategic behavior is such that the investment
strategy of each rm depends on the intra-industry standard deviation in .
The corresponding testable implication is that rms investment strategies depend on the
intra-industry value spread. In industries with low value spread, rms are closer competitors
with similar market to book ratios, and hence the cost of preempting each other by investing
aggressively is too high. Firms investment strategies are more similar: rms increase their
capacity by similar amounts, and their investments cluster. Conversely, in industries with
high value spread, rms are more distant competitors, and rms with higher market to book
invest earlier and more than their competitors. These results are consistent with the empirical
literature on mergers, which nds that rms with high market to book are more capable of
preempting others.
6
Our second set of results relates to the predictions on how rms strategic interaction
aects the dynamics of expected returns. Consistent with the results for monopolies discussed
in Carlson et al (2004), we nd that under imperfect competition rms undergo a period of
high expected returns before their own investment, and a period of low expected returns upon
investment. Yet we add that, in industries with high value spread, rms also undergo a period
3
We elaborate on the link between industry dynamics and static measures of competition below.
4
See Weeds(2002), Grenadier (2002), and Aguerrevere (2009).
5
See, for instance, Hayashi (1982).
6
See Andrade et al (2001) and the discussion in Jovanovic and Rousseau (2002).
2
of low expected returns before their competitors invest. The insight by asset pricers that a
rms exposure to systematic risk is signicantly related to its own investment is incomplete in
industries with high value spread, and also in industries with low demand growth, low demand
volatility, and high demand elasticity.
The model provides several empirical implications on the relation between rms expected
returns and market to book ratios. In the model, a rms market to book ratio reects its
ability to increase its market share in subsequent periods. As a result, the model implies
that the market to book sorts used in the empirical asset pricing literature are eectively
aggregating rms according to their relative position in the industry.
7
Similarly, we nd that
the dynamics of the expected returns of each rm are aected by the intra-industry value
spread. This suggests why Cohen et al (2003) nd that the value spread of US rms is
predominantly intra-industry.
The core testable asset pricing implication is that rms betas and excess returns correlate
more positively in industries with low value spread. Firms strategic interaction aects the
intra-industry correlation of their expected returns, even when all rms in the industry are
subject to no idiosyncratic shocks, and there is a single source of systematic risk. In more
homogeneous industries with low value spread, rms have investment dynamics, and their
expected returns correlate positively over time. Conversely, in more heterogeneous industries
with high value spread, there are leaders which invest earlier and more than other rms, and
the betas of leaders and laggards correlate negatively over time.
The model also shows that those industries with low value spread usually have lower stan-
dard deviation in mark-ups, and lower concentration as measured by the Herndahl-Hirshman
Index (HHI). This allows us to formulate testable predictions on how the spread in mark-ups
and the concentration of an industry aect the dynamics of expected returns. In particular,
our testable implications on the HHI rationalize the recent empirical evidence provided by
Hoberg and Phillips (2010). Consistent with our model, Hoberg and Phillips (2010) nd that
in less concentrated industries have more predictable average industry returns, in which peri-
ods of high market to book ratios, high returns, high betas and high investment, are followed
by periods of lower market to book ratios, lower investment, lower returns and lower betas.
A related implication of our model, however, is that measures of competition commonly
used in the literature such as the spread in mark-ups or the HHI may prove insucient to study
industry dynamics precisely because they are static. In our model, industries with low HHI
may have low intra-industry value spread, but this need not apply to all cases; for instance, a
deconcentrating industry may have a high value spread, and a concentrating industry may have
7
See Fama and French (1992).
3
a lower value spread.
8
As a result, the key sorting variable to test for the impact of rms
strategic interaction on industry dynamics is the intra-industry value spread. The intra-
industry value spread is forward looking, and captures current and expected future dierences
in mark-ups and market shares.
The third main contribution of our paper relies on its empirical evidence. Consistent with
the model, we rst document that rms investments are signicantly related to the intra-
industry value spread, both at the rm and industry level. To test our predictions on industry
dynamics, we construct a measure of comovement which captures the average pairwise cor-
relation in rms investments, betas and excess returns by industry.
9
Consistent with the
assumption that corporate investment aects rms expected returns, we report a signicant
relation between the intra-industry comovement in investment and the intra-industry comove-
ment in betas or excess returns. More importantly, we provide supporting empirical evidence
on the prediction that betas and excess returns correlate more positively in industries with
lower value spread, lower standard deviation in mark-ups, and lower concentration.
Lastly, our model provides additional results on how the number of rms aects the dy-
namics of investment strategies and expected returns in heterogeneous industries. Consistent
with studies by Grenadier (2002) and Aguerrevere (2009) for industries with identical rms, a
higher number of rms erodes the values and betas of all rms in the industry. However, in
industries with high value spread, we add that a higher number of rms erodes more severely
the betas of those rms with lower market to book. We also nd that the stylized prediction
in Grenadier (2002) that a higher number of rms induces rms to accelerate investment need
not apply to all types of industries. In industries with high value spread, a higher number of
rms may induce rms with lower to delay their investment.
The paper is organized in ve sections. Section 1 describes the set up and the equilibrium
investment strategies. Section 2 provides the corresponding asset pricing implications. Section
3 elaborates on comparative statics. Section 4 reports the supporting empirical evidence.
Section 5 concludes.
Related literature
The model in our paper is closely related to Carlson et al (2004), who study the dynamics
of rms exposure to systematic risk in a cross section of rms in which each rm has a
single growth option to invest. In their model, rms are monopolists which do not invest
strategically. We build on their framework to study how rms strategic interaction under
8
See discussion in Section 1.4.
9
Our methodology follows Khanna and Thomas (2009). See Section 4.
4
imperfect competition aects their investment decisions and expected returns.
The model is closely related to the studies on strategic investment by Fundenberg and
Tirole (1985), Grenadier (1996), Weeds (2002) and Mason and Weeds (2010). While these
papers focus on duopolies in which rms only decide when to invest, we study industries with
multiple, heterogeneous rms which decide when and how much to invest. The solution
approach relies on sorting conditions and incentive compatibility constraints as in Maskin and
Tirole (1988). The approach is consistent with Back and Paulsen (2009), who require dynamic
investment models of oligopoly to account for rms incentives to preempt each other.
The theme of our study also relates to Grenadier (2002) and Aguerrevere (2009), who
consider real options models to study how competitive pressures aect corporate investment
and expected returns. Their papers focus on imperfectly competitive industries in which all
rms are identical, and assume that rms invest simultaneously and continuously at each point
in time. In contrast, we allow rms to be heterogeneous, and we elaborate on the equilibrium
eects of strategic interaction under imperfect competition.
Our model is also close to Carlson et al (2009) and Bena and Garlappi (2011). These
papers consider models of investment timing with two rms, in which the beta of the leader is
dampened by the expected reduction in prots once the follower invests. Bena and Garlappi
(2011) provide evidence on this result in the context of patent races. We observe that, in
industries with high value spread, the beta of any operating rm is dampened by expected
reductions in prots due to investments by other rms.
The asset pricing implications and empirical tests are heavily related to Cohen et al (2003)
and Hoberg and Phillips (2010). The model suggests why the value spread in Cohen et al
(2003) is predominantly intra-industry, and why Hoberg and Phillips (2010) nd that more
predictable patterns in the average returns of less concentrated industries. The empirical results
in this paper further complement the evidence in Hoberg and Phillips (2010) that rms returns
comove more positively in less concentrated industries.
In addition, Pastor and Veronesi (2009) show how investors learning about new technolo-
gies can explain bubbles in stock prices in US industries. Spiegel and Tookes (2011) study how
product market competition aects merger activity. Garlappi (2004) and Novy Marx (2011)
explore the link between industry dynamics and expected returns. Other papers include Hou
and Robinson (2006), Bulan et al (2009), Ali et al (2009), Opp et al (2011) and Lyandres and
Watanabe (2011); and studies on microfounded explanations on stock return dynamics such
as Pastor and Veronesi (2003).
5
1 Strategic interaction and investment dynamics
We begin by studying the simplest type of industry which conveys the core predictions of
our paper. For this sake, we consider an industry with = 2 rms in which rms have a
single growth option to invest, and in which rms only dier in their costs of production after
investment.
1.1 Basic model
The basic model builds on the framework by Carlson et al (2004), and we use their notation
where possible. We consider an imperfectly competitive industry with = 2 rms. Each
rm is all equity nanced and run by a manager who is the single shareholder.
Firms compete in capacity and produce an homogeneous good which they sell in the market
at a price j
t
. We assume that rms operate at full capacity at any point in time. The demand
function requires that the product market price j
t
equals
j
t
= A
t
1

1
s
t
(1)
where - 1 is the elasticity of demand and A
t
is a systematic multiplicative shock, and the
industry output 1
t
is the sum of the production at time t.
10
The demand shock A
t
follows a
geometric Brownian motion with drift j
a
and volatility o
a
such that
dA
t
= j
a
A
t
dt +o
a
A
t
d.
t
(2)
where .
t
is a standard Wiener process, and A
0
is strictly positive.
11
We denote the two rms in the industry by , = 1, '. Each rm has assets in place,
and a single growth option to increase their capacity. Managers maximize shareholder value
by choosing when to invest, and also how much to increase the scale of production upon
investment. The resulting strategy
)
= r
)
;
)
species a stopping rule specifying the
critical value r
)
for the stochastic demand shock A
t
at which rm , exercises its growth
option, and also the optimal scale of production
)
1 upon investment.
In the basic model, we assume that both rms have the same initial installed capacity before
investment 1.
12
Firms increase their installed capacity from 1 to
)
1 upon investment. We
10
Firms under imperfect competition do not operate in the range where - < 1.
11
We further assume that A0 is suciently low such that the growth options of all rms in the industry are
strictly positive at time t = 0. Hence A0 < r
s
1
in the basic model; we dene r
s
1
in Section 1.3.
12
We allow rms to dier in their installed capacity before investment in Section 1.4.
6
further set the output of each rm , to be equal to its installed capacity, such that the total
production 1
t
in (1) is the sum of the installed capacity of both rms at any point in time.
13
The decision to invest is irreversible, and entails benets and costs.
14
Upon investment,
rms benet from a lower instantaneous marginal cost of production. In the basic model, we
assume that rms have the same instantaneous marginal cost of production c before investment,
and have lower and dierent marginal costs of production c
)
< c upon investment. We denote
rm 1 as the rm with the lowest instantaneous marginal costs upon investment, such that
c
1
< c
A
.
Upon investment, rms incur a xed cost )1. We do not consider variable costs of invest-
ment for the sake of tractability; the qualitative predictions of the model remain unchanged if
rms are subject to linear cost of investment.
15
We also assume that the costs of production
are linear in A
t
; this enables us to compare more easily rms capacity choices in our model
to those predicted by static games of strategic interaction.
16
Given all our assumptions, the instantaneous prots of rm , before its own investment

)t
are given by

)t
=
_
j

t
cA
t
_
1 (3)
where the superscript

denotes the cashows before investment. The instantaneous prots
of rm , after its own investment
+
)t
are equal to

+
)t
=
_
j
+
t
c
)
A
t
_

)
1 (4)
where the superscript
+
denotes the cashows after investment.
1.2 Valuation
The value of any rm , equals the expected present value of its risky prots. Using a similar
argument in Carlson et al (2004), we determine the value of the rm using a replicating portfolio
with weights on a risk free and a risky asset.
13
Hence YI = 21 if both rms have not invested, YI = (1 A) 1 if one of the two rms has invested, and
YI = (A1 AT ) 1 if both rms have invested.
14
The irreversibility of investment implies a commitment by rms not to adjust their capacity upon a reduction
in market prices.
15
We can extend the model to incorporate a linear cost of capital j1 (A 1) 1 as in neoclassical investment
models, with j1 0 as the purchase price of capital. For optimal investment timing, adding this cost operates
as redening xed costs as

)1, with

) = ) j1 (A 1). Optimal capacity decisions would depend on the
alternative marginal cost c = cAI j1.
16
See Fundenberg and Tirole (1991).
7
We let 1
t
denote the price of a riskless bond with dynamics d1
t
= r1
t
dt, and we let
o
t
be a risky asset with dynamics do
t
= j
c
o
t
dt + o
c
o
t
d.
t
. The risky asset o
t
has a drift
j
c
j
a
= c 0, and its diusion is such that no
c
= o
a
. The parameter n may vary by
industry, and can be interpreted as the weight of the stocks of the corresponding industry on
the market portfolio. We assume n = 1 to ease on notation.
17
We use the traded assets 1
t
and o
t
to dene a risk neutral measure, under which the demand shock A
t
follows a geometric
Brownian motion with drift r c and volatility o
a
.
18
The value of the rm \
)t
for any investment strategy
)
= r
)
;
)
contains two well known
components. The rst is the value of a growing perpetuity of cash ows generated by its assets
in place. The second is the value of its investment opportunities or growth options. This
provides the standard prediction in real options models that the value of the rm depends on
its lifestage. By considering \
)t
under imperfect competition, we allow for a third component
which reects the impact of rms strategic interaction on their value.
In our model, all rms sell their products at the common market price j
t
. Since the
investment of any rm aects the market value of all rms through the market price j
t
, rms
have incentives to invest strategically. At any point in time, the market price j
t
at which
rm , sells its production depends the capacity decisions of all its competitors. Whenever
a competitor of rm , invests, the market price j
t
goes down, and the current and expected
future prots of rm , are also lower.
To formalize these arguments, we denote by

)t
the expected change in instantaneous
prots of rm , due to investments by other rms before rm , invests. We denote by
+
)t
the expected change in instantaneous prots of rm , due to investments by other rms after
rm , invests.
Proposition 1 [Firm value under imperfect competition] The value of rm , for any invest-
ment strategy
)
= r
)
;
)
is given by
\
)t
=
_

I
c
+

I
c
+
_
_

+
I
c
+

+
I
c

I
c
_

AI=a

)1
_
_
AI
a

_
u
if A
t
6 r
)

+
I
c
+

+
I
c
if A
t
r
)
(5)
where 1 is dened in the Appendix.
17
The assumption that n = 1 does not aect any of the qualitative implications in the paper. Yet a
calibration for dierent industries should take into account that each industry has a dierent weight n in the
market portfolio.
18
See Carlson et al (2004). The dynamics of the demand shock under the risk neutral measure are dAI =
(v c) AI oiAId :I, where :I = :I

s
r
x
t.
8
Proof. See Appendix A.
The value of any rm , as stated in Proposition 1 depends on the assets in place of rm ,,
its growth opportunities, and the investments of its competitor. The fact that the investments
of the competitor of rm , also aect \
)t
is reected in the expected reductions in future prots
denoted by

I
c
and

+
I
c
, in (5). The signs and magnitudes of the terms

)t
and
+
)t
depend on rms investment strategies in equilibrium. In particular, the signs of

)t
and

+
)t
depend on whether rms invest sequentially or simultaneously.
We illustrate numerically how strategic interaction aects rm value in Figure 1. For this
sake, we simulate multiple paths of the Brownian demand shocks, we compute rms values
using the denition in Proposition 1, and we report the average rm value at each instant t.
19
In Figure 1, we consider the special case in which rms invest sequentially, such that rm 1
invests earlier than rm '. A strategy in which r
1
< r
A
need not be an equilibrium outcome;
other equilibria may exist where r
1
= r
A
or r
1
r
A
. As we discuss below, however, there
exists an equilibrium outcome in which r
1
< r
A
.
20
In the example of Figure 1, the value of each rm goes above the value of its assets in
place when its own growth option is in the money, and yet it goes below the value of its assets
in place when its competitor is about to invest. This second eect is entirely due to rms
strategic interaction. In the rst panel of Figure 1, we note that when r
1
< A
t
< r
A
and
rm ' is about to invest, the value of rm 1 is dampened by the expected reduction in prots
once rm ' invests

1I
c
< 0. Similarly, in the second panel of Figure 1, when A
t
< r
1
and
rm 1 is about to invest, the value of rm ' is dampened by the expected reduction in prots
once rm 1 invests

LI
c
< 0.
1.3 Equilibrium investment strategies
The basic model has two alternative industry equilibria in pure strategies: a sequential equilib-
rium and clustering equilibrium. We denote by
c
)
the investment strategy of any rm , in the
sequential equilibrium with leaders and followers. We denote by
c
)
the investment strategy
of any rm , in the clustering equilibrium in which rms invest simultaneously.
Anticipating, the equilibrium outcome depends on the cross sectional dierences in rms
production technologies. With = 2 rms, these cross sectional dierences are summarized
by the standard deviation of rms instantaneous marginal costs of production or o
c
=
jc
L
c
1
j
2
.
We characterize the equilibrium strategies as a function of o
c
. We elaborate on the intra-
19
We provide further details of the parameters used in numerical examples in Appendix J.
20
In Figure 1, we actually depict rms values under the sequential equilibrium strategies I
s

described in
Section 1.8.
9
industry value spread as a general measure of intra-industry heterogeneity in Section 1.4.
1.3.1 Equilibrium concept
The equilibrium concept is Bayes-Nash. The state of the industry is described by the history of
the stochastic demand shock A
t
. At any point in time, a history is the collection of realizations
of the stochastic process A
c
, : 6 t, and the actions taken by all rms in the industry. The
investment strategy
)
maps the set of histories of the industry into the set of actions r
)
;
)

for rm ,. Before investment, rm , responds immediately to its competitors investment


decisions. This yields Nash equilibria in state dependent strategies of the closed-loop type.
21
Upon investment, rm , cannot take any other action.
We follow Weeds (2002) and we assume that rms follow Markov strategies such that their
actions are functions of the current state A
t
only. As discussed in Weeds (2002), other non-
Markov strategies may also exist; however, if one rm follows a Markov strategy, the best
response of the other rm is also Markov. We consider the set of subgame perfect equilibria
in which each rms investment strategy, conditional on its competitors strategy, is value
maximizing. A set of strategies that satises this condition is Markov perfect. The initial
demand shock A
0
is suciently low to focus on equilibria in pure strategies.
22
Subgame
perfection requires that each rms strategy maximizes its value conditional on its competitors
strategy.
1.3.2 Solution approach
The solution approach relies on sorting conditions and incentive compatibility constraints
(ICCs). The sorting conditions of the multiple action strategy r
)
;
)
indicate which rms in
the industry have the comparative advantage to invest earlier and have a larger market share
than their competitors. When rms dier in their future costs of production c
)
, we prove in
Appendix B that for any strategy more ecient rms nd it less costly to invest earlier and
more since
0
0c

_
0\
I
0a

_
0,
0
0c

_
0\
I
0

_
< 0 (6)
21
A closed-loop equilibrium is a Nash equilibrium in state-dependent strategies. See Chapter 13 of Fundenberg
and Tirole (1991), Weeds (2002) and Back and Paulsen (2009) for related discussions on closed-loop strategies.
22
When rms are identical, the equilibrium may involve mixed strategies, whose formulation is complicated
by the continuous time nature of the game, as noted by Fundenberg and Tirole (1985) and Weeds (2002). When
rms have dierent production technologies, however, Mason and Weeds (2010) shows that a sucient condition
to avoid these concerns is to assume that A0 is suciently low. Ao is assumed strictly lower than the lowest
optimal investment threshold in the industry. Hence A0 < r
s
1
for ` = 2; we dene r
s
1
later on.
10
The sorting conditions in (6) have important implications for rms strategic behavior.
First, (6) implies that there are no sequential equilibria in which less ecient rms invest
earlier. Since rm 1 has a comparative advantage to invest earlier and more, rm ' does not
become a leader in equilibrium even if it has incentives to preempt rm 1. Consequently, rm
1 is the only potential leader when rms invest sequentially. Second, the sorting conditions
imply that rms invest simultaneously when the more ecient rm has incentives to do so. If
rm 1 does not have an incentive to become a leader, neither does rm ', whose ability to
invest earlier and more is comparatively lower.
In equilibrium, we account for rms incentives to preempt each other using ICCs. Due to
the dierences in rms production technologies, the sorting conditions in (6) show that more
ecient rms with lower marginal costs of production c
)
nd it less costly to invest earlier and
more than their competitors. Yet less ecient rms may still want to invest as if they had
lower future costs of production. In particular, rm ' has incentives to invest earlier and
more than rm 1 whenever its value as a leader is higher than its value as a follower.
To express this intuition more formally, we denote by \

A
is the value of rm ' in a
Stackelberg game in which rm 1 is the leader by assumption; and we denote by r

1
the
investment threshold of rm 1 in such Stackelberg game.
23
We denote by

\
A
the value of
the less ecient rm ' when it deviates and pursues the optimal investment strategy of rm
1 as a leader.
24
We conclude that rm ' has incentives to become a leader whenever

\
A

AI=a

1
> \

A
[
AI=a

1
(7)
The inequality in (7) provides an upper bound o
c
such that rm ' has no incentives to
become a leader and preempt rm 1 if o
c
o
c
. When o
c
o
c
, the investment strategy of rm
1 is that of a standard Stackelberg game in which rm 1 always invests rst. Conversely, when
o
c
< o
c
, the inequality in (7) imposes an incentive compatibility constraint to the maximization
problem of rm 1.
Given (7), rm ' is indierent between following the strategy of rm 1 or following its
own when its value is the same under the two alternative strategies. The corresponding
complementary slackness condition is given by
`
_
\
c
A

\
c
A
_

AI=a
s
1
= 0 (8)
23
The Stackelberg strategy of rm 1 is such that I

= I
s

(` = 0).
24
Due to the sorting conditions, the strategy to deviate in timing only is dominated by the strategy to deviate
in both timing and capacity.
11
where the multiplier ` in (8) relates to Posner (1975), and measures to what extent the contest
for monopoly power between rms 1 and ' hinders the value of rm 1.
25
Put together, the conditions in (6) and (7) imply that there are two dierent types of
sequential equilibria. The exists one sequential equilibrium in which o
c
< o
c
and ` 0; and
there exists an alternative sequential equilibrium in which o
c
o
c
and ` = 0. Since the focus
of our paper is on the impact of strategic interaction on investment and expected returns, we
assume for simplicity o
c
is suciently low such that (7) holds throughout the paper. The
sequential equilibrium in which ` = 0 has qualitatively the same properties of that in which
` 0.
As a remark, we note that the impact of the assumption in (7) on the equilibrium outcome
is captured through the multiplier ` 0. The threshold o
c
implied by (7) and the multiplier `
in (8) are mechanically related, and the economic intuition behind both concepts is the same.
o
c
denes the threshold up to which rm ' has incentives to preempt rm 1; conversely,
` captures how costly it is for rm 1 to deter the investments by rm '. We provide the
comparative statics of ` in Section 3.
26
1.3.3 Equilibrium outcome
The equilibrium outcome depends critically on the intra-industry heterogeneity in rms mar-
ginal costs of production, which is captured by o
c
. To see this, we note that rm 1 may
become a market leader, enjoy early monopoly rents and yet pay the shadow cost of preemp-
tion ` 0. Alternatively, rm 1 may allow the follower to invest simultaneously, attain lower
duopoly rents from the start, and yet avoid any cost of preemption. Hence rm 1 becomes a
leader in equilibrium if and only if o
c
is suciently high.
We denote o
c
the minimum value of o
c
for which rm 1 nds it optimal to become a leader
in equilibrium. The threshold o
c
obtains when rm 1 is indierent between pursuing the
investment strategies
c
1
and
c
1
at A
t
= r
c
1
, such that
\
c
1
[
AI=a
s
1
= \
c
1
[
AI=a
s
1
(9)
where (9) is evaluated at r
c
1
since the rm 1 invests earlier in the sequential equilibrium (i.e.
r
c
1
< r
c
).
We summarize the main properties of rms equilibrium strategies in Proposition 2. We
denote by :
)t
=

I
jI
0 the mark-up in prots of rm , at time t.
25
Due to the sorting conditions of the game in (6), the ICC of rm A in (8) is binding as long as (7) holds.
The ICC of rm 1 is not binding in equilibrium.
26
In Section 3, we discuss the comparative statics of ` with respect to -, j
i
, oi and `. We also provide the
comparative statics of ` with respect to oc in Figure 1.
12
Proposition 2 [Investment under imperfect competition] The subgame perfect industry equi-
librium for = 2 with c
1
< c
A
is such that
if o
c
< o
c
, rms invest simultaneously, r
c
)
= r
c
,
c
A
<
c
1
, and
c
)t
= 0;
if o
c
> o
c
, rms invest sequentially, r
c
1
< r
c
1
,
c
A
<
c
1
, and
c
)t
6 0; and
if A
t
r
c
1
, o
c
n,t
< o
c
n,t
and HHI
c
<HHI
c
.
Proof. See Appendix C.
Proposition 2 characterizes the two alternative subgame perfect equilibria which arise in our
baseline model. When o
c
< o
c
, rms are close competitors, neither rm has incentives to lead,
rms investments cluster in equilibrium, and there is lower standard deviation in mark-ups and
lower concentration. Conversely, when o
c
> o
c
and rms are more distant competitors, rm
1 invests earlier than rm 1, and the industry has higher standard deviation in mark-ups and
it is more concentrated.
In the basic model, the result that the intra-industry standard deviation in mark-ups o
n,t
and the HHI under sequential investment are strictly higher than under simultaneous invest-
ment holds weakly at any point in time, and strictly when A
t
r
c
1
. By assumption and when
A
t
6 r
c
1
, o
n,t
and the HHI of both industries is the same. Similarly, by construction, when
the leader in the more heterogeneous industry invests at A
t
= r
c
1
, it holds that o
c
n,t
and HHI
c
are higher than o
c
n,t
and HHI
c
thereafter.
Figure 2 illustrates the equilibrium outcome of the model as a function of the parameter
o
c
. In the upper left hand side panel, we observe that rm 1 is more valuable in the clustering
equilibrium when o
c
< o
c
, and it is more valuable in the sequential equilibrium otherwise. In
the upper right hand side panel, we note that rm ' is more valuable under simultaneous
investment for any o
c
. Put together, these two results imply that the clustering equilibrium
obtains when o
c
< o
c
, i.e. when both rms strictly prefer to invest simultaneously.
The remaining charts in Figure 2 illustrate how the equilibrium investment strategies of each
rm depend on o
c
. In the middle charts of Figure 2, we observe that rm 1 invests signicantly
earlier under sequential investment (o
c
o
c
) than under simultaneous investment (o
c
< o
c
);
the reverse applies to rm '. Similarly, rm 1 invests more under sequential investment
(o
c
o
c
) than under simultaneous investment (o
c
< o
c
). This explains why in the bottom
left hand side panel of Figure 2 the market share of rm 1 is higher when o
c
o
c
, and also
why the HHI of the industry is higher when o
c
o
c
. Lastly, Figure 2 shows the shadow cost
of preemption for rm 1 or ` is decreasing in o
c
. This result relates to the inequality in (7),
and shows that it is less costly for rm 1 to deter the investment of rm ' as o
c
increases.
13
Table 1 complements Figure 2 as it provides a numerical example of the sequential and
simultaneous investment strategies for the same o
c
. The purpose of this example is two-fold.
First, since we consider all possible investment strategies for the same o
c
, we can compute the
value of all rms under each strategy and predict the equilibrium outcome. In the example,
the equilibrium outcome is such that all rms invest simultaneously since o
c
is relatively low.
Second, Table 1 illustrates numerically how rms optimal investment strategies aect their
values for the same o
c
. When rms invest sequentially, rm 1 behaves more aggressively, and
invests earlier and more compared to Stackelberg games.
27
The binding ICC in (8) makes rm
1 invest more aggressively to deter rm '. Consistent with Figure 2, Table 1 also shows that
rm 1 invests earlier in the sequential equilibrium than in the clustering equilibrium, while
rm ' does the reverse. Firms optimal scale upon investment are such that the standard
deviation in rms mark-ups and the HHI are higher under sequential investment.
Lastly, the example in Table 1 illustrates how the magnitude and sign of the expected
reduction in prots
)t
depend on rms investment strategies in equilibrium. When o
c
< o
c
,
rms invest simultaneously,
c
)t
= 0, and the characterization of rms values is the same as
that predicted for idle rms by Dixit and Pyndick (1994). Conversely, when o
c
o
c
, rms
invest sequentially, and each rm expects a reduction in its prots when its competitor invests
such that
c
)t
_ 0. Table 1 compares such expected reductions in prots
)t
at A
t
= A
0
.
1.4 Imperfect competition, marginal q, and market to book
An alternative interpretation of Proposition 2 is in the context of neoclassical investment
models. Neoclassical investment models such as Hayashi (1982) predict that the optimal
investment of any rm depends exclusively on its own marginal product of capital or = \
1
.
When rms behave strategically, we add that the investment strategy of any rm also depends
on the intra-industry standard deviation in . As we argue below, this result applies not only
to the industry in our basic model, but also to industries with multiple sources of heterogeneity
in rms production technologies.
We denote the intra-industry standard deviation in at A
t
= A
0
by o
q,0
. We evaluate
at A
t
= A
0
below for the sake of simplicity; all we need is to compare rms marginal products
of capital before rms invest (i.e.A
t
< r
c
1
).
Proposition 3 [-theory of imperfect competition] Under imperfect competition, rms invest-
ment strategies are such that:
27
In Stackelberg games, rm 1 does not have to deter rm A from investing earlier as in our model; rm 1
invests earlier than rm A by assumption.
14
if o
q,0
is low, rms investments cluster in equilibrium;
if o
q,0
is high, the rm with the highest at A
t
= A
0
invests earlier and more.
Proof. See Appendix D.
To understand the intuition behind Proposition 3, we consider rst the link between the
sorting conditions of the game, rm type and marginal . When rms dier exclusively in
their marginal costs of production after investment, rm type is given by c
)
, and the more
ecient rm has the ability to invest earlier and more. The marginal product of capital of
any rm , is strictly decreasing in c
)
. Consequently, for any strategy , it holds that rms
with a higher at A
t
= A
0
have the ability to invest earlier and more. Table 1 illustrates
how the more ecient rm 1 has the highest at A
t
= A
0
.
As we discuss in Appendix D, a similar intuition holds in an industry in which rms dier
in their installed capacity before investment 1
)
, and rms have the same marginal cost of
production after investment c. In this case, rms dier in the value of their assets in place
and have the same growth option. The option to invest is relatively more valuable for the
smaller rms, and hence smaller rms are willing to invest earlier and more. Given that the
marginal product of capital is strictly decreasing in 1
)
, those rms with a lower 1
)
have a
higher , and the ability to invest earlier and more.
More generally, the insight behind Proposition 3 is that, under imperfect competition, a
rms comparative advantage to preempt its competitors or type is captured by its marginal
product of capital . This result is particularly useful in industries with multiple sources of
cross sectional heterogeneity, in which it becomes less clear which are the important dierences
in rms production technologies determine rms investment strategies. When rms dier
both 1
)
and c
)
, we show in Appendix D that a rm with a higher has the ability to invest
earlier and more than its peers.
Since we can redene rms type in terms of their , the predictions of the model on industry
dynamics for the general case in which rms dier in 1
)
and c
)
depend on the intra-industry
standard deviation in or o
q,0
. In industries with low o
q,0
, rms invest similar amounts and
their investments cluster in equilibrium. Conversely, in industries with high o
q,0
, the rm
with the highest invests earlier and more than its competitor.
When we consider industries in which rms dier in 1
)
and c
)
, we also nd that the result
in Proposition 2 that industries with high value spread always have higher standard deviation
in mark-ups o
n,t
and higher concentration is not general. For instance, we can consider an
industry with leaders and followers (i.e. high value spread) in which smaller rms invest and
catch up in market share with larger rivals. Over time, this industry may have a lower HHI
15
than another industry with lower value spread.
28
Lastly, we note that the choice of 1
)
and c
)
as the sources of heterogeneity across rms
is just for the sake of convenience. While we solve the game when rms dier in these two
specic dimensions, we argue that Proposition 3 applies to industries in which rms dier in
other aspects as well -i.e. dierences in productivity, investment costs, management skills, etc.
The case in which rms dier in c
)
refers more generally to industries in which rms dier
in their growth opportunities. Conversely, the case in which rms dier in 1
)
refers more
generally to industries in which rms dier in their assets in place.
1.5 The intra-industry value spread
Since rms marginal is not observable, and for the sake of empirical tests, we restate the
results in Proposition 3 in terms of testable implications by considering the identity between
and the market value to book ratio
\
1
. For any strategy and A
t
< r
)
, it is straightforward
to show that

)t
=
\
)t
1
)t

1
-c
_
j

t
1

t
+
_
j
+
t
1
+
t

j

t
1

t
__
A
t
r
)
_
u1
_
(10)
such that the marginal of rm , equals its market to book ratio, minus an additional term
which captures the net present value of the marginal income of rm , per unit of capital.
29
The identity in (10) implies that the relative comparative advantage of rm , to invest
earlier and more than its competitors can be restated in terms of its market to book ratio
\
1
.
Note that even if the market to book ratio of a rm is a noisy proxy of its as seen in (10),
the second term in (10) is common to all rms in the industry. Consequently, the observable
measure of the cross sectional variation in within an industry in the model is such that o
q,t
= o 1
1
,t
.
We refer to o 1
1
,t
as the intra-industry value spread at time t. Our denition is of the value
spread is mechanically related to that in Cohen et al (2003), who dene the value spread as
the standard deviation of the logarithm of rms rms book to market ratios or o
ln
1
1
,t
. Using
a rst order Taylor approximation, we show in Appendix D that
o
ln
1
1
,t
- j1
1
,t
o 1
1
,t
(11)
Corollary 1 Under imperfect competition, rms investment strategies depend on their own
market to book ratio, their current and future expected extraordinary prots, and the intra-
industry value spread.
28
We elaborate on this example in Appendix D.
29
The identity in (10) indicates that 1I may vary by rm type since it applies all types of industries, including
those in which rms dier in their installed capacity. See Appendix D for an example of these industries.
16
Proof. See Appendix D.
Corollary 1 has several implications which are relevant for empirical tests. The rst relates
to the underlying determinants of corporate investment. Expression (10) is consistent with
Hayashi (1982) and shows that investment decisions depend on rms market to book ratio and
their market power. It is also consistent with the empirical literature on investment equations,
which uses a rms market to book ratios as a good proxy for growth opportunities.
Since we derive rms investment strategies in an industry equilibrium, we add however
that the intra-industry value spread captures to what extent rms investment prospects are
aected by rms strategic interaction. A marginal increase in the capacity of rm , does not
only reduce the future market price for rm ,, as seen in the second term of (10); it also aects
the market share and market prices of all its competitors. As a result, the model predicts that
o 1
1
,t
is signicant in explaining investment.
The insight that rms preemptive behavior is aected by the intra-industry value spread is
consistent with earlier studies on mergers. The theoretical study on mergers by Jovanovic and
Rousseau (2002) suggests that rms with high market to book are more capable of preempting
others. Similarly, Andrade et al (2001) document that rms with high market to book regularly
acquire rms with low market to book.
Lastly, a relevant empirical implication of the model is that standard measures of com-
petition such as the intra-industry standard deviation in mark-ups o
n,t
and the HHI prove
insucient to capture the degree of competition in an industry because they are static. Firms
investment decisions depend not only on the current HHI, but also on the expected future
changes in the HHI. In contrast, the intra-industry value spread is an observable industry
characteristic which captures the unobserved heterogeneity in rms production technologies
over time.
2 Industry risk dynamics
In this section, we study the impact of imperfect competition on rms expected returns by
analyzing their exposure to systematic risk or betas. In the model, the conditional CAPM
holds since rms are subject to a single source of systematic risk. The source of systematic
risk is given by the demand shock in (1). The riskless rate of return r is exogenously specied,
and the market price of risk is constant and exogenously given.
While several papers in the literature consider single factor models to explain the evidence
on the cross section of returns,
30
Fama and French (1992) and subsequent papers show that
30
See, for instance, Berk et al (1999), Carlson et al (2004), Zhang (2005), and Aguerrevere (2009).
17
rms betas are a poor measure of rms exposure to systematic risk, and that asset pricing
models with multiple risk factors may have higher explanatory power. This gives rise to two
related concerns on the asset pricing implications in this section.
The rst concern is whether our results prevail in a framework with multiple sources of
systematic risk. In Appendix E, we discuss an extension with multiple risk factors in which
the testable implications of our single factor model still prevail. The second related concern
is whether our predictions on rms betas also hold for returns. We address this concern
empirically in Section 4, by testing our asset pricing predictions on both rms betas and
excess returns.
2.0.1 Firms betas
We denote the beta of rm , at time t by ,
)t
. To determine ,
)t
, we follow Carlson et al (2004)
and infer expected returns from replicating portfolios composed of a risk free asset and a risky
asset that exactly reproduce the dynamics of rm value. The proportion of the risky asset
held in the replicating portfolio at any time t yields ,
)t
.
Proposition 4 [Firms betas under imperfect competition] For any strategy , the beta of rm
, at time t is given by
,
)t
= 1 + I
t
( 1)
_
1
1
c

)t
1
)t

1
)t
\
)t
_
(12)
where I
t
is an indicator function which is equal to 0 if all rms have invested at time t and is
equal to 1 otherwise.
Proof. See Appendix F.
The identity in (12) for rms betas under imperfect competition resembles that in Carlson
et al (2004) for monopolistic rms. As in their paper, the exposure to systematic risk of
any rm , depends on the relative contribution of its own growth opportunities to total rm
value. Under imperfect competition, however, we add that ,
)t
also depends on the growth
opportunities of other competing rms in the industry.
In our model, the terms
)t
and \
)t
in (12) are aected by rms strategic interaction
in product markets. Whenever one rm in the industry invests, total production increases,
the market price j
t
goes down, and there is a discrete change in the beta of all rms in the
industry. This explains why the indicator function I
t
in (12) equals zero only when all rms
in the industry have invested.
18
2.1 Intra-industry correlation of betas
The identity in (12) relates a rms beta to its instantaneous cashows and market to book
ratios for any strategy
)
. The more interesting aspect of the model is yet how imperfect
competition aects the industry dynamics of expected returns. We predict that rms betas
correlate more positively in industries with low o
c
or low value spread.
Proposition 5 [Intra-industry correlation of betas] The equilibrium dynamics of ,
)t
and

I
\
I
depend on o
c
such that:
if o
c
< o
c
, rms betas correlate positively;
if o
c
> o
c
, the betas of leaders and followers correlate negatively.
Proof. See Appendix G.
To understand the mechanism behind Proposition 5, consider rst the dynamics of ,
)t
when o
c
< o
c
. In industries with low value spread in which rms invest simultaneously, the
value of each rm is larger than the value of its assets in place before investment, and equal
to the value of its assets in place thereafter. This implies that ,
c
)t
is higher than one before
investment and equal to one thereafter. This is illustrated numerically in the example of Table
1.
More importantly, the dynamics of ,
c
)t
are qualitatively the same as those that obtain in a
real options model in which rms do not invest strategically, such as Carlson et al (2004). This
holds because strategic interaction has no equilibrium eects when rms are close competitors
(i.e.
c
)t
= 0). In more homogeneous industries, ,
c
)t
increases before the investment of rm
, and decreases thereafter, as if rm , were an idle rm.
Consider now the dynamics of ,
)t
when o
c
o
c
. In industries with high value spread,
there are leaders and followers, and strategic interaction aects the conditional dynamics of
betas since rms expect a reduction in their prots once their competitors invest (i.e.
c
)t
6 0).
Under sequential investment, when one rm expects to increase its market share, the other
expects a reduction in its own. The betas of rm 1 and rm H correlate negatively in
industries with high value spread.
We illustrate the dynamics rms betas in the sequential and clustering equilibria in Figure
3. For this sake, we simulate multiple paths of the Brownian demand shocks, we compute
rms betas using the denition in (12), and we report the average rm beta at each instant t.
We also consider the limit case in which o
c
= o
c
; the limit case in which o
c
= o
c
highlights
that even for the same set of parameters, and when all rms in the industry are subject to
19
the same systematic shock, the intra-industry correlation of betas depends on the dynamics of
investment. The corresponding average intra-industry correlation in rms betas equals 0.9
when rms invest simultaneously, and equals 0.9 when rms invest sequentially.
31
2.2 Predicting average industry betas
Consistent with Carlson et al (2004), rms betas increase before rms exercise their own
investment opportunity, and decrease upon investment. It need not follow, however, that the
same dynamics apply to the average industry beta j
o,t
. The stylized real options prediction
that a rms beta increases before investment and decreases upon exercise only applies to j
o,t
in industries with low value spread.
In industries with low value spread, we nd that the average industry beta is more pre-
dictable, since a period of high market to book ratios, high investment, and high betas is
followed by a period of lower market to book ratios, lower investment and lower betas. This
pattern does not hold in industries with high value spread, in which the dynamics of the aver-
age industry beta are not representative of the dynamics of the dynamics of the beta of each
rm in the industry.
Corollary 2 In industries with low value spread, periods of high market to book ratios, high
investment and high betas are followed by periods of lower market to book ratios, lower invest-
ment and lower betas.
In industries in which o
c
< o
c
, Corollary 2 implies that the threshold r
c
provides a common
reference point for all rms in the industry at which they trigger their investments. Hence j
c
o,t
increases before all rms invest and decreases sharply upon investment. Conversely, in indus-
tries in which o
c
> o
c
, rms betas are negatively correlated, and the industry has multiple
investment thresholds. The discrete changes in j
c
o,t
when rms add capacity have no clear
sign; they are either positive or negative depending on whether it is rm 1 or rm ' invests.
Combining Proposition 2 and Corollary 2, a related testable implication is that average
industry expected returns should be more predictable in less concentrated industries, unless
these industries are undergoing deep transitions from high to low competition or viceversa.
The empirical evidence in Hoberg and Phillips (2010) is consistent with this result. Hoberg
and Phillips (2010) observe that US rms returns are more predictable and comove more
positively in less concentrated industries.
32
31
In line with empirical practices and our tests in Section 4, we compute the average intra-industry correlation
in betas over the entire time span of Figure 8.
32
Notably, Hoberg and Phillips (2010) study concentrating and deconcentrating industries separatedly.
20
In particular, Hoberg and Phillips (2010) nd that in less concentrated industries periods
of high market to book ratios, high returns, high betas and high investment, are followed by
periods of lower market to book ratios, lower investment, lower returns and lower betas. They
also nd no predictable pattern in the average industry returns of industries with high HHI.
Our model explains each of their ndings given that rms strategic interaction only aects
the dynamics of j
c
o,t
in more concentrated industries.
2.3 The value spread and the cross section of betas
In a related empirical study, Cohen et al (2003) show how the value spread of the entire
cross section of US public rms depends on the standard deviation in expected returns, and
the standard deviation in rms prots. We derive a similar identity for the intra-industry
standard deviation of the logarithm of rms book to market ratios o
ln
1
1
,t
.
Proposition 6 [Intra-industry value spread and o
o,t
] Under imperfect competition, o
ln
1
1
,t
is
mechanically related to o
o,t
and o
r
1
,t
such that
o
2
ln
1
1
,t
-
t
o
2
o,t
o
2
ln
r
1
,t
2j
t
(13)
where
t
is dened in Appendix H, and j
t
is the covariance between ln
1
\
and ln

1
.
Proof. See Appendix H.
Given (10), Proposition 6 indicates that the intra-industry value spread is mechanically
related to the cross section of expected returns and the cross sectional dierences in rms
cashow to assets ratios. The fundamental dierence between (13) in Proposition 6 and the
identity for the value spread in Cohen et al (2003) is that our measure is purely intra-industry,
and it is based on the premise that rms betas are interrelated under imperfect competition.
Cohen et al (2003) show that the value spread of US public rms in multiple industries
is mostly explained by the intra-industry cross sectional variation in returns, and the intra-
industry cross sectional variation in rms earnings. This result is highly consistent with
the predictions of our model. We argue that the intra-industry value spread determines the
dynamics of rms investments and rms betas in equilibrium.
Put together, the ndings in Cohen et al (2003) and the predictions of our model provide
an empirical interpretation on the role of book to market sorts commonly used in the asset
pricing literature.
33
Given that the value spread is predominantly intra-industry, market
to book ratios reect the relative ability of one rm to increase market share relative to its
33
See Fama and French (1992) and related papers.
21
competitors. Consequently, the book to market sorts typically used by asset pricers aggregate
stocks into portfolios according to rms relative position in the industry.
A related and yet mechanical testable prediction of Propositions 2 and 6 is that industries
with high value spread have higher intra-industry standard deviation in betas or excess returns.
We nd supporting empirical evidence on this in Section 4. Figure 3 shows that, even at the
threshold o
c
= o
c
, the cross sectional variance of rms betas is higher in industries with
leaders and followers.
3 Inter-industry testable implications
In Sections 1 and 2, we keep all exogenous parameters related to the industrial organization
constant except for o
c
. As a result, our focus is on how the intra-industry cross sectional vari-
ation in rms technologies aects the industry dynamics of investment and risk. In practice
however, industries dier in multiple characteristics which also inuence these dynamics. We
hereby study how inter-industry dierences in the demand growth j
a
, demand volatility o
a
,
demand elasticity -, and the number of rms aect the main predictions of our model.
3.1 Product Market Demand
The comparative statics on the industry dynamics of investment and betas with respect to -,
j
a
, and o
a
are twofold. On one hand, we characterize how industry dynamics are aected by
-, j
a
, and o
a
directly. These results provide new testable implications with respect to our
results in Sections 1 and 2.
On the other hand, we study how the industry dynamics of investment and betas depend
on -, j
a
, and o
a
indirectly through the intra-industry value spread, which is itself a function
of -, j
a
, and o
a
. This second set of results is useful to understand whether our testable
predictions in Sections 1 and 2 still hold when we compare industries with dierences in -, j
a
,
and o
a
.
We illustrate the direct and indirect implications on product market demand in the context
of the basic model. In the basic model, the dynamics of investment and expected returns
depend critically on the threshold o
c
. The threshold o
c
at which rm 1 is indierent between
pursuing strategies
c
1
and
c
1
depends on the underlying parameters of market demand. In
particular, o
c
is decreasing in -, and increasing in demand growth j
a
and volatility o
a
. We
illustrate these properties numerically in Figure 4.
The economic rationale behind these results is consistent with the industrial organization
literature. Ivaldi et al (2003) show that, for a xed number of rms, tacit coordination is more
22
likely in growing industries, in which current prots are low relative to future prots. Boyer et
al (2001) suggest that demand uncertainty induces coordination as it boosts the growth option
value of all rms. Similarly, Ivaldi et al (2003) and Motta (2004) suggest that coordination is
more likely in industries with low demand elasticity.
Figure 4 also provides comparative statics on the shadow cost of preemption `. These
comparative statics show to what extent the underlying determinants of demand aect the
preemptive behavior of rm '. ` is increasing in j
a
and o
a
; since j
a
and o
a
boosts the
growth option value of all rms, leaders nd it more costly to deter the investments of their
competitors in industries with high demand growth and high demand volatility. ` is decreasing
in -; since an increase in - implies a reduction in the market power of rm ', leadership is
relatively less costly for rm 1 as - increases.
The direct testable implication of the comparative statics in Figure 4 is that in industries
with low demand elasticity -, high demand growth j
a
, and high demand volatility o
a
rms
have more similar investment patterns, rms betas correlate more positively, and the average
industry beta is more predictable. Furthermore, the stylized assumption in production based
asset pricing that a rms beta is only aected by its own investment decisions is incomplete
in industries with high demand elasticity -, low demand growth j
a
, and low demand volatility
o
a
.
The comparative statics of o
c
also have an important indirect empirical implication. In
particular, rms investment strategies and betas correlate more positively in industries with
low value spread not only when we compare industries with a dierent o
c
(as in Proposition
2), but also when we compare heterogeneous industries with dierences in -, j
a
and o
a
.
To see this, we note that the threshold o
c
is decreasing in -; industries are more likely
to have leaders and followers as - increases. By construction, and all else equal, the intra-
industry value spread is higher under sequential investment. Hence our prediction of more
comovement in betas in industries with low value spread holds for industries with dierent
demand elasticities -.
The threshold o
c
is also increasing in j
a
and o
a
; industries are less likely to have leaders
and followers as j
a
and o
a
increase. Once again, and all else equal, we know that the intra-
industry value spread is higher under sequential investment. Hence our prediction of more
comovement in betas in industries with low value spread holds for industries with dierences
in j
a
and o
a
.
The practical implication of these results is that we need not control for dierences across
industries in -, j
a
and o
a
when we test whether the dynamics of rms betas are more similar
in industries with low value spread. This facilitates our task of testing the main predictions
23
of the model in Section 4.
3.2 Industries with N > 2 rms
The model in Sections 1 2 considers the simpler case of = 2. In practice, however,
industries may have more than two rms. We thus explore how the number of rms in the
industry aects the predictions of the baseline model. By construction, a higher number of
rms increases the set of potential industry equilibria. When 2, some rms may invest
sequentially, while some others may nd it optimal to cluster instead.
The approach to solve the game is the same as in Section 1, and relies on sorting conditions
and ICCs. The sorting conditions of the game are the same for any value of , and hence
also in oligopolies rms with higher market to book have the ability to invest earlier and more.
Furthermore, the sorting conditions facilitate the analysis when 2, insofar they constrain
the set of possible equilibria to those in which rms with high market to book invest earlier or
in tandem with rms with lower market to book. We elaborate on the solution approach in
Appendix I.
The overall prediction in Proposition 3 for = 2 that rms investment patterns are more
similar in industries with low value spread also holds with 2. In oligopolies, rms decide
when and how much to invest based on the dierence between their market to book ratio and
that of their competitors. Hence when rms have more similar market to book ratios, rms
have more similar investment patterns, and the intra-industry comovement in rms betas is
higher.
As we discuss in Appendix I, however, while the intra-industry value spread is key in
determining the equilibrium outcome, in oligopolies the entire distribution of rms market
to book ratios is necessary to characterize the equilibrium. Intuitively, in an industry with
rms, the intra-industry value spread captures how distant each rm is from the average
rm in the industry. In the model, however, each rm is constrained by the behavior of their
closest competitor.
34
Consequently, if the distribution of rms market to book ratios is fairly
skewed, the equilibrium outcome may combine early investments by leaders with late, clustered
investments by laggards.
We illustrate mechanics of the case of 2 rms with a numerical example for = 3
rms. We label rms by 1, ' and 1, where rm 1 can be interpreted as a single rm or more
broadly as a fringe of rms with very similar costs of production whose investments cluster in
equilibrium. We assume that rms have the same installed capacity 1 before investment, and
uniformly distributed marginal costs of production after investment such that c
1
< c
A
< c
1
.
34
This is the result of having sorting conditions and binding ICCs in equilibrium.
24
We report all potential equilibria in pure strategies in Tables 2 and 3. Firms may invest
sequentially or simultaneously as in the case of = 2 (Table 2). Furthermore, two of the
three rms may cluster, and the remaining rm may either lead or follow (Table 3).
The numerical example in Tables 2 and 3 provides several important insights. Consistent
with our ndings as in Section 1, we observe that since the intra-industry value spread is
relatively low, the three rms optimally invest simultaneously (Table 2). Also, just as in
Section 2, since the model has a single stochastic shock, the instantaneous correlation between
the betas of any pair of rms is either 1 or 1. However, the average intra-industry correlation
in rms betas over time is aected by .
To see this, we use the parametrization in the numerical example of Table 2, we compute
rms betas over time using simulated Brownian paths, and we compute the average intra-
industry correlation in rms betas in the fully sequential equilibrium in which = 3 and
r
1
< r
A
< r
1
. We compare our results to those in the fully sequential equilibrium for
= 2 depicted in Figure 3. When = 2, the average intra-industry correlation in rms
betas equals 0.9. When = 3, the average intra-industry correlation in rms betas
is 0.32. The absolute average intra-industry correlation is lower when = 3, since the
betas of two non-investing rms are always positively correlated over time.
35
The numerical example also illustrates how a higher number of rms aects rms
investment strategies and betas. For this sake, we compare our numerical examples for = 2
in Table 1 and = 3 in Table 2.
36
Consistent with Grenadier (2002), Aguerrevere (2009)
and Bulan et al (2009), Tables 1 and 2 show that a higher erodes the values and betas of
all rms in the industry. We add to the their ndings that a higher need not erode the
values and betas of all rms evenly. In heterogenous industries, an increase in aects more
severely the betas of those rms with lower marginal in the industry. The beta of rm ' in
Panel (B) of Table 2 at A
t
= A
0
is lower than the corresponding beta of rm ' at A
t
= A
0
in Panel (C) of Table 1.
We also nd that an increase in need not induce all rms to accelerate investment. In
Grenadier (2002) and in the industries with low value spread in this paper, rms optimally
invest earlier and less as increases. However, in industries with high value spread, an
increase in may induce rms with high to invest earlier and more to preserve their position
as leaders, forcing rms with lower to delay their investment. In Tables 1 and 2, and when
rms invest sequentially (Panels B and A, respectively), rm 1 invests earlier when = 3;
35
When r1 < rL < rT , and rm 1 invests, the betas of rms A and 1 correlate positively as they both
expect a reduction in their prots. When rm A invests, the betas of A and 1 comove positively for the same
reason. Lastly, when rm 1 invests, the betas of 1 and A comove positively as well.
36
Tables 1 and 2 use the same underlying parameters. See Appendix K.
25
conversely, rm ' invests earlier when = 2.
37
4 Investment and risk dynamics in US industries
The model provides qualitative predictions on how rms strategic interaction aect the intra-
industry dynamics of investments and betas. A reasonable concern, however, is whether these
eects are economically signicant. Given the nature of the problem of study, it is reasonable to
argue that strategic interaction plays a bigger role in determining the dynamics of investment
and risk in some industries and not in others.
A natural experiment to tackle this concern would be to calibrate the model to match the
investment and risk dynamics of dierent industries. Complicating the task of calibration,
however, the parameters which characterize the organization of an industry in our model
are empirically unobservable, or require at least a thorough empirical study to infer their
magnitude. These parameters include rms marginal costs of production, rms costs of
investment, and the underlying determinants of product market demand -, j
a
, and o
a
.
We pursue an alternative approach and assess whether the main testable predictions of our
model hold on average for the cross section of US industries. Our tests rely on similar datasets
used in previous studies such as Hou and Robinson (2006) and Hoberg and Phillips (2010). In a
nutshell, our empirical tests provide supporting empirical evidence on the following predictions:
Firms investment strategies are signicantly related to the intra-industry value spread;
Firms betas and excess returns correlate more positively in industries with low intra-
industry value spread; and
Firms betas and excess returns correlate more positively in industries with low intra-
industry standard deviation in mark-ups, and low HHI.
4.1 Working database and summary statistics
We dene an industry by its four-digit SIC code. This is the nest available industry classi-
cation that is available in our merged CRSP-COMPUSTAT dataset. We prefer such measure
as opposed to a broader industry denition since our testable predictions rely on the impact
of rms strategic interaction in product markets.
We include all NYSE, AMEX, and NASDAQ-listed rms in the intersection of the CRSP
monthly returns le and the COMPUSTAT annual le between January 1968 and December
37
See Panel B in Table 1 and Panel A in Table 2.
26
2008. We estimate the beta of equity of each rm as the sum of the coecients of monthly
returns on lagged, lead and contemporary market returns of the stock return of each rm in
the sample. We compute betas at a monthly frequency.
We follow Fama and French (1992) and match each rms CRSP excess return and beta
from July of year t until June of year t + 1 to the corresponding accounting information in
COMPUSTAT for the scal year ending in year t 1. We construct the database at a monthly
frequency. In the empirical tests, we run the empirical tests on the asset pricing implications
of the model at a monthly frequency. We use data at annual frequency to run the empirical
tests on investment equations.
We denote the explanatory variables in our working sample as the equity beta ,; the excess
return 1; the market to book asset ratio
\
1
; the book leverage ratio
1
1
; the market to book
equity ratio
\ 1
11
; the cashow to assets ratio

1
; the investment rate
1
1
; and the mark-up in
prots :. We elaborate on the database construction in Appendix J.
To test our predictions on industry dynamics, we construct a measure of comovement which
captures the average pairwise correlation in rms investments, market to book equity ratios,
market to book asset ratios, betas and excess returns by industry. We denote the intra-
industry comovement of variable r in month-year t as .
a,t
. The methodology follows Khanna
and Thomas (2009). For each variable, and for each month, we compute .
a,t
as the average
of the pairwise correlations of variable r for each unrepeated pair of rms within the industry.
We compute the pairwise correlation coecients using a rolling window of 60 months.
38
We also consider the two static measures of competition discussed in Section 1. One is
the intra-industry deviation in mark-ups or o
n,t
, which we construct using the COMPUSTAT
annual les. The other is the logarithm of the HHI index by four-digit SIC code reported by
the US Census Bureau or lnHHI, which is limited to manufacturing industries only.
39
In line
with Ali et al (2009), we do not compute the HHI using CRSP-COMPUSTAT sales data since
such index is not highly correlated with the US Census Bureau concentration index.
40
In Table 4, we report the working sample averages of all variables by rm-year and industry-
year. We report the working sample statistics by industry-year for the intra-industry standard
deviations o
a,t
, and the measures of intra-industry comovement .
a,t
.
38
The length of the rolling window is the same as the one we use to compute rms betas. See Appendix J
for details.
39
We use logs for scaling purposes only; results are qualitatively the same when we use the HHI.
40
In untabulated tests, we also consider the proxy of the HHI recently proposed by Hoberg and Phillips (2010).
While we obtain similar results, the industry denition of their proxy is noisy for the sake of our study, as it
relies on 3-digit SIC codes.
27
4.2 Empirical approach
We apply the same empirical methodology to test all our implications on investment and risk.
Given that in our model the underlying industry determinants of demand and the number of
rms are constant, we run all tests using cross sectional regressions as in Fama and MacBeth
(1973). To account for serial correlation, we consider Newey West standard errors.
For the sake of completeness, we have also run all empirical tests in this section using OLS
regressions with year dummies. We do not tabulate these results as they are all very similar
to those reported using Fama and MacBeth (1973) regressions.
The model further assumes that rms are unlevered, while most rms in our working sample
are levered. We thus run our tests on (equity) betas and stock returns using two alternative
denitions of the intra-industry value spread: one based on the asset value spread or o 1
1
,t
,
and another based on the equity value spread o 1 T
1T
,t
. We elaborate further on how leverage
aects (equity) betas and excess (stock) returns in Appendix K.
4.3 Investment
We rst evaluate the prediction that rms investment decisions relate signicantly to the intra-
industry value spread. We provide the corresponding empirical evidence in Table 5. We nd
that the intra-industry value spread is signicant in explaining investment, both at the rm
level (Panels B and C) and industry level (Panel E and F). We obtain similar results when
using the intra-industry asset value spread (Panels B and E), and the intra-industry equity
value spread (Panels C and F).
4.4 Betas and excess returns
The null hypotheses of the model on industry risk dynamics rely on two important working
assumptions. The rst is that rms investment decisions aect their exposure to systematic
risk. The second is that the single factor model in our paper provides relevant testable
implications on the cross section of returns. We therefore run simple preliminary tests to
assess whether these assumptions are fairly consistent with the evidence in our working sample.
We rst focus on the underlying assumption that rms investment decisions aect their
exposure to systematic risk. The related testable implication is that the intra-industry dynam-
ics of betas are signicantly related to the corresponding dynamics of investment and market
to book ratios. We provide the supporting empirical evidence in Panels A to F of Table 6.
Consistent with the model, the intra-industry comovement in betas and excess returns are
signicantly related to the intra-industry comovement in investment. Similarly, the intra-
28
industry comovement in betas and excess returns are signicantly related to the intra-industry
comovement in market to book ratios.
The second important implicit assumption is that the predictions of our single factor model
may apply to both betas and excess returns. Like other papers, we acknowledge that our single
factor model does not explain why there exist value and size premia in excess returns. However,
both in the model and in the data, the intra-industry comovement in betas is signicantly
related to the intra-industry comovement in excess returns. The average R-square in Panel G
of Table 6 indicates that the intra-industry comovement in betas explains on average 37% of
the intra-industry comovement in excess returns.
41
4.5 Industry dynamics and competition
The main asset pricing implication of our model is that the intra-industry dynamics of be-
tas and excess returns are more similar in industries with low intra-industry value spread.
Consequently, we predict a negative and signicant correlation between the intra-industry
comovement in betas and excess returns, and the intra-industry value spread.
Table 7 provides the corresponding empirical evidence using Fama and MacBeth (1973)
regressions. Consistent with the model, we nd a negative and signicant correlation between
the intra-industry comovement in betas and the intra-industry value spread (Panels A and
B). We also nd a negative and signicant relation between the intra-industry comovement
in excess returns and the intra-industry value spread (Panels E and F).
Proposition 2 further suggests that those industries with low value spread may also have
low standard deviation in mark-ups, and low HHI; this holds when the intra-industry value
spread is positively correlated with the intra-industry standard deviation in mark-ups, and
with the HHI. In our dataset, we observe a signicant and positive correlation between the
intra-industry asset value spread, the equity value spread, the standard deviation in mark-ups,
and the log of the HHI. The pairwise correlation between the asset (equity) value spread and
the dispersion in mark-ups is 17.55% (resp. 19.73%). The pairwise correlation between the
asset (equity) value spread and the logarithm of the HHI is 16.51% (resp. 16.46%).
The corresponding testable implication is that of a negative and signicant correlation
between these static measures of competition and the intra-industry comovement in betas or
excess returns. As suggested by the model, we report in Table 7 a negative and signicant
relation between the comovement in betas, and the static measures of competition given by
o
n,t
and lnHHI (Panels C and D). We also nd a negative and signicant relation between
41
The corresponding adjusted R-square using an OLS regression with year dummies is 80%.
29
the comovement in excess returns, and the static measures of competition given by o
n,t
and
lnHHI (Panels G and H).
The evidence in Table 7 complements the empirical ndings in Hoberg and Phillips (2010).
Using an alternative empirical approach, Hoberg and Phillips (2010) conclude that returns
comove more positively in industries with low HHI. We report that the intra-industry co-
movement in betas and excess returns is higher in more competitive industries, in which rms
have more similar market to book ratios, more similar operating mark-ups, and more similar
market shares. We obtain the same qualitative results in Table 7 when we run OLS regressions
with year dummies.
5 Conclusion
This paper provides a model of industry equilibrium to study how strategic interaction aects
the intra-industry dynamics of corporate investment and expected returns. Under imperfect
competition, the fundamental insight in the asset pricing literature that a rms exposure to
systematic risk or beta is aected by its own investment decisions is incomplete. In industries
with high value spread, low demand growth, low demand volatility, and high demand elasticity,
a rms beta is sometimes better explained by the investment of its peers.
In imperfectly competitive industries, we predict that the investment strategy and exposure
to systematic risk of each rm is aected by marginal product of capital of all its competitors;
this suggests why the value spread in Cohen et al (2003) is predominantly intra-industry. We
nd theoretically and empirically that rms betas and excess returns correlate more positively
in industries with low value spread. We also explain why rms betas and excess returns
correlate more positively in industries why low HHI, and low intra-industry standard deviation
in mark-ups. Our ndings on industry dynamics and the HHI are highly consistent with the
evidence in Hoberg and Phillips (2010).
To conclude, we highlight that the fundamental insight of our paper is that product markets
have non trivial eects on rms investment decisions and their expected returns. In this
context, dynamic models of strategic interaction typically studied in the industrial organization
literature become a useful tool to explain empirical regularities in the cross section of returns.
The model can be extended in many ways.
30
References
Ali, Ashiq, Sandy Klasa, Eric Yeung, 2009, The limitations of Industry Concentration Measures Constructed
with Compustat Data: Implications for Finance Research, Review of Financial Studies, 3839-3871.
Andrade, Gregor, Mark Mitchell, and Erik Staord, 2001, New Evidence and Perspectives on Mergers, Journal
of Economic Perspectives 15, 103-120.
Aguerrevere, Felipe. 2009, Real Options, Product Market Competition, and Asset Returns, Journal of Finance
64, 957-983.
Back, Kerry and Dirk Paulsen, 2009, Open Loop Equilibria and Perfect Competition in Option Exercise Games,
Review of Financial Studies 22, 4531-4552.
Bena, Jan and Lorenzo Garlappi, 2011, Technological Innovation, Industry Rivalry, and the Cross Section of
Expected Returns, Working Paper.
Berk, Jonathan, Richard Green and Vasant Naik, 1999, Optimal Investment, Growth Options and Security
Returns, Journal of Finance 54, 11531607.
Bolton, Patrick, and David Scharfstein, 1990, A Theory of Predation Based on Agency Problems in Financial
Contracting, American Economic Review 80, 93-106.
Boyer, Marcel, Pierre Lasserre, Thomas Mariotti, and Michel Moreaux, 2001, Real Options, Preemption, and
the Dynamics of Industry Investments, Working Paper.
Bulan, Laarni, Christopher Mayer and Tsuriel Somerville, 2009, Irreversible investment, real options, and
competition: Evidence from real estate development, Journal of Urban Economics 65, 237-251.
Bustamante, Maria Cecilia, 2012, The Dynamics of Going Public, Review of Finance 16, 577-618.
Carlson, Murray, Adlai Fisher, and Ron Giammarino, 2004, Corporate Investment and Asset Price Dynamics:
Implications for the Cross-Section of Returns, Journal of Finance 59, 2577-2603.
Carlson, Murray, Engelbert Dockner, Adlai Fisher, Ron Giammarino, 2009. Leaders, Followers, and Risk
Dynamics in Industry Equilibrium. Working Paper.
Cho, In-Koo and Joel Sobel, 1990, Strategic Stability and uniqueness in signaling games, Journal of Economic
Theory 50, 381-413.
Cohen, Randolph, Christopher Polk, and Tuomo Vuolteenaho, 2003, The value spread, Journal of Finance 58,
609-641.
Cohen, Randolph, and Christopher Polk, 1996, An investigation of the impact of industry factors in asset pricing
tests, Working Paper.
Dixit, Avinash and Robert Pyndick, 1994, Investment under Uncertainty, Princeton University Press (Boston,
MA.)
Fama, Eugene, and Kenneth French, 1992, The Cross-Section of Expected Stock Returns, Journal of Finance
47, 427-465.
Fama, Eugene, and James MacBeth, 1973, Risk, Return, and Equilibrium: Empirical Tests, Journal of Political
Economy 81, 607-636.
Fulghieri, Paolo, and S. Nagarajan, 1996, On the strategic role of high leverage in entry deterrence, Journal of
Banking and Finance 20, 1-23.
31
Fundenberg, Drew and Jean Tirole, 1985, Preemption and Rent Equalization in the Adoption of New Technology,
Review of Economic Studies 52, 383-401.
Fundenberg, Drew and Jean Tirole, 1991, Game Theory, MIT Press (Boston, MA.)
Garlappi, Lorenzo, 2004, Risk Premia and Preemption in R&D Ventures, Journal of Financial and Quantitative
Analysis 39, 843-872.
Greene, William, 2003, Econometric Analysis. (Pearson).
Grenadier, Steve, 1996, The Strategic Exercise of Options: Development Cascades and Overbuilding in Real
State Markets, Journal of Finance 51, 1653-1679.
Grenadier, Steve, 2002, Option Exercise Games: An Application to the Equilibrium Investment Strategies of
Firms, Review of Financial Studies 15, 691-721.
Hayashi, Fumio, 1982, Tobins Marginal q and Average Q: A Neoclassical Interpretation, Econometrica 50,
213-224.
Hennessy, Christopher, 2004, Tobins Q, Debt Overhang and Investment, Journal of Finance 59, 1717-1741
Hoberg, Gerard and Gordon Phillips, 2010, Real and Financial Industry Booms and Busts, Journal of Finance,
forthcoming.
Hou, Kewei, and David Robinson, 2006, Industry concentration and average stock returns, Journal of Finance
61, 1927-1956.
Ivaldi, Marc, Bruno Jullien, Patrick Rey, Paul Seabright, and Jean Tirole, 2003, The Economics of Tacit
Collusion. Final Report for DG Competition, European Commission.
Jovanovic, Bojan and Peter Rousseau, 2002, The Q-theory of Mergers, American Economic Review 92, 198-
204.
Khanna, Tarun, and Catherine Thomas, 2009, Synchronicity and rm interlocks in an emerging market, Journal
of Financial Economics 92, 182204.
Lambrecht, B. and Perraudin, W., 2003. Real options and preemption under incomplete information. Journal
of Economic Dynamics & Control 27, 619643.
Lyandres, Evgeny and Masa Watanabe, 2011. Product Market Competition and Equity Returns, Working
Paper.
Moskowitz, Tobias and Mark Grimblatt, 1999. Do industries explain momentum? Journal of Finance 54,
1249-1290.
Novy Marx, Robert, 2011, Operating Leverage, Review of Finance 15, 103-134.
Maskin, Eric and Jean Tirole, 1988. A Theory of Dynamic Oligopoly I: Overview and Quantity Competition
with Large Fixed Costs. Econometrica 56, 549-569.
Mason, Robin and Helen Weeds, 2010, Investment, Uncertainty and Preemption, International Journal of
Industrial Organization 28, 278-287.
Motta, Massimo, 2004, Competition policy: theory and practice. Cambridge University Press.
Opp, Marcus, Christine Parlour, and Johan Walden, 2011. Industrial Asset Pricing, Working Paper.
Pastor, Lubos and Pietro Veronesi, Stock Valuation and Learning about Protability, Journal of Finance 58,
1749-1789.
32
Pastor, Lubos and Pietro Veronesi, Technological Revolutions and Stock Prices, American Economic Review
99, 14511483.
Posner, Richard, 1975. The social costs of monopoly and regulation. Journal of Political Economy 83, 807-827.
Spiegel, Matthew, and Heather Tookes, 2011. Dynamic Competition, Valuation and Merger Activity, Journal
of Finance, forthcoming.
Weeds, Helen, 2002, Strategic Delay in a Real Options Model of R&D Competition, Review of Economic Studies
69, 729-747.
Zhang, Lu, 2005, The Value Premium, Journal of Finance 60, 67-103.
33
Appendix
A Proposition 1
For any strategy I = |r; A, we denote

I
=
r

the value of the assets in place of rm ) before


investment,
+
I
=
r
+
j


r
+
j

the value of the assets in place of rm ) after investment. At the investment


threshold AI = r, the rm can pay )1 to increase the value of its assets in place from

I
to
+
I
. Given
exercise at AI > r, the value of the growth option to invest is calculated as a perpetual binary option with
payo \
/
+
I
\
/

I
)1. We then observe
42
that the expected value of the growth option to invest is given by
GI =
_

+
I

I
)1
_ _
^
t
i
j
_
r
, where
_
^
t
i
j
_
r
is the price of a contingent claim that pays 1 if the rm invests
and 0 otherwise, and the parameter u 1 is such that
u =
1
2

r

2
x

_
_
r

2
x

1
2
_
2

2r

2
x
_1
2
For any strategy I = |r; A, we conclude that \I equals

I
GI if AI < r, and
+
I
if AI > r.
B Sorting conditions on x
)
and
)
The strategy pursued by rms is a multiple-action pair such that I = |r; A. The proof of the sorting
conditions on I follows Bustamante (2012) and consists of two steps. The rst step is to show that if the
value function \I complies the conditions in Cho and Sobel (1990), then the sorting condition of the action
pair I corresponds to the sorting conditions of each action in isolation. The second step is to derive the sorting
conditions of r and A separately.
We denote by AI

1
"

the expected price by rm ) at time t. In equilibrium, AI

1
"

is equal to the market


price jI when .I = 0; we use a more general notation since the sorting conditions should hold for any strategy
I. In the basic model, the value function \I rm ) given the set of actions I is such that
\I =
_

1
"
1
^
t

c
^
t

1
_
_

Y
+

1
" i
j

A1
_

1
"
^
t

1 (cA c)
i
j

1 )1
_
_
^
t
i
j
_
r
We consider the case of ` = 2 in which 1 = 1 and c1 < cL. In line with Cho and Sobel (1990), \I
is continuous in I and for any type ). Furthermore, if r1 < rL and A1 AL, then it must be the case that
\LI

\LI implies \1I

\1I. This last condition ensures that if rm A has incentives to deviate, rm 1
will pay a cost to ensure incentive compatibility. Denote d =
i
L
i
M
< 1. We also denote by

Y,. the expected
production of the industry when rm ) deviates and pretends to be rm i. Then the condition \LI

\LI
implies cL Uc where Uc is given by
Uc =
_
A1 ALd
r1
_
1
_
A1
_

Y
+
L,1
_

1
"
d
r1
AL
_

Y
+
L
_

1
"

_
1 d
r1
_
_
_

L
_

1
"
c
]
i
M
__
Similarly, the condition \1I

\1I implies c1 Uc. Therefore if c1 < cL and cL Uc, it holds that
c1 < Uc for any parameter value.
Consider now the sorting condition for each action r and A separately. The sorting conditions reect
that, all else equal, more ecient rms nd it less costly to invest earlier and more, namely
42
See, for example, Dixit and Pyndick (1994). The details of the derivation of the parameter u 1 are
provided in Chapter 5.
34
O
Oc
j
_
O1
jt
Oi
j
_
=
(1r)

A1
_
^
t
i
j
_
r
0 and
O
Oc
j
_
O1
jt
O
j
_
=
1

1
_
^
t
i
j
_
r
< 0
Put together, these inequalities ensure that the incentive compatibility constraint of rm A is binding and
that there exists a sequential equilibrium when ` = 2.
The sorting conditions described for ` = 2 also apply for the more general case of ` 2. This is because
the suciency conditions in Cho and Sobel (1990) apply for games with ` types. All conditions above hold
when c1 < ... < c < ....c^.
C Proposition 2
C.1. Sequential equilibrium
In the sequential equilibrium, the manager of rm ) chooses the strategy I
s

to maximize \
s
I
, where ) = 1, A.
The functional form of \
s
I
is that provided in Proposition 1, where the terms .
s
I
can be explicitly dened
given r
s
1
< r
s
L
. Given r
s
1
< r
s
T
, it holds that .
s
1,I
= .
s+
LI
= 0, and the expressions for .
s
L,I
< 0 and
.
s+
1,I
< 0 are such that
.
s
LI
=
_
(1 A
s
1
)

1
"
2

1
"
_
r
s
1
1
1
1
"
_
^
t
i
s
L
_
r
if AI 6 r
s
1
.
s+
1,I
=
_
(A
s
L
A
s
1
)

1
"
(1 A
s
1
)

1
"
_
r
s
L
A
s
1
1
1
1
"
_
^
t
i
s
M
_
r
if r
s
1
< AI 6 r
s
L
For the sake of convenience, we also use the notation in Appendix A, and we denote
s

I
as assets in place
before investment,
s
+
I
as assets in place after investment, and G
s
I
as the value of the growth option to invest
of rm ) in the sequential equilibrium. Then \
s
I
equals
s

I
G
s
I
if AI < r
s

, and
s
+
I
if AI > r, where

LI
=
r


r
s
Mt

,
s

1I
=
r

,
s
+
LI
=
r
+
M

and
s
+
1I
=
r
+
L


r
s+
Lt

.
Consider rst the optimization problem of rm A. To ensure that r
s
L
is chosen optimally, the derivative of
G
s
LI
with respect to r must be zero for all values of AI. To ensure that A
s
L
is chosen optimally, the derivative
of
s+
LI
with respect to A
s
L
must be zero for all values of AI. The corresponding optimal investment strategy
I
s
L
is such that
r
s
L
= )
r
1r
__
(A
s
L
A
s
1
)

1
"
A
s
L
(1 A
s
1
)

1
"
_
1

1
"
(A
s
L
cL c)
_
1
and
cL = (A
s
L
A
s
1
)

1
"
1

1
"
_
1
1
s

s
M

s
M
+
s
L
_
Consider now the optimization problem of rm 1. The optimality conditions of rm 1 are dierent from
those of rm 1 since in the sequential equilibrium rm 1 is subject to the complementary slackness condition
in (8). We solve for I
s
1
using Kuhn-Tucker. The value function of the manager of rm 1 at AI = r
s
1
is such
that
/ = \
s
1I
`
_
\
s
L

\
s
L
_
and the corresponding optimality conditions are given by
OL
O^
t

^
t
=i
s
L
= 0;
OL
O
s
L

^
t
=i
s
L
= 0 and
OL
OA

^
t
=i
s
L
= 0
where the Lagrange multiplier ` 0 due to the sorting conditions of the game. The optimal threshold r
s
1
is given by
35
r
s
1
= )
r
1r
__
(1 A
s
1
) 1

1
"
,c1
_
A
s
1

_
(21)

1
"
c
__
1
where , =
c
L
Ac
M
c
L
(1A)
< 1. The optimal scale upon investment A
s
1
1 is such that
(1 r
s
)
_
1
1
s
_
r
s
1
(1 A
s
1
)

1
"
1

1
"
r
s
r
s
1
(A
s
L
A
s
1
)

1
"
1

1
"
_
1
1
s

s
M

s
M
+
s
L
_
1
O
s
M
O
L
1
__
r
s
(1 u)
_
Oi
s
M
O
s
L

s
L
i
s
M
_
r
s
1
_
(A
s
L
A
s
1
)

1
"
(1 A
s
1
)

1
"
_
1

1
"
=,c1r
s
1
where r
s
=
_
i
s
L
i
s
M
_
r1
.
C.2. Clustering equilibrium
We denote the value of rm ) in the clustering equilibrium as \
c
I
. Given r
c

= r
c
, we characterize \
c
I
as in
Proposition 1 such that .
c
,I
= .
c+
I
= 0. Also, using the notation in Appendix A, we denote
c

I
as assets
in place before investment,
c
+
I
as assets in place after investment, and G
c
I
as the value of the growth option
of rm ) in the clustering equilibrium. Hence \
c
I
equals
c

I
G
c
I
if AI < r
c

, and
c
+
I
thereafter.
To obtain the optimal equilibrium strategies of rms in the clustering equilibrium, the proof consists of two
steps. First, we show that the Markov-perfect clustering equilibrium is such that both rms are better o by
investing simultaneously. This is consistent with Fundenberg and Tirole (1985) and Weeds (2002). Second,
we derive the optimal investment threshold r
c
and the increases in capacity A in the clustering equilibrium.
In our model, both rms are better o by investing simultaneously when the value of rm 1 is higher under
the investment strategies I
c

. The sorting conditions of the game in Appendix B on \1I and \LI imply that all
rms have incentives to invest simultaneously when rm 1 has incentives to do so. If rm 1 does not have an
incentive to invest sequentially, neither does rm A, whose ability to invest earlier and more is comparatively
lower. As a result, the clustering equilibrium obtains when rm 1 does not exercise its option to become a
leader.
The equilibrium investment timing r
c
is such that it maximizes the value of rm 1, and the corresponding
optimality conditions is such that the derivative of G
c
1I
with respect to r
c
equals zero for all values of AI.
Given the asymmetry in rms production technologies, each rm would attain its maximum value by clustering
at investment thresholds. A priori, this might lead to a range of potential equilibrium thresholds r
c
. The
lowest demand threshold r
c
would correspond to that of the more ecient rm 1; conversely, the upper bound
r
c
would correspond to the optimal demand threshold for the less ecient rm A. However, rm 1 has no
incentives to wait further than its own optimal threshold r
c
. Meanwhile, rm A still has incentives to invest at
r
c
not to become a follower. As a result, the equilibrium investment threshold r
c
maximizes the value of rm
1, and is given by
r
c
= )
r
1r
__
(A
c
L
A
c
1
)

1
"
A
c
1
2

1
"
_
1

1
"
(c1A
c
1
c)
_
1
The optimal scale A
c

1 is such that it maximizes the assets in place


c+
I
of each rm ) upon investment,
namely
c = (A
c
L
A
c
1
)

1
"
1

1
"
_
1
1
s

c
j

c
M
+
c
L
_
36
D Proposition 3
The proof of Proposition 3 is done in two steps. First, we derive the sorting conditions of the game when rms
dier both in their installed capacities before investment 1, and their costs of production upon investment c.
Second, we solve for the optimal investment strategies when rms dier in these two dimensions.
To derive the sorting conditions when rms dier in 1 and c, we rst consider the sorting conditions of
the two extreme cases: one in which rms dier only in c, and another in which rms only dier on 1. The
sorting conditions for the extreme case in which only dier in c is already discussed in Appendix B.
Consider then the case in which rms dier only in 1 such that 11 < 1L and c = c. When 11 < 1L
and c = c, the functional form of \I is the same as in Appendix B. We rst prove that if r1 < rL and
A1 AL, then \LI

\LI implies \1I

\1I. The condition \LI

\LI implies c Uc (1L), where in
Uc (1L) the denition of

Y
+
L,1
is such that when rm A deviates its production upon investment is A11L.
The condition \1I

\1I implies c Uc (11) , where in Uc (1L) the denition of

Y
+
1,L
is such that when rm
1 deviates its production upon investment is AL11. Since the market demand in (1) is strictly decreasing in
1, it holds that if 11 < 1L then Uc (1L) < Uc (11). Therefore if 11 < 1L, c Uc (1L) < Uc (11), it
holds that c < Uc (11) for any parameter value.
The expressions for the marginal sorting conditions on r and A with respect to 1 are not as simple
as those with respect to c in Appendix B. As a result, we consider additional assumptions to ensure that
the corresponding sorting conditions always have the same sign. To ease on exposition, we rst provide their
expression and required sign; we then consider sucient conditions under which the indicated sign holds for any
investment strategy. The economic rationale behind the marginal sorting conditions with respect to 1 relates
to the study by Boyer et al (2001).
We denote the market share of rm ) before investment by s

=
1
j
b
Y

j
, and upon investment by s
+

=

j
1
j
b
Y
+
j
.
Consider rst the marginal sorting condition on r. We require that, all else equal, rms with more 1 wait
longer to invest, namely
O
O1
j
_
O1
jt
Oi
j
_
= (u 1)
__
_

1
"
_
1
1
s
s

_
c
_

_
_

Y
+

1
"
_
1
1
s
s
+

_
c
_
A
_
1
i
j
_
^
t
i
j
_
r
1

0
The economic intuition behind this sorting condition is that the net gain from investing in capital for rm
) is decreasing in 1. Since the relative gain from investing is larger for smaller rms, smaller rms are willing
to invest earlier. A sucient yet not necessary assumption such that this sorting condition is always positive
is that the net decrease in marginal costs upon investment c c < 0 is relatively large.
Consider now the sorting condition on A. We require that, all else equal, rms with less 1 are willing to
invest more, namely
O
O1
j
_
O1
jt
O
j
_
=
_
_

Y
+

1
"

1
1
s
s
+


1
s
_
1
1
s
_
s
+

_
1 s
+

__
c
_

j
i
j

_
^
t
i
j
_
r
< 0
This sorting condition implies that the marginal product of capital of any rm is decreasing in 1. When
there are decreasing returns to scale, smaller rms are willing to invest more than larger rms. A sucient
yet not necessary assumption such that this sorting condition holds is that the elasticity of demand - 1 is
relatively low. The positive relation between returns to scale and demand elasticity is discussed in neoclassical
investment papers such as Hayashi (1982).
We refer now to the general case in which rms dier both in 1 and c. The joint implication of the two
extreme cases discussed so far is that rms with lower installed capacity before investment 1 and lower costs
of production after investment c have the ability to invest earlier and more.
37
We rst show that if r1 < rL and A1 AL, then \LI

\LI implies \1I

\1I. The condition
\LI

\LI implies cL Uc (1L). The condition \1I

\1I implies c1 Uc (11) . Since 11 < 0 and also
11 < 1L, we know that Uc (1L) < Uc (11). Therefore if c1 < cL, 11 < 1L, cL Uc (1L) < Uc (11) it
holds that c1 < Uc (11).
We use our previous results on the marginal sorting conditions on r and A to show that, when rms dier
in 1 and c, the corresponding marginal sorting conditions are given by
O
Oc
j
_
O1
jt
Oi
j
_

O
O1
j
_
O1
jt
Oi
j
_
0 and
O
Oc
j
_
O1
jt
O
j
_

O
O1
j
_
O1
jt
O
j
_
< 0
The pair which determines rm type |1; c can restated in terms of rms marginal product of capital
q = \1 in (10). For any strategy I = |r; A, we note that the marginal product of capital q equals (11).
The sorting conditions with respect to 1 ensure that \1 is strictly decreasing in the capacity of rms such
that \11 < 0. The sorting conditions on c also imply that \1 is strictly decreasing in c: rms with higher
marginal costs of production are willing to invest less. Consequently, rms with more installed capacity and
higher future costs of production have a lower q before investment, for any strategy I.
We dene a scalar q such that q is the marginal product of capital of rm ), evaluated at AI = A0 and
some strategy I = |r; A. The choice of the strategy I to dene q is without loss of generality; we use the
same I for all rms; the choice of I does not aect the sorting of q. Similarly, we use A0 for the sake of
simplicity; however, any AI 6 r
s
1
is suitable to dene q. We predict that rms higher q have the comparative
advantage to invest earlier and more, since
O
Oq
j
_
O1
jt
Oi
j
_
=
O

@V
jt
@x
j

Oc
j
Oc
j
Oq
j

O

@V
jt
@x
j

O1
j
O1
j
Oq
j
0, and
O
Oq
j
_
O1
jt
O
j
_
=
O

@V
jt
@
j

Oc
j
Oc
j
Oq
j

O

@V
jt
@
j

O1
j
O1
j
Oq
j
< 0
The suciency conditions described for ` = 2 also apply for ` 2. The suciency conditions for Cho
and Sobel (1990) apply for games with ` types, and all conditions above hold when c1 < ... < c < ....c^.
We derive the equilibrium investment strategies when rms dier in 1 and c using the same approach as in
Appendix C.
To prove (11), we use the property that the Taylor approximation of ov (j ()) such that ov (j ()) -
[j
0
(1 ())[
2
ov () , where ov () is the cross sectional variance of in a given industry. The proof of this result
follows the derivation of the delta method in Greene (2003) and relies on two equations. Equation 1 is the rst
order Taylor series expansion of j () around the mean of or j

such that j () = j
_
j

_
j
0
_
j

_ _
j

_
.,
where . 0 is the approximation error. Equation 2 is the property that 1 [j ()[ - j
_
j

_
when . - 0.
Using Equation 2, we approximate the variance of j () such that ov [j ()[ - 1
_
_
j () j
_
j

__
2
_
. We use
Equation 1 and operate to get ov (j ()) - [j
0
(1 ())[
2
ov (). Given =
1
1
and ) () = ln(), we conclude
that o
ln
K
V
,I
- jV
K
,I
o V
K
,I
.
Finally, to illustrate that industries with high value spread may have a lower HHI than industries with low
value spread, we reconsider the case of two industries in which rms have dierent installed capacities 1, and
the same cost of production c upon investment. We assume further that one industry has higher standard
deviation in rms installed capacities than the other, such that rms invest sequentially in one industry (high
o1), and simultaneously in the other (low o1).
In all industries, the sorting conditions indicate that rms with lower 1 have the ability to invest earlier
and more. By construction, then, small rms (low 1) catch up with large rms (high 1) in both industries,
38
and the HHI in both industries decreases once rms begin to invest. Also by construction, once all rms invest
and AI r
s
T
, the concentration in the industry with low o1 or HHI
c
is strictly lower than the concentration
in the industry with high o1 or HHI
s
. However, depending on the parameter specication, if HHI
s
falls
signicantly when rm 1 invests (i.e. A
s
1
1 is large), we may observe that the industry with higher o1 is less
concentrated, such that HHI
s
<HHI
c
in the range r
s
1
< AI < r
s
T
.
E Multiple risk factors
We rst note that the intra-industry variance in expected returns at time t or o
2
T,I
in the model is a linear
function of the cross sectional variance in betas o
2
c,I
. The return of rm ) at time t is given by
u1
1
. Applying
Itos lemma to the valuation equation in Proposition 1, we get d\ =
_
j
i
AI\i

2
x
2
A
2
I
\ii
_
AI\id1I. As in
Carlson et al (2004), an investment in A\i units of \ instantaneously replicates rm value. Multiplying by
S
1
gives the proportion of the replicating portfolio invested in the risky asset or beta , =
i1x
1
. Applying the
variance operator to
u1
1
yields
o
2
T,I
= o
2
i
o
2
c,I
The result above is not surprising insofar the risk free rate and the market price of risk are held constant in
the paper. A less obvious result, however, is that the qualitative results of our model still hold if \I depends
on multiple sources of risk, and rms investment strategies are orthogonal to all shocks but the demand shock
AI.
To see this, we consider an alternative set-up in which \I depends on the demand shocks AI as in (2),
and an additional source of systematic risk WI. We assume that the shock WI follows a geometric Brownian
motion with diusion ou, is uncorrelated with the demand shock AI, and that rms determine their investment
strategies contingent on AI only.
Let
i
be the proportion of the replicating portfolio invested in the risky asset AI, and let
u
be the
proportion of the replicating portfolio invested in the risky asset WI. Using the same arguments as before, the
corresponding expression for o
2
T,I
yields
o
2
T,I
= o
2
i
o
2

x
,I
o
2
u
o
2

w
,I
such that rms strategic investment decisions aect their expected returns through o
2

x
,I
.
F Proposition 4
The derivation of ,
I
follows that of Proposition 2 in Carlson et al (2004). Applying Itos lemma to \ , we note
that the exposure to systematic risk of the rm equals the proportion of the replicating portfolio invested in the
risky asset, such that , =
i1x
1
. The exact expression for ,
I
depends on the equilibrium outcome. If oc 6 oc,
,
c
I
equals 1 (u 1)
1

_
j

I
cAI
_
1
1
c
t
1 if AI 6 r
c
and 1 otherwise, where j

I
= (21)

1
"
1. If oc oc, ,
s
1I
equals 1 (u 1)
1

_
j

I
cAI
_
1
1
s
Lt
1 if AI 6 r
s
1
, 1 (u 1)
1

( jI c1AI)
1
1
s
Lt
< 1 if r
s
1
< AI 6 r
s
L
, and is
equal to 1 otherwise, where jI = (1 A
s
1
1)

1
"
1. If oc oc, ,
s
LI
equals 1 (u 1)
1

_
j

I
cAI
_
1
1
s
Mt
< 1 if
AI 6 r
s
1
, 1 (u 1)
1

( jI cAI)
1
1
s
Mt
1 if r
s
1
< AI 6 r
s
L
, and is equal to 1 otherwise.
39
G Proposition 5
The sign of the covariance between the beta of rm 1 and that of rm A in Proposition depends on oc. For
any investment strategy I, the denition of rms betas in Proposition 8 implies that the covariance in rms
betas depends on the covariance in rms cashow to value ratios, and hence
cijn[co (,
1I
, ,
LI
)[ = cijn

co
_
\1I
r
Lt

, \LI
r
Mt

__
When oc 6 oc, both rms expect an increase in value upon investment, and .
c
,I
= .
c+
I
= 0. This
implies that, before investment, \I
r
jt

= GI, where GI is the value of the growth option of rm ), and


co (G1I, GLI) = !
c
1
!
c
L
o
2
i
A
2r
I
0
where we dene ! 0 such that GI = !A
r
I
. Hence co (,
1I
, ,
LI
) 0 if AI < r
c
.
Conversely, when oc oc, each rm expects a reduction in its prots upon the investment of its competitor,
where .
s
L,I
< 0 and .
s+
1,I
< 0. Consider rst the interval r
s
1
< AI < r
s
L
. In this case, rm 1 only expects
a reduction in its prots, while rm A only expects an increase in its prots upon investment. As a result,
\1I
r
Lt

= .
s+
1,I
< 0, while \LI
r
Mt

= GLI. Put together, this implies that co (,


1I
, ,
LI
) < 0 if
r
s
1
< AI < r
s
L
since
co
_
.
s+
1,I
, GLI
_
= +
s
1
!
s
L
o
2
i
A
2r
I
< 0
where +
s
1
= .
s+
1,I
A
r
I
< 0. Similarly, since .
s
L,I
< 0, the same argument applies to show that
co (,
1I
, ,
LI
) < 0 when AI < r
s
1
.
H Proposition 6
To prove Proposition 6, we apply the variance operator to ,
I
in (12) such that o
2
c,I
= c
2
(u 1)
2
o
2

V
,I
. We
then use the property that the rst Taylor approximation of ov () (r)) such that ov () (r)) - [)
0
(1 (r))[
2
ov (r) ,
where ov (r) is the cross sectional variance of r in a given industry. We derive this rst order approximation
in Appendix D. Using this result, the intra-industry variance at time t of ln
r
1
is such that o
2
ln

V
,I
- j
2

V
,I
o
2

V
,I
.
We note that o
2
ln

V
,I
= o
2
ln

K
,I
o
2
ln
K
V
,I
2j
I
, where j
I
is the covariance between ln
r
t
1
t
and ln
1
t
1
t
. Reordering
terms, we get the expression in Proposition 6, where
I
equals c
2
j
2

V
,I
(u 1)
2
.
I The case of N > 2 rms
The solution approach for ` 2 relies on sorting conditions and ICCs just at the case of ` = 2. In oligopolies,
the use of sorting conditions facilitates the analysis, insofar they constrain the set of possible equilibria to those
in which rms with higher marginal q invest earlier or in tandem with other rms. Just as signalling games
with multiple discrete types,
43
each rm cares about its closest and strongest competitor. Consequently, if rm
) invests earlier and more than its closest competitor rm i in equilibrium, the only binding ICC for rm ) is
that of rm i.
To illustrate the main properties of the case of ` 2, we consider an the extension of the basic model
with ` = 8 rms. We assume that rms have the same installed capacity before investment, and they have
43
See, for instance, Fundenberg and Tirole (1991), Chapter 7.
40
heterogeneous marginal costs of production upon investment. The sorting conditions of the game are the same
for any value of `, and hence they are equal to those in Appendix B. We label rms by 1, A and 1. We
assume c1 < cL < cT , c1 = cL o, cT = cL /, o 0, and / 0. Given this notation, cL is the average
marginal cost of production upon investment, and oc =
_
a
2
+l
2
3
is the standard deviation in rms marginal
costs of production.
The key departure of the case of ` 2 relative to the case of ` = 2 is that rms strategies in equilibrium
cannot be characterized uniquely in terms of oc. In the example with ` = 8, the parameters o and / are both
necessary to determine the equilibrium outcome of the game. More intuitively, this implies that higher order
moments of the distribution of c also matter- i.e. the skewness of rms marginal costs of production. An
important exception is the case in which c is uniformly distributed such that o = /, and oc =
_
2
3
o. Just as
in Proposition 2, however, even if o and / are dierent, rms investments are more clustered in equilibrium if
o and / are relatively small (i.e. low oc). Table 2 illustrates an example with ` = 8 in which oc is small and
the equilibrium outcome is that all rms invest at the common investment threshold r
c
.
Put dierently, the equilibrium outcome of the game with ` = 8 depends on the ability of both rm 1 and
A to invest earlier than their closest competitor. Even if rm 1 may nd it more convenient to invest earlier
than rm A, the strategy of rm A also depends on the preemptive behavior of rm 1. For instance, if rms
A and 1 are close competitors (i.e. / 0 is relatively small), rm A may nd it too costly to invest earlier
than rm 1. This, in turn, may constrain the ability of rm 1 to invest earlier than rms A and 1 (i.e if.
o 0 is not suciently large).
Consider now the more general case in which rms dier in their production technologies before and after
investment. Just as in Proposition 8, the more general testable implication of the model for the case of ` 2
rms is that rms investment dynamics are more similar in industries with low value spread. We can extend
the insight in Proposition 8 to the case of ` 2 rms since the sorting conditions in Appendix D apply for any
number of rms `.
J Database construction
The working sample is drawn from a merged CRSP-COMPUSTAT database, considering the 1968-2008 period.
We estimate the beta of equity of each rm as the sum of the coecients of monthly returns on lagged, lead and
contemporary market returns of the stock return of each rm in the sample. We compute betas at a monthly
frequency, using ve-year rolling windows containing the previous 60 observations. We compute stock returns
in excess of the risk free rate reported in CRSP.
We match each rms CRSP stock return and betas from July of year t until June of year t 1 to the
corresponding accounting information in COMPUSTAT for the scal year ending in year t 1. With the
exception of lnHHI, we construct the remaining explanatory variables using COMPUSTAT tapes. lnHHI is the
logarithm of the HHI for manufacturing industries reported by the US Census Bureau; since the HHI is reported
every ve years, we repeat the HHI of year t over the next four years for every industry.
The market value of equity is the product of item PRCC_F times CSHO. The market value of assets \ is
the market value of equity plus total liabilities. The total liabilities 1 are computed as AT minus CEQ minus
TXDB. Operating cashows are the sum of SALE minus COGS minus XSGA. Investment 1 = (A 1) 1 is
CAPX. We consider 1 to be total assets AT, with the exception of
1
1
where 1 is set as lagged PPENT. The
operating mark-up i is the ratio of over SALE. All COMPUSTAT variables are winsorized at 1%.
We construct the intra-industry comovement in variable r at time t or .i,I as in Khanna and Thomas
41
(2009). For variables r =
_
,; 1;
1
1
;
1 T
1T
;
1
1
_
, and for each month, we consider the average of the correlation
coecients C. between the variable r of each unrepeated pair of rms i and ) within the same industry, such
that
C. =
Cor(.,)
_
1 ar(.)1 ar()
where Co (i, )) is the covariance between the variable r of rms i and ) during the window between month
t and month t 60, \ ov(i) is the variance of rm is variable r in such window, and \ ov()) is the variance of
rm )s monthly variable r. To compute the comovement in the ratios
1
1
and
1 T
1T
, we compute the market
value of equity at a monthly frequency, using the time series of PRCC and CSHO reported in CRSP.
K Levered rms
When rms are unlevered, managers determine the investment strategy that maximizes the market value of
assets, and the beta of the rm captures the exposure to systematic risk of the market value of assets. Conversely,
the managers of levered rms choose the investment policies that maximize the market value of equity, and the
(equity) beta of a levered rm captures the exposure to systematic risk of shareholder value. We hereby discuss
the alternative cases of risk free and risky debt.
When we extend the basic model such that all rms in the industry hold the same perpetual risk free debt
contract, we nd that the testable implications in 1 4 remain the same. Just as in Section 1, the dynamics of
rms investment decisions depend on the intra-industry asset value spread. Similarly, the dynamics of rms
equity betas also depend on the intra-industry asset spread, but they are also aected by the leverage ratio.
To see this, we denote rms pre-existing risk free, perpetual debt coupon /. We denote the market value
of equity by 1 = \
l
r
. Since / is exogenously given, q is such that to 11 = \1, and the identity in (10)
still holds. Following the derivation in Appendix E, the equity beta equals
^1x
1
. Reordering terms, the equity
beta equals
,
I
= 1 II (u 1)
_
1
1

r
jt
1
jt
_
[1 II (u 1)[
b
r
1
jt
An important insight from this simple example with risk free debt is yet that, since the investment oppor-
tunity set of the industry is changing over time, the standard Modigliani-Miller argument that asset betas are a
linear function of equity betas and leverage ratios does not hold. When the investment opportunity set of the
industry changes over time, the levered equity beta of any rm in the industry is non linear in their leverage
ratio.
When rms hold risky debt in perfectly competitive industries, Hennessy (2004) shows that the wedge
between q and the market to book asset ratio in (10) also depends on debt overhang eects. Bolton and
Scharfstein (1990) and Fulghieri and Nagarajan (1996) further show that risky debt aects rms strategic
behavior under imperfect competition. Put together, these results suggest that, when rms hold risky debt,
the dynamics of investment, equity betas and excess returns depend jointly on the intra-industry equity value
spread, and the intra-industry standard deviation in leverage ratios.
As a result, we abstract from correcting rms betas and excess returns for leverage in the empirical tests,
and we use two alternative measures of the intra-industry value spread. One is based on equity market to
book ratios, and the other is based on asset market to book ratios. We obtain very similar results with both
measures.
42
L Parameter choice in numerical examples
The parameters in Tables 1-3 are v = 6.%, c = 2.%, o = 2%, - = 2.8, ) = 1, A0 = 0.01, c1 = 0.1,
cL = 0.1117, c = 0.110, and 1 = 1. The technology of rm A is such that cT = 0.128. The parameters
in Figures 1 4 are the same with exception of cL = 0.1077. The numerical example in Figure 1 uses the
equilibrium investment strategies I
s

. Figures 1 and 8 represent rms expected values and betas by reporting


the average of rms values and betas out of 00 simulations of the Brownian demand shocks.
43
Table 1: Investment strategies when N = 2
Stackelberg Sequential Clustering
Strategies Strategies Strategies
(A) (B) (C)
L M L M L M
Strategies
r 0.032 0.125 0.019 0.655 0.090 0.090
A 98.88 50.88 162.07 25.68 64.42 60.96
` 0.000 0.000 0.939 0.000 0.000 0.000
Valuation at A0
Firm Value 0.585 0.202 0.270 0.060 0.433 0.395
.,0 -0.405 -0.169 -0.119 -0.205 0.000 0.000
q,0 0.543 0.160 0.199 0.011 0.371 0.333
Firms Betas
At A0 1.176 0.923 1.021 0.012 1.129 1.112
At r
s
1
0.828 1.291 0.943 1.313 1.157 1.139
At r
c
0.706 1.296 0.879 1.311 1.000 1.000
At r
s
L
1.000 1.000 1.000 1.000 1.000 1.000
Indicators at r
s
L
HHI 0.551 0.551 0.764 0.764 0.501 0.501
or 0.104 0.104 0.115 0.115 0.096 0.096
This table illustrates the potential industry equilibria when ` = 2 and rms dier in their future production
technologies. The investment strategy |r; A consists of the demand threshold at which rm ) invests r,
and the scale of the rm A 1 upon investment. The superscript c corresponds to sequential strategies;
the superscript c corresponds to clustering strategies. ` is the multiplier of the ICC of rm 1. .,0 is the
expected reduction in future prots of rm ) before other rms invest, evaluated at AI = A0. q,0 is the
marginal product of capital of rm ) at AI = A0. HHI is the Herndahl-Hirshman Index. or is the standard
deviation in the operating mark-ups. The outcome is the clustering equilibrium in Panel (C).
44
Table 2: Sequential and clustering strategies when N = 3
Sequential Clustering
Strategies Strategies
(A) (B)
L M F L M F
Strategies and Valuation
r 0.014 0.093 0.801 0.212 0.212 0.212
A 113.30 46.81 21.05 56.21 51.01 45.81
` 0.959 0.920 0.000 0.000 0.000 0.000
Firm Value at A0 0.237 0.069 0.033 0.257 0.233 0.211
Firms Betas
At A0 1.035 0.360 -0.603 1.063 1.038 1.009
At r
s
1
0.835 1.058 -2.900 1.083 1.051 1.013
At r
s
L
0.927 0.916 1.309 1.000 1.000 1.000
Table 3: Other investment strategies when N = 3
Mixed Cases
L leads, M and F follow L and M lead, F follows
(A) (B)
L M F L M F
Strategies and Valuation
r 0.023 1.113 1.113 0.092 0.092 0.447
A 159.45 19.19 12.74 120.41 34.97 22.79
` 0.918 0.000 0.000 0.000 0.702 0.000
Firm Value at A0 0.314 0.054 0.049 0.338 0.216 0.126
Firms Betas
At A0 1.107 0.134 0.021 1.122 1.007 0.809
At r
s
1
0.942 1.013 1.004 0.974 0.997 1.060
At r
s
L
1.000 1.000 1.000
These tables illustrate the potential equilibrium outcomes with ` = 8 when rms have the same parameters
as those in Table 1. The investment strategy |r; A consists of the demand threshold at which rm ) invests
r, and the scale of rm ) A 1 after investment. The superscript c corresponds to sequential investment;
the superscript c corresponds to clustered investment. ` is the multiplier of the binding ICC on rm ). .,0
is the expected reduction in prots of rm ) before other rms invest, evaluated at AI = A0. The outcome is
the clustering equilibrium in Panel (B) of Table 2.
45
Table 4: Working sample statistics
Firm level Industry level
mean sd N mean sd N
1
1
0.360 0.520 113,007 0.324 0.293 14,745
, 1.102 0.947 115,702 1.040 0.547 15,014
1 0.082 0.564 115,765 0.073 0.366 15,077
1
1
1.477 0.826 110,355 1.407 0.525 14,931
1 T
1T
2.085 1.543 109,797 1.985 1.050 14,836
T
1
0.526 0.230 115,702 0.544 0.145 15,014
r
1
0.082 0.219 115,633 0.110 0.099 15,013
i 0.144 0.110 115,419 0.129 0.075 14,779
o I
K
0.274 0.353 12,584
o
c
0.635 0.417 12,815
oT 0.374 0.279 12,815
o V
K
0.530 0.394 12,693
o V B
KB
1.088 0.718 12,523
o B
K
0.178 0.081 12,815
o
K
0.111 0.116 12,811
or 0.058 0.047 12,782
lnHHI 5.645 1.185 8,539
. I
K
0.031 0.066 14,812
.
c
0.026 0.032 14,857
.T 0.016 0.012 14,857
. V
K
0.107 0.213 14,849
. V B
KB
0.178 0.201 14,244
This table reports the summary statistics of our CRSP-COMPUSTAT working sample of US public rms
from 1968 to 2008.
1
1
is the investment rate; , is the equity beta; 1 is the excess stock return, which is
annualized in this table since all statistics are reported in annual terms;
1
1
is the market to book asset ratio;
1 T
1T
is the market to book equity ratio;
T
1
is the book leverage ratio;
r
1
is operating cashows to assets; i is
the operating mark-up on prots; oi denotes the intra-industry standard deviation in variable r; lnHHI is the
logarithm of the US Census HHI; and .i denotes the intra-industry comovement in variable r. The details on
database construction are provided in Appendix J.
46
Table 5: Investment and the intra-industry value spread
Firm level Investment Industry level Investment
(A) (B) (C) (D) (E) (F)
1
1
0.131*** 0.115*** 0.121*** 0.151*** 0.145*** 0.138***
(0.010) (0.009) (0.009) (0.012) (0.016) (0.012)
r
1
-0.084* -0.067 -0.067 -0.182*** -0.123*** -0.120***
(0.046) (0.044) (0.044) (0.064) (0.053) (0.059)
o V
K
0.090*** 0.013***
(0.012) (0.002)
o V B
KB
0.039*** 0.017***
(0.005) (0.005)
N 107,749 105,633 105,243 14,672 12,552 12,385
Avg. 1
2
0.053 0.057 0.054 0.081 0.097 0.097
This table reports the Fama and MacBeth (1973) regressions on the investment to capital ratios
1
1
at the
rm and industry level. The data used is in annual frequency.
1
1
is the market to book asset ratio;
1 T
1T
is
the market to book equity ratio;
r
1
is operating prots to assets; and oi denotes the intra-industry standard
deviation in variable r. Newey-West corrected standard errors are reported in parentheses. ***j < 0.01,
**j < 0.0 and *j < 0.1.
47
Table 6: Betas, excess returns and investment dynamics
Comovement in Betas .
c
Comovement in Excess Returns .T
(A) (B ) (C) (D) (E) (F) (G )
. I
K
0.055*** 0.036***
(0.002) (0.001)
. V
K
0.028*** 0.013***
(0.003) (0.001)
. V B
KB
0.020*** 0.013***
(0.002) (0.001)
.
c
0.372***
(0.038)
N 173,434 173,939 163,932 173,434 173,939 163,932 174,002
Avg. 1
2
0.034 0.062 0.034 0.056 0.062 0.062 0.366
This table reports the Fama and MacBeth (1973) regressions on comovement in betas and excess returns.
The data used is in monthly frequency. .i denotes the intra-industry comovement in variable r; , is the
equity beta; 1 is the excess stock return;
1
1
is the investment rate;
1 T
1T
is the market to book equity ratio;
and
1
1
is the market to book asset ratio. Newey-West corrected standard errors are reported in parentheses.
***j < 0.01, **j < 0.0 and *j < 0.1.
48
Table 7: Industry dynamics,
the intra-industry value spread, and competition
Comovement in Betas .
c
Comovement in Excess Returns .T
(A) (B) (C) (D) (E) (F) (G) (H)
o V
K
-0.0024*** -0.0012***
(0.0008) (0.0003)
o V B
KB
-0.0016*** -0.0009***
(0.0004) (0.0002)
or -0.0316*** -0.0132***
(0.0018) (0.0008)
lnHHI -0.0004*** -0.0001***
(0.0001) (0.0000)
N 147,243 145,245 148,412 113,178 147,243 145,245 148,412 113,178
Avg. 1
2
0.023 0.021 0.032 0.015 0.026 0.023 0.031 0.012
This table reports the Fama and MacBeth (1973) regressions on comovement measures as a function of
the intra-industry value spread and static measures of competition. The data used is in monthly frequency.
.i denotes the intra-industry comovement in variable r; , is the equity beta; 1 is the excess stock return;
1
1
is the market to book asset ratio;
1 T
1T
is the market to book equity ratio; i is the mark-up on operating
prots; and lnHHI is the logarithm of the US Census HHI. Newey-West corrected standard errors are reported
in parentheses. ***j < 0.01, **j < 0.0 and *j < 0.1.
49
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
-20
0
20
40
60
80
100
t
Value of firm L as a leader
x
L
Firm L invests
x
M
Firm M invests
Assets in place |
Lt
/
Expected reduction in profits |
Lt
/
Total firm value | V
Lt
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
-1
0
1
2
3
4
5
6
7
t
Value of firm M as a follower
x
M
Firm L invests
x
M
Firm M invests
Assets in place |
Mt
/
Expected reduction in profits |
Mt
/
Total firm value | V
Mt
Figure 1. This gure illustrates how rms strategic interaction aect their values for the special case in
which rm 1 invests earlier than rm A such that r1 < rL. The total value of any rm consists of its assets
in place, its growth options, and the expected reduction in future prots due to investments by its competitors.
50
\1 at AI = r
s
1
\L at AI = r
s
1
0.5 1 1.5 2
1
1.2
1.4
1.6
1.8
2
2.2

c
0.5 1 1.5 2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8

c
r1 rL
0.5 1 1.5 2
0.02
0.03
0.04
0.05
0.06
0.07
0.08

c
0.5 1 1.5 2
0.2
0.4
0.6
0.8
1
1.2
1.4

c
c1 at AI = r
s
1
`
0.5 1 1.5 2
0.55
0.6
0.65
0.7
0.75
0.8
0.85
0.9

c
0.5 1 1.5 2
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8

c
Figure 2. This gure illustrates how the intra-industry standard deviation in costs of production oc aects
rms investment strategies in equilibrium. The red color relates to the sequential strategies I
s

; the blue color


relates to the clustering strategies I
c

. The black dotted line reects rms investment strategies in equilibrium.


\ is the value of rm ); r is the demand threshold at which rms invest; ` is the multiplier of the ICC of rm
); c1 is the market share of rm 1 when all rms have invested. oc is expressed in %.
51
Industry dynamics of ,
I
0.5 1 1.5 2 2.5 3 3.5 4 4.5
0
0.2
0.4
0.6
0.8
1
1.2
1.4

jt
t
x
L
s
x
M
s
x
c

L
s
if sequential eq.

M
s
if sequential eq.

L
c
if clustering eq.

M
c
if clustering eq.
Figure 3. This gure shows the dynamics of the beta of rm ) at time t or ,
I
in the basic model when
oc = oc. ,
s

is the beta of rm ) when rms invest sequentially in equilibrium; ,


c

is the beta of rm ) when


rms investments cluster in equilibrium.
52
Demand elasticity -
0.7 0.75 0.8 0.85 0.9 0.95 1 1.05 1.1 1.15
2
2.05
2.1
2.15
2.2
2.25
2.3
2.35
2.4
2.45
2.5
0.7 0.71 0.72 0.73 0.74 0.75 0.76
2
2.05
2.1
2.15
2.2
2.25
2.3
2.35
2.4
2.45
2.5
Demand growth j
i
0.58 0.59 0.6 0.61 0.62 0.63 0.64 0.65 0.66 0.67
0.02
0.022
0.024
0.026
0.028
0.03
0.032
0.034
0.036
0.038
0.04
0.719 0.7195 0.72 0.7205 0.721 0.7215 0.722
0.02
0.022
0.024
0.026
0.028
0.03
0.032
0.034
0.036
0.038
0.04
Demand volatility oi
0.57 0.58 0.59 0.6 0.61 0.62 0.63 0.64 0.65
0.05
0.06
0.07
0.08
0.09
0.1
0.11
0.12
0.13
0.14
0.15
0.7185 0.719 0.7195 0.72 0.7205 0.721 0.7215
0.05
0.06
0.07
0.08
0.09
0.1
0.11
0.12
0.13
0.14
0.15
Figure 4. This gure illustrates how the underlying determinants of market demand aect the threshold
oc and the shadow cost of preemption `. The threshold oc is the minimum value of the intra-industry standard
deviation in rms costs of production oc at which rm 1 prefers to cluster. ` is the Lagrange multiplier on
the incentive compatibility constraint of rm A under sequential investment.
53

Anda mungkin juga menyukai