SSRN Id3182158

Product Life Cycles in Corporate Finance
Gerard Hoberg
Marshall School of Business, University of Southern California
Vojislav Maksimovic
Smith School of Business, University of Maryland
We are especially grateful to Wei Jiang (the editor) and two anonymous referees for extensive and very
thoughtful suggestions. We thank Christopher Ball at metaHeuristica for providing terrific text analyt-
ics capabilities that made this project possible. We also thank AJ Chen, Espen Eckbo, Laurent Fresard,
Don Lessard, William Mann, Lee Pinkowitz, Ed Rice, Berk Sensoy, Xunhua Su, Rene Stulz, Luke Tay-
lor, Karin Thornburn, Xiaoli Tian, and Paolo Volpin and seminar participants at Case Western Reserve,
Georgetown University, INSEAD, the Norwegian School of Economics, the Securities and Exchanges Com-
mission, Swiss Finance Institute (Lausanne), Swiss Finance Institute (Lugano), Temple University, Toulouse
Business School, University of Maryland, University of Southern California, University of Virginia, Vander-
bilt University, the 2018 Financial Intermediation Research Society conference, the 2019 London Business
School Corporate Finance Conference, the University of Washington 4th Summer Finance Conference and
the 2019 Mitsui Symposium on Comparative Corporate Governance and Globalization at Michigan for ex-
cellent comments. All authors contributed equally to this paper. Send correspondence to Gerard Hoberg,
hoberg@marshall.usc.edu.
Electronic copy available at: https://ssrn.com/abstract=3182158

ABSTRACT
We develop a novel 10-K text-based model of product life cycles and examine firm
investment policies. Conditioning on the life cycle substantially improves the power
of q to explain investment and reveals a natural ordering of investments over the life
cycle. While R&D and CAPX sensitivity are high early in the cycle, acquisitions arise
as firms mature, and divestitures and product extension investments arise as firms
decline. q-sensitivities that condition on the life cycle can vary by as much as 400%
from traditional sensitivities. The life cycle framework further reveals an enriched
relationship between competition, investment, and corporate profits.
2
Numerous studies in economics, strategy, and finance suggest that a company and its prod-
ucts go through life cycles, and this progression is important in understanding a myriad of
issues facing the firm.1 Issues of relevance include how a firm interacts with rivals, investment
decisions, including both organic investment and acquisitions, and the firm’s ability to re-
main flexible as its organizational capital ages. Although these pioneering studies have made
important contributions, we posit that the importance of life cycles in fields such as corporate
finance has been significantly understated, likely because of challenges in measurement.
We define a four-stage product life cycle following Abernathy and Utterback (1978):
product innovation, process innovation, maturity, and decline. For parsimony, we will refer to
these stages as Life1, Life2, Life3, and Life4, respectively. We apply computational linguistic
methods to 10-Ks based on “Chained Context Discovery” (see Cimiano 2010) that focus
on multiple search terms that are proximate in text. We use these anchor phrase methods
to compute a four-element vector for each firm in each year, with elements summing to
one. Intuitively, firms with multiple products will have positive exposures to more than one
stage of the life cycle. The firm-year nature of our measures allows us to dynamically track
firm product portfolios through the life cycle, and to include firm fixed effects in all of our
regressions. These measures should also prove useful in many other contexts.
In this paper, we examine hypotheses motivated by a simple q-model of investment that

incorporates the life cycle, and linearly links investment to Tobin’s q over the stages of
the life cycle. We argue that (A) different stages of the product life cycle require different
prime inputs and (B) the firm’s incentives to invest vary over the product cycle as uncertainty
resolves. We show in our simple model that these variations predict differences in investment
sensitivities to q along the cycle. This basic theory also motivates the conditional empirical
q-model we use in our tests.
Our study makes two main empirical contributions. First, we find a natural ordering of
investment sensitivities to q over the life cycle. Firms with exposure to the earliest product
innovation stage heavily invest in R&D when their valuations rise. In the second process
1
See, for example, Abernathy and Utterback (1978), Klepper (1996), Klepper and Thompson (2006),
Loderer, Stulz, and Waelchli (2016), and Arikan and Stulz (2016).
3
innovation stage, their CAPX becomes more sensitive to Tobin’s q. Our most novel results
are those in the later stages. Firms with products in the third mature stage focus more on
acquisitions when q increases. Finally, firms with products entering decline disinvest (sell
more assets) when q declines, and also attempt to “regain youth” via product extension
strategies and acquisitions when q increases. Our evidence of strategies to “regain youth”
when products decline is further reinforced by tests examining financing policies, as firms
exposed to this stage raise more capital (both equity and debt) when their q rises. Overall,
the natural ordering indicates a progression from organic investment to inorganic investment
to, finally, disinvestment and extension strategies.
Overall, we document economically large variation in q-sensitivities across investment

policies that are not observable using the baseline q-model used in the literature. For exam-
ple, when going from the lowest to the highest q tercile, the baseline q-model shifts R&D
by 3.5% of assets and CAPX by 0.8% of assets. In contrast, our conditional life cycle model
shifts R&D by 10% and CAPX by 4.9% of assets. These results show that earlier concerns
that stock market valuations are unconnected to investment, as expressed by Morck, Shleifer,
and Vishny (1990) and Blanchard, Rhee, and Summers (1993), and the counterarguments
summarized in Bond, Edmans, and Goldstein (2012) and Bai, Phillipon, and Savov (2016),
might be revisited in a life cycle context. Moreover, the explanatory power of the condi-
tional q-model has increased in recent years. Also, the cycle’s predicted progression is not
inevitable, as some shocks can accelerate or reverse the cycle’s course.
Our second major contribution is to use the conditional life cycle model to enrich our
understanding of the link between competition, investment, and corporate profits. We doc-
ument a more refined set of economically large channels that build on those discussed in the
investment-q literature (e.g., Gutierrez and Philippon 2018; Akdogu and McKay 2008) and
in the IO-q literature that focuses on the link between q and corporate profits (see, e.g.,
Lee, Shin, and Stulz 2021; Lindenberg and Ross 1981; Stigler 1964). A central theme in our
findings is that interacting life cycle stages with competition produces large amplifications
in q-sensitivities that are not observable without the life cycle model.
We first show that although competition is first-order important, that life cycle effects
4
have larger economic magnitudes. Life cycle effects are distinct from competition effects,
and additionally there are interactive effects. On the investment-q side, tercile sorts show
that the conditional life cycle model alone shifts R&D by 9% and CAPX by 4.9% of assets.
Conditioning on competition alone only shifts R&D by 5.5% of assets and CAPX by 2.1%
of assets. Together, models including competition and life cycles illustrate that the highest
levels of q-sensitivity arise for earlier-stage life cycle firms that face high competition, and
the lowest sensitivities arise for late-stage life cycle firms that face low competition.
A central idea implied by the IO-q literature is that q comprises the capitalized value of
the firm’s assets in place and the present value of the firm’s investment opportunities. When
the firm’s market power is high, the firm’s value will weigh more on profits from assets in
place, and Lee, Shin and Stulz (2021) suggest that the relationship between investment and
q will weaken and the link between q and realized profits will strengthen. In contrast, when
competition is fierce and q is less profit driven, then the relation between q and investment
will be stronger. The authors show that IO-q considerations are growing over time, and for
larger firms, q has become more related to profits than to investment opportunities.2 This
framework motivates our analysis.
The product life cycle has direct implications for the relative importance of IO-q and
investment-q-sensitivities. For example, IO-q-sensitivity should be relevant only for life
cycle stages, where the firm actually has products in the market. Hence, Life1 exposure
should generate investment-q- sensitivities, but not IO-q-sensitivities. In contrast, Life2 to
Life4 indicate products in the market, and both IO-q and investment-q should be relevant.
The IO-q predictions are particularly strong for Life2 (Abernathy and Utterback indicate
that improving profits through efficiency is the core purpose of Life2) and for Life3 (positive
shocks to barriers to entry or demand can improve profits and increase Q). Predictions are
mixed for Life4 as firms are selling assets.
Our results, which can only be observed using the life cycle model, support these predic-
2
To understand the intuition behind this relation, one could contrast a stable monopoly, whose value
depends on earning rents from assets in place, with a competitive market in which firms create value by
exploiting transient investment opportunities. The latter case has a high q-sensitivity of investment. The
former case has a stronger relation between realized current profits and q.
5
tions. IO-q-sensitivities are positive and significant only in the middle Life2 and Life3 stages,
where sensitivities are 200% to 500% larger than those for the standard IO-q-model. We also
confirm that IO-q-sensitivities are sharper in less competitive markets, and investment-q-
sensitivities are sharper in more competitive markets. These results are economically large,
and suggest that firms begin the life cycle focused on organic growth options, which generate
large investment-q-sensitivities that depend on the life cycle stage. Later in the cycle, growth
options mature into assets in place and IO-q-sensitivities increase. Regarding competition,
the path through the life cycle is a low-investment high-rent path in concentrated industries
and a high-investment low-rent path in competitive industries.
We also document novel instances of negative q-sensitivities for investments in some life
cycle stages. Although firms in the early stages of the life cycle have strong positive q-
sensitivities to R&D, firms exposed to the mature third stage have a negative sensitivity to
R&D. Because this is surprising given the existing literature, we explore this finding. We
find that this result is focused on the healthcare sector and has roots in a particularly strong
interaction between investment-q and IO-q sensitivities in Life3. Unique to this sector,
we find that a high q for Life3 firms indicates not only increased profits but also stronger
barriers to entry as the product market fluidity of rival firms declines and the Life3 stage
itself becomes more stable. These findings suggest that these firms can reduce R&D given
the stronger barriers to entry, and can enjoy high and stable rents, consistent with theoretical
ideas in Sutton (1989) and Aghion et al. (2005) and confirmatory results in Garfinkel and
Hammoudeh (2021) in the healthcare sector.
Although we are not able to fully establish causality, we note three aspects of our tests
that support that our results reflect genuine life cycle effects. First, our panel structure allows
us to include firm and year fixed effects in all models, which absorb any firm-specific omitted
variables. Second, we take the foundational core of the Abernathy and Utterback life cycle
theory seriously (which links product life cycles to technological capabilities as a primitive)
and develop a set of instrumental variables rooted in the technological characteristics of each
focal firm’s distant peers.3 These instruments include measures of the extent to which distant
3
Distant peers are those with somewhat-related products, but not very-related products (those in the
6
peer firms are using digital technology, whether distant peers are experiencing technological
change, whether they are in highly fluid product markets, and the extent to which they are
focused on optimizing their supply chains. Our use of four instruments allows us to rerun
our main analysis using instrumented life cycle stages based on technological characteristics.
The results support our baseline noninstrumented results. Third, we consider an alternative
set of instruments based on using the average life cycle stages of a focal firm’s distant product
market peers, and again find supportive results. The use of distant peers is motivated by the
network econometrics literature as a means of reducing the scope for endogeneity (especially
limiting the scope for alternative explanations that rely on the focal firm or its close rivals)
to drive our results.4
An array of additional tests further supports our life cycle interpretation. First, although
we include firm fixed effects, our results are additionally robust to including the lagged invest-
ment dependent variable, reducing the scope for mechanical influences from past investment
that might relate to textual content. Second, we explore a number of placebos and find
that using our life cycle instruments to instrument for competition (instead of for the life
cycle) produces no results. Third, we find that our results do not obtain if one uses firm
age to construct an alternative four-stage life cycle. Fourth, we examine financing policies
in addition to investment policies and find consistent interpretations. Finally, we consider
an array of validation tests that illustrate that the primary signal in our variables is con-
sistent with Abernathy and Utterback. One validation test taps into plausibly exogenous
variation from the technology bust of 2001 and the financial crisis of 2008, where we find
that these shocks are associated with expected changes in firms’ life cycle stages. Although
substantial evidence supports the existence of genuine life cycle effects, we cannot rule out
that endogeneity might explain some of our findings.5
same TNIC-2 industry, but not the same TNIC-3 industry).
4
See Bramoulle, Djebbari and Fortin (2009) for theory and Cohen-Cole, Kirilenko, and Patacchini (2014)
for a recent application to finance. These studies indicate that using the attributes of more distant peers
can produce exogenous variation that can be used as an instrument.
5
Although these tests provide some assurance that our variables measure genuine life cycle effects, they
do rely on the management’s perceptions. This limitation is not unique to our study as it spans studies
based on survey evidence and also Kaplan and Zingales (1997) and Hadlock and Pierce (2010), which are
based on human and machine interpretations of 10-K content, respectively.
7
Aside from a few noteworthy exceptions, decades of empirical research have relied on
highly aggregated measures of investment opportunities. The product life cycle offers a new
approach to disaggregate and refine these opportunities, and in turn, to refine empirical
predictions in settings beyond our own. One recent exception is Peters and Taylor (2017),
who argue that the calculation of Tobin’s q should be updated to incorporate estimates of
intangible capital. We will discuss their innovative approach below, and we note that our
approach is distinct but yet complementary to theirs. In contemporaneous work, Andrei et
al (2018) propose a learning model in which investors observe realized cash flows but learn
about the firm’s long-run cash flows. In their model, the explanatory power of the simple q
equation is higher for more R&D-intensive industries. This might help to explain our finding
of increased explanatory power of q equations over time.
1. Related Literature and Theory

Creating value in a product market often requires going through a set of predictable stages
in which the relation between q and different types of investment changes. For example,
consider a new commercial airliner manufacturer. Initially, the firm will invest in design
and development. Over time, the firm will shift investment to plant and process efficiency.
Thereafter, the mature firm’s value will come from sales in a continuous and stable fashion.
Finally, as new competitors arise, the focus will be on supporting products still in service and
phasing out obsolete models. Managers can create value in each stage, but such strategies
are state-specific and entail different relations between q and investment in R&D, physical
plant, and acquisitions.
Our analysis of the relation between q and investment builds on Abernathy and Utter-
back’s (1978) highly cited classification of product life cycle stages. They argue that projects
traverse a set of stages: (1) product innovation, (2) process innovation, (3) stability and ma-
turity, and, finally, (4) product discontinuation. We take these stages as given, and model
a firm as a portfolio of products, each potentially being in a different life cycle stage.6 Our
6
Klepper (1996), and Klepper and Thompson (2006) suggest that industries consist of submarkets. We
posit that participation in each submarket can be viewed as a distinct project, each of which cycles through
the Abernathy and Utterback stages.
8
hypotheses then rest on managers maximizing firm value given trade-offs. This foundation
is akin to the Jensen and Meckling (1976) view of the firm as a ”nexus of contracts.” In
our setting, these contracts incentivize the optimal set of activities given the firm’s life cycle
stages. The resultant contracts and activities will then vary over the life cycle.7
1.1 Related literature
Our paper is related to recent work on life cycles measured using firm age. Loderer, Stulz
and Waelchli (2016) argue that, as firms age, they become more rigid and less able to respond
to growth opportunities.8 Product market competition slows this process whereas investor
monitoring speeds aging as firms prioritize investor relationships. Arikan and Stulz (2016)
show that acquisition activity follows a U-shaped pattern with respect to age. We find many
results that are consistent with these studies: age is relevant empirically and life cycle effects
are pervasive. We also find that issuance and investment are codetermined, reinforcing the
need for capital as a primary issuance motive (see DeAngelo, DeAngelo, and Stulz 2010).
However, we also show that a comprehensive model of product life cycles generates many
novel and economically important findings.
Our simple model and subsequent analysis are motivated by the q-theory of investment
(Hayashi 1982). This theory predicts that the firm’s investment opportunities can be mea-
sured as the ratio of the firm’s market value to the cost of reproducing the firm’s assets.9
q-theory model has been widely studied in finance, both in structural models, such as Hen-
nessy, Levy, and Whited (2007), and in reduced-form models, such as Chen and Chen (2012),
Erickson and Whited (2000), Peters and Taylor (2017), and Harford (2007). Given assump-
tions about firm homogeneity and competition in the market for outputs and inputs, the
standard predictions regarding investment and q obtain. One maintained assumption is a
positive relation between future cash flows and ex ante capital stock. However, this might
not hold in practice. For example, an R&D firm might have a high market value but might
7
A recent paper by Hajda (2019) analyzes product introductions in consumer industries using supermarket
scanner data. We analyze a broader range of activities and industries as our primary data are drawn from
10-Ks.
8
Maksimovic and Phillips (2008) explore industry life cycles and capital expenditures.
9
See Hassett and Hubbard (1997), Caballero (1999), and Philippon (2009) for reviews.
9
not purchase production facilities before it has a product (or even afterward if it outsources
production). Also, a mature firm can increase its market value, and hence its q, by shutter-
ing inefficient operations. Scholars agree on such variation, but such cases are not reflected
in the workhorse q-model due to tractability. We construct a simple life cycle model that
quantifies this heterogeneity.
1.2 An illustrative model
To model the full life cycle, we consider five dates. At time t0 , a product opportunity
arrives. The first life cycle stage Life1 begins at t0 and ends at t1 , by which time the firm
knows whether or not the project will be successful. At time t0 , the firm simultaneously (A)
commits a fixed sum R > 0 to product development, in which case the probability of success
is π > 0 and (B) decides how much capacity k1 to build at cost γ2 k12 .10 If the project fails,
the capacity k1 has zero scrap value. The success of the project is then revealed just after
t1 . The subscript “1” on k1 indicates that the investment was made in the first-stage Life1.
The firm reports its investment k1 at t1 , and the project’s first-stage q (q1 ) is recorded.
Just after the end of the first stage at t1 , if the project is successful, the firm enters Life2 and
decides how much additional productive capacity k2 to build at cost γ2 k22 .11 The additional
capacity becomes available at the end of Life2 at t2 . The firm then releases its financial
statements and q2 is reported. This staging of capital investment (some investment is made
in each life cycle stage) using the standard convex cost structure creates an incentive to
smooth investment over Life1 and Life2. In the spirit of AU, Life1 capacity is risky and must
be acquired before the outcome of product development is known, whereas in Life2, the firm
knows the outcome of product development before it acquires productive capacity.
At the end of Life2, at t2 , the firm has exhausted its organic growth opportunities (as
indicated by AU) and the following sequence occurs. The firm enters Life3 and can add k3
10
It is straightforward to make the probability of success endogenous by allowing the firm to trade off
the cost R against the probability of success π = f (R) to optimize the net value of the project. If the firm
decides not to invest in development, the project does not proceed.
11
It is straightforward to specify a lower γ in Life2 relative to Life1 to account for process learning that
occurs over the second stage. Following Hayashi (1992), we use quadratic capital adjustment costs, which
are standard in the literature. Adjustment costs also can be scaled so that it is less costly to add k2 units
of capital to a larger preexisting capital base. Throughout, we focus on interior optima.
10
µγ 2
more units of additional capacity through inorganic acquisitions at cost k ,
2 3
where µ > 0
allows for a differential cost of acquisition of capacity and its integration.12 The project
comes online and the firm produces. It then realizes cash flow φ(k1 + k2 + k3 ), where φ is
the net price of output. The firm then releases its financial statements and q3 is reported.13
After time t3 , the project enters Life4. To produce in Life4, the aging capacity requires
product extension investments to extend the declining product line. The firm can choose to
ωγ 2
revitalize k4 units of existing capacity at a cost k ,
2 4
where ω reflects the differential cost
of revitalization. At the same time the remaining units are disposed at zero cost.14 At the
end of time t4 , all profits are distributed and any remaining capital depreciates to zero an
instant later. This simplifying formulation obviates the need to formally introduce another
date and to further discuss resale or redeployment values at the end of the project. We also
assume a constant discount factor δ across the dates.
1.2.1 Model solution.
We solve for the investment decisions at each stage using backward induction starting with
Life4. We initially treat each project as a separate firm, but relax this restriction later. At
the end of Life4, the firm will have a value v4 = φk4 , where k4 is the number or revitalized
capacity units and φ is the price of the output each unit of capital has produced. At time
t3 , the beginning of Life4, the firm chooses k4 to maximize value v3 :
1
v3 = δφk4 − ωγk42 (1)
2
The expression indicates the value realized from refurbishing previously installed capacity
δφ
and selling the output produced. Solving gives k4∗ = ωγ
. Substituting back into v3 gives the
optimal valuation at time t3 :
12
Such acquisitions may not always be possible, as there has to be a match with a seller. We could model
this by assigning a probability of such a match and solving the model accordingly. This does not yield
additional insights, and we omit this step.
13
As in Hayashi (1982), the production function has constant returns to scale. In principle, φ can be
stochastic and time dependent (φ̃t ). We would then distinguish between Et (φ̃t+τ ) that the firm uses in its
decision making at time t, and the realization φ̃t+τ . For simplicity, we hold φ constant.
14
This is a simplifying assumption to avoid additional complexity. In a more general model, there would
be a salvage value.
11
δ 2 φ2 δφ2
v3∗ = , v4∗ = (2)
2ωγ ωγ
At the beginning of Life3 at time t2 , the firm has preinstalled capacity k1 + k2 . In Life3,
the firm will then add k3 units of capacity through acquisitions. At the end of the period, the
value of the installed capacity is φ(k1 + k2 + k3 ) and the continuation value of this capacity
in Life4 is v3∗ . For notational ease, we define φ∗ as the total end of Life3 value per unit of
installed capacity (k1 + k2 + k3 ),
v3∗
φ∗ = φ + (3)
k1∗ + k2∗ + k3∗
The amount of capacity that the firm obtains through acquisitions and installs in Life3
µγ 2
is k3 , which has a cost k .
2 3
At the end of Life3 at time t3 , the firm produces one unit
of output per unit of installed capacity. The total value of this capacity is comprised of
production revenues realized at time t3 (with revenue φ per unit) and the continuation value
v3∗ (both parts together valued at φ∗ per unit as defined above). The firm thus solves for k3
by maximizing
µγ 2
− k + δφ∗ (k1 + k2 + k3 ) (4)
2 3
δφ∗
Solving gives k3∗ = µγ
. Investment increases with the price of output and falls with the cost
of acquiring and installing capital. Substituting, we obtain the firm value at the beginning
of Life3, at t2 :
2
δφ∗ δφ∗ δφ∗

µγ
v2∗ =− ∗
+ δφ k1 + k2 + ∗
= δφ k1 + k2 + (5)
2 µγ µγ 2µγ
At the beginning of Life2 (time t1 ), the firm selects k2 to maximize value
δφ∗
γ
γ 2
v1∗ 2 ∗ 2 ∗
= max − k2 + δv2 = max − k2 + δ φ (k1 + k2 + ) (6)
k2 2 k2 2 2µγ
δ 2 φ∗
Solving gives k2∗ = γ
. Investment increases with the price of output and falls with the
12
cost of installing capital. Substituting, we obtain the firm value at t1 :
2
δ 2 φ∗ δ 2 φ∗ δφ∗ δ 2 φ∗ δφ∗

γ
v1∗ =− + δφ k1 +∗
+ ∗
= δφ k1 + + (7)
2 γ γ 2µ 2γ 2µγ
At time t0 , the firm selects k1 before product development uncertainty is resolved.
γ
v0∗ = max − k12 + δπv1∗ (8)
k1 2
δ 3 φ∗ π
Solving, we obtain k1∗ = γ
. As expected, k1∗ is higher when the probability of success
is higher (i.e., when π is larger).
Next, we obtain the q values for each stage. q4 can be calculated directly from Equation
v4∗
(2) and equals k4∗
= φ. Given the constant returns to scale assumption, the project’s q
observed by the market at time t3 (just before the firm cash flows are distributed from
φ∗ (k1∗ +k2∗ +k3∗ )
production in Life3) is q3 = (k1∗ +k2∗ +k3∗ )
= φ∗ . This valuation incorporates the continuation
value at time t3 , φ∗ rather than φ. To relate q4 to q3 , the Internet Appendix derives the
relation ζ (as a function of primitives) between φ and φ∗ , where φ∗ = ζφ, so that q4 = ζ1 q3 .
In calculating q1 the market takes into account the expected value of the project at time
t2 divided by the installed capital base k1 . To calculate q1 , the market divides the value of
the project v1∗ given by Equation (7) by the number of installed units at time t1 , which is
given by k1∗ . Thus, q1 = (δ 2 + 2π
δ 1
+ 2µπ )φ∗ . Note that we use “A1 ” henceforth to refer to the
δ 1
quantity (δ 2 + 2π + 2µπ ) for convenience. When the project is very risky (i.e., π is low), upon
successful completion, q1 will be high.15 Similarly, to obtain q2 , we divide v2∗ by k1∗ + k2∗ to
∗
1
obtain q2 = δ + 2µ(δπ+1) φ = A2 φ∗ .
Thus far, our analysis has occurred at the project level. We define the firm as a portfolio
of projects at different stages in their life cycles. Without loss of generality, assume that the
firm has ni projects at each stage. Then the value of the firm, V , is the sum of the values of
15
If q1 is measured before the uncertainty is resolved, the expression for A1 is multiplied by π in all
subsequent analysis. In that case, risky projects are associated with a low q1 .
13
each of the firm’s projects.
V = n1 v1∗ + n2 v2∗ + n3 v3∗ + n4 v4∗ (9)
The firm’s total investment is also the sum over projects.
1.3 Model implications
To take our model to data, we aggregate our product-level model of investment to firm-level
observables. Our focus is on understanding the investment policies of firms with exposure
to different stages of the life cycle. Thus, firm activities are being driven by two separate
factors. The first, as in the classic q-model, is temporal shocks to firm value.16 Second, as
each product progresses through the life cycle, holding φ constant, the incentives to invest
change. As a result, there is a different value function and a different relation between the
firm’s activity and each individual stage’s stand-alone qi , i = 1, .., 4 (the q that would be
observed if the project was a separate firm at each stage). We now investigate how this
source of variation, which is new to the literature, affects the q-sensitivity of investment.
To obtain the relation between aggregate investment at the firm level and the firm’s
overall q, we have to aggregate the q 0 s of the individual projects. The aggregation is in two
dimensions. All products in the same stage have the same marginal valuations of invest-
ment and therefore the same q’s. We adopt the identifying assumption that our text-based
exposures to the life cycle stages (denoted θk ) measure the firm’s focus and are proportional
to the number of products in each stage. Thus, using our earlier results that q1 = A1 q3 ,
q2 = A2 q3 , and q4 = ζ1 q3 , we obtain
θ1 k1 + θ2 k2 + θ3 k3 + θ4 k4 = a + θ1 b1 q1 + θ2 b2 q2 + θ3 b3 q3 + θ4 b4 q4
b4
= a + θ1 b1 A1 q3 + θ2 A2 b2 q3 + θ3 b3 q3 + θ4 q3 ,
ζ
While we do not observe q3 directly, we can proceed by denoting q3 = λQ, where Q is the
16
As noted above, this channel is shut down for the purposes of exposition but can be made explicit by
making the value φ (and thereby also φ∗ ) depend on time, φt .
14
firm’s total observed q, and substitute to obtain17
T otal Investment = a + β1 θ1 Q + β2 θ2 Q + β3 θ3 Q + β4 θ4 Q, (10)
δ 3 πλ δ2 λ δλ δλ
where β1 = γ
, β2 = γ
, β3 = µγ
, and β4 = µγζ
. Firm-level level investment is a func-
tion of q conditioned on the life cycle stages. In general, the βk coefficients (q-sensitivities)
will be different for the different stages.
This equation can be estimated in aggregate or separately for each component.
1.3.1 Competition and q -sensitivities.
We posit that the nature of investment through the cycle might vary across competitive and
less competitive markets. Multiple studies have suggested that managers in less-competitive
markets might prefer the “quiet life” and therefore might be less willing to invest than
are managers running firms in more competitive markets. For example, Hart (1983) shows
that competition can act as a governance mechanism, and its absence can increase quiet-life
agency conflicts. On the empirical side, Giroud and Mueller (2010) find that firms in compet-
itive industries maintain their operating performance even when acquisition threats decline
(and hence competition likely disciplines managers and reduces agency conflicts), whereas
firms in concentrated industries do not. Nickell (1996) finds that total factor productivity
growth is higher in competitive industries. We follow Hart (1983) and model quiet life agency
conflicts through the cost function, as frictions created by agency conflicts have the same
impact as do frictions created by higher costs. We thus assume that our cost parameter γ
has two parts (one due to operational adjustment costs and another due to quiet life agency
costs): γ = γoper + γquiet . Following Hart (1983) and the broader literature cited above, we
assume that agency conflicts are more severe in less competitive industries.
The sensitivity of investment to q in the first and second stages are thus
δ 3 πλ δ2λ δλ δλ
β1 = , β2 = , β3 = , β4 = .
γoper + γquiet γoper + γquiet µ(γoper + γquiet ) µ(γoper + γquiet )ζ
17
While we do not need to obtain the parameter λ, it can be derived using the fact that life cycle exposures
sum to unity, as shown in the Internet Appendix.
15
These expressions indicate that q-sensitivities are strictly decreasing as quiet life agency
conflicts γquiet increase. This result obtains in all stages of the life cycle and motivates a
final prediction we later test: we expect q-sensitivities to be higher in competitive markets
than in less competitive markets.
Competition can also reduce profits as firms compete for rents. In our setting, this would
imply lower prices φ in competitive relative to concentrated markets. As the equation above
illustrates, however, our model predicts that q-sensitivities are not sensitive to φ. This
is intuitive as a lower φ would result in both a lower q and a lower level of investment,
and, hence, sensitivity is invariant to φ. As a result, our model predicts no price effects,
indicating that the adjustment cost effects discussed above should be more important in
predicting q-sensitivities for high versus low competition markets.
We also note that Abel and Eberly (2011) provide an alternative theory that explores
competition and the q-sensitivity of investment. Although their framework departs from
standard q-theory by making alternative assumptions about the firm’s production function
and adjustment costs, their model directly generates the prediction that q-sensitivities are
higher in more competitive industries.
2. Data and Methods

Our new life cycle variables are purely derived from publicly available 10-K text using text
processing software provided by metaHeuristica LLC. This software employs “chained con-
text discovery” (see Cimiano 2010) and has prebuilt modules for fast and flexible querying
and visual interpretations.
2.1 Data
Our sample begins with the universe of Compustat firm-years with adequate 10-K data
available between 1997 and 2017. We exclude financial firms (those with SIC codes in the
range [6000,6999]). After further limiting the sample to U.S. publicly traded firm-years with
10-Ks in the metaHeuristica system (both current and lagged), nonmissing data on operating
16
income and Tobin’s q, sales of at least $1 million, and assets of at least $1 million, we are left
with 68,798 firm-years. Our sample of 10-Ks is extracted using metaHeuristica and covers
all filings that appear as “10-K,” “10-K405,” “10-KSB,” or “10-KSB40.” We query each
document for its fiscal year, filing date, and the central index key (CIK) and link each 10-K
document to the CRSP/COMPUSTAT database using the central index key (CIK), and the
mapping table provided in the WRDS SEC Analytics package.
2.2 The product life cycle
Our goal is to use direct textual queries to identify the life cycle stage of a firm’s product
portfolio. This “anchor-phrase” method differs from simple bag-of-words methods because it
uses textual proximity to provide context. This methodology has been used in past studies
including Hoberg and Maksimovic (2015) and Hoberg and Moon (2017). Our proposed
product life cycle has four stages: (1) product innovation, (2) process innovation, (3) stability
and maturity, and (4) product discontinuation. For parsimony, we will refer to these stages
as Life1, Life2, Life3, and Life4, respectively. Critically, our research requires that firms
discuss these stages in their 10-K. Here, we point readers to Regulation S-K, where Item
101, for example, requires that firms provide “an explanation of material product research
and development to be performed during the period covered” by the 10-K. A substantial
amount of such text would indicate a firm with a high loading on the product innovation
stage. Regarding process innovation, the same disclosure rules require the firm to disclose
its results from operations, of which discussions of the costs of production are a significant
component. A firm in the third maturity stage should be characterized by discussions of
continuation and market share, but without reference to product or process innovation.
Finally, a firm in the fourth stage will discuss obsolescence and product discontinuation.
We empirically model the stages of a firm’s product portfolio as a four element vector
{Life1, Life2, Life3, Life4}, such that each of the four elements is bounded in [0,1], and
the sum of the four components is unity. We expect firms to have nonzero loadings on
more than one of these stages in any given year, and the relative intensities of each stage
indicate the firm’s product portfolio exposure to the cycle. For example, a firm with a
17
vector {.6,.3,.1,0} would overall be seen as earlier in the life cycle than a firm with weights
{.1,.3,.3,.3}. However, both firms have some exposure to the product innovation, process
innovation and maturity stages.
We construct our measures of product life cycle to ensure that they identify the life cycle
exposures of the firm’s products, and that they are not mechanically related to investment
activities. To do so, we exclude all 10-K paragraphs that explicitly mention capital expen-
ditures or R&D. In particular, we exclude paragraphs if they contain the following phrases
(our results are robust to skipping this step):
General exclusions: capital expenditure* OR research and development
To measure the firm’s loading on the first-stage “Life1,” we identify all paragraphs in a
firm’s 10-K (after applying the above exclusion) that contain at least one word from each of
the following two lists (an “and” condition, not an “or” condition).18
Life1 list A: product OR products OR service OR services

Life1 list B: development OR launch OR launches OR introduce OR introduction OR
introductions OR new OR introducing OR innovation OR innovations OR expansion OR
expanding OR expand
To measure the firm’s loading on “Life2,” we identify all paragraphs in a firm’s 10-K
(after above exclusions) that contain at least one word from the following lists.
Life2 list A: cost OR costs OR expense OR expenses

Life2 list B: labor OR employee OR employees OR wage OR wages OR salary OR salaries
OR inventories OR inventory OR warehouse OR warehouses OR warehousing OR trans-
portation OR shipping OR freight OR materials OR overhead OR administrative OR man-
ufacturing OR manufacture OR production OR equipment OR facilities OR facility
To measure the firm’s loading on “Life3,” we require three lists. A firm’s 10-K must
contain at least one word from each of the first two lists (lists A and B below), and must
not contain any words from the third list below (list C). The exclusion ensures that Life3
18
Note that Life1 is focused on providing a metric for changes in the firm’s product line rather than inputs
or expenditures like R&D or advertising expenditures.
18
is characterized as the static state of product maturity as the exclusion list is based on the
union of the other three dynamic life cycle stages.
Life3 list A: product OR products OR service OR services

Life3 list B: line OR lines OR offerings OR mix OR existing OR portfolio OR current
OR categories OR category OR continue OR group OR groups OR customer OR customers
OR core OR consists OR continue OR provide OR providing OR provided OR provider OR
providers OR includes OR continued OR consist
Life3 list C (exclusions): development OR launch OR launches OR introduce OR intro-
duction OR introductions OR new OR introducing OR innovation OR innovations OR ex-
pansion OR expanding OR expand OR future OR obsolete OR obsolescence OR discontinued
OR discontinue OR discontinuance OR discontinuation OR discontinues OR discontinuing
OR cost OR costs AND expense OR expenses
To measure the firm’s loading on “Life4,” we identify all paragraphs in a firm’s 10-K that
contain at least one word from each of the following two lists.
Life4 list A: product OR products OR service OR services OR inventory OR inventories

OR operation OR operations
Life4 list B: obsolete OR obsolescence OR discontinued OR discontinue OR discontinuance
OR discontinuation OR discontinues OR discontinuing
The above queries result in a count of the number of paragraphs that hit on each of
the four stages Life1 to Life4. We then compute our firm-year life cycle exposure vector
by dividing the four paragraph counts by the summed paragraph counts over all four. The
result is a four-element vector for each firm-year {Lif e1, Lif e2, Lif e3, Lif e4} that sums to
one with nonnegative elements in [0, 1].
To add intuition for the content of firm life cycle disclosures, we report the top-60 terms
that cluster in paragraphs associated with each life cycle stage in Table IA.1 in the Internet
Appendix. The vocabularies are intuitive and support our interpretations.19 We also measure
19
The word lists are generated by using tfidf weights to compare word frequencies in the query paragraphs
to the unconditional word frequencies in 10-Ks overall. We thank Christopher Ball at metaHeuristica for
providing this tool kit, and we thank an anonymous referee for this suggestion.
19
10-K document length (“Whole 10-K Size”) as the natural logarithm of the number of
paragraphs in the given firm’s 10-K. Our results are robust to including or excluding this
variable as a control.
2.3 Measuring Q
The literature has developed multiple measures of Tobin’s q, with each perhaps being ideal
for different applications. We compute q following Gutierrez and Philippon (2018) as the
market value of the firm divided by book assets.20 We are ultimately agnostic on the broader
debate regarding which q is most broadly “the best.” Instead, our goal is to choose a method
for q that is most consistent with our goal of testing product life cycles over a broad array
of investment policies.
Recently, Peters and Taylor (2017) use estimates of intangible capital investment to pro-
vide novel measures of q that take into account capital stocks of both tangible and intangible
capital. Investment in intangibles consists of 20% of SG&A expenses and 100% of R&D ex-
penses in each year, and these stocks then depreciate at 15% to 20% per year. This approach
has many advantages, but it can also confound interpretations in our context. For example,
we expect the nature of R&D and SG&A to vary over the life cycle. High SG&A in the
early stages might build organizational capital, whereas it might reflect high costs of sales in
the later stages. Also, a firm with high recent investments in intangible capital might tran-
sition to maturity, making the rolling adjustments used in this calculation potentially stale
or inadequate regarding their predictive power. To avoid confounding interactions between
the product life cycle itself and measures of q, we estimate q using a generic approach, as
discussed above. However, in our Internet Appendix, we show that we obtain similar results
if we instead use the q from Peters and Taylor (2017). Our results are also robust to the
Erickson and Whited (2000) measurement error adjustment.
20
We compute the market value of the firm as the sum of three parts: (1) equity value computed as shares
outstanding (CSHO) times the share price (PRCC F), (2) book debt (DLC+DLTT), and (3) preferred stock
PSTKL replaced with a zero if missing.
20
2.4 Policy and outcome variables
We examine four investment policies: R&D/assets, CAPX/assets, SDC $ acquisitions/assets,

and SDC $ asset sales/assets. We scale all variables by beginning of period total assets (AT).
The R&D (XRD) and CAPX variables obtain from COMPUSTAT. When R&D is missing,
we assume it to be zero. Both investment ratios are winsorized within each year at the 1%
and 99% levels. We obtain acquirer and target data using both full-firm and partial-firm
asset acquisition data from SDC Platinum. SDC $ Acquisitions/Assets is the transaction
value from SDC divided by the firm’s Compustat assets at the beginning of the fiscal year.
When a firm has multiple transactions, the SDC transaction value is the summed value over
all transactions in the given fiscal year. We include both public and private firm acquisitions
as long as they have nonmissing transaction value information. Analogously, SDC $ Asset
Sales/Assets is the transaction value of assets sold from SDC (computed in an analogous
way to the SDC acquisitions variable). As these SDC-based variables have more extreme
distributions, we winsorize at both the 5% and 95% levels. We also note that our results are
robust if we instead use dummy variables to identify acquisition and asset sale activity.
2.5 Summary statistics and correlations
Table 1 displays summary statistics for our 1997 to 2017 panel of 68,798 firm-year observa-
tions. Panel A reports statistics for our new life cycle variables. We first note that the values
of Life1 to Life4 sum to unity by construction. The table also shows that textual prevalence
is highest for process innovation (Life2), followed by maturity (Life3) and product innovation
(Life1). Discussions of product decline are less common and make up 6.7% of the total text
devoted to all four stages.
Investment rates are also consistent with existing studies. The average firm spends 5.8%
of its assets on R&D, and 6.0% on CAPX annually. Firms acquire assets worth roughly 4.1%
of assets (partial or full acquisitions), and sell roughly 1.1% of assets. The average Tobin’s
q in our sample is 1.84.
Table IA.2 in the Internet Appendix displays the life cycle progressions for two sample
21
companies. Panel A displays Amazon, which begins our sample in 1998 with a high Life1
exposure of 0.44, which gradually declines to less than half this level (0.19) by the end of
our sample. During this time, Life2 has more than doubled from 0.23 to 0.48. These trends
document Amazon’s trajectory through the Abernathy and Utterback cycle as Amazon first
converged its platform (Life1 declines) and increased its Life2 focus on processes, such as
warehousing, supply chain, and package delivery. Panel B displays the results for Tesla, which
also experiences a significant drop in Life1 since its initial public offering (IPO) in 2010 and a
corresponding rise in Life3 as more products have come online and realized relative stability
in the market. Tesla also maintains a heavy focus on Life2 given its well-known focus on
process.
Panel A of Table 2 reports Pearson correlation coefficients. Because they sum to unity,
the Life1 to Life4 variables are negatively pairwise correlated. We also observe that Life1 is
negatively associated with firm age (-22.3%) and Life4 is positively associated with firm age
(15.2%).21 This corroborates a primary prediction of the product life cycle theory. Firms
generally begin life with a large fraction of their product portfolio in the product innova-
tion stage and end life with product discontinuation and eventual delisting. However, one
surprising result is that process innovation (Life2) is positively correlated with age whereas
product maturity (Life3) has close to zero correlation. Results later in the paper will show
that these univariate findings are purely driven by cohort effects, and the ordering of the life
cycle stages relative to aging becomes closer to the theoretical predictions when we focus on
within-firm variation (and control for firm fixed effects). For example, for a given firm in
time series, process innovation precedes product maturity on average.
The table also echoes our finding that firms in different stages of the life cycle focus on
very different investments. Life1 firms focus heavily on R&D (56.1% correlation) and Life2
firms focus on CAPX (27.6% correlation). As we would expect given their product maturity
and potential lack of internal growth options, Life3 and Life4 firms correlate negatively with
both of these forms of investment.
Acquisitions are positively associated with Life3, indicating that mature firms focus
21
Consistent with the literature, firm age is the Compustat listing vintage.
22
on acquisition-based investment options when as their internal growth options (R&D and
CAPX) decline. Life4 firms, in contrast, are negatively correlated with all three forms of
investment (R&D, CAPX, acquisitions) and are positively correlated with being targets of
acquisitions. Hence, the option to sell and transfer assets externally is one way that declining
firms can create value for their shareholders as their products become obsolete.
Panel B of Table 2 reports the autoregressive coefficients of our four life cycle variables.
All four stages are roughly 80% persistent, with Life4 being least persistent at 76.4%. These
results indicate that a firm’s life cycle exposure is stable over time and that movement
through the cycle is a relatively slow process.
Figure 1 illustrates how Life1 to Life4 vary over our sample period for large and small
firm quartiles (based on total assets, sorted annually). As we expect, small firms have
higher values of Life1 than large firms and large firms have higher values of Life2 than small
firms. Life2 is also rising over our sample period for larger firms, indicating more focus on
process. Figure 1 also shows that Life3 is initially much higher for large firms, but it declines
significantly over time. By the end of our sample, the gap between the large and small firms
has essentially closed. Our findings indicate a major transition for large firms that is new to
the literature.
Intuitively, Life4 increases dramatically after the technology bust of 2001 and then only
gradually declines during our sample. This is consistent with higher obsolescence and failure
during this period.
The shift away from the inactive Life3 stage is consistent with larger firms becoming
more dynamic. As Life1 and Life2 are particularly dynamic, we define a firm’s Dynamism
Index as the summed exposure to these first two stages (Lif e1 + Lif e2). Figure 2 shows
how this index changes over time for both small and large firms. At the beginning of our
sample, small firms are more dynamic than large firms, but this gap later vanishes. We
conclude that large firms have undergone a major transformation during our sample.
23
3. Validation
Our life cycle measures are derived using anchor-phrase queries, which require key concepts
to appear in close proximity, ensuring our interpretation through texture. Nevertheless, we
consider two important validation tests.
The first examines the timing predictions of Abernathy and Utterback (1978). The pre-
diction is that product innovation (Life1) should precede process innovation (Life2), which
should precede maturity (Life3), and ultimately decline (Life4). We thus regress each life
cycle variable on firm age (measured as Compustat listing vintage). We note that it is
particularly important to include firm fixed effects in these tests, as only then can we draw
conclusions regarding whether individual firms specifically make transitions over time consis-
tent with the predicted cycle. We cluster standard errors by firm, and results are presented
in Table 3 .
The results for firm age in panel A support the Abernathy and Utterback (1978) life
cycle. Life1 and Life2 are both negatively related to firm age, and, thus, product and process
innovation appear most when firms are young. Life3 and Life4 (stability and decline) are
more prevalent for older firms. Our inferences are little changed with additional controls in
panel B. The only unexpected finding is that the coefficient for Life2 is more negative than
that for Life1. One explanation is that much product innovation occurs when firms are still
private, which we do not observe, which can influence the computed link to firm age. Also,
young firms face financial constraints and might need to pay attention to cost cutting early
in order to preserve liquidity. Consistent with this view, Hoberg and Maksimovic (2015) and
Hadlock and Pierce (2010) show that younger and more innovative firms have more financial
constraints.
Table 4 further explores evolution through transition matrix analysis. We first assign
each firm to the life cycle stage that is most prevalent in its 10-K, and explore how firms that
are in a given life cycle stage in year t-1 transition to other life cycle stages in year t. Panel
A shows results for the full sample and illustrates that the AU life cycle holds on average,
with some subtle deviations. The Life1 stage is reasonably sticky, as it persists 78.2% of
24
the time. When Life1 firms do transition, they are most likely to transition to Life2 (11.7%
likely), although they sometimes transition all the way to Life3 (9.7% likely). Transitions
to Life4, not surprisingly, are uncommon. We also find that the Life2 stage is sticky (91.7%
persistent), and shifts from Life2 favor Life3 supporting AU on average. The Life3 stage is
less persistent and Life3 firms often shift back toward Life2 (17.8% likely). These results
illustrate that cost cutting is likely not a one-time focus as the baseline AU theory might
suggest, but rather, firms optimize costs in a more continuous and ongoing way.
Panels B and C of Table 4 show, consistent with our main thesis, that life cycle effects
interact with competition. In high (low) competition markets, firms initially in earlier stages
of the life cycle tend to stay in earlier stages for longer (shorter) times. These results are
consistent with firms in competitive markets favoring earlier life cycle stages to innovate
and build differentiation from rivals and escape competition. Panels D and E show that the
value of the firm, measured by Tobin’s q, also affects transitions. Firms with lower (higher)
q transition to more mature (less mature) stages of the life cycle. The remaining panels
show results for young versus old firms, small versus large firms, and firms in different Fama-
French five industries. These results show that the AU progression generally holds across
sectors.
Figure 3 reports average life cycle exposures versus firm age percentiles on the x -axis.
The solid line plots the variable’s raw average, and the dotted line represents the average net
of firm and year fixed effects (within-firm variation). Interestingly, the relationship between
Life2 and Life3 and firm age switches sign. This reflects significant cohort effects, with older
cohorts being more process oriented. We conclude that it is important to include firm fixed
effects when making life cycle inferences.22
In a second validation test reported in Table 5 , we examine whether the life cycle predicts
changes in the size of the firm’s product portfolio in the next year. Following Hoberg and
Phillips (2010), we measure product portfolio growth as the logarithmic growth in the size of
the 10-K business description. We predict and find that Life1 positively predicts and Life4
22
Controlling for firm fixed effects is also important given the Cohen, Malloy, and Nguyen (2018) finding
that even small changes in disclosure are highly informative about future outcomes.
25
negatively predicts product description growth. These results are significant at well-beyond
the 1% level despite the inclusion of controls and firm fixed effects. Also, unlike Life1, lagged
R&D expenditures does not predict product portfolio expansion, further illustrating that our
life cycle variables are unique.
4. NBER Recessions and Life Cycle Dynamics

Next, we examine whether major exogenous shocks can affect the positions of product port-
folios within the life cycle. We consider two well-known NBER recessions. The first is the
technology bust, which began in March 2001 and ended in November 2001. As our sample
is yearly, we compare 2001 (recession period) to the prior 3-year period (1998 to 2000). The
second is the financial crisis, which began in December of 2007 and ended in June of 2009.
We thus compare 2008 to 2009 (recession period) to the prior 3-year period (2004 to 2006).
Although it does not affect our results, we omit 2007 because the NBER recession began at
the very end of 2007.
We examine the impact on life cycle stages using both a transition matrix and a regression-
based test. We identify a firm’s ex ante life cycle stage as Life1 if its de-meaned value of
Life1 is higher than the de-meaned values of the other four life stages (we create similar ex
ante dummies for the other stages). To examine ex post shifts, for each firm, we compute
the difference between the current firm-year’s four-element life vector minus the firm-year’s
life vector in the previous year. “Toward Life1” is then a dummy set to one if this difference
for Life1 is more positive than this difference for the other three stages. We compute similar
dummy variables for the other three stages.
In our multivariate regressions displayed on the left-hand side of Table 6 , we regress

these ex post transition dummy variables on the ex ante life cycle stages of the firm, their
interactions with the post-treatment recession dummy, and controls for size, age, and Fama-
French-48 industry and time fixed effects. All standard errors are clustered by firm. To
preserve space, we only report the coefficients and t-statistics for the key interaction terms.
We separately report raw transition matrix changes. For firms binned into each of the
26
four ex ante stages, we report the average values of the four ex post transition dummy
variables (“Toward Life 1,” etc.). The result is an annual 4x4 directed transition matrix. To
assess the impact of each NBER recession, we simply consider the difference in transitions
computed as the values of this matrix in the recession period for each shock minus the values
of this matrix in the pretreatment period.
Panel A of Table 6 displays the results for the technology bust of 2001. We find two
primary effects. First, the technology bust was “destabilizing” (stages became less sticky)
as the “Life1 to Life1“ coefficient is negative and significant with a t-statistic of -4.64 and
the “Life2 to Life2” coefficient is also negative and significant (t-statistic of -2.24). Second,
we find a uniform shift away from Life1 and toward Life4 (all four “to Life1” coefficients
are negative and significant and three of the four “to Life4” coefficients are positive and
significant).
The transitions on the right side of Table 6 show that these results are economically
large. For example, the likelihood that an ex ante Life1 firm will transition toward Life2
or Life4 increases by 6.4 and 8.2 percentage points, respectively. Overall, the destabilizing
nature of the technology bust, and its push toward Life4 are intuitive given the nature of
this event and the failure of many online products. This analysis validates our life cycle
measures and illustrates the real consequences of major recessions.
Panel B of Table 6 shows that, unlike the technology bust, the financial crisis of 2008
to 2009 was stabilizing (increased stage stickiness) as all four reflexive-stage coefficients are
positive and the “Life2 to Life2“ coefficient is also highly significant. The second main result
is a broad shift toward Life3, which differs from the shift toward Life4 for the technology
bust. Because the financial crisis was mainly about financial liquidity rather than economic
distress, these results illustrate that financial constraint shocks tend to slow firm progressions
whereas economic distress can speed progressions. Also, financial constraints lead firms to
transition to the inactive stage rather than obsolescence. The inactive mature stage can be
attractive because it requires no investment and favors accumulation of liquidity (through
operations). Intuitively, this is consistent with liquidity-preservation during the crisis.
27
5. Investment and the Product Life Cycle
In this section, we consider panel data regressions to examine ex post firm investments. The
dependent variables in Table 7 are R&D/assets (panel A), CAPX/Assets (panel B), SDC
Acquisitions/Assets (panel C), and SDC Asset Sales/Assets (panel D). We report results
for the full sample and subsamples based on competition terciles measured using text-based
network industry classifications (TNIC)-3 total similarity (see Hoberg and Phillips 2016).23
The first three columns show basic q-model regressions, where Tobin’s q is the key right-
hand-side (RHS) variable, and we include controls for size, age, firm µi fixed effects, and
year λt fixed effects:
CAP X
[ ]i,t = α0 + α1 Qi,t + α2 log[Assetsi,t ] + α3 log[Age] + λt + µi + i,t (11)
Assets
The latter columns show results for our conditional model, which adds the life-cycle-stage
interactions with Tobin’s q. This is a conditional model because the life cycle stages sum
to unity, and hence the four interaction terms are a valid decomposition of Tobin’s q itself.
This model is motivated by the basic theory in Section 1.:
CAP X
[ ]i,t = α0 + α1 Lif e1i,t Qi,t + α2 Lif e2i,t Qi,t + α3 Lif e3i,t Qi,t
Assets
+α4 Lif e4i,t Qi,t + α8 Log[Assetsi,t ] + α9 Log[Agei,t ] + λt + µi + i,t (12)
Panel A of Table 7 shows that Life1 firms invest more heavily in R&D when the q rises.
This intuitive result is positive and significant across the subsamples and is stronger for
firms in more competitive markets. Most novel, we also find that only Life1 and (to a lesser
extent) Life4 firms have a positive q-sensitivity to R&D. In contrast, Life3 has a negative and
significant q-sensitivity and Life2 sensitivity is near zero. Because only 31% of all products
are in Life1 or Life4 (see Table 1), this implies that positive q-sensitivity to R&D is quite
conditional on the life cycle. The negative sensitivity for Life3 is particularly novel (we are
23
We measure competition using total similarity because its roots are innovation based, and, hence, it is
more closely linked with our life cycle hypothesis. Results are similar using the TNIC HHI.
28
unaware of other studies noting negative sensitivities), and the subsample results indicate
that this result is strongest for firms in competitive product markets. We explore this novel
finding in detail in Section 7.1, and find that it has roots in market structure effects unique
to the healthcare sector.
Turning to CAPX in panel B, we find that Life2 firms have the highest q-sensitivity.
Life3 and Life4 firms have positive but weaker q-sensitivities for CAPX that are one-third
to one-half as large. In contrast, we find a negative and significant Life1 q-sensitivity to
CAPX, which is strongest in the most competitive product markets. This finding is novel
and is consistent with competitive threats occasionally disrupting product development. In
this scenario, q will decline, and the firm might reduce R&D and shift toward tangible
CAPX to begin commercializing any products created thus far. Reinforcing this potential
interpretation, we find that life1 firms experiencing drops in q are indeed more likely to
transition toward Life2 (see Table IA.3 in Internet Appendix).
Panel C of Table 7 examines acquisition investments and shows that Life3 firms have the
highest q-sensitivity and Life4 firms have weak positive sensitivity to q. The high acquisition
sensitivity of Life3 firms is consistent with mature firms facing a dry-well problem on organic
investment, and hence they grow through inorganic investment.24 We also find positive and
significant sensitivities for Life4 firms across all investment policies in panels A to C, and
hence Life4 firms appear to seek opportunities to shift their products back to sustainable life
cycle stages (product extension strategies), and q naturally rises in such cases.
Panel D shows the results for asset sales/assets, and we find that Life4 firms have a signif-
icant and negative sensitivity. Unlike panels A to C, where we expect positive sensitivities,
in panel D the negative sensitivity is natural because asset sales are a form of disinvestment.
The results for Life4 indicate that firms with declining products disinvest more when their
q decreases, which is consistent with “giving up” when the outlook for product extension
strategies is poor. In these scenarios, unwinding the firm’s assets is likely optimal. When q
rises in contrast, as noted above, Life4 firms instead increase all other investments consistent
24
Because Tobin’s q contains information about valuation, a concern is that our results might be driven in
part by motives to use overvalued stock as a means of payment. In unreported tests, we remove transactions
that use stock, and our results are fully robust.
29
with product-extension strategies and cost cutting. Auxiliary tests in Section 7.2 examining
financing policies will further reinforce our evidence of the possibility of “regaining youth”
when q rises for Life4 firms. In that section, we will also document that firms exposed to
Life4 additionally raise more capital in the form of both equity and debt when q rises.
Table 7 also documents an important link between competition and q-sensitivities.

Across all four panels, we find that q-sensitivities are larger for firms operating in more
competitive markets. These findings are consistent with our predictions in Section 1: com-
petition increases the sensitivity to growth opportunities. Table 7 overall shows that both
life cycles and competition matter, and each is distinct. Later, we will examine the economic
impact of both and find that life cycle effects are 1.5x to 2x larger in magnitude.
There are interesting exceptions to the “average” tendency to follow the natural ordering
of investments through the product life cycle. Consider, for example, firms with nontrivial
exposures to Life1, but that are not actively patenting. We conjecture that such firms can
achieve new product introductions through alternatives, such as acquisitions (for evidence on
new product synergies, see Hoberg and Phillips 2010). In Table IA.4 in Internet Appendix,
we rerun the tests in Table 7 for firms that patented in year t − 1 (patenting firms) and
for those that did not (nonpatenting firm). We find that the nonpatenting firms indeed
have a positive and significant q-sensitivity to acquisitions when they have more exposure
to Life1. This supports our prediction that nonpatenting firms might achieve Life1 goals
through acquisitions.
The Internet Appendix shows robustness to (1) using Tobin’s q as measured by Peters
and Taylor (2017) (Table IA.5), (2) using the Erickson and Whited (2017) measurement
error correction (Table IA.6), (3) using a stricter anchor-phrase query method that excludes
more action words (Table IA.7), (4) requiring anchor-phrase word lists appear within a 10-
word window or a 5-word window instead of in the same paragraph (Table IA.8 or IA.9,
respectively), (5) including controls for financial constraints (Table IA.10) and (6) including
lagged investments and their interactions with q as controls using both least squares and the
Blundell and Bond (1998) correction for the correlation between lagged dependent variable
30
and the firm fixed effect (Table IA.11).25 This latter test ensures that our life cycle variables
contain novel content that is not mechanically related to investment policies. In unreported
tests, we also confirm that our results are robust to either including or excluding size and age
as controls. Finally, Table IA.12 examines robustness across the Fama-French five sectors,
and illustrates that our main results are general across sectors.
We conclude that investment follows a natural ordering through the product life cycle,
and this determines the relevance of investment-q-models across various types of invest-
ment. Life1 and Life2 are associated with organic investment in the form of R&D and
CAPX, respectively. As firms mature to Life3, they focus on inorganic investment, and
when products enter decline (Life4), they focus on asset sales (when opportunities are poor)
and product extension strategies (when opportunities are better). Regarding competitive ef-
fects, q-sensitivities are uniformly larger in competitive industries likely due to more dynamic
responses to opportunities.
5.1 Age or competition alone
Table 8 examines whether our main results in Table 7 can be observed without utilizing
the text-based life cycle. We thus examine whether conditional q-models based on firm
age or measures of competition can produce the highly heterogeneous and even negative q-
sensitivities we report. We first sort age into quartiles using annual sorts. We use quartiles
so that each variable can be divided into four stages as is the case for our life cycle model.
Rows 1, 4, 7, and 10 of Table 8 show that a four-stage firm age model generates very little
variation in q-sensitivities for all four investment variables. For example, R&D sensitivity
is nearly unchanged at 0.008 for age bins one to three, and only declines to a still positive
0.003 in the last bin.
The remaining rows of Table 8 examine annual quartile bins based on competition mea-
sured as TNIC total similarity or the TNIC HHI. The results support our earlier-mentioned
finding that q-sensitivities are broadly higher in competitive product markets. However, the
q-sensitivity coefficients remain highly homogeneous and never switch signs. We conclude
25
Additional details are explained in the table header for each test.
31
that neither age nor competition generates our main results for the text-based life cycle in
Table 7 .
5.2 Instrumental variable models
We consider two instrumental variable (IV) models to illustrate the likely mechanisms that
drive our life cycle results. A second goal is to further illustrate that our results are not
mechanically related to potential limitations in the text-based variables. For example, one
concern is that (despite our careful construction to avoid this) the life cycle queries might
pick up mechanical references to investment actions rather than to the life cycle itself. We
thus consider instruments that are constructed without using information about the focal
firm’s life stages, thus mitigating this concern.
Our approach relies on the foundation that firm life cycles are influenced at least in part
by either broad sectoral trends in technological adoption (our first set of instruments) or peer
firm life cycle stages (our second set). In constructing instruments, we follow prior studies
in the network econometrics literature26 and identify peer influences using only the more
distant peers-of-peers. Since peers of peers are not themselves direct peers of the focal firm,
their influence on the focal firm is plausibly exogenous from the perspective of the focal firm.
Our first set of instruments reflects the fact that the AU life cycle is based on technological
capabilities as primitives. We consider four ex ante measurable technological characteristics
defined below, where each is averaged over the focal firm’s distant peers. We define distant
peers as firms that are in the focal firm’s broad TNIC-2 industry, but not its TNIC-3 industry
(see Hoberg and Phillips 2016).
Distant peer “digital focus”: The number of paragraphs in a firm’s 10-K that contain
the word root of either “digital*” or “digitiz*,” divided by the total number of paragraphs
in the 10-K. We average this ratio over the firm’s distant peers.
Distant peer “technological change”: The number of paragraphs in a firm’s 10-K

that contain both the word root “technol*” and the term “change,” divided by the total
26
See Bramoulle, Djebbari and Fortin (2009) for theory and Cohen-Cole, Kirilenko, and Patacchini (2014)
for a recent application in finance.
32
number of paragraphs in the 10-K. We average this ratio over the firm’s distant peers.
Distant peer “supply chain”: The number of paragraphs in a firm’s 10-K that contain
the phrase “supply chain,” divided by the total number of paragraphs in the 10-K. We average
this ratio over the firm’s distant peers.
Distant peer “fluidity”: We start with self-fluidity for each firm, which is the cosine
similarity between the given firm’s 10-K business description in year t relative to its business
description in the prior year (Hoberg, Phillips, and Prabhala 2014). A high fluidity indicates
a rapidly evolving product portfolio and hence a highly malleable technology. We average
self-fluidity over the firm’s distant peers.
We use these four variables as instruments for the focal firm’s life cycle exposures. We
first consider regressions where the focal firm’s life cycle stages in year t+1 are the dependent
variable, and the four ex ante measurable instruments along with controls for size, age, firm
fixed effects, and year fixed effects are the RHS variables.
Table 9 displays the first-stage results and illustrates that the instruments are strong and
intuitive predictors of the focal firm’s life cycle stages. Firms with distant peers focusing on
digital technology are more likely to be in Life1 and are less likely to be in Life4. Technological
change also predicts more Life1 along with less exposure to Life3. This result is sensible given
that technological change is a form of variance, and Life3 is a stage of stability (absence of
variance). Distant Peer Fluidity is positively related to Life1 and Life4, and negatively
related to Life3. This is also sensible as Life1 and Life4 indicate flux on the product side
(introduction or discontinuation), and Life3 is static. Finally, distant peer supply chain focus
strongly predicts more Life2 and negatively predicts Life3. Because supply chain indicates
process optimization, its link to Life2 is intuitive and consistent with AU. These results are
highly significant, indicating relevance of these instruments.
Table 10 reports the second-stage results, where the dependent variable includes the
four investment policies from our main tests. In all, we consider two sets of instruments for
our key life cycle interactions with the focal firm’s Tobin’s q. The first is the four distant peer
technological characteristics noted above, all interacted with the focal firm’s Tobin’s q. Our
33
second set of instruments is simply the average life cycle stages of the focal firm’s distant
peers, again all interacted with the focal firm’s Tobin’s q. The goal is to test robustness
using instruments that focus our identification on the primitive technological underpinnings
of the life cycle constructed using less endogenous distant peers.
The odd-numbered rows in Table 10 are based on the technological characteristic in-
struments, and the even-numbered rows are based on the distant life cycle instruments. The
results are similar for both, and all results strongly reinforce our main findings in Table 7
. In particular, we continue to find our natural ordering of q-sensitivities throughout the
life cycle. q-sensitivities also remain heterogeneous across the life cycle, and we find zero
and negative q-sensitivities as before. The negative sensitivity of life3 firms to R&D is par-
ticularly robust using either set of instruments. The negative sensitivity of life1 firms to
CAPX is robust using the distant peer life cycle instruments, but is not significant using
technological characteristics.
In a final falsification test, we rerun our IV tests, but we replace the four focal firm’s
life cycle stages (what we instrument for) with competition quartiles based on TNIC total
similarity. Table 11 shows that all results vanish. These results further reinforce that our
results do not obtain unless we can condition on life cycles states. Overall, our IV tests
support that our main results are likely driven by genuine life cycle effects, and that they
are unlikely to be driven by unintended mechanical links to the investment policies of the
focal firm. However, we are unable to fully prove causality, as true natural experiments
are not available. Yet these tests illustrate that technological characteristics are empirically
relevant, as suggested by AU.
5.3 Economic magnitudes
We now evaluate the economic significance of our main results. We first compare our con-
ditional life cycle q-model to the basic q-model in the literature, and later compare the
conditional life cycle model to a conditional competition q-model. A key challenge is that
our life cycle model is four-dimensional, as is the case for the quartile-based competition
q-model. Therefore, it is difficult to use sorts due to the large number of sort dimensions.
34
To solve this problem, we follow Back et al (2017) and Hoberg and Welch (2009), who note
similar difficulties with multiple sorts, and we focus on model projections.27 Specifically, we
first compute fitted values from the following two models being compared: the conditional
life cycle q-model and the basic q-model, where Ii,t denotes a given investment policy by firm
i in year t:
Ii,t = a + b1 Lif e1i,t−1 Qi,t−1 + ... + b4 Lif e4i,t−1 Qi,t−1 + i,t (13)
Ii,t = a + b1 Qi,t−1 + i,t . (14)
We run the two models in Equations (13) and (14) annually in the cross-section. We
then perform independent separate annual tercile sorts using the ex ante year t − 1 fitted
values from each of these two models. Finally, we sort firms into nine bins based on low,
medium, and high predicted values (three bins for each model, so nine total bins). We then
report the average ex post year t investment policies for firms in each bin in Table 12 . We
also report intertercile ranges for each sort dimension and use this information to compare
the relative economic impact of the two models.
For R&D, Table 12 shows that the average intertercile range for the life cycle conditional
model is 10.5% of assets, which is larger than the 3.5% of assets attributable to the basic
model. The life cycle conditional model thus has 2.88x more economic impact than does the
basic model. For CAPX, the conditional model’s intertercile range is 4.9% of assets, which is
6.7x more informative than the basic model. For acquisitions, the intertercile range is 1.8%
of assets, which is roughly equal to that of the basic model. For asset sales, the intertercile
range is 0.7% of assets, which fully dominates the basic model, which does not have a positive
intertercile range in these two-way sorts. Overall, these shifts are economically large.
Figure 4 reports analogous time-series evidence regarding the relative explanatory power
of the two models for R&D and CAPX. This is done by reporting annual adjusted R2 based
on running the two models in Equations (13) and (14) annually in the cross-section. The
figure illustrates that not only does the conditional model have a systematically higher R2 for
27
For robustness, we also consider a simpler but less comprehensive sort method (discussed at the end of
this section).
35
both policies but also its improvement relative to the basic model is also increasing during
our sample.
In Table 13 , we compare the life cycle conditional model to the following competition
conditional model (as used in Table 8 ):
Ii,t+1 = a + b1 Comp1i,t Qi,t + ... + b4 Comp4i,t Qi,t+1 + i,t (15)
Comp1 to Comp4 are dummies indicating the TNIC total similarity competition quartile
of the focal firm in year t. As above, we conduct annual independent tercile sorts using the
predicted values of our conditional life cycle model in Equation (13) and for the conditional
competition model in Equation (15). Table 13 shows that the conditional life cycle model
has information that is distinct from the conditional competition model (as both sorts create
significant predictive ability). Moreover, using the same calculations as above, we find that
the life cycle model is 1.64x as informative as the competition model for R&D, and 2.28x
more informative for CAPX. These results highlight the importance of both life cycle and
competition effects. Regarding acquisitions and asset sales, we find that both models are
independently important, and each contributes similar economic magnitudes.
Figure 5 reports analogous time-series evidence when the life cycle model in Equation (13)
is run separately for four subsamples based on above versus below median sorts of TNIC total
similarity and dynamism (Life1+Life2). The highest explanatory power is strongly focused
on the subsample that has both high levels of dynamism and high competition, reinforcing
that both are important.
For robustness, Tables IA.13 and IA.14 in the Internet Appendix report the economic
impact of our life cycle conditional model compared to the basic model and the competition
conditional model using an alternative simpler nonprojection sort method. In particular, for
R&D, CAPX, acquisitions, and asset sales, respectively, we sort using q interacted with the
life stage that it is most sensitive to for the given policy. Our results are similar to those
using the projection sort method above.
36
6. Product Life Cycles Stages, Operating Cash Flows,
and Financing
In recent work, Lee, Shin and Stulz (2021) explore the link between operating profits and
q. The authors refer to the relation between q and operating profits as IO-q, so that
operating prof its = a + bIO−q × q. The IO-q relationship depends on how much profit
the firm extracts from existing assets and the expectations of investors about its ability to
continue to do so.28
The relation between economic rents, competition and Tobin’s q was explored earlier by
Lindenberg and Ross (1981) and Lee, Shin and Stulz (2021). Low competition generates
higher rents, higher market values, and higher q. In low competition environments firms
preserve their rents for longer periods. Thus, an absence of competition leads to a higher q
for a given level of current operating profits, and a higher IO-q. When competition is more
fierce, rents are competed away faster and q is lower for a given level of profits, predicting a
lower IO-q.
The product life cycle has further implications for IO-q. In particular, IO-q can only be
relevant for stages of the life cycle where the firm has products in the market. Hence, we
predict that Life1 exposure can only relate to investment-q and not IO-q. In contrast, stages
Life2 to Life4 are associated with products in the market, and hence these later stages can
have both IO-q and investment-q influences. These predictions are particularly strong for
Life2 (Abernathy and Utterback indicate that improving efficiency and therefore profits is
the core purpose of Life2) and for Life3 (positive shocks to barriers to entry or demand can
improve profits and increase Q).
The predictions regarding IO-q for Life4 are mixed. On one hand, Life4 firms have mature
products in the market, suggesting a positive link to IO-q as is the case in Life3. On the
28
The relationship between operating income and q is conceptually separate from the relationship between
investment and q studied in earlier sections. The former is a relationship between the level of current
operating income and the level economic rents accruing to the firm and is mediated by the level of competition.
The latter is a relationship between the marginal changes in the firm’s assets and the marginal change in its
expected future value.
37
other hand, Life4 firms sell assets, thereby both reducing the associated operating income,
and increasing q.
To test these predictions, we consider return on assets computed as operating income

before depreciation (OIBDP) plus R&D (XRD), all scaled by assets. We adjust operating
income for R&D as we treat R&D as an investment, and not an operating expense (and
OIBDP is defined to be net of expenses and R&D).29 We consider our baseline regressions
from Table 7 but with ROA as the dependent variable.
Table 14 displays the results and confirms our core predictions. In particular, q does
not predict profitability when Life1 is high. However, Life2 and Life3 do have significant
and positive sensitivities to profitability. Life4 is positive but mostly insignificant, consistent
with its mixed predictions.
The results also support our auxiliary prediction that the sensitivity of profitability to
q is stronger in low competition markets than in high competition markets, especially for
firms exposed to the Life3 stage, where AU argue that the link between stable products
and profitability is particularly direct. These findings can also inform results in Gutierrez
and Philippon (2017), who suggest that investment sensitivities are lower in high-profit less
competitive markets. Our results suggest that their findings are linked to firms with stable
and mature products and few organic growth opportunities. These firms have high IO-q,
and as discussed in earlier sections, these firms also have low investment-q.
Putting these results together with our previous findings, investment is more responsive
to changes in q in competitive markets. In contrast, operating income generated by assets in
place is more strongly related to q in less-competitive markets. Overall, we uniformly find
that results for both investments and profitability are highly conditional on the life cycle.
6.1 Negative Life3 R&D sensitivity
Next, we will examine whether the negative R&D q-sensitivity reported in Table 7 for Life3
can be explained at least in part by IO-q considerations.
29
Our results are also robust to following the formulation in Lee, Shin, and Stulz (2021), who do not adjust
for R&D but instead truncate operating income at zero.
38
As a starting point, our earlier analysis in Table IA.12 indicated that the negative life3
R&D sensitivity is unique to the healthcare industry. This sector is further interesting from
an IO-Q perspective due to the well-known role of patents being important in protecting
firms in the drug and medical devices industry (see, e.g., Austin 1993). In these markets,
Life3 corresponds to products that have been fully developed (Life1) and commercialized
(Life2), and their primary role is to generate stable economic rents. Table 15 explores the
relationship between q, profitability, barriers to entry, and the stability of the Life3 stage
itself. We use the Fama-French five industries to identify the healthcare sector, and we report
results for the healthcare sector (panel A) and for all nonhealthcare firms (panel B).
Rows 1 and 5 confirm our earlier finding that the negative q-sensitivity to R&D for Life3
firms is indeed limited to the healthcare sector as the corresponding coefficient is positive
and insignificant in panel B for nonhealthcare firms. Row 2 confirms that Life3 is indeed the
primary profit-generating stage for healthcare firms as the Life3 IO-q-sensitivity is stronger
than for any other life cycle stage. Moreover, this finding is sharper for healthcare firms in
panel A than for other firms in panel B. Outside of healthcare, the link between Life2 and
profitability is stronger, consistent with the expected role of cost-cutting generating profits
in these markets.30
Rows 3 and 4 explore competitive threats (product market fluidity) and the stability of
Life3 itself. When barriers to entry are strong, we expect both lower product market fluid-
ity and increased Life3 stability.31 We find that a high q for Life3 healthcare firms indeed
predicts both lower ex post product market fluidity and increased persistence of the Life3
stage. Comparing with other industries, Life3 firms in healthcare (panel A) are both better
protected from competitive threats and more stable than those in other industries (panel B).
This interpretation is further supported by evidence in Garfinkel and Hammoudeh (2021),
who show that breakthrough therapies in healthcare induce rivals to reduce innovation spend-
ing, consistent with a reduced fluidity, a strengthening of barriers to entry, and a reduced
30
This is further consistent with our earlier findings that even older companies often focus on Life2, as
cost-cutting opportunities can dynamically arise as new technologies emerge.
31
Product market fluidity measures the extent to which rivals are churning their products and hence posing
competitive threats (see Hoberg, Phillips, and Prabhala 2014).
39
need to invest in R&D to escape competition. In our context, their findings are also con-
sistent with our earlier-mentioned negative R&D q-sensitivity in healthcare markets as new
therapies would increase the firm’s q, while diminishing the need for R&D to protect Life3
products.
Note that successful stable firms in healthcare might alternatively want to increase R&D
and develop more new products given the success of their recent innovations. The life cycle
model is crucial to address and test this hypothesis, as this alternative proposed activity
would be in the Life1 category and not the Life3 category. Our results in Row 1 illustrate
that indeed life1 continues to have a strong positive q-sensitivity to R&D in healthcare.
Hence, it is both true that the firm will increase R&D for any Life1 projects when q increases,
and yet reduce R&D for Life3 projects. These richer conclusions are not possible without
the conditional life cycle model. In particular, the first column in Table 15 shows that the
basic q-model can only show that q-sensitivity to R&D is broadly positive, which masks the
much richer state-dependent nature of q-sensitivities in this market.
Taken together, our healthcare results from Table 15 indicate that high Life3 × q
predicts not only more profits but also that profit-generating Life3 products have increased
stability. Because R&D spending is often done to defend markets against rivals (Sutton
1989; Aghion et al. 2005), R&D spending would be increasingly less important as barriers
to entry increase, explaining at least in part why the sensitivity of R&D to q is negative in
these markets.
6.2 Financing policies
Next, we explore auxiliary predictions related to our earlier findings that firms in late stages
of the life cycle (especially Life4) seek product extension strategies to “regain youth” in
the product life cycle when their q rises. To ensure that our inferences regarding evidence
of increased real investment in the form of R&D, CAPX, and acquisitions is real and sub-
stantive, we now examine whether firms in this scenario need to raise additional capital
to fund these strategies. Table 16 runs the model used for our main results in Table 7
, except we consider five new left-hand side variables (Compustat definitions in parenthe-
40
ses):32 equity issuance/assets (SSTK/AT), SEO issuance/assets (SDC Platinum SEO dol-
lars raised/Compustat assets), debt issuance/assets (DLTIS/AT), equity repurchases/assets
(PRSTK/AT), and dividends/assets (DVC/AT).
Table 16 shows that both Life3 and Life4 firms increase payouts when q increases.
These findings are intuitive as mature-stage firms likely have larger surplus cash flows. More
striking, Life4 firms also have a positive q-sensitivity to issuing debt (sharpest in less com-
petitive markets), and a positive q-sensitivity to issuing equity (sharpest in more competitive
markets). These findings reinforce our earlier conclusion that Life4 firms search for product
extension strategies that can help to regain youth in the product cycle when q rises, and
these strategies typically require both new financing and additional investment in the form
of R&D, CAPX, and acquisitions. The shift toward equity in competitive markets likely
reflects higher risk.
Our results add new context to the existing literature. Deangelo, Deangelo, and Stulz
(2010) examine firm age life cycle effects and SEO issuance. Our strong link between equity
issuance and Life1 complements their finding that younger firms issue more equity. Yet our
results are also distinct as we control for firm age, and our focus is on the product life cycle
rather than the firm’s age life cycle. Our results help explain the puzzling finding in DSS
that firm age effects are modest and that many older firms still issue equity. In particular, we
noted earlier that many older firms have high Life1 exposures due to shocks or new products,
and these Life1 firms issue equity. Our paper also speaks to the model of financing growth
by Frank and Sanati (2020). Their model combines trade-offs and collateral requirements
to predict equity issuances early and debt issuance later. We find these later stages require
further differentiation. Debt issuance occurs more especially in Life2 and Life4, and not
in Life3, which is static, and hence firms have little need for capital. Overall, our results
suggest that product life cycle effects are important, and they motivate future research in
32
All variables are from Compustat, except for SEO issuance, which is from SDC as noted.
41
this area.33
7. Conclusion
Motivated by a simple extension to the q-theory of investments rooted in product life cycle
dynamics, we develop a novel four-stage text-based model of the product life cycle that
aggregates to the firm-level. We construct our text-based life cycle model using anchor-phrase
technologies applied on annual firm 10-Ks, which yields a firm-year panel of life-cycle-stage
exposures. The stages are product innovation, process innovation, maturity, and decline.
Our first main result is evidence of a natural ordering of investment policies over the life
cycle. Firms initially have high q-sensitivities for R&D. q-sensitivities then rise for CAPX as
firms shift from product to process innovation. As products reach maturity, firms shift from
organic to inorganic investment in the form of acquisitions. Finally, firms with products
entering decline disinvest (sell more assets) when q declines, and favor product extension
strategies and acquisitions when q increases.
Our second main result is that life cycle and competitive effects are distinct, and their in-
teractions produce economically large insights regarding investment-q and IO-q-sensitivities.
Early in the life cycle, investment-q-sensitivities dominate whereas profits and IO-q dominate
as products mature. Regarding competition interactions, the path of sensitivities through the
life cycle is a low-investment high-rent path in concentrated industries and a high-investment
low-rent path in competitive industries. Inferences regarding both main conclusions are not
possible to observe without the life cycle model.
Our measures and overall platform for testing life cycle hypotheses should prove useful in
many settings where the firm’s policies depend on its exposure and those of its competitors to
the product life cycle. For example, Chen, Hoberg and Maksimovic (2020) use our life cycle
model to show that both firm disclosure policies (such as IP disclosures and information
33
Our results also contribute to a larger literature investigating how factors, such as asset tangibility, size,
and Tobin’s q, affect financing (e.g., Rajan and Zingales 1995; Frank and Goyal 2009). Two seminal theories
have direct implications for life cycles. First, asymmetries of information following Myers and Majluf (1984)
suggest that equity is more likely for early-stage firms. As noted above, our evidence supports this prediction.
Second, trade-offs (Modigliani and Miller 1958) might vary over the life cycle, as partly developed by Frank
and Sanati (2020).
42
redactions in 10-Ks) and actual firm-pairwise internet searches vary with the life cycle.
We also expect that the life cycle model and our dynamism measures should be useful in
understanding major trends and firm responses to ongoing technological and trade shocks.
More broadly, firms also face different risks through the cycle, indicating applications in asset
pricing. Finally, we believe our framework has natural applications to other disciplines, such
as marketing, management, and strategy.
43
References
Abel, A., and J. Eberly. 2011. How q and cash flow affect investment without frictions: An analytic expla-
nation. Review of Economic Studies 78:1179–200.
Abernathy, W. J., and J. M. Utterback. 1978. Patterns of industrial innovation. Technology Review 80:41–47.
Aghion, P., N. Bloom, R. Blundell, R. Griffith, and P. Howitt. 2005. Competition and innovation: An
inverted-U relationship. Quarterly Journal of Economics 120:701–28.
Akdogu, E., and P. McKay. 2008. Investment and competition. Journal of Financial and Quantitative Anal-
ysis 43:299–330.
Andrei, D., W. Mann, and N. Moyen. 2019. Learning and the improving relationship between investment
and q. Journal of Financial Economics 133:251–72.
Arikan, A., and R. M. Stulz. 2016. Corporate acquisitions, diversification, and the firm’s lifecycle. Journal
of Finance 71:139–94.
Arellano, M., and S. Bond. 1991. Some tests of specification for panel data: Monte Carlo evidence and an
application to employment equations. Review of Economic Studies 58:277–97.
Austin, D. 1993. An event-study approach to measuring innovative output: The case of biotechnology.
American Economic Review 83:253–58.
Back K., N. Kapadia, and B. Ostdiek. 2018. Testing factor models on characteristic and covariance pure
plays. Working Paper, Rice University.
Bai, J., T. Phillipon, and A. Savov. 2016. Have financial markets become more informative? Journal of
Financial Ecomomics 122:635–54.
Blanchard, O., C. Rhee, and L. Summers. 1993. The stock market, profit, and investment. Quarterly Journal
of Economics 108:115–36.
Blundell, R., and S. Bond. 1998. Initial conditions and moment restrictions in dynamic panel data models.
Journal of Econometrics 87:115–43.
Bond P., A. Edmans, and I. Goldstein. 2012. The real effects of financial markets. Annual Review of Financial
Economics 4:339–60.
Bramoulle, Y., H. Djebbari and B. Fortin. 2009. Identification of peer effects through social networks. Journal
of Econometrics 113:41–55.
44
Chen, H., and S. Chen. 2012. Investment-cash flow sensitivity cannot be a good measure of financial con-
straints: Evidence from the time series. Journal of Financial Economics 103:393–410.
Chen, Y., G. Hoberg, and V. Maksimovic. 2020. Life cycles of firm disclosures. Working Paper, University
of Southern California.
Chen, Q., I. Goldstein, and W. Jiang. 2007. Price informativeness and investment sensitivity to stock price.
Review of Financial Studies 20:619–50.
Cimiano, P. 2010. Ontology learning and population from text: algorithms, evaluation and applications. New
York: Springer.
Cohen, L., C. Malloy, and Q. Nguyen. 2018. Lazy prices. Journal of Finance 75:1371–415.
Cohen-Cole, E., A. Kirilenko, and E. Patacchini. 2014. Trading networks and Liquidity provision. Journal
of Financial Economics 113:235–51.
DeAngelo, H., L. DeAngelo, and R. Stulz. 2010. Seasoned equity offerings, market timing, and the corporate
lifecycle. Journal of Financial Economics 95:275–95.
Erickson, T., and T. Whited. 2000. Measurement error and the relationship between investment and q.
Journal of Political Economy 108:1027–57.
Frank, M., and V. Goyal. 2009. Capital structure decisions: Which factors are reliably important? Financial
Management 38:1–37.
Frank, M., and A. Sanati. 2020. Financing corporate growth. Review of Financial Studies 34:4926–98.
Garfinkel, J., and M. Hammoudeh. 2021. Competition threats and rival innovation responses: Evidence from
breakthrough therapies. Working Paper, University of Iowa.
Giroud, X., and H. Mueller. 2010. Does corporate governance matter in competitive industries? Journal of
Financial Economics 95:312–31.
Gutierrez, G., and T. Philippon. 2017. Investment-less growth: An empirical investigation. Brookings Papers
on Economic Activity Fall 2017:89–190.
———. 2018. Declining competition and investment in the US. Working Paper, New York University.
Hassett, K., and R. G. Hubbard. 1997. Tax policy and investment, in Fiscal Policy: Lessons from the
Literature, ed. A. Auerbach. Cambridge: MIT Press.
45
Hadlock, C., and J. Pierce. 2010. New evidence on measuring financial constraints: Moving beyond the KZ
index. Review of Financial Studies 23:1909–40.
Hajda, J. 2019. Product market strategy and corporate policies. Working Paper, Swiss Finance Institute.
Hart, Oliver. 1983. The market mechanism as an incentive scheme. Bell Journal of Economics 14:366–82.
Hayashi, Fumio. 1982. Tobin’s marginal q and average q: A neoclassical interpretation. Econometrica 50:213–
24.
Hennessy, C., A. Levy, and T. Whited. 2007. Testing Q theory with financing frictions. Journal of Financial
Hoberg, Gerard, and Vojislav Maksimovic. 2015. Redefining financial constraints: A text-based analysis.
Hoberg, G., and K. Moon. 2017. Offshore activities and financial vs operational hedging. Journal of Financial
Hoberg, G., and G. Phillips. 2010. Product market synergies in mergers and acquisitions: A text based
analysis. Review of Financial Studies 23:3773–811.
———. 2016. Text-based network industries and endogenous product differentiation. Journal of Political
Economy 124:1423–65.
Hoberg, G., G. Phillips, and N. Prabhala. 2014. Product market threats, payouts, and financial flexibility.
Journal of Finance 69:293–324.
Hoberg G., and I. Welch. 2009. Optimized vs. sort-based portfolios. Working Paper, University of Southern
California.
Lee, D. W., H. Shin, and R. Stulz. 2021. Why does equity capital flow out of high Tobin’s q industries?
Lindenberg, E., and S. Ross. 1981. Tobin’s Q ratio and industrial organization. Journal of Business 54:1–32.
Jensen, M., and W. Meckling. 1976. Theory of the firm: Managerial behavior, agency costs and ownership
structure. Journal of Financial Economics 3:305–60.
Klepper, S. 1996. Entry, exit, growth, and innovation over the product life cycle. American Economic Review
86:562–83.
46
Klepper, S., and P. Thompson. 2006. Submarkets and the evolution of market structure. RAND Journal of
Loderer, C., R. Stulz, and U. Waelchli. 2016. Firm rigidities and the decline in growth opportunities. Man-
agement Science 63:1–20.
Maksimovic, V. 1988. Capital structure in repeated oligopolies. RAND Journal of Economics 19:389–407.
Maksimovic, V., and G. Phillips. 2008. The industry life cycle, acquisitions and investment: Does firm
organization matter? Journal of Finance 63:673–708.
Miller, M., and F. Modigliani. 1958. The cost of capital, corporate finance and the theory of investment.
American Economic Review 48:261–97.
Morck R., A. Shleifer, and R. Vishny. 1990. The stock market and investment: Is the market a sideshow?
Brookings Papers on Economic Activity 2:157–215
Myers, S., and N. Majluf. 1984. Financing and investment decisions when firms have information that
investors do not have. Journal of Financial Economics 13:187–221.
Nickell, S. 1996. Competition and corporate performance. Journal of Political Economy 104:724–46.
Peters, R., and L. Taylor. 2017. Intangible capital and the investment-q relation. Journal of Financial
Philippon, T. 2009. The bond market’s Q. Quarterly Journal of Economics 124:1011–56.
Rajan, R., and L. Zingales. 1995. What do we know about capital structure? Some evidence from interna-
tional data. Journal of Finance 50:1421–60.
Stigler, G. 1964. A theory of oligopoly. Journal of Political Economy 72:44–61.
Sutton, J. 1989. Endogenous sunk costs and market structure. In Industrial structure in the new industrial
economics, eds. G. Bonanno and D. Brandolini. Oxford, UK: Oxford University Press.
47
Table 1
Summary statistics
Variable Mean SD Minimum Median Maximum # obs.
A. Life cycle variables

Life1 0.242 0.135 0.000 0.223 1.000 68,798
Life2 0.418 0.175 0.000 0.398 1.000 68,798
Life3 0.274 0.127 0.000 0.264 1.000 68,798
Life4 0.067 0.087 0.000 0.032 0.633 68,798
B. Investment and Tobin’s q
R&D/assets 0.058 0.113 0.000 0.000 0.842 68,798
CAPX/assets 0.060 0.076 0.000 0.036 0.589 68,798
SDC $ acquisitions/assets 0.041 0.119 0.000 0.000 0.870 68,798
SDC $ asset sales/assets 0.011 0.045 0.000 0.000 0.418 68,798
Tobin’s q 1.843 1.861 0.198 1.253 24.890 68,798
Summary statistics are reported for our sample of 68,798 observations based on annual firm observations from 1998 to 2017.
The variables Life1-Life4 are based on textual queries to firm 10-Ks in each year. Life1 measures the intensity of product
innovation; Life2 measures the intensity of process innovation; Life3 measures the intensity of stable and mature products; and
Life4 measures the intensity of product decline (discontinuation). Section 2 describes all variables in detail.
48
Table 2
Pearson correlation coefficients
R&D/ CAPX/ SDC SDC Tobin’s log
Row Variable Life1 Life2 Life3 Life4 assets assets acquirer target q age
A. Correlation coefficients
Life2 -0.648
Life3 -0.004 -0.601
Life4 -0.250 -0.118 -0.254
R&D/assets 0.561 -0.337 -0.077 -0.085
CAPX/assets -0.151 0.276 -0.135 -0.120 -0.101
SDC $ acquisitions/assets 0.037 -0.048 0.057 -0.043 0.043 0.102
SDC $ asset sales/assets -0.034 -0.010 -0.015 0.094 -0.017 0.024 0.050
Tobin’s q 0.342 -0.210 0.009 -0.124 0.408 0.118 0.149 -0.028
log firm age -0.223 0.096 0.001 0.152 -0.204 -0.092 -0.091 0.006 -0.210
log assets -0.247 0.149 0.029 0.043 -0.308 0.029 -0.014 0.005 -0.169 0.388
49
Row statistic Life1 Life2 Life3 Life4
B. Persistence Statistics
AR(1) coefficient 0.861 0.874 0.812 0.764

Pearson correlation coefficients (panel A) and autoregressive coefficients (panel B) are reported for our sample of 68,798 observations based on annual firm observations from 1998 to
2017. The variables Life1-Life4 are based on textual queries to firm 10-Ks in each year. Life1 measures the intensity of product innovation; Life2 measures the intensity of process
innovation; Life3 measures the intensity of stable and mature products; and Life4 measures the intensity of product decline (discontinuation). The autoregressive coefficients in panel B
are equal to the ordinary least squares (OLS) coefficient obtained when regressing each variable on its lagged value. Section 2 describes all variables in detail.
Table 3
Product life cycle and firm age
Dependent log log Tobin’s 10-K
Row variable age assets q size Adj R2 obs.
A. Firm and year fixed effects
(1) life1 –0.007 .80 68,965

(–1.88)
(2) life2 –0.065 .77 68,965
(–11.75)
(3) life3 0.057 .65 68,965
(11.37)
(4) life4 0.015 .41 68,965
(4.38)
B. Firm and year fixed effects plus controls
(6) life1 –0.008 0.004 0.003 0.000 .80 68,899
50
(–1.99) (3.90) (10.90) (–12.91)
(7) life2 –0.062 –0.006 –0.002 0.000 .77 68,899
(–11.35) (–3.84) (–4.61) (11.17)
(8) life3 0.051 0.009 0.000 0.000 .65 68,899
(10.49) (6.75) (0.22) (–11.53)
(9) life4 0.018 –0.007 –0.001 0.000 .42 68,899
(5.25) (–6.64) (–6.45) (11.67)

The table reports OLS estimates for our sample of annual firm observations from 1998 to 2017. An observation is one firm in 1 year. The dependent variable is a life cycle variable and
is indicated in the first row. All rows include firm and year fixed effects, and standard errors are clustered by firm. Panel A reports results for a pure life cycle versus firm age model
(where firm age is measured as Compustat listing vintage), and panel B adds key control variables. t-statistics are in parentheses.
Table 4
Life cycle transitions (various subsamples)
Ex
post Ex ante Ex ante Ex ante Ex ante
Row Sample Life Life1 Life2 Life3 Life4
A. Full sample
1 All firms Life1 0.782 0.030 0.076 0.070

2 All firms Life2 0.117 0.917 0.178 0.418
3 All firms Life3 0.097 0.044 0.736 0.081
4 All firms Life4 0.004 0.008 0.010 0.431
B. Low competition
5 Low competition Life1 0.661 0.024 0.061 0.060

C. High competition
9 High competition Life1 0.848 0.040 0.095 0.103

D. Low Tobin’s q
13 Low q Life1 0.693 0.017 0.051 0.033

14 Low q Life2 0.182 0.933 0.222 0.447
15 Low q Life3 0.114 0.039 0.710 0.068
16 Low q Life4 0.011 0.011 0.018 0.452
E. High Tobin’s q
17 High q Life1 0.825 0.058 0.105 0.168

18 High q Life2 0.089 0.885 0.143 0.324
19 High q Life3 0.083 0.052 0.749 0.123
20 High q Life4 0.002 0.005 0.004 0.385
F. Young firms
21 Young Life1 0.805 0.041 0.085 0.104

22 Young Life2 0.103 0.905 0.157 0.370
23 Young Life3 0.089 0.049 0.750 0.067
24 Young Life4 0.002 0.004 0.008 0.459
G. Old firms
25 Old Life1 0.714 0.020 0.062 0.056

26 Old Life2 0.150 0.930 0.197 0.443
27 Old Life3 0.127 0.038 0.729 0.060
28 Old Life4 0.009 0.011 0.012 0.441
The table reports life cycle transition matrix probabilities for our sample of annual firm observations from 1998 to 2017. One
observation is one firm in 1 year. We first assign each firm in each year to a single life cycle stage based on which of its
four exposures is the highest. We then consider each firm’s life cycle stage in year t-1 and year t, and compute an empirical
transition matrix. For example, for the ex ante Life1 column, the first four rows show the fraction of firms that were initially
in Life1 in year t-1 that ended up in each of the four life cycle stages in year t. For example, we observe that 11.7% of these
initial Life1 firms transitioned to the Life2 stage and 9.7% transitioned to the Life3 stage. We report transition probabilities
for the overall sample in panel A and also consider subsamples based on high versus low terciles based on competition (total
similarity), Tobin’s q, firm age, and firm size. Finally, we report transitions for the Fama-French five industry sectors.
51
Table 4
Life cycle transitions (various subsamples) (cont.)
Ex
post Ex ante Ex ante Ex ante Ex ante
Row Sample Life Life1 Life2 Life3 Life4
H. Small firms
29 Small Life1 0.814 0.055 0.100 0.108

30 Small Life2 0.104 0.880 0.170 0.378
31 Small Life3 0.078 0.054 0.717 0.100
32 Small Life4 0.004 0.011 0.013 0.414
I. Large firms
33 Large Life1 0.713 0.015 0.062 0.039

34 Large Life2 0.138 0.938 0.180 0.477
35 Large Life3 0.144 0.039 0.749 0.074
36 Large Life4 0.005 0.008 0.009 0.410
J. FF5 consumer industry
37 Consumer Life1 0.660 0.019 0.056 0.036

38 Consumer Life2 0.220 0.938 0.226 0.523
39 Consumer Life3 0.117 0.034 0.708 0.067
40 Consumer Life4 0.004 0.009 0.010 0.374
K. FF5 health industry
41 Health Life1 0.905 0.116 0.093 0.145

42 Health Life2 0.062 0.799 0.179 0.349
43 Health Life3 0.030 0.077 0.715 0.060
44 Health Life4 0.003 0.008 0.012 0.446
L. FF5 manufacturing industry
45 Manuf Life1 0.621 0.012 0.043 0.060

46 Manuf Life2 0.219 0.952 0.256 0.527
47 Manuf Life3 0.145 0.030 0.690 0.093
48 Manuf Life4 0.014 0.006 0.010 0.320
M. FF5 misc+finance industry
49 Misc+Fin Life1 0.641 0.012 0.045 0.038

50 Misc+Fin Life2 0.201 0.939 0.225 0.389
51 Misc+Fin Life3 0.144 0.037 0.712 0.063
52 Misc+Fin Life4 0.014 0.012 0.018 0.510
N. FF5 tech industry
53 Tech Life1 0.741 0.069 0.101 0.119

54 Tech Life2 0.116 0.842 0.116 0.299
55 Tech Life3 0.140 0.081 0.776 0.114
56 Tech Life4 0.003 0.008 0.007 0.468
52
Table 5
Validation: Product market fluidity and product description growth
Business Whole obs/
log log descr. 10-K R&D Tobin’s adj
Row Life1 Life2 Life4 age assets size size assets q R2
A. Dependent variable = Product description growth
(1) 0.116 0.023 –0.088 –0.333 0.000 63,776

(5.08) (1.35) (–4.22) (–59.66) (4.89) .19
(2) 0.095 0.009 –0.078 –0.062 0.018 –0.343 0.000 0.004 0.006 63,713
(4.18) (0.53) (–3.71) (–7.51) (6.33) (–59.58) (3.87) (0.21) (8.21) .20
B. Dependent variable = Product market fluidity
53
(3) 1.194 0.433 0.317 2.072 0.000 64,419
(6.66) (3.17) (1.99) (42.38) (4.45) .82
(4) 1.008 0.313 0.425 –0.624 0.194 1.977 0.000 0.612 0.037 64,356
(5.70) (2.29) (2.66) (–8.47) (7.79) (40.58) (3.18) (3.27) (6.15) .83
The table reports OLS estimates for our sample of annual firm observations from 1998 to 2017. An observation is one firm in 1 year. The dependent variable is product market fluidity
(see Hoberg, Phillips, and Prabhala (2014) or product description growth (see Hoberg and Phillips 2010) in panels A and B, respectively. All specifications include firm and year fixed
effects. Standard errors are clustered by firm. t-statistics are in parentheses.

Table 6
Tech bust and financial crisis and life cycle transitions
Treatment regression coefficients Treatment regression t-statistics Change in transition matrix probabilities
Dep. variable Life1 Life2 Life3 Life4 Life1 Life2 Life3 Life4 R2 Transition Life1 Life2 Life3 Life4
A. Compare 1998 to 2000 (pretreatment) to 2001 (tech bust NBER recession)

1 Toward Life1 -0.061 -0.073 -0.093 -0.075 -4.64 -4.45 -6.14 -3.25 .293 Toward Life1 -0.063 -0.074 -0.094 -0.072
2 Toward Life2 0.028 -0.037 0.040 0.039 1.87 -2.24 2.44 1.53 .298 Toward Life2 0.064 -0.001 0.078 0.073
3 Toward Life3 -0.040 -0.009 0.007 0.022 -2.64 -0.52 0.51 0.86 .341 Toward Life3 -0.084 -0.052 -0.038 -0.024
4 Toward Life4 0.073 0.118 0.046 0.014 6.64 8.66 3.99 0.69 .130 Toward Life4 0.082 0.127 0.054 0.022
B. Compare 2004 to 2006 (pretreatment) to 2008 to 2009 (fin crisis NBER recession)
5 Toward Life1 0.022 -0.055 0.027 -0.024 1.33 -3.64 1.53 -1.32 .293 Toward Life1 0.024 -0.055 0.028 -0.027
6 Toward Life2 -0.038 0.065 -0.043 0.049 -2.28 4.30 -2.42 2.65 .317 Toward Life2 -0.097 0.011 -0.104 -0.005
7 Toward Life3 0.035 0.014 0.028 0.006 2.00 0.93 1.67 0.34 .317 Toward Life3 0.079 0.057 0.079 0.050
8 Toward Life4 -0.019 -0.024 -0.013 -0.032 -1.69 -2.16 -1.06 -2.33 .123 Toward Life4 -0.007 -0.013 -0.003 -0.018
54
The table reports difference-in-differences OLS estimates and economic magnitudes for three shock periods. We consider the two NBER recessions. In panel A, we compare the
technology bust NBER recession in 2001 to the 3-year period prior (1998 to 2000). In panel B, we compare the financial crisis NBER recession from 2008 to 2009 to the prior 3-year
period (2004 to 2006). Although it does not materially affect our results, we omit 2007 for this shock as this NBER recession officially began at the very end of 2007 in December making
it ambiguous. In all regressions, one observation is one firm in 1 year. The key RHS variables are four dummies indicating the ex ante life stage of the firm (and their interactions with
the post-treatment dummy). To determine the binary life stage of a given firm, we first de-mean the four life stages. A firm is deemed to be in life1 if its demeaned life1 is larger than
its de-meaned values of the other three stages. We perform a similar calculation for the other three stages. The dependent variable is a dummy indicating the ex post change in the life
cycle stage. To compute this dummy variable, we first compute the ex post change in the four life cycle stages for each firm. If the change in life1 is more positive than the change in
the other four stages, the dummy (Toward Life1) is set to one. We compute similar change variables for the other three stages. The table depicts the results of regressions of these ex
post change dummies on the four ex ante life stage dummies and interactions of these variables with the post-treatment dummy for each difference-in-difference test. We additionally

include controls for log assets, log age, and Fama-French-48 industry fixed effects. To conserve space, we only report the coefficients and the t-statistics for the important post-treatment
x initial life-cycle-stage variables. All standard errors are clustered by firm. In the last four columns, we display economic magnitudes based on a transition matrix interpretation of the
same regressions. In particular, we report the difference (post-treatment minus pretreatment) changes in transition probabilities from the ex ante life cycle stages (noted in the column
headers) toward the ex post stages noted in the first column “Transition.” For example, the figure in the fourth row first column of 0.082 indicates that the probability that a life1 firm
moves toward life4 is 8.2 percentage points higher during the tech bust recession than it was during the prior 3-year pretreatment period.
Table 7
Investment panel data regressions
Basic model Conditional model
Tobin’s log log TobQ x TobQ x TobQ x TobQ x log log Adj #
Row Sample q assets age Life1 Life2 Life3 Life4 assets age R2 obs
A. R&D/assets
(1) Full sample 0.007 -0.027 0.007 0.029 -0.002 -0.011 0.014 -0.027 0.005 .833 68,798
(15.58) (-21.53) (2.63) (12.41) (-1.45) (-4.38) (2.38) (-21.90) (2.13)
(2) TSIMM tercile 1 0.004 -0.011 0.006 0.019 -0.001 -0.003 0.007 -0.011 0.006 .815 22,848
(3.15) (-7.35) (2.32) (2.63) (-0.82) (-1.08) (1.22) (-7.42) (2.21)
(3) TSIMM tercile 2 0.004 -0.019 0.003 0.014 0.001 -0.004 0.011 -0.019 0.003 .851 22,837
(5.38) (-10.94) (1.19) (2.89) (0.54) (-1.02) (1.98) (-10.99) (1.10)
(4) TSIMM tercile 3 0.009 -0.048 0.010 0.030 -0.001 -0.016 0.025 -0.047 0.007 .829 22,814
(12.80) (-18.33) (1.69) (9.18) (-0.25) (-3.47) (2.24) (-18.64) (1.16)
B. CAPX/assets
(5) Full sample 0.008 -0.011 -0.018 -0.007 0.024 0.008 0.015 -0.011 -0.018 .585 68,798
(23.87) (-14.93) (-8.10) (-4.14) (14.65) (4.12) (4.00) (-14.83) (-8.04)
(6) TSIMM tercile 1 0.009 -0.012 -0.011 -0.002 0.015 0.008 0.020 -0.012 -0.011 .515 22,848
(10.11) (-9.82) (-2.97) (-0.66) (6.40) (2.34) (3.11) (-9.86) (-2.98)
(7) TSIMM tercile 2 0.010 -0.014 -0.012 -0.004 0.023 0.007 0.015 -0.014 -0.012 .564 22,837
(12.41) (-9.73) (-2.70) (-1.15) (7.27) (2.15) (1.90) (-9.68) (-2.73)
(8) TSIMM tercile 3 0.006 -0.010 -0.028 -0.010 0.030 0.008 0.010 -0.009 -0.026 .637 22,814
(15.16) (-6.90) (-5.86) (-3.97) (10.18) (2.60) (1.47) (-6.77) (-5.37)
C. SDC $ acquisitions/assets
(9) Full sample 0.010 -0.024 -0.016 0.004 0.002 0.026 0.014 -0.024 -0.016 .174 68,798
55
(13.30) (-16.08) (-4.21) (1.39) (0.81) (6.82) (2.03) (-16.26) (-4.07)
(10) TSIMM tercile 1 0.005 -0.025 -0.005 -0.003 0.007 0.008 0.018 -0.025 -0.005 .150 22,848
(4.64) (-9.52) (-0.73) (-0.63) (2.07) (1.73) (1.77) (-9.51) (-0.71)
(11) TSIMM tercile 2 0.009 -0.031 -0.002 0.004 0.007 0.016 0.018 -0.031 -0.002 .206 22,837
(5.96) (-10.91) (-0.32) (0.58) (1.68) (2.16) (1.56) (-10.95) (-0.33)
(12) TSIMM tercile 3 0.011 -0.021 -0.023 0.000 0.003 0.034 0.016 -0.022 -0.023 .222 22,814
(9.61) (-7.80) (-2.93) (-0.05) (0.71) (5.58) (1.15) (-8.14) (-2.81)
D. SDC $ asset sales/assets
(13) Full sample -0.001 0.000 0.003 0.000 -0.001 0.000 -0.005 0.000 0.004 .238 68,798
(-4.42) (1.07) (3.15) (0.26) (-2.09) (-0.48) (-2.99) (1.07) (3.20)

(14) TSIMM tercile 1 -0.001 0.001 0.001 -0.001 -0.001 0.000 -0.002 0.001 0.001 .259 22,848
(-3.08) (1.12) (0.59) (-0.95) (-1.08) (0.17) (-0.63) (1.15) (0.62)
(15) TSIMM tercile 2 0.000 0.000 0.002 0.000 0.000 0.000 -0.005 0.000 0.003 .266 22,837
(-1.46) (0.60) (1.25) (0.05) (-0.45) (-0.16) (-1.28) (0.61) (1.29)
(16) TSIMM tercile 3 -0.001 0.000 0.005 0.001 -0.001 -0.001 -0.011 0.000 0.006 .275 22,814
(-2.69) (-0.06) (2.50) (0.88) (-0.91) (-0.97) (-2.66) (-0.13) (2.52)
The table reports results from firm-year panel data OLS investment-q regressions from 1998 to 2017. The dependent variable is ex post R&D/assets (panel A), CAPX/assets (panel B),
SDC $ acquisitions/assets (panel C), and SDC $ asset sales/assets (panel D). The key RHS variables are the lagged life cycle variables interacted with Tobin’s q. All regressions include
controls for size, age, firm fixed effects, and year fixed effects. All RHS variables are ex ante measurable and are observable in year t − 1. In each panel, we consider four subsamples:
the full sample, and tercile-based subsamples based on annual sorts of TNIC total similarity (see Hoberg and Phillips 2016). All ratio variables are winsorized at the 1% and 99% levels.
The last two columns indicate the adjusted R2 and the number of observations. In the adjusted R2 column, we also include likelihood ratio tests examining if the conditional model’s
R2 is significantly larger than that of the basic model, and all results are significant at the 1% level as indicated by the three stars. All regressions include firm and year fixed effects.
t-statistics (clustered by firm) are reported in parentheses.
Table 8
Investment panel data regressions (age or competition instead of life cycle)
Tobin’s log log TobQ x TobQ x TobQ x TobQ x Log Log Adj #
Row Sample q assets age Quartile 1 Quartile 2 Quartile 3 Quartile 4 assets age R2 obs.
A. R&D/assets
(1) Age quartiles 0.007 -0.027 0.007 0.008 0.008 0.008 0.003 -0.027 0.007 .830 68,798
(15.57) (-21.53) (2.63) (11.75) (8.41) (8.22) (3.13) (-21.54) (2.76)
(2) TNIC TSIMM quartiles 0.007 -0.027 0.007 0.003 0.004 0.005 0.010 -0.027 0.006 .831 68,798
(15.57) (-21.53) (2.63) (3.21) (5.84) (6.65) (14.01) (-21.61) (2.59)
(3) TNIC HHI quartiles 0.007 -0.027 0.007 0.008 0.009 0.006 0.005 -0.027 0.006 .830 68,798
(15.57) (-21.53) (2.63) (10.20) (11.05) (9.47) (4.77) (-21.55) (2.49)
B. CAPX/assets
(4) Age quartiles 0.008 -0.011 -0.018 0.007 0.008 0.007 0.012 -0.011 -0.019 .580 68,798
(23.87) (-14.93) (-8.10) (16.05) (16.05) (11.56) (10.95) (-14.95) (-8.39)
(5) TNIC TSIMM quartiles 0.008 -0.011 -0.018 0.009 0.009 0.009 0.006 -0.011 -0.018 .580 68,798
(23.87) (-14.93) (-8.10) (11.99) (12.04) (16.53) (14.98) (-14.93) (-8.07)
(6) TNIC HHI quartiles 0.008 -0.011 -0.018 0.008 0.007 0.008 0.009 -0.011 -0.018 .580 68,798
(23.87) (-14.93) (-8.10) (16.46) (14.52) (15.60) (12.74) (-15.11) (-8.11)
C. SDC acquisitions/assets
(7) Age quartiles 0.010 -0.024 -0.016 0.012 0.009 0.008 0.012 -0.024 -0.015 .174 68,798
(13.30) (-16.08) (-4.21) (10.25) (7.37) (6.36) (6.43) (-16.07) (-3.89)
56
(8) TNIC TSIMM quartiles 0.010 -0.024 -0.016 0.007 0.008 0.011 0.012 -0.024 -0.016 .174 68,798
(13.30) (-16.08) (-4.21) (5.70) (6.66) (8.90) (9.78) (-16.07) (-4.20)
(9) TNIC HHI quartiles 0.010 -0.024 -0.016 0.012 0.012 0.009 0.008 -0.024 -0.016 .174 68,798
(13.30) (-16.08) (-4.21) (9.07) (9.90) (7.79) (7.04) (-16.13) (-4.17)
D. SDC asset sales/assets
(10) Age quartiles -0.001 0.000 0.003 0.000 -0.001 0.000 -0.001 0.000 0.004 .238 68,798
(-4.42) (1.07) (3.15) (-1.94) (-4.99) (-1.78) (-1.58) (1.11) (3.32)
(11) TNIC TSIMM Quartiles -0.001 0.000 0.003 -0.001 -0.001 -0.001 0.000 0.000 0.003 .238 68,798
(-4.42) (1.07) (3.15) (-1.94) (-3.52) (-2.75) (-2.36) (1.07) (3.18)

(12) TNIC HHI quartiles -0.001 0.000 0.003 -0.001 -0.001 -0.001 0.000 0.000 0.003 .238 68,798
(-4.42) (1.07) (3.15) (-2.32) (-3.10) (-3.79) (-1.07) (1.09) (3.17)
The table reports results from firm-year panel data OLS investment-q regressions from 1998 to 2017. The dependent variable is ex post R&D/assets (panel A), CAPX/assets (panel B),
SDC $ acquisitions/assets (panel C), and SDC $ asset sales/assets (panel D). The key RHS variables are a set of placebo life cycle variables interactions with Tobin’s q (the life cycle
variables used here are based on age or competition variables, not our text-based life cycle variables). In rows 1, 4, 7, and 10, the placebo life cycle variables are dummies indicating
age quartiles, where quartiles are formed in each year. In rows 2, 3, 8, and 11, the placebo life cycle variables are dummies indicating TNIC total similarity quartiles, where quartiles
are formed in each year. In rows 3, 6, 9, and 12, the placebo life cycle variables are dummies indicating NAICS HHI quartiles, where quartiles are formed in each year. All regressions
include controls for size, age, firm fixed effects, and year fixed effects. All RHS variables are ex ante measurable and are observable in year t − 1. All ratio variables are winsorized at the
1% and 99% levels. The last two columns indicate the adjusted R2 and the number of observations. All regressions include firm and year fixed effects. t-statistics (clustered by firm)
are reported in parentheses.
Table 9
Life cycles and broad technology traits
Distant Distant Distant Distant
Dependent peer peer peer peer log log Adj #
Row variable digital tech. change fluidity supply chain assets age R2 obs.
(1) Life1 0.358 1.493 0.000 -1.114 0.001 -0.006 .80 68,146
(2.11) (4.86) (3.81) (-1.81) (0.47) (-1.45)
(2) Life2 0.083 -0.640 0.000 4.691 -0.002 -0.064 .77 68,146
(0.40) (-1.53) (-1.06) (5.33) (-1.44) (-11.80)
(3) Life3 0.085 -0.859 -0.001 -3.522 0.006 0.050 .65 68,146
(0.43) (-2.27) (-4.49) (-4.54) (4.87) (10.41)
(4) Life4 -0.525 0.005 0.000 -0.055 -0.004 0.019 .42 68,146
(-2.88) (0.02) (3.14) (-0.09) (-4.16) (5.50)
57
The table reports results from firm-year panel data regressions from 1998 to 2017. The dependent variable is ex post R&D/assets (panels A), CAPX/assets (panel B), SDC acquisitions
(panel C), and SDC targets (panel D). All dependent variables are the focal firm’s ex post life cycle stages. All RHS variables are ex ante measurable and are observable in year t − 1.
In all models, the dependent variable is regressed on four ex ante technological characteristics and controls for Tobin’s q, size, and age. Technological characteristics are computed as an
average of a given quality over the given focal firm’s distant peers (firms that are in the focal firm’s TNIC-2 industry, but not its TNIC-2 industry). A firm’s digital focus is the number
of paragraphs that contain the term of either “digital*” or “digitiz*” divided by the total number of paragraphs in the 10-K. Technological change is similarly defined but paragraphs
must have “technol*” and “change.” Supply chain is similarly defined but focuses must have “supply chain.” Fluidity is based on the Hoberg, Phillips, and Prabhala (2014) definition
and measures the extent to which firms are changing their product features. All ratio variables are winsorized at the 1% and 99% levels. The last two columns indicate the adjusted R2
and the number of observations. All regressions include firm and year fixed effects. t-statistics (clustered by firm) are reported in parentheses.

Table 10
Instrumental variable investment regressions
TobQ x TobQ x TobQ x TobQ x log log #

Row Instruments used Life1 Life2 Life3 Life4 Assets Age obs.
A. R&D/assets
(1) Tech instruments 0.053 -0.008 -0.031 -0.039 -0.027 0.004 66,797
(5.72) (-0.92) (-2.12) (-0.56) (-23.28) (1.75)
(2) Distant life instruments 0.054 -0.001 -0.044 0.012 -0.027 0.004 66,850
(7.53) (-0.36) (-4.58) (2.07) (-23.64) (1.61)
B. CAPX/assets
(3) Tech instruments 0.006 0.066 -0.022 -0.188 -0.011 -0.015 66,797
(0.59) (5.79) (-1.52) (-2.05) (-12.77) (-5.59)
(4) Distant life instruments -0.031 0.039 0.024 0.010 -0.011 -0.016 66,850
(-6.91) (12.80) (4.02) (2.93) (-15.50) (-7.76)
(5) Tech instruments 0.008 -0.004 0.056 -0.178 -0.024 -0.012 66,797
(0.52) (-0.24) (2.56) (-1.42) (-16.53) (-2.76)
58
(6) Distant life instruments -0.022 -0.003 0.066 0.017 -0.024 -0.014 66,850
(-2.74) (-0.80) (5.42) (2.52) (-17.45) (-3.71)
D. SDC $ asset sales/assets
(7) Tech instruments 0.000 -0.004 0.001 0.005 0.000 0.003 66,797
(0.00) (-1.24) (0.26) (0.24) (1.03) (3.20)
(8) Distant life instruments 0.001 -0.001 -0.001 -0.005 0.000 0.004 66,850
(0.42) (-1.19) (-0.45) (-3.19) (1.16) (3.40)
The table reports results from firm-year instrumental variables investment-q regressions from 1998 to 2017. The dependent variable is ex post R&D/assets (panel A), CAPX/assets

(panel B), SDC $ Acquisitions/assets (panel C), and SDC $ Asset sales/assets (panel D). All dependent variables are based on the focal firm’s investment policies. The key RHS variables
being instrumented for are the life cycle interactions with Tobin’s q. The odd-numbered rows use the technology instruments from Table 9 interacted with Tobin’s q as the instruments.
The even-numbered rows use the life cycle stages of distant peers interacted with Tobin’s q as the instruments. Distant peers are firms that are in the focal firm’s TNIC-2 industry but
are not in the firm’s TNIC-3 industry. All RHS variables and instruments are ex ante measurable and are observable in year t − 1. All ratio variables are winsorized at the 1% and 99%
levels. The last two columns indicate the adjusted R2 and the number of observations. All regressions include firm and year fixed effects. t-statistics (clustered by firm) are reported in
parentheses.
Table 11
Instrumental variable investment regressions (instrument for competition quartiles instead of life cycles)
TobQ x TobQ x TobQ x TobQ x
Compet Compet Compet Compet log log #
Row Instruments used quartile 1 quartile 2 quartile 3 quartile 4 assets age obs.
A. R&D/assets
(1) Tech instruments -0.061 0.043 -0.011 0.017 -0.027 0.003 66,797
(-0.72) (0.35) (-0.16) (1.27) (-18.13) (0.44)
(2) Distant life instruments -0.135 0.243 -0.148 0.039 -0.027 -0.008 66,850
(-0.79) (0.78) (-0.77) (1.14) (-8.01) (-0.41)
B. CAPX/assets
(3) Tech instruments 0.017 0.090 -0.109 0.036 -0.012 -0.025 66,797
(0.11) (0.38) (-0.82) (1.19) (-5.20) (-1.95)
(4) Distant life instruments 0.068 -0.060 0.096 -0.025 -0.011 -0.011 66,850
(0.92) (-0.44) (1.13) (-1.67) (-7.05) (-1.33)
(5) Tech instruments 0.137 -0.193 0.115 -0.004 -0.024 -0.005 66,797
(0.62) (-0.58) (0.58) (-0.09) (-6.74) (-0.26)
59
(6) Distant life instruments 0.134 -0.252 0.189 -0.020 -0.024 0.000 66,850
(0.72) (-0.74) (0.89) (-0.52) (-6.27) (0.00)
D. SDC $ assets sales/assets
(7) Tech instruments 0.000 -0.007 0.006 -0.002 0.000 0.004 66,797
(0.01) (-0.38) (0.54) (-0.72) (1.08) (2.75)
(8) Distant life instruments -0.002 0.000 -0.001 0.000 0.000 0.003 66,850
(-0.37) (-0.03) (-0.07) (-0.23) (1.19) (2.84)
The table reports results from firm-year instrumental variables investment-q regressions from 1998 to 2017. The dependent variable is ex post R&D/assets (panel A), CAPX/assets

(panel B), SDC $ acquisitions/assets (panel C), and SDC $ asset sales/assets (panel D). All dependent variables are based on the focal firm’s investment policies. The key RHS variables
being instrumented for are TNIC total similarity quartile dummy interactions with Tobin’s q (competition placebo life cycle variables). The odd-numbered rows use the technology
instruments from Table 9 interacted with Tobin’s q as the instruments. The even-numbered rows use the life cycle stages of distant peers interacted with Tobin’s q as the instruments.
Distant peers are firms that are in the focal firm’s TNIC-2 industry but are not in the firm’s TNIC-3 industry. All RHS variables and instruments are ex ante measurable and are
observable in year t − 1. All ratio variables are winsorized at the 1% and 99% levels. The last two columns indicate the adjusted R2 and the number of observations. All regressions
include firm and year fixed effects. t-statistics (clustered by firm) are reported in parentheses.
Table 12
Investment policy economic magnitudes: Life cycle conditional model versus basic q -model
Life vs.
Life Basic Basic Basic Basic basic
Dependent model model model model model model
Row variable tercile tercile 1 tercile 2 tercile 3 T3−T1 ratio
1 R&D/assets Tercile 1 0.004 0.007 0.015 0.011

4 R&D/assets High − Low 0.073 0.086 0.143 2.88
5 CAPX/assets Tercile 1 0.034 0.038 0.043 0.010

8 CAPX/assets High − Low 0.048 0.054 0.046 6.69
9 SDC $ acq/assets Tercile 1 0.022 0.038 0.034 0.012

12 SDC $ acq/assets High − Low 0.016 0.012 0.026 1.01
13 SDC $ divest/assets Tercile 1 0.007 0.009 0.005 -0.002

14 SDC $ divest/assets Tercile 2 0.009 0.011 0.011 0.002
15 SDC $ divest/assets Tercile 3 0.014 0.015 0.014 -0.000
16 SDC $ divest/assets High − Low 0.007 0.005 0.009 ∞
The table reports economic magnitudes of the relationship between our conditional life cycle q-model and investment policies,
and additionally compares the explanatory power to the basic q-model. In particular, we first run the following models separately
in each year:
Investment Variablei,t+1 = a + b1 Lif e1i,t Qi,t + b2 Lif e2i,t Qi,t + b3 Lif e3i,t Qi,t + b4 Lif e4i,t Qi,t + i,t ,
Investment Variablei,t+1 = a + b1 Qi,t + i,t .

All variables on the RHS are lagged and thus ex ante measurable, as is the case in our baseline regressions. The first model is
our conditional life cycle q-model. The second model is the basic q-model. For each model, we compute the predicted value, and
we then form two-way independent sorts of all observations in to terciles based on each of the two predicted values. The table
reports the average value of the investment variable for each of the nine (3x3) bins. We also report the average high minus low
difference for the two directions of the sorts, indicating the total ability of each model to explain the given investment variable
used in the regression. Finally, in the last column we report the ratio of the average high minus low portfolio differences for
the life cycle model divided by the same average high minus low difference for the competition portfolio. A value exceeding
one, intuitively, indicates that the total ability of the life cycle model to explain the given investment policy exceeds that of
the basic model. A value of one indicates that both are equally important. If the value becomes negative, indicating one sort
variable fully subsumes the other, we report a value of ∞.
60
Table 13
Investment policy economic magnitudes: Life cycles versus competition
Life Compet Compet Compet Compet Life v.
Dependent model model model model model compet
Row variable tercile tercile 1 tercile 2 tercile 3 T3−T1 ratio

4 R&D/assets High − Low 0.036 0.062 0.172 1.64

8 CAPX/assets High − Low 0.029 0.057 0.061 2.28

12 SDC $ acq/assets High − Low 0.019 0.019 0.017 0.99

16 SDC $ divest/assets High − Low 0.005 0.004 0.004 1.20
The table reports economic magnitudes of the relationship between our conditional life cycle q-model and investment policies,
and additionally compares the explanatory power to a conditional competition q-model. In particular, we first run the following
models separately in each year:
Investment Variablei,t+1 = a + b1 Lif e1i,t Qi,t + b2 Lif e2i,t Qi,t + b3 Lif e3i,t Qi,t + b4 Lif e4i,t Qi,t + i,t ,
Investment Variablei,t = a + b1 Comp1i,t Qi,t + b2 Comp2i,t Qi,t + b3 Comp3i,t Qi,t + b4 Comp4i,t Qi,t+1 + i,t .
All variables on the RHS are lagged and thus ex ante measurable, as is the case in our baseline regressions. The first model is
our conditional life cycle q-model. The second model is based on four competition dummies Comp1 to Comp4, which are each
equal to one if the firm is in the first to the fourth quartile, respectively, based on annual sorts of each firm’s TNIC-3 total
similarity. We use four groups to allow consistency across the two models. For each model, we compute the predicted value, and
we then form two-way independent sorts of all observations in to terciles based on each of the two predicted values. The table
reports the average value of the investment variable for each of the nine (3x3) bins. We also report the average high minus low
difference for the two directions of the sorts, indicating the total ability of each model to explain the given investment variable
used in the regression. Finally, in the last column we report the ratio of the average high minus low portfolio differences for
the life cycle model divided by the same average high minus low difference for the competition portfolio. A value exceeding
one, intuitively, indicates that the total ability of the life cycle model to explain the given investment policy exceeds that of
the competition model. A value of one indicates that both are equally important. As competition and life cycle stages are only
modestly correlated, the explanatory power of the two models is mostly distinct and additive.
61
Table 14
IO-q regressions
Adj
Tobin’s log log TobQ x TobQ x TobQ x TobQ x log log #
Row Sample q assets age Life1 Life2 Life3 Life4 assets age R2 obs.
Dependent variable: Operating income/assets
(1) Full sample 0.007 -0.006 0.039 -0.020 0.036 0.012 0.007 -0.006 0.040 .600 68,797
(8.86) (-2.82) (7.22) (-5.74) (11.58) (3.26) (0.69) (-2.75) (7.48)
(2) Tercile 1 0.015 0.005 0.017 -0.016 0.033 0.025 0.002 0.005 0.017 .664 22,849
(5.42) (1.34) (1.87) (-1.44) (5.20) (3.54) (0.15) (1.40) (1.90)
(3) Tercile 2 0.013 -0.007 0.038 -0.020 0.034 0.020 0.041 -0.007 0.037 .668 22,836
(7.82) (-2.18) (4.04) (-2.18) (6.34) (2.68) (2.36) (-2.19) (4.06)
62
(4) Tercile 3 0.004 -0.016 0.045 -0.018 0.033 0.009 0.007 -0.016 0.048 .587 22,813
(3.97) (-3.99) (4.17) (-3.83) (6.05) (1.68) (0.31) (-3.97) (4.55)
The table reports results from firm-year panel data OLS investment-q regressions from 1998 to 2017. The dependent variable is ex post operating income scaled by assets:
(OIBDP+XRD)/assets. We adjust operating income for R&D as we treat R&D as an investment, and not an operating expense (and OIBDP is defined to be net of expenses
and R&D). Our results are also robust following the formulation in Lee, Shin, and Stulz (2021), who do not adjust for R&D but instead truncate operating income at zero. The key RHS
variables are the lagged life cycle variables interacted with Tobin’s q. All regressions include controls for size, age, firm fixed effects, and year fixed effects. In each panel, we consider
four samples: the full sample and tercile-based subsamples based on annual sorts of TNIC total similarity (see Hoberg and Phillips 2016). All RHS variables are ex ante measurable and
are observable in year t − 1. All ratio variables are winsorized at the 1% and 99% levels. The last two columns indicate the adjusted R2 and the number of observations. All regressions

Table 15
Healthcare versus nonhealthcare industry q regressions
Dependent Tobin’s log log TobQ x TobQ x TobQ x TobQ x log log Adj #
Row aariable q assets age Life1 Life2 Life3 Life4 assets age R2 obs.
A. Healthcare industry (dependent variables as specified)

(1) R&D/assets 0.014 -0.059 -0.022 0.035 0.009 -0.022 0.015 -0.058 -0.022 .797 9,262
(11.57) (-14.72) (-1.64) (7.17) (1.09) (-2.92) (0.86) (-14.78) (-1.64)
(2) OI/assets 0.007 0.017 0.060 -0.006 0.008 0.042 -0.033 0.016 0.060 .580 9,262
(3.65) (2.46) (2.81) (-0.83) (0.70) (3.50) (-1.19) (2.41) (2.84)
(3) Prod mkt fluidity 0.024 0.345 -1.324 0.213 -0.053 -0.250 -0.115 0.349 -1.320 .821 9,228
(1.30) (4.46) (-4.32) (3.07) (-0.53) (-2.20) (-0.42) (4.56) (-4.34)
(4) Life3 0.001 0.010 0.043 -0.004 -0.013 0.029 -0.002 0.009 0.041 .714 9,262
(0.81) (3.38) (3.58) (-2.00) (-3.01) (4.28) (-0.24) (3.37) (3.38)
B. Nonhealthcare industry (dependent variables as specified)
63
(5) R&D/assets 0.005 -0.019 0.007 0.016 -0.002 0.001 0.011 -0.019 0.007 .823 59,535
(11.58) (-16.62) (3.49) (7.02) (-2.34) (0.39) (2.16) (-16.75) (3.49)
(6) OI/assets 0.007 -0.011 0.036 -0.026 0.040 0.011 0.017 -0.011 0.036 .605 59,535
(8.33) (-5.38) (6.76) (-5.80) (12.43) (2.73) (1.69) (-5.09) (6.74)
(7) Prod mkt fluidity 0.079 0.378 -1.289 0.139 -0.058 0.146 0.067 0.376 -1.287 .756 58,494
(8.95) (11.58) (-13.43) (3.32) (-1.58) (2.92) (0.69) (11.53) (-13.41)
(8) Life3 0.000 0.005 0.053 0.002 0.015 -0.023 0.052 0.004 0.052 .648 59,535
(0.58) (3.22) (9.88) (0.57) (5.20) (-3.77) (5.98) (3.18) (9.88)

The table reports results from firm-year panel data OLS investment-q regressions from 1998 to 2017. The table shows results for the health industry (panel A) as defined by the
Fama-French five industry classification and all nonhealthcare firms (panel B) (we thank Ken French for providing this classification information on his website). The dependent variable
is specified in the first column. The key RHS variables are the lagged life cycle variables interacted with Tobin’s q. All regressions include controls for size, age, firm fixed effects, and
year fixed effects. All RHS variables are ex ante measurable and are observable in year t − 1. All ratio variables are winsorized at the 1% and 99% levels. The last two columns indicate
the adjusted R2 and the number of observations. All regressions include firm and year fixed effects. t-statistics (clustered by firm) are reported in parentheses.
Table 16
Financing panel data regressions
Row Sample q assets age Life1 Life2 Life3 Life4 assets age R2 obs.
A. equity issuance/assets
(1) Full sample 0.037 -0.081 -0.032 0.088 0.026 -0.021 0.075 -0.081 -0.036 .397 68,798
(17.21) (-27.23) (-4.88) (8.65) (4.00) (-1.89) (3.26) (-27.30) (-5.53)
(2) Tercile 1 0.031 -0.050 -0.024 0.053 0.020 0.024 0.041 -0.050 -0.024 .352 22,848
(5.91) (-9.76) (-2.73) (2.13) (1.99) (1.71) (1.53) (-9.73) (-2.73)
(3) Tercile 2 0.031 -0.066 -0.029 0.057 0.025 0.013 0.015 -0.066 -0.029 .393 22,837
(7.95) (-12.40) (-2.65) (2.91) (2.92) (0.80) (0.58) (-12.33) (-2.68)
(4) Tercile 3 0.037 -0.131 -0.044 0.099 0.031 -0.055 0.109 -0.129 -0.053 .413 22,814
(11.67) (-21.47) (-2.75) (6.74) (2.50) (-2.81) (2.54) (-21.05) (-3.38)
B. SDC SEOs/assets
(5) Full sample 0.020 -0.031 -0.040 0.042 0.009 0.007 0.003 -0.031 -0.041 .226 68,798
(10.10) (-15.50) (-8.06) (4.45) (1.64) (0.67) (0.26) (-15.50) (-8.31)
(6) Tercile 1 0.007 -0.012 -0.027 0.001 0.010 0.008 0.001 -0.012 -0.027 .224 22,848
(3.91) (-4.88) (-4.63) (0.09) (2.40) (1.86) (0.11) (-4.86) (-4.64)
(7) Tercile 2 0.015 -0.023 -0.041 0.014 0.018 0.020 -0.031 -0.023 -0.040 .196 22,837
64
(4.44) (-6.75) (-4.48) (1.04) (2.83) (1.27) (-1.86) (-6.79) (-4.43)
(8) Tercile 3 0.024 -0.057 -0.059 0.041 0.015 -0.001 0.076 -0.057 -0.062 .248 22,814
(7.52) (-13.19) (-4.90) (2.90) (1.30) (-0.03) (1.77) (-13.00) (-5.24)
C. Debt issuance/assets
(9) Full sample 0.009 -0.049 0.007 -0.006 0.027 0.003 0.048 -0.049 0.006 .358 68,798
(8.04) (-13.51) (0.69) (-1.19) (5.16) (0.59) (3.43) (-13.51) (0.66)
(10) Tercile 1 0.015 -0.048 0.010 0.005 0.020 0.008 0.055 -0.048 0.009 .404 22,848
(5.19) (-5.68) (0.47) (0.35) (2.22) (0.73) (2.29) (-5.76) (0.46)
(11) Tercile 2 0.006 -0.052 0.028 -0.014 0.018 0.005 0.073 -0.052 0.027 .400 22,837
(2.93) (-7.33) (1.40) (-1.19) (2.07) (0.46) (2.73) (-7.35) (1.36)

(12) Tercile 3 0.008 -0.056 0.003 -0.015 0.042 0.006 0.055 -0.055 0.006 .310 22,814
(5.12) (-9.48) (0.17) (-2.08) (3.90) (0.60) (1.89) (-9.40) (0.34)
The table reports results from firm-year panel data OLS financing-policy-q regressions from 1998 to 2017. The dependent variable is ex post Compustat equity issuance/assets (panel A),
SDC SEOs/assets (panel B), Compustat debt issuance/assets (panel C), Compustat equity repurchases/assets (panel D), and SDC dividends/assets (panel E). The key RHS variable in
these “OLS” panel data regressions are the lagged life cycle variables interacted with Tobin’s q. All regressions include controls for size, age, firm fixed effects, and year fixed effects. All
RHS variables are ex ante measurable and are observable in year t − 1. In each panel, we consider four subsamples: the full sample, and tercile-based subsamples based on annual sorts
of TNIC total similarity (see Hoberg and Phillips 2016). All ratio variables are winsorized at the 1% and 99% levels. The last two columns indicate the adjusted R2 and the number of
observations. All regressions include firm and year fixed effects. t-statistics (clustered by firm) are reported in parentheses.
Table 16
Financing panel data regressions (continued)
Row Sample q assets age Life1 Life2 Life3 Life4 assets age R2 obs
D. Equity repurchases/assets
(13) Full sample 0.002 0.004 0.013 -0.005 0.009 0.004 0.007 0.004 0.013 .328 62,561
(8.68) (6.30) (7.99) (-3.90) (8.11) (2.40) (2.50) (6.42) (8.12)
(14) Tercile 1 0.005 0.004 0.008 0.000 0.009 0.006 -0.004 0.004 0.008 .313 21,561
(6.28) (3.78) (2.53) (0.11) (3.90) (1.96) (-1.41) (3.90) (2.54)
(15) Tercile 2 0.004 0.006 0.011 -0.004 0.008 0.007 0.018 0.006 0.011 .391 21,005
(7.14) (5.01) (3.80) (-1.41) (3.94) (1.94) (3.45) (5.01) (3.79)
65
(16) Tercile 3 0.001 0.003 0.016 -0.002 0.004 0.001 0.009 0.003 0.016 .321 19,718
(3.09) (2.94) (5.47) (-1.18) (2.64) (0.49) (1.59) (3.03) (5.57)
E. Dividends/assets
(17) Full sample 0.001 -0.001 0.003 -0.003 0.004 0.001 0.003 -0.001 0.003 .550 68,671
(6.28) (-3.10) (2.98) (-5.39) (6.79) (0.96) (2.41) (-3.03) (3.08)
(18) Tercile 1 0.002 0.001 0.004 -0.001 0.004 0.002 0.004 0.000 0.004 .565 22,821
(5.30) (0.95) (1.94) (-0.73) (3.70) (0.98) (1.78) (0.89) (1.93)
(19) Tercile 2 0.001 0.000 0.005 -0.004 0.005 0.002 0.002 0.000 0.005 .520 22,808
(4.47) (-0.57) (3.55) (-2.67) (3.38) (1.71) (1.36) (-0.55) (3.62)
(20) Tercile 3 0.000 -0.002 -0.003 -0.002 0.003 0.000 0.001 -0.002 -0.002 .612 22,746

(0.63) (-3.41) (-1.55) (-3.82) (3.90) (-0.30) (0.81) (-3.35) (-1.38)
Figure 1
Life cycle stages over time
Life1 vs Time (Large vs Small Firms) Life2 vs Time (Large vs Small Firms)
0.33 0.49
0.31 0.47
0.29
0.45
0.27
0.25 0.43
0.23 0.41
0.21 0.39
0.19
0.37
0.17
0.15 0.35
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
Life1 (Large Firms) Life1 (Small Firms) Life2 (Large Firms) Life2 (Small Firms)
Life3 vs Time (Large vs Small Firms) Life4 vs Time (Large vs Small Firms)
66
0.37 0.11
0.35 0.1
0.09
0.33
0.08
0.31
0.07
0.29
0.06
0.27 0.05

0.25 0.04
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
Life3 (Large Firms) Life3 (Small Firms) Life4 (Large Firms) Life4 (Small Firms)
The figure plots mean values of Life1 to Life4 for firms in the bottom and top quartiles of firms by asset size, computed annually.
Figure 2
Firm dynamism over time
Dynamism vs Time (Large vs Small Firms)
0.69
0.68
0.67
0.66
0.65
0.64
67
0.63
0.62
0.61
0.6
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Dynamism (life1+life2) (Big Firms) Dynamism (life1+life2) (Small Firms)

The figure plots the average Firm Dynamism index, which is defined as (Lif e1 + Lif e2), for firms in the bottom and top quartiles of firms by asset size,
computed annually.
Figure 3
Life cycles versus firm age
Life1 vs Age Life2 vs Age
0.55
0.29
0.5
0.27
0.25 0.45
Life2
0.23
Life1
Raw Life1 0.4 Raw Life2
0.21 Adj Life1 Adj Life2
0.35
0.19
0.17 0.3
3%
3%
10%
20%
30%
40%
45%
53%
58%
63%
68%
73%
78%
83%
88%
93%
98%
10%
20%
30%
40%
45%
53%
58%
63%
68%
73%
78%
83%
88%
93%
98%
Age Quantile Age Quantile
Life3 vs Age Life4 vs Age
0.11
0.38
0.1
68
0.36
0.34 0.09
0.32 0.08
0.3
Life4
0.07
Life3
0.28 Raw Life3 Raw Life4
0.06
0.26
Adj Life3 Adj Life4
0.24 0.05
0.22 0.04
3%
0.2
10%
20%
30%
40%
45%
53%
58%
63%
68%
73%
78%
83%
88%
93%
98%
3%
10%
20%
30%
40%
45%
53%
58%
63%
68%
73%
78%
83%
88%
93%
98%
Age Quantile

Age Quantile
The figure plots two series for each life-cycle-stage variable. The solid line in each figure plots the mean raw values against percentiles based on age for all
sample firms. The dotted line in each figure represents the mean values after each life variable is regressed on both firm and year fixed effects and then
plotted against percentiles based on age. The former thus plots both within and across firm variation, and the latter focuses on within-firm variation only.
Figure 4
Regression R2 over time (basic vs. conditional model)
RSQ vs Time: Conditional vs Basic CAPX Q‐Models
0.25
0.2
0.15
0.1
0.05
0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
CAPX (conditional model) CAPX (basic model)
RSQ vs Time: Conditional vs Basic R&D Q‐Models
0.6
0.5
0.4
0.3
0.2
0.1
0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
R&D (conditional model) R&D (basic model)
The figure plots the R2 of the annual cross-sectional regressions in Tables IA.15 and IA.16 in the Internet
Appendix. The basic model does not adjust for differences in the investment-q relationship for different
values of the life variables. The conditional model includes the level of the Life variables interacted with
Tobin’s q.
69
Figure 5
Conditional model regression R2 over time (competition and dynamism subsamples)
RSQ vs Time: Conditional CAPX Q‐Model: High vs Low Dynamism and Competition
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Dynamic & Competitive Dynamic & Concentrated Static & Competitive Static & Concentrated
RSQ vs Time: Conditional R&D Q‐Model: High vs Low Dynamism and Competition
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Dynamic & Competitive Dynamic & Concentrated Static & Competitive Static & Concentrated
The figure plots the R2 of the annual cross-sectional regressions using the conditional model as shown in
Tables IA.15 and IA.16 in the Internet Appendix. The conditional model includes the Life variables
interacted with Tobin’s q. The upper figure displays results for the CAPX-q-model and the lower figure
displays results for the R&D-q-model. Dynamism is defined as (lif e1 + lif e2) and competition is defined
as TNIC total similarity from Hoberg and Phillips (2016). In each year, we perform independent sorts of
the full sample into above and below median values of dynamism and TNIC total similarity. Note that the
four subsamples are quite evenly balanced in terms of number of observations, which arises because
dynamism and HHIs are less than 5% correlated.
70
Internet Appendix
Not Intended for Publication

Additional Model Derivations: ζ, λ
We first derive an explicit expression for ζ.
At t3 , the value of firm’s installed capacity is the sum of the value of output produced
and sold in this stage φ(k1∗ + k2∗ + k3∗ ), and the continuation value of this capacity in Life4 is
v3∗ . Here φ∗ is the total end of period value per unit of the installed capacity (k1∗ + k2∗ + k3∗ ),
v3∗
φ∗ = φ + (16)
k1∗ + k2∗ + k3∗
The first term on the right hand side is the per-unit proceeds of the sale of output at
t3 and the second term is the per-unit continuation value. Substituting for φ∗ = ζφ, where
ζ > 1, and solving for ζ, gives:
p
1 ρ(2δµ + ρ)
ζ= +
2 2ρ
where ρ = (δ 2 µπ + δµ + 1)ω.
While we do not need to obtain the parameter λ, used in equation (10), it can be
computed using the fact that life cycle exposures sum to unity, so that the firm’s overall Q
can directly decomposed into
k1∗ k ∗ + k2∗ k ∗ + k2∗ + k3∗ k∗

Q = θ1 q1 + θ2 1 q2 + θ 3 1 q3 + θ4 4 q4 (17)
K K K K
where K = θ1 k1∗ + θ2 (k1∗ + k2∗ ) + θ3 (k1∗ + k2∗ + k3∗ ) + θ4 k4 . Substituting for the the ki∗ s and qi s,
yields:
ζ(µ(ωδζ(δπ(θ1 + θ2 + θ3 ) + θ2 + θ3 ) + θ4 ) + ωθ3 ζ)
λ= >0 (18)
µ (ωδζ 2 (δ 2 π(θ1 + θ2 ) + δ(πθ3 + θ2 ) + θ3 ) + θ4 ) + ωθ3 ζ 2
The quantities ζ > 0 and λ > 0 are used in the derivations in the text. These quantities
are not functions of γ.
72
Table IA.1
Life Cycle Vocabularies
Row Life1 Vocabulary Score Life2 Vocabulary Score Life3 Vocabulary Score Life4 Vocabulary Score
1 introductions 75.83 spoilage 76.16 persuasive 67.44 discontinue 65.81
2 introduce 74.46 wasted 76.14 dsl 66.33 discontinuation 65.11
3 introducing 74.02 overheads 76.01 resellers 66.27 obsolete 64.92
4 launches 74.02 fifo 75.21 dial 65.64 discontinuance 64.69
5 innovation 73.68 idle 74.83 undelivered 65.6 obsolescence 64.32
6 innovations 73.28 abnormal 74.78 fob 65.42 discontinuing 62.09
7 entrants 72.74 arb 74.1 tailored 65.4 noncompetitive 57.7
8 launch 72.17 lifo 73.67 oems 65.31 render 56.43
9 introduction 72.04 rehandling 73.52 bundled 65.25 discontinued 56.07
10 enhancements 71.37 overhead 73.05 passes 65.24 slow 55.01
11 adapt 71.33 betterments 73.04 wholesalers 65.23 moving 54.53
12 commercialized 71.3 headcount 72.69 personalized 65.17 discontinues 54.16
13 announcements 71.29 wages 72.56 shipped 65.17 divest 50.68
14 versions 71.23 clarify 72.26 shipment 65.09 decided 50.4
15 expanding 71.13 criterion 72.18 chains 64.95 divestiture 50.14
16 product’s 70.92 freight 71.75 customized 64.93 uncompetitive 50
17 collaborations 70.58 reconcile 71.41 merchandisers 64.77 divested 49.85
18 evolving 70.49 deduct 71.37 integrators 64.72 fifo 49.72
19 commercialization 70.43 wage 71.35 distance 64.71 audiovisual 49.2
20 commercialize 70.31 supersedes 71.22 vsoe 64.41 rapid 48.96
21 noncompetitive 70.26 obsolescence 71.16 distributes 64.41 eliminations 48.93
22 collaborators 70.2 salaries 71.16 accessories 64.39 downs 48.76
23 commercializing 70.16 depreciated 70.87 array 64.39 realizable 48.68
24 productive 70.06 repairs 70.7 diverse 64.31 technological 48.6
25 intensely 69.89 fringe 70.58 manufactures 64.28 unmarketable 48.4
26 preclinical 69.87 revises 70.22 marketers 64.27 finished 48.38
27 clearances 69.78 relocation 70.09 hosting 64.23 exit 48.2
28 broaden 69.76 intrinsic 70.01 voice 64.15 continuing 48.18
29 innovative 69.67 postage 69.97 sized 64.15 disposal 48.03
30 intensify 69.66 expensing 69.84 reseller 64.12 reconcile 48
31 collaborative 69.57 supercedes 69.71 packaged 64.1 scrapped 47.77
32 biopharmaceutical 69.54 shipping 69.5 oem 64.09 inventory 47.69
33 pace 69.41 disposals 69.19 teller 64.09 hand 47.64
34 nda 69.39 shortages 69.02 boxes 63.94 ercot 47.62
35 alliances 69.34 occupancy 69 retailers 63.85 inventories 47.43
36 lengthy 69.27 abandonment 68.99 custom 63.77 reclassified 47.38
37 prototype 69.21 guarantorsubsidiaries 68.96 branded 63.66 write 47.38
38 feasibility 69.14 costing 68.94 customer’s 63.65 extinguishment 47.25
39 complementary 69.14 inefficiencies 68.94 sells 63.55 overhead 47.24
40 technological 69.13 unmarketable 68.94 breadth 63.54 shrinkage 47.11
41 biotechnology 69.12 obsolete 68.93 assortment 63.52 scrap 47.03
42 unmarketable 69.1 mcfe 68.93 competes 63.43 erosion 46.87
43 penetrate 69.09 warehousing 68.89 messaging 63.42 ceased 46.73
44 technologically 69.03 handling 68.87 groups 63.42 salability 46.55
45 labeling 68.99 depletion 68.82 mail 63.38 parts 46.46
46 render 68.98 overruns 68.72 catalogs 63.32 minority 46.3
47 redesign 68.97 eliminations 68.72 connectivity 63.31 segregated 46.09
48 characterized 68.96 rationalization 68.69 assemblies 63.19 quantities 46.06
49 generations 68.94 interpretations 68.65 broadband 63.18 forecasted 45.76
50 penetration 68.9 exit 68.64 mass 63.15 unsaleable 45.55
51 launching 68.88 moving 68.63 telephony 63.13 down 45.48
52 recalls 68.87 closure 68.63 phone 63.03 disposition 45.32
53 devote 68.87 workforce 68.57 delivers 63.03 rohs 45.28
54 histories 68.85 prepaids 68.52 ship 63.02 forecasts 45.14
55 milestone 68.84 noncash 68.48 fragmented 63.02 warranty 45.12
56 candidate 68.83 incurs 68.41 grocery 63.01 exited 45.11
57 recall 68.74 labor 68.39 networking 63.01 disposed 44.98
58 formulation 68.71 valuing 68.32 voip 62.94 manufactured 44.84
59 investigational 68.7 payroll 68.32 servers 62.94 evolving 44.82
60 frequent 68.64 closures 68.28 wireline 62.9 unconsolidated 44.82
The table displays vocabularies that cluster with each life cycle stage. We use metaHeuristica’s significant terms analysis tool kit
to develop vocabularies that strongly associate with each life cycle stage in the corpus overall. Words are selected by comparing
the word distribution of paragraphs that hit on each anchor-phrase query with the distribution of words in 10-Ks overall. This
comparison is made using tfidf weighting to ensure common words are not scored excessively. We report the 60 terms that are
most represented in the distribution of words for each life cycle query based on these tfidf weights (relative to the 10-K overall)
in the table below.
73
Table IA.2
Amazon and Telsa Life Cycle Progressions
Row Year Life1 Life2 Life3 Life4
Panel A: Amazon Life Cycle Exposures
1998 0.44 0.23 0.32 0.01

1999 0.40 0.26 0.31 0.02
2000 0.31 0.30 0.37 0.01
2001 0.38 0.29 0.30 0.03
2002 0.31 0.34 0.33 0.03
2003 0.23 0.37 0.38 0.02
2004 0.26 0.34 0.38 0.02
2005 0.25 0.35 0.38 0.02
2006 0.25 0.35 0.39 0.01
2007 0.25 0.35 0.40 0.00
2008 0.24 0.36 0.39 0.00
2009 0.24 0.36 0.40 0.00
2010 0.25 0.40 0.35 0.00
2011 0.24 0.39 0.36 0.00
2012 0.23 0.39 0.39 0.00
2013 0.25 0.41 0.34 0.01
2014 0.22 0.43 0.34 0.01
2015 0.22 0.45 0.34 0.00
2016 0.21 0.47 0.32 0.00
2017 0.19 0.48 0.32 0.00
Panel B: Tesla Life Cycle Exposures
2011 0.38 0.53 0.08 0.01

2012 0.41 0.47 0.10 0.02
2013 0.40 0.49 0.09 0.02
2014 0.35 0.50 0.12 0.02
2015 0.33 0.50 0.14 0.03
2016 0.29 0.49 0.19 0.03
2017 0.29 0.48 0.20 0.02
The table displays the life cycle exposures of Amazon (Panel A) and Tesla (Panel B) over the years in our sample period.
74
Table IA.3
Life2 Regressions (3 Bins by Competition)
Basic Model Conditional Model
Tobins log log TobQ x TobQ x TobQ x TobQ x log log Adj #
Row Sample Q assets age Life1 Life2 Life3 Life4 assets age R2 obs
Dependent Variable: ex post Life2

(1) High Competition -0.003 -0.006 -0.029 -0.009 0.004 -0.004 0.036 -0.005 -0.029 0.904 15,408
tercile (-6.30) (-2.87) (-3.18) (-4.52) (1.18) (-1.47) (3.76) (-2.72) (-3.17)
(2) Mid Competition -0.001 0.000 -0.060 -0.003 -0.016 0.007 0.074 -0.001 -0.061 0.790 25,382
tercile (-1.46) (-0.15) (-6.86) (-0.48) (-2.90) (0.79) (5.82) (-0.23) (-7.04)
75
(3) Low Competition 0.000 0.003 -0.080 0.028 -0.060 0.043 0.043 0.004 -0.079 0.682 27,709
tercile (-0.14) (1.04) (-7.94) (3.25) (-7.11) (3.91) (3.17) (1.25) (-8.17)
The table runs regressions like those in Table 6 except that we change the dependent variable to ex-post Life2 rather than an investment policy. The goal is to test the unique prediction
regarding the negative Q-sensitivity of life1 firms regarding CAPX investment for competitive industry firms in Panel B of Table 6. Our hypothesis is that the negative sensitivity
indicates that low Q Life1 firms increase their CAPX more than high Q life1 firms, and this is because these firms are experiencing losses in their product development growth options,
and hence are transitioning to Life2. We thus predict a negative and significant TobQ * Life1 coefficient, but only in competitive industries as is the case with the negative Q-sensitivity
noted above. We include all controls and fixed effects as in Table 6 and compute standard errors using the same methods. t-statistics (clustered by firm) are reported in parentheses.

Table IA.4
Investment Panel Data Regressions (Non-patenting Firms vs Patenting Firms)
Panel A: R&D/ assets

(1) Full Sample 0.007 -0.027 0.007 0.029 -0.002 -0.011 0.014 -0.027 0.005 0.833 68,798
(15.58) (-21.53) (2.63) (12.41) (-1.45) (-4.38) (2.38) (-21.90) (2.13)
(2) No Patent Firms 0.006 -0.015 0.007 0.025 -0.004 -0.006 0.007 -0.015 0.006 0.856 47,197
(10.62) (-14.17) (3.59) (9.09) (-4.17) (-2.11) (1.83) (-14.40) (3.28)
(3) Patenting Firms 0.008 -0.050 -0.003 0.029 0.004 -0.020 0.018 -0.049 -0.003 0.821 21,601
(9.73) (-18.93) (-0.39) (7.76) (1.03) (-5.05) (1.52) (-18.99) (-0.52)
Panel B: CAPX/ assets
(4) Full Sample 0.008 -0.011 -0.018 -0.007 0.024 0.008 0.015 -0.011 -0.018 0.585 68,798
(23.87) (-14.93) (-8.10) (-4.14) (14.65) (4.12) (4.00) (-14.83) (-8.04)
(5) No Patent Firms 0.009 -0.014 -0.021 -0.008 0.026 0.009 0.016 -0.014 -0.020 0.595 47,197
(19.15) (-13.76) (-6.85) (-3.99) (11.90) (3.77) (3.63) (-13.49) (-6.81)
(6) Patenting Firms 0.006 -0.005 -0.012 -0.001 0.018 0.004 0.008 -0.005 -0.012 0.490 21,601
(13.67) (-4.38) (-3.75) (-0.31) (7.53) (1.54) (1.40) (-4.39) (-3.71)
Panel C: SDC $ Acquisitions/ assets
76
(7) Full Sample 0.010 -0.024 -0.016 0.004 0.002 0.026 0.014 -0.024 -0.016 0.174 68,798
(13.30) (-16.08) (-4.21) (1.39) (0.81) (6.82) (2.03) (-16.26) (-4.07)
(8) No Patent Firms 0.011 -0.027 -0.018 0.011 0.001 0.021 0.005 -0.027 -0.017 0.210 47,197
(9.84) (-14.07) (-3.59) (2.37) (0.46) (4.35) (0.63) (-14.13) (-3.52)
(9) Patenting Firms 0.008 -0.018 -0.007 -0.002 0.002 0.025 0.035 -0.019 -0.008 0.137 21,601
(6.98) (-7.39) (-0.90) (-0.34) (0.30) (3.90) (2.45) (-7.58) (-0.98)
Panel D: SDC $ Asset Sales/ assets
(10) Full Sample -0.001 0.000 0.003 0.000 -0.001 0.000 -0.005 0.000 0.004 0.238 68,798
(-4.42) (1.07) (3.15) (0.26) (-2.09) (-0.48) (-2.99) (1.07) (3.20)
(11) No Patent Firms -0.001 0.001 0.003 -0.001 -0.001 0.000 -0.006 0.001 0.003 0.256 47,197

(-4.28) (1.80) (1.86) (-0.55) (-2.28) (0.30) (-2.72) (1.81) (1.90)
(12) Patenting Firms 0.000 -0.001 0.005 0.001 -0.001 -0.001 -0.004 -0.001 0.005 0.242 21,601
(-1.74) (-1.70) (2.57) (0.55) (-0.88) (-0.53) (-0.84) (-1.69) (2.60)
The table reports results from firm-year panel data OLS investment-Q regressions from 1998 to 2017. These regressions are run in the same way as our main panel-data investment
regressions in Table 7 with one exception. In particular, we use different subsamples and focus on the main sample plus the following two subsamples: (1) the sub sample of firm-years
that had zero patents in the ex-ante year t − 1 and (2) the subsample of firm-years that had at least one patent in the ex-ante year t − 1. We refer to these two subsamples as the “No
Patent Firms” and the “Patenting Firms”, respectively.
Table IA.5
Investment Panel Data Regressions (Robustness using Peters and Taylor (2017) Q)

(1) Full Sample 0.007 0.001 -0.042 0.021 -0.004 0.000 0.008 0.001 -0.044 0.792 68,763
(21.74) (0.56) (-15.96) (11.58) (-3.31) (-0.12) (2.76) (0.68) (-17.15)
(2) Tercile 1 0.003 0.000 -0.011 0.011 0.001 -0.003 0.001 0.000 -0.011 0.777 22,848
(5.06) (0.31) (-3.56) (2.80) (1.17) (-1.27) (0.36) (0.24) (-3.64)
(3) Tercile 2 0.004 0.000 -0.022 0.017 -0.002 -0.001 0.008 0.000 -0.023 0.811 22,826
(6.66) (-0.22) (-6.51) (5.60) (-1.96) (-0.41) (1.62) (-0.12) (-6.85)
(4) Tercile 3 0.008 0.003 -0.082 0.019 0.000 0.000 0.023 0.003 -0.084 0.791 22,794
(19.20) (1.57) (-13.79) (5.97) (-0.14) (-0.07) (2.24) (1.50) (-14.34)
(5) Full Sample 0.008 -0.006 -0.050 0.000 0.021 0.008 0.013 -0.007 -0.048 0.551 68,763
(24.12) (-6.74) (-16.07) (-0.27) (11.24) (3.63) (2.34) (-7.12) (-15.65)
(6) Tercile 1 0.010 -0.006 -0.037 0.005 0.016 0.004 0.022 -0.006 -0.037 0.499 22,848
(11.83) (-3.74) (-7.04) (1.28) (5.14) (1.15) (2.60) (-4.22) (-7.07)
(7) Tercile 2 0.010 -0.012 -0.043 0.007 0.020 0.004 0.007 -0.012 -0.042 0.560 22,826
(9.34) (-6.07) (-7.04) (1.26) (5.25) (0.71) (0.65) (-6.14) (-6.97)
(8) Tercile 3 0.007 -0.004 -0.062 -0.004 0.023 0.010 0.011 -0.004 -0.057 0.596 22,794
(16.71) (-2.11) (-9.19) (-1.47) (7.07) (3.05) (0.99) (-2.25) (-8.72)
Panel C: SDC Acquisitions/ assets
(9) Full Sample 0.010 -0.031 -0.023 0.010 0.002 0.016 0.013 -0.031 -0.024 0.221 68,763
(15.53) (-17.00) (-4.50) (3.11) (0.94) (4.44) (1.74) (-16.89) (-4.67)
77
(10) Tercile 1 0.004 -0.029 -0.017 -0.003 0.004 0.010 0.008 -0.029 -0.017 0.179 22,848
(3.73) (-8.62) (-2.09) (-0.46) (1.17) (1.84) (1.28) (-8.61) (-2.04)
(11) Tercile 2 0.010 -0.039 -0.009 0.009 0.007 0.012 0.024 -0.039 -0.009 0.241 22,826
(6.02) (-10.60) (-0.83) (1.11) (1.49) (1.36) (1.48) (-10.66) (-0.89)
(12) Tercile 3 0.010 -0.029 -0.026 0.008 0.003 0.018 0.015 -0.029 -0.028 0.295 22,794
(12.61) (-8.83) (-2.55) (1.80) (0.77) (3.59) (0.89) (-8.77) (-2.77)
Panel D: SDC Asset Sales/ assets
(13) Full Sample 0.000 0.001 0.002 -0.001 0.000 0.000 -0.001 0.001 0.002 0.258 68,763
(-1.00) (3.49) (2.07) (-1.09) (0.13) (0.76) (-0.83) (3.53) (2.09)
(14) Tercile 1 0.000 0.002 0.001 0.000 -0.001 0.001 0.000 0.002 0.001 0.275 22,848

(-1.27) (2.45) (0.23) (0.36) (-1.72) (0.67) (-0.03) (2.46) (0.25)
(15) Tercile 2 0.000 0.001 0.002 -0.001 0.000 0.001 0.002 0.001 0.001 0.291 22,826
(0.07) (1.64) (0.69) (-0.54) (-0.28) (0.57) (0.61) (1.60) (0.67)
(16) Tercile 3 0.000 0.001 0.004 0.000 0.000 -0.001 -0.004 0.001 0.004 0.309 22,794
(-1.48) (1.26) (1.88) (0.37) (0.14) (-0.90) (-0.92) (1.26) (1.87)
regressions in Table 7 with one exception. In particular, we use different subsamples and focus on the main sample plus the following two subsamples: (1) the sub sample of firm-years
that had zero patents in the ex-ante year t − 1 and (2) the subsample of firm-years that had at least one patent in the ex-ante year t − 1. We refer to these two subsamples as the “No
Patent Firms” and the “Patenting Firms”, respectively. The table reports results from firm-year panel data investment-Q regressions from 1998 to 2017. The dependent variable is ex
post R&D/assets (Panel A), CAPX/Assets (Panel B), SDC Acquisitions/Assets (Panel C), and SDC Asset Sales/Assets (Panel D). The key RHS variables in these “OLS” panel data
regressions are the lagged life cycle variable interactions with Tobins’ Q (Tobins Q is computed as in Peters and Taylor 2017). All regressions include controls for size, age, firm fixed
effects and year fixed effects. All RHS variables are ex ante measurable and are observable in year t − 1. In each panel, we consider three subsamples: the full sample, and those with
above or below median competition levels as measured using the TNIC HHI (see Hoberg and Phillips 2016). All ratio variables are winsorized at the 1/99% level. The last two columns
indicate the adjusted R2 and the number of observations. All regressions include firm and year fixed effects. t-statistics (clustered by firm) are reported in parentheses.
Table IA.6
Investment Panel Data Regressions (Measurement-Error Corrected using Erickson and Whited (2000))

(1) Base Q 0.076 -0.008 0.015 0.166 -0.033 -0.091 0.033 -0.013 -0.010 0.339 68,798
(40.31) (-12.79) (9.05) (26.05) (-5.84) (-14.22) (2.98) (-25.03) (-9.04)
(2) P &T Q 0.029 -0.009 0.001 0.040 -0.022 0.000 0.027 -0.007 -0.021 0.245 68,763
(33.41) (-22.01) (0.76) (14.57) (-9.38) (-0.10) (4.76) (-18.86) (-20.48)

(3) Base Q 0.011 0.005 -0.006 -0.035 0.068 -0.004 -0.022 0.005 -0.007 0.067 68,798
(9.52) (12.65) (-5.61) (-7.62) (11.61) (-1.26) (-4.10) (13.06) (-6.37)
(4) P &T Q 0.012 0.005 -0.013 -0.035 0.061 0.012 -0.020 0.006 -0.013 0.201 68,763
(44.30) (15.02) (-12.12) (-6.64) (14.49) (2.68) (-2.67) (15.33) (-10.85)

(5) Base Q 0.036 0.006 0.001 -0.002 -0.013 0.042 0.015 0.003 -0.010 0.060 68,798
(51.18) (14.63) (1.12) (-0.54) (-2.19) (13.08) (1.92) (10.08) (-12.19)
(6) P &T Q 0.017 0.002 -0.004 -0.010 -0.012 0.046 0.019 0.003 -0.014 0.092 68,763
78
(80.62) (6.31) (-3.62) (-2.15) (-1.89) (14.03) (2.50) (9.71) (-11.92)

(7) Base Q 0.020 0.003 0.010 -0.002 0.002 -0.001 -0.010 0.001 0.001 0.033 68,798
(11.61) (9.17) (10.74) (-1.28) (1.24) (-2.85) (-4.60) (5.69) (4.25)
(8) P &T Q 0.004 0.000 0.004 0.000 0.000 0.000 -0.002 0.001 0.001 0.036 68,763
(12.68) (4.04) (10.28) (-0.23) (0.05) (-0.22) (-1.51) (9.18) (2.20)

This table is run in the same way that Table 7 in the main paper (titled “Investment Panel Data Regressions”) is run with two modifications: (1) we run measurement-error-corrected
regression model from Erickson and Whited (2000) and (2) we consider both the basic measure of Tobin’s Q as well as the measure suggested by Peters and Taylor (2017) as noted
in the first column (we thank the authors for providing their data via WRDS). The measurement error model is implemented using the Stata command “xtewreg” and we thank Toni
Whited for providing the code for this routine on her website. In all models, we assume the 4 cumulants. In the conditional model, we assume 2 mismeasured variables: the life cycle
and Tobin’s Q. In the basic model, we assume only one mismeasured variable: Tobin’s Q.
The table reports results from firm-year panel data investment-Q regressions from 1998 to 2017. The dependent variable is ex post R&D/assets (Panels A), CAPX/Assets (Panel
B), SDC $ Acquisitions/Assets (Panel C), and SDC $ Asset Sales/Assets (Panel D). All dependent variables are based on the focal firm’s investment policies. All RHS variables are ex
ante measurable and are observable in year t − 1. In all models, the dependent variable is regressed on ex-ante life cycle variable interactions with Tobin’s Q and size plus age controls.
All ratio variables are winsorized at the 1/99% level. The last two columns indicate the adjusted R2 and the number of observations. All regressions include year fixed effects.
Table IA.7
Investment Panel Data Regressions (Using Narrower Exclusion Queries)

(1) Full Sample 0.007 -0.027 0.007 0.029 0.000 -0.008 0.011 -0.027 0.006 0.833 68,505
(15.59) (-21.53) (2.67) (12.28) (-0.21) (-3.78) (1.69) (-21.85) (2.27)
(2) Tercile 1 0.004 -0.011 0.006 0.019 0.000 -0.002 0.005 -0.011 0.006 0.816 22,751
(3.12) (-7.38) (2.26) (2.54) (0.04) (-0.92) (0.79) (-7.41) (2.21)
(3) Tercile 2 0.004 -0.019 0.004 0.013 0.002 -0.002 0.009 -0.019 0.004 0.851 22,739
(5.40) (-10.94) (1.36) (2.28) (1.03) (-0.41) (1.34) (-11.00) (1.32)
(4) Tercile 3 0.009 -0.048 0.010 0.030 0.002 -0.011 0.022 -0.047 0.007 0.829 22,718
(12.82) (-18.30) (1.72) (9.65) (0.57) (-3.11) (1.73) (-18.54) (1.22)
(5) Full Sample 0.008 -0.011 -0.018 -0.004 0.023 0.007 0.020 -0.011 -0.017 0.584 68,505
(23.82) (-14.83) (-7.97) (-2.40) (14.05) (4.54) (4.51) (-14.76) (-7.93)
(6) Tercile 1 0.009 -0.012 -0.011 0.000 0.013 0.008 0.020 -0.012 -0.011 0.515 22,751
(10.06) (-9.73) (-2.96) (0.11) (6.35) (2.74) (2.76) (-9.78) (-2.97)
(7) Tercile 2 0.010 -0.014 -0.011 -0.002 0.022 0.007 0.018 -0.014 -0.011 0.564 22,739
(12.41) (-9.82) (-2.56) (-0.65) (7.06) (2.45) (2.02) (-9.77) (-2.61)
(8) Tercile 3 0.006 -0.009 -0.028 -0.007 0.028 0.007 0.020 -0.009 -0.026 0.635 22,718
(15.11) (-6.82) (-5.88) (-2.55) (9.46) (2.70) (2.55) (-6.67) (-5.49)
(9) Full Sample 0.011 -0.024 -0.016 0.001 0.002 0.026 0.012 -0.024 -0.016 0.175 68,505
(13.36) (-16.09) (-4.15) (0.26) (0.64) (8.28) (1.39) (-16.36) (-4.00)
79
(10) Tercile 1 0.006 -0.026 -0.004 -0.003 0.008 0.008 0.020 -0.026 -0.004 0.150 22,751
(4.71) (-9.64) (-0.62) (-0.61) (2.17) (1.85) (1.80) (-9.63) (-0.61)
(11) Tercile 2 0.009 -0.031 -0.003 -0.001 0.006 0.019 0.015 -0.031 -0.003 0.208 22,739
(5.93) (-11.03) (-0.36) (-0.17) (1.42) (2.83) (1.22) (-11.10) (-0.35)
(12) Tercile 3 0.011 -0.021 -0.024 -0.002 0.002 0.032 0.006 -0.022 -0.022 0.223 22,718
(9.64) (-7.74) (-2.95) (-0.49) (0.47) (6.60) (0.39) (-8.20) (-2.79)
(13) Full Sample -0.001 0.000 0.003 0.000 -0.001 -0.001 -0.006 0.000 0.003 0.241 68,505
(-4.37) (0.97) (3.09) (0.26) (-2.27) (-0.94) (-2.63) (0.96) (3.11)
(14) Tercile 1 -0.001 0.001 0.001 -0.001 -0.001 -0.001 -0.003 0.001 0.001 0.262 22,751

(-3.05) (1.11) (0.61) (-0.46) (-1.02) (-0.47) (-0.73) (1.13) (0.62)
(15) Tercile 2 0.000 0.000 0.003 0.000 0.000 0.000 -0.006 0.000 0.003 0.268 22,739
(-1.45) (0.50) (1.29) (-0.33) (-0.39) (0.00) (-1.29) (0.50) (1.32)
(16) Tercile 3 -0.001 0.000 0.005 0.001 -0.001 -0.001 -0.011 0.000 0.005 0.277 22,718
(-2.70) (-0.16) (2.42) (0.68) (-1.17) (-1.18) (-2.08) (-0.23) (2.41)
regressions in Table 7 with one exception. In particular, we use more narrow anchor-phrase queries to define our main life cycle stage variables. Please note that, in Section 2., queries
are constructed by linking a list of product words to a list of action words. Anchor-phrase technology then relies on proximity searches to identify specific instances where the firm is
discussing issues that place it in one of the four life cycle stages.
In the alternative anchor-phrase formulation used to construct the life cycle stages for this table, we additionally require that paragraph anchor-phrase query hits for Life1, Life2,
and Life4 must not have content associated with any other life stage. For example, our refined query for Life1 not only requires the presence of the action words for Life1 summarized
in Section 2., but it also requires that any action words from Life2 and Life4 must not be present (note that Life3 is excluded from this list as it has no action words, and our main
methodology already ensures Life3 is free of action words). The queries used in this robustness test are stricter, and hence we observe fewer anchor-phrase hits overall. Note that we use
the broader queries in our main tests because firms can discuss products in more than one life stage in a single paragraph, and our main method would count all such mentions, and
thus offers more power. The results below are very similar to those in our main Table 7 and hence this distinction is not particularly important.
Table IA.8
Investment Panel Data Regressions (Using 10-Word Windows Instead of Paragraph Queries)

(1) Full Sample 0.007 -0.028 0.007 0.026 -0.001 -0.013 0.010 -0.028 0.005 0.832 67,468
(15.47) (-21.55) (2.72) (14.39) (-0.54) (-5.85) (2.50) (-21.91) (2.01)
(2) Tercile 1 0.004 -0.011 0.007 0.017 0.000 -0.004 0.006 -0.012 0.006 0.815 22,413
(3.10) (-7.34) (2.39) (2.70) (0.06) (-1.56) (1.38) (-7.43) (2.23)
(3) Tercile 2 0.004 -0.019 0.004 0.015 0.001 -0.006 0.008 -0.019 0.003 0.849 22,392
(5.44) (-10.87) (1.26) (4.03) (0.71) (-1.66) (1.92) (-10.92) (1.03)
(4) Tercile 3 0.009 -0.049 0.011 0.027 0.000 -0.018 0.020 -0.048 0.007 0.827 22,378
(12.70) (-18.35) (1.70) (10.28) (0.14) (-4.44) (2.21) (-18.60) (1.08)
(5) Full Sample 0.007 -0.011 -0.017 -0.005 0.022 0.009 0.016 -0.011 -0.017 0.589 67,468
(23.49) (-14.72) (-7.75) (-3.59) (14.51) (4.81) (5.44) (-14.67) (-7.55)
(6) Tercile 1 0.009 -0.012 -0.010 -0.001 0.014 0.009 0.019 -0.012 -0.010 0.515 22,413
(9.97) (-9.75) (-2.77) (-0.46) (6.59) (2.83) (4.31) (-9.82) (-2.74)
(7) Tercile 2 0.009 -0.014 -0.012 -0.001 0.022 0.005 0.013 -0.014 -0.012 0.568 22,392
(12.14) (-9.71) (-2.78) (-0.21) (7.28) (1.57) (2.30) (-9.71) (-2.78)
(8) Tercile 3 0.006 -0.009 -0.027 -0.008 0.027 0.010 0.015 -0.009 -0.024 0.640 22,378
(14.92) (-6.79) (-5.45) (-3.80) (9.57) (3.50) (2.84) (-6.68) (-4.90)
80
(9) Full Sample 0.011 -0.024 -0.017 0.006 0.000 0.028 0.016 -0.024 -0.016 0.176 67,468
(13.22) (-15.88) (-4.21) (2.41) (-0.13) (7.77) (2.78) (-16.11) (-4.05)
(10) Tercile 1 0.006 -0.026 -0.005 0.001 0.005 0.007 0.019 -0.026 -0.005 0.150 22,413
(4.67) (-9.45) (-0.81) (0.31) (1.48) (1.59) (2.38) (-9.45) (-0.79)
(11) Tercile 2 0.009 -0.031 -0.003 0.002 0.006 0.020 0.016 -0.031 -0.003 0.207 22,392
(5.90) (-10.91) (-0.43) (0.31) (1.70) (2.99) (1.74) (-10.97) (-0.38)
(12) Tercile 3 0.011 -0.021 -0.025 0.003 -0.003 0.037 0.024 -0.022 -0.024 0.224 22,378
(9.51) (-7.65) (-2.97) (0.87) (-0.60) (6.39) (2.17) (-8.07) (-2.88)
(13) Full Sample -0.001 0.000 0.003 0.000 -0.001 0.000 -0.005 0.000 0.003 0.241 67,468

(-4.36) (0.97) (2.98) (-0.09) (-1.61) (-0.58) (-3.31) (0.98) (3.03)
(14) Tercile 1 -0.001 0.001 0.001 0.000 -0.001 -0.001 -0.003 0.001 0.001 0.264 22,413
(-3.30) (0.97) (0.50) (-0.06) (-1.25) (-0.66) (-1.05) (1.00) (0.50)
(15) Tercile 2 0.000 0.000 0.002 0.000 0.000 0.000 -0.003 0.000 0.003 0.266 22,392
(-1.39) (0.54) (1.21) (-0.43) (-0.46) (0.29) (-1.13) (0.55) (1.25)
(16) Tercile 3 -0.001 0.000 0.006 0.000 0.000 -0.001 -0.008 0.000 0.006 0.274 22,378
(-2.63) (-0.04) (2.45) (0.23) (-0.47) (-0.73) (-2.59) (-0.11) (2.48)
regressions in Table 7 with one exception. In particular, we use 10-word window anchor-phrase queries to define our main life cycle stage variables instead of paragraph-based queries.
In particular, the key terms that define each life cycle query must appear within a 10-word window of one another within the 10-K in order to count as a valid hit on the given life cycle
stage. This is in contrast to our baseline results, which instead require that the key words appear within the same paragraph. The results below are very similar to those in our main
Table 7 and hence this distinction is not particularly important.
Table IA.9
Investment Panel Data Regressions (Using 5-Word Windows Instead of Paragraph Queries)

(1) Full Sample 0.008 -0.029 0.007 0.024 0.000 -0.011 0.010 -0.029 0.005 0.828 63,644
(15.17) (-21.77) (2.61) (13.83) (0.19) (-5.22) (2.63) (-22.07) (1.92)
(2) Tercile 1 0.005 -0.012 0.007 0.016 0.001 -0.002 0.006 -0.012 0.007 0.808 21,148
(3.02) (-7.25) (2.35) (2.43) (0.32) (-0.98) (1.52) (-7.34) (2.22)
(3) Tercile 2 0.004 -0.021 0.004 0.014 0.001 -0.005 0.008 -0.021 0.003 0.845 21,130
(5.09) (-11.05) (1.25) (3.81) (0.39) (-1.35) (2.14) (-11.15) (1.01)
(4) Tercile 3 0.009 -0.052 0.010 0.023 0.003 -0.015 0.018 -0.051 0.005 0.819 21,112
(12.25) (-18.88) (1.39) (9.48) (0.93) (-4.15) (2.35) (-19.07) (0.80)
(5) Full Sample 0.007 -0.011 -0.015 -0.003 0.020 0.008 0.015 -0.011 -0.015 0.593 63,644
(22.87) (-14.74) (-6.91) (-2.42) (13.50) (4.82) (5.81) (-14.67) (-6.70)
(6) Tercile 1 0.009 -0.012 -0.008 -0.001 0.013 0.010 0.018 -0.012 -0.008 0.518 21,148
(9.82) (-9.49) (-2.20) (-0.56) (6.05) (3.49) (4.69) (-9.55) (-2.16)
(7) Tercile 2 0.009 -0.013 -0.012 0.001 0.021 0.004 0.012 -0.013 -0.011 0.580 21,130
(12.10) (-9.60) (-2.72) (0.50) (7.48) (1.33) (2.52) (-9.54) (-2.73)
(8) Tercile 3 0.006 -0.009 -0.023 -0.006 0.024 0.010 0.014 -0.009 -0.020 0.644 21,112
(14.27) (-6.84) (-4.51) (-2.86) (8.84) (3.45) (3.11) (-6.79) (-3.95)
81
(9) Full Sample 0.011 -0.024 -0.017 0.007 -0.002 0.028 0.016 -0.024 -0.016 0.179 63,644
(12.95) (-15.10) (-3.91) (2.85) (-0.72) (8.13) (3.18) (-15.33) (-3.74)
(10) Tercile 1 0.006 -0.026 -0.006 0.001 0.003 0.011 0.016 -0.026 -0.005 0.153 21,148
(4.42) (-9.13) (-0.77) (0.18) (1.00) (2.41) (2.22) (-9.11) (-0.72)
(11) Tercile 2 0.010 -0.030 -0.001 0.001 0.006 0.023 0.019 -0.030 -0.001 0.214 21,130
(5.92) (-10.25) (-0.16) (0.11) (1.62) (3.32) (2.20) (-10.30) (-0.09)
(12) Tercile 3 0.011 -0.020 -0.026 0.007 -0.007 0.034 0.024 -0.021 -0.026 0.226 21,112
(9.23) (-7.23) (-2.86) (1.93) (-1.55) (6.34) (2.41) (-7.59) (-2.85)
(13) Full Sample -0.001 0.000 0.003 0.000 0.000 0.000 -0.004 0.000 0.003 0.242 63,644

(-4.24) (0.59) (2.65) (-0.47) (-0.87) (-0.74) (-3.51) (0.60) (2.69)
(14) Tercile 1 -0.001 0.001 0.000 0.000 -0.001 -0.002 -0.003 0.001 0.000 0.271 21,148
(-3.56) (1.01) (0.11) (0.29) (-1.15) (-1.19) (-1.41) (1.04) (0.08)
(15) Tercile 2 0.000 0.001 0.003 -0.001 0.000 0.001 -0.003 0.001 0.003 0.259 21,130
(-1.08) (0.72) (1.51) (-0.90) (-0.33) (1.02) (-1.23) (0.73) (1.57)
(16) Tercile 3 0.000 0.000 0.006 0.000 0.000 -0.001 -0.007 -0.001 0.006 0.274 21,112
(-2.49) (-0.73) (2.47) (-0.32) (0.14) (-0.63) (-2.62) (-0.81) (2.53)
regressions in Table 7 with one exception. In particular, we use 5-word window anchor-phrase queries to define our main life cycle stage variables instead of paragraph-based queries.
In particular, the key terms that define each life cycle query must appear within a 5-word window of one another within the 10-K in order to count as a valid hit on the given life cycle
stage. This is in contrast to our baseline results, which instead require that the key words appear within the same paragraph. The results below are very similar to those in our main
Table 7 and hence this distinction is not particularly important.
Table IA.10
Investment Panel Data Regressions (add controls for financial constraints)
Conditional Model
No
TobQ x TobQ x TobQ x TobQ x log log Delay Cap.+Liq. Adj #
Row Sample Life1 Life2 Life3 Life4 assets age Constraint Dummy R2 obs

(1) Full Sample 0.029 -0.002 -0.011 0.014 -0.027 0.006 0.021 0.000 0.833 68,798
(12.39) (-1.47) (-4.39) (2.38) (-21.92) (2.41) (3.34) (0.02)
(2) TSIMM Tercile 1 0.019 -0.001 -0.003 0.007 -0.011 0.006 0.015 -0.001 0.815 22,848
(2.65) (-0.82) (-1.10) (1.21) (-7.42) (2.30) (2.62) (-0.97)
(3) TSIMM Tercile 2 0.014 0.001 -0.004 0.011 -0.019 0.003 0.019 0.001 0.851 22,837
(2.90) (0.51) (-1.01) (2.00) (-11.03) (1.25) (2.05) (0.55)
(4) TSIMM Tercile 3 0.030 -0.001 -0.016 0.025 -0.047 0.008 0.028 0.001 0.829 22,814
(9.17) (-0.25) (-3.48) (2.26) (-18.67) (1.38) (1.89) (0.28)
(5) Full Sample -0.007 0.024 0.008 0.015 -0.011 -0.018 0.000 -0.001 0.585 68,798
(-4.14) (14.66) (4.12) (4.01) (-14.83) (-7.94) (-0.01) (-1.09)
(6) TSIMM Tercile 1 -0.002 0.015 0.008 0.020 -0.012 -0.011 -0.003 0.000 0.515 22,848
(-0.66) (6.40) (2.34) (3.11) (-9.88) (-2.97) (-0.42) (-0.13)
(7) TSIMM Tercile 2 -0.004 0.023 0.007 0.015 -0.014 -0.012 -0.014 -0.002 0.564 22,837
(-1.14) (7.28) (2.14) (1.89) (-9.69) (-2.77) (-1.48) (-1.15)
(8) TSIMM Tercile 3 -0.010 0.030 0.008 0.010 -0.009 -0.025 0.012 -0.001 0.637 22,814
(-3.98) (10.18) (2.61) (1.49) (-6.77) (-5.21) (1.10) (-0.28)
82
(9) Full Sample 0.004 0.002 0.026 0.014 -0.024 -0.016 -0.016 0.000 0.174 68,798
(1.40) (0.82) (6.82) (2.03) (-16.27) (-4.14) (-1.63) (-0.14)
(10) TSIMM Tercile 1 -0.003 0.007 0.008 0.018 -0.025 -0.005 -0.008 0.001 0.150 22,848
(-0.63) (2.07) (1.74) (1.77) (-9.50) (-0.73) (-0.50) (0.21)
(11) TSIMM Tercile 2 0.004 0.007 0.016 0.018 -0.031 -0.002 -0.003 -0.005 0.206 22,837
(0.57) (1.68) (2.16) (1.57) (-10.91) (-0.29) (-0.18) (-1.61)
(12) TSIMM Tercile 3 0.000 0.003 0.034 0.015 -0.022 -0.025 -0.036 0.007 0.223 22,814
(-0.01) (0.70) (5.58) (1.10) (-8.17) (-3.01) (-1.84) (1.74)
(13) Full Sample 0.000 -0.001 0.000 -0.005 0.000 0.004 0.000 0.000 0.238 68,798

(0.26) (-2.08) (-0.48) (-2.99) (1.08) (3.20) (-0.04) (-0.68)
(14) TSIMM Tercile 1 -0.001 -0.001 0.000 -0.002 0.001 0.001 -0.003 0.000 0.259 22,848
(-0.96) (-1.08) (0.17) (-0.62) (1.14) (0.62) (-0.56) (-0.13)
(15) TSIMM Tercile 2 0.000 0.000 0.000 -0.005 0.000 0.003 0.002 0.001 0.266 22,837
(0.05) (-0.45) (-0.15) (-1.28) (0.59) (1.28) (0.30) (0.58)
(16) TSIMM Tercile 3 0.001 -0.001 -0.001 -0.011 0.000 0.006 0.001 -0.002 0.275 22,814
(0.86) (-0.90) (-0.96) (-2.65) (-0.12) (2.51) (0.21) (-1.20)
regressions in Table 7 with one exception. In particular, we add a control for financial constraints based on the delay-investment query in Hoberg and Maksimovic (2015). We also
include a dummy variable indicating firms that do not have a Capitalization and Liquidity section in their 10-K (as advised by the authors). These regressions also include controls for
size and age, which are also measures of constraints as noted in Hadlock and Pierce (2010). All regressions include firm and year fixed effects. t-statistics (clustered by firm) are reported
in parentheses.
Table IA.11
Investment Panel Data Regressions (with Lagged Dependent Variable)
Lagged Lagged
TobQ x TobQ x TobQ x TobQ x log log Dep Dep Var Adj #
Row Model Life1 Life2 Life3 Life4 assets age Var x TobQ R2 obs

(1) Baseline 0.010 -0.002 0.001 0.007 -0.010 -0.003 0.547 0.024 0.895 68,798
(5.56) (-2.84) (0.49) (1.84) (-15.79) (-2.43) (38.40) (9.17)
(2) Arrellano- 0.010 -0.002 0.001 0.007 -0.010 -0.003 0.547 0.024 59,916
Bond (6.08) (-3.08) (0.53) (1.95) (-17.50) (-2.67) (42.92) (10.22)

(3) Baseline -0.006 0.016 0.004 0.012 -0.010 -0.013 0.344 0.055 0.622 68,798
(-3.84) (11.44) (2.12) (3.60) (-16.77) (-7.15) (25.41) (8.67)
(4) Arrellano- -0.006 0.016 0.004 0.012 -0.010 -0.013 0.344 0.055 59,916
Bond (-4.16) (12.44) (2.30) (3.87) (-18.19) (-7.78) (28.18) (9.45)

(5) Baseline 0.003 0.002 0.019 0.015 -0.024 -0.016 0.006 0.008 0.177 68,798
(0.89) (0.72) (4.89) (2.09) (-16.81) (-4.08) (4.76) (6.15)
(6) Arrellano- 0.003 0.002 0.019 0.015 -0.024 -0.016 0.006 0.008 59,916
83
Bond (0.96) (0.77) (5.27) (2.28) (-18.26) (-4.43) (5.15) (6.61)

(7) Baseline 0.000 -0.001 0.000 -0.006 0.000 0.004 -0.001 0.001 0.238 68,798
(0.14) (-2.11) (-0.62) (-3.35) (1.19) (3.24) (-1.86) (1.43)
(8) Arrellano- 0.000 -0.001 0.000 -0.006 0.000 0.004 -0.001 0.001 59,916
Bond (0.16) (-2.28) (-0.67) (-3.66) (1.31) (3.52) (-2.02) (1.54)

The table reports results from firm-year panel data investment-Q regressions from 1998 to 2017. The dependent variable is ex post R&D/assets (Panels A), CAPX/ assets (Panel B),
SDC Acquisitions/ assets (Panel C), and SDC Sales as Targets/ assets (Panel D). In each panel, we consider two models. The “Baseline” model uses OLS includes firm and year fixed
effects. The “Arrellano-Bond” model is similar and further ensures econometric biases associated with the joint use of firm fixed effects and lagged dependent variables do not induce
bias. All dependent variables are based on the focal firm’s investment policies. All RHS variables are ex ante measurable and are observable in year t − 1. In all models, the dependent
variable is regressed on ex-ante life cycle variables, interactions with Tobin’s Q, and size plus age controls. Tobin’s Q is re-centered at the sample mean prior to running the regressions
so that the Life1, Life2, and Life4 coefficients are interpretable as the impact of one sigma of the given variable on the dependent variable for a firm having an average Q. All ratio
variables are winsorized at the 1/99% level. The last two columns indicate the adjusted R2 and the number of observations. All regressions include firm and year fixed effects. t-statistics
(clustered by firm) are reported in parentheses.
Table IA.12
Investment Panel Data Regressions by Fama-French-5 Industry Sectors

(1) Full Sample 0.007 -0.027 0.007 0.029 -0.002 -0.011 0.014 -0.027 0.005 0.833 68,798
(15.58) (-21.53) (2.63) (12.41) (-1.45) (-4.38) (2.38) (-21.90) (2.13)
(2) Tech 0.005 -0.040 0.009 0.011 -0.002 0.003 0.017 -0.041 0.009 0.772 18,905
(9.19) (-18.37) (1.68) (3.24) (-0.64) (0.80) (1.51) (-18.45) (1.64)
(3) Manufacturing 0.004 -0.004 0.005 0.018 -0.001 -0.002 0.004 -0.004 0.005 0.825 15,971
(4.13) (-4.22) (2.78) (3.36) (-0.52) (-0.58) (0.65) (-4.34) (2.81)
(4) Health 0.014 -0.059 -0.022 0.035 0.009 -0.022 0.015 -0.058 -0.022 0.797 9,262
(11.57) (-14.72) (-1.64) (7.17) (1.09) (-2.92) (0.86) (-14.78) (-1.64)
(5) Consumer 0.004 -0.004 0.001 0.021 -0.001 -0.005 0.004 -0.004 0.001 0.807 14,097
(2.08) (-2.49) (0.50) (2.32) (-0.61) (-1.83) (1.05) (-2.47) (0.51)
(6) Miscellaneous 0.003 -0.005 0.005 0.012 -0.002 0.000 0.007 -0.005 0.005 0.764 10,563
(2.59) (-2.98) (1.88) (1.87) (-1.27) (0.01) (1.34) (-3.03) (1.91)
(7) Full Sample 0.008 -0.011 -0.018 -0.007 0.024 0.008 0.015 -0.011 -0.018 0.585 68,798
84
(23.87) (-14.93) (-8.10) (-4.14) (14.65) (4.12) (4.00) (-14.83) (-8.04)
(8) Tech 0.006 -0.011 -0.005 -0.004 0.018 0.008 0.009 -0.011 -0.005 0.550 18,905
(15.14) (-9.99) (-1.37) (-1.60) (6.52) (2.69) (1.88) (-9.84) (-1.43)
(9) Manufacturing 0.019 -0.014 -0.024 -0.009 0.037 0.009 0.031 -0.013 -0.025 0.654 15,971
(8.92) (-5.79) (-4.44) (-1.05) (7.98) (1.25) (2.21) (-5.71) (-4.70)
(10) Health 0.006 -0.005 -0.012 0.000 0.016 0.002 0.015 -0.005 -0.011 0.384 9,262
(8.12) (-4.68) (-2.52) (-0.14) (4.22) (0.65) (1.40) (-4.76) (-2.26)
(11) Consumer 0.012 -0.013 -0.014 0.000 0.021 0.011 0.019 -0.013 -0.015 0.562 14,097
(9.92) (-8.23) (-3.12) (-0.10) (7.64) (2.87) (2.45) (-8.38) (-3.18)
(12) Miscellaneous 0.007 -0.015 -0.025 -0.003 0.016 0.004 0.019 -0.015 -0.025 0.527 10,563

(6.85) (-5.78) (-3.45) (-0.49) (4.06) (0.76) (1.98) (-5.80) (-3.45)
The table reports results from firm-year panel data OLS investment-Q regressions from 1998 to 2017. The table shows results for the overall sample and separately for each of the
Fama-French-5 industry sectors (we thank Ken French for providing this classification mapping on his website). The dependent variable is ex post R&D/assets (Panel A), CAPX/Assets
(Panel B), SDC $ Acquisitions/Assets (Panel C), and SDC $ Asset Sales/Assets (Panel D). The key RHS variables are the lagged life cycle variables interacted with Tobins’ Q. All
regressions include controls for size, age, firm fixed effects and year fixed effects. All RHS variables are ex ante measurable and are observable in year t − 1. All ratio variables are
winsorized at the 1/99% level. The last two columns indicate the adjusted R2 and the number of observations. All regressions include firm and year fixed effects. t-statistics (clustered
by firm) are reported in parentheses.
Table IA.12
Investment Panel Data Regressions by Fama-French-5 Industry Sectors (Continued)

(13) Full Sample 0.010 -0.024 -0.016 0.004 0.002 0.026 0.014 -0.024 -0.016 0.174 68,798
(13.30) (-16.08) (-4.21) (1.39) (0.81) (6.82) (2.03) (-16.26) (-4.07)
(14) Tech 0.012 -0.022 -0.024 0.013 -0.003 0.022 0.011 -0.022 -0.025 0.206 18,905
(9.89) (-8.82) (-2.73) (1.88) (-0.43) (3.14) (0.78) (-8.96) (-2.80)
(15) Manufacturing 0.009 -0.038 0.003 0.010 -0.002 0.027 0.020 -0.038 0.004 0.161 15,971
(3.74) (-9.74) (0.39) (0.93) (-0.52) (2.75) (1.40) (-9.78) (0.46)
(16) Health 0.005 -0.011 -0.031 0.001 -0.005 0.021 0.031 -0.011 -0.034 0.094 9,262
(4.11) (-4.11) (-2.90) (0.22) (-0.59) (2.74) (1.49) (-4.11) (-3.18)
(17) Consumer 0.007 -0.028 -0.005 0.015 0.005 0.000 0.016 -0.028 -0.005 0.130 14,097
(3.61) (-7.85) (-0.63) (1.21) (1.26) (0.03) (1.25) (-7.86) (-0.63)
85
(18) Miscellaneous 0.012 -0.029 -0.025 0.017 0.012 0.007 0.010 -0.029 -0.025 0.215 10,563
(4.94) (-7.02) (-2.60) (1.53) (2.18) (0.78) (0.79) (-7.04) (-2.60)
(19) Full Sample -0.001 0.000 0.003 0.000 -0.001 0.000 -0.005 0.000 0.004 0.238 68,798
(-4.42) (1.07) (3.15) (0.26) (-2.09) (-0.48) (-2.99) (1.07) (3.20)
(20) Tech -0.001 0.000 0.002 0.000 -0.002 0.000 -0.002 0.000 0.002 0.263 18,905
(-2.66) (-0.37) (1.09) (-0.17) (-1.56) (0.15) (-0.67) (-0.40) (1.09)
(21) Manufacturing -0.001 0.002 0.003 -0.002 -0.001 0.000 -0.008 0.002 0.003 0.247 15,971
(-3.64) (2.55) (1.30) (-0.65) (-0.72) (-0.14) (-1.85) (2.57) (1.35)
(22) Health -0.001 0.000 0.006 0.001 0.000 -0.002 -0.009 0.000 0.006 0.176 9,262

(-2.13) (0.14) (1.95) (0.58) (-0.25) (-0.89) (-1.89) (0.10) (2.03)
(23) Consumer -0.001 0.001 0.004 -0.001 -0.002 0.000 -0.005 0.001 0.004 0.253 14,097
(-2.43) (0.82) (1.57) (-0.28) (-1.66) (0.27) (-0.96) (0.89) (1.59)
(24) Miscellaneous -0.001 -0.001 0.007 -0.001 0.001 -0.002 -0.005 -0.001 0.006 0.201 10,563
(-1.78) (-0.64) (2.12) (-0.38) (0.62) (-1.04) (-0.97) (-0.64) (2.12)
Table IA.13
Investment Policy Economic Magnitudes: Life Cycle Conditional Model versus Basic Q
Model (Interaction Term Sorts)
Life vs
Life Basic Basic Basic Basic Basic
Dependent Model Model Model Model Model Model
Row Variable Tercile Tercile 1 Tercile 2 Tercile 3 T3−T1 Ratio
1 R&D/ assets Tercile 1 0.010 0.005 0.004 -0.006

3 R&D/ assets Tercile 3 0.118 0.102 0.136 0.018
4 R&D/ assets High − Low 0.108 0.097 0.132 ∞
5 CAPX/ assets Tercile 1 0.039 0.039 0.043 0.004

6 CAPX/ assets Tercile 2 0.063 0.052 0.044 -0.019
7 CAPX/ assets Tercile 3 0.146 0.106 0.074 -0.072
8 CAPX/ assets High − Low 0.107 0.067 0.031 ∞
9 SDC $ Acq/ assets Tercile 1 0.023 0.040 0.041 0.018

12 SDC $ Acq/ assets High − Low 0.009 0.009 0.019 0.64
13 SDC $ Divest/ assets Tercile 1 0.011 0.010 0.007 -0.004

16 SDC $ Divest/ assets High − Low 0.006 0.004 0.001 ∞
The table reports economic magnitudes of the relationship between life cycle Q-sensitivities and investment policies, and
additionally compares the economic impact to that to the basic Q model. We form two-way independent sorts of all observations
into terciles based on each of the two dimensions: life cycle Q interactions and Tobins Q alone. The first sort variable,
representing the life cycle impact, is one of the four life cycle variables interacted with Tobins’ Q. For R&D, CAPX, acquisitions
and asset sales, we use life1, life2, life3 and life4 for the life cycle variable, respectively. This reflects the natural ordering of
investments found in earlier tables. The second sort variable is Tobins Q alone. We use independent annual sorts of both the
life cycle Q interactions and the raw Tobins Q alone and the table below reports the average value of the investment variable for
each of the nine (3x3) bins. We also report the average high minus low difference for the two directions of the sorts, indicating
the total ability of each sort variable to explain the given investment variable. Finally, in the last column we report the ratio of
the average high minus low portfolio differences for the life cycle model divided by the same average high minus low difference
for the competition portfolio. A value exceeding one, intuitively, indicates that the total ability of the life cycle model to explain
the given investment policy exceeds that of the competition model. A value of one indicates that both are equally important.
If the value becomes negative, indicating one sort variable fully subsumes the other, we report a value of ∞.
86
Table IA.14
Investment Policy Economic Magnitudes: Life Cycles versus Competition (Interaction Term
Sorts)
Life Compet Compet Compet Compet Life vs
Dependent Model Model Model Model Model Compet
Row Variable Tercile Tercile 1 Tercile 2 Tercile 3 T3−T1 Ratio

4 R&D/ assets High − Low 0.031 0.052 0.167 1.80

8 CAPX/ assets High − Low 0.031 0.024 0.052 2.73

12 SDC $ Acq/ assets High − Low 0.015 0.025 0.015 0.94
13 SDC $ Divest/ assets Tercile 1 0.010 0.009 0.011 0.001

14 SDC $ Divest/ assets Tercile 2 0.011 0.011 0.011 0.001
16 SDC $ Divest/ assets High − Low 0.006 0.003 -0.000 ∞
The table reports economic magnitudes of the relationship between life cycle Q-sensitivities and investment policies, and
additionally compares the economic impact to that of competition. We form two-way independent sorts of all observations
into terciles based on each of the two dimensions: life cycle and competition. The first sort variable, representing the life cycle
impact, is one of the four life cycle variables interacted with Tobins’ Q. For R&D, CAPX, acquisitions and a asset sales, we
use life1, life2, life3 and life4 for the life cycle variable, respectively. This reflects the natural ordering of investments found in
earlier tables. The second sort variable is TNIC-3 Total Similarity multiplied by Tobins Q. We use independent annual sorts of
both the life cycle and competition variables and the table below reports the average value of the investment variable for each
of the nine (3x3) bins. We also report the average high minus low difference for the two directions of the sorts, indicating the
total ability of each sort variable to explain the given investment variable. Finally, in the last column we report the ratio of the
average high minus low portfolio differences for the life cycle model divided by the same average high minus low difference for
the competition portfolio. A value exceeding one, intuitively, indicates that the total ability of the life cycle model to explain
the given investment policy exceeds that of the competition model. A value of one indicates that both are equally important.
If the value becomes negative, indicating one sort variable fully subsumes the other, we report a value of ∞. As competition
and life cycle stages are only modestly correlated, the explanatory power of the two models is mostly distinct and additive.
87
Table IA.15
CAPX Investment-Q Regressions
Tobin’s log log Adj. TobQ x TobQ x TobQ x TobQ x Adj

Row Year Q age assets R2 Life1 Life2 Life3 Life4 R2
(1) 1998 0.005 0.006 -0.017 0.027 -0.019 0.049 0.002 -0.022 0.089
1998 (5.83) (8.09) (-8.56) (-8.18) (19.4) (0.74) (-2.65)
(2) 1999 0.005 0.005 -0.010 0.028 -0.005 0.026 0.006 -0.034 0.051
1999 (8.12) (7.54) (-6.25) (-2.43) (11.9) (2.26) (-4.73)
(3) 2000 0.005 0.003 -0.009 0.050 -0.009 0.030 0.007 -0.009 0.083
2000 (13.2) (3.86) (-4.62) (-5.97) (14.5) (3.86) (-1.82)
(4) 2001 0.004 0.003 -0.000 0.023 -0.012 0.037 0.002 -0.031 0.098
2001 (9.12) (5.09) (-0.26) (-8.04) (20.6) (1.19) (-5.39)
(5) 2002 0.005 0.002 0.001 0.021 -0.008 0.033 -0.002 -0.013 0.087
2002 (8.53) (4.44) (0.87) (-5.11) (18.7) (-1.12) (-2.45)
(6) 2003 0.008 0.002 0.000 0.028 -0.008 0.045 -0.008 -0.028 0.109
2003 (9.90) (4.48) (0.12) (-3.32) (20.6) (-2.82) (-4.76)
(7) 2004 0.005 0.001 -0.002 0.013 -0.016 0.039 -0.005 -0.017 0.102
2004 (6.75) (2.29) (-1.16) (-7.16) (19.8) (-1.68) (-3.00)
(8) 2005 0.003 0.002 -0.004 0.007 -0.022 0.044 -0.013 -0.023 0.126
2005 (3.89) (3.59) (-1.93) (-9.18) (21.5) (-4.25) (-3.69)
(9) 2006 0.006 0.004 -0.008 0.017 -0.025 0.063 -0.028 -0.031 0.180
2006 (5.92) (4.88) (-3.63) (-8.98) (26.2) (-7.49) (-3.78)
(10) 2007 0.005 0.006 -0.016 0.028 -0.028 0.066 -0.032 -0.028 0.174
2007 (4.59) (7.24) (-6.82) (-8.71) (24.2) (-7.52) (-3.07)
(11) 2008 0.004 0.006 -0.014 0.029 -0.030 0.063 -0.027 -0.029 0.184
2008 (4.06) (7.95) (-6.58) (-10.1) (24.7) (-6.80) (-3.35)
88
(12) 2009 0.005 0.005 -0.006 0.035 -0.020 0.050 -0.020 -0.022 0.147
2009 (5.12) (9.74) (-4.90) (-7.59) (20.8) (-5.38) (-3.08)
(13) 2010 0.006 0.005 -0.012 0.038 -0.028 0.066 -0.020 -0.028 0.201
2010 (6.09) (7.47) (-6.93) (-9.59) (25.0) (-4.76) (-3.31)
(14) 2011 0.003 0.005 -0.013 0.024 -0.025 0.060 -0.028 -0.030 0.170
2011 (2.92) (6.74) (-6.13) (-8.14) (21.8) (-6.47) (-3.27)
(15) 2012 0.003 0.005 -0.017 0.032 -0.033 0.069 -0.027 -0.036 0.188
2012 (2.25) (6.98) (-8.03) (-9.20) (22.2) (-5.56) (-3.01)
(16) 2013 0.003 0.004 -0.014 0.028 -0.025 0.060 -0.024 -0.036 0.169
2013 (2.56) (6.53) (-7.05) (-8.24) (21.0) (-5.57) (-3.48)

(17) 2014 0.001 0.005 -0.016 0.025 -0.022 0.056 -0.032 -0.024 0.154
2014 (0.94) (6.12) (-7.38) (-8.06) (19.9) (-7.49) (-2.46)
(18) 2015 0.001 0.004 -0.010 0.029 -0.018 0.041 -0.014 -0.019 0.138
2015 (1.35) (7.87) (-7.02) (-10.0) (18.1) (-4.62) (-2.69)
(19) 2016 0.001 0.003 -0.005 0.018 -0.015 0.034 -0.013 -0.017 0.116
2016 (1.95) (6.94) (-3.73) (-8.09) (16.8) (-4.19) (-2.48)
(20) 2017 0.001 0.003 -0.005 0.012 -0.017 0.047 -0.028 -0.022 0.147
2017 (1.88) (5.21) (-3.67) (-7.67) (19.9) (-7.69) (-2.62)
The table reports results from annual OLS investment-Q regressions from 1998 to 2017. Regressions are run separately in each year and each regression is purely cross sectional, as one
observation is one firm. The dependent variable in all models is ex post CAPX/assets in year t. All RHS variables are ex ante observable in year t − 1. In all, the results below are based
on two distinct Q-models. The first block of four columns is the basic investment-Q regression where CAPX/assets is regressed on ex-ante Tobin’s Q and basic controls. The second
block of 9 columns is the conditional model, where CAPX/assets is regressed on the life variable interactions with Tobin’s Q (here controls for log age and log assets are included but
are not reported to conserve space). t-statistics are in parentheses.
Table IA.16
R&D Investment-Q Regressions

(1) 1998 0.027 -0.011 -0.004 0.211 0.096 -0.026 -0.023 0.022 0.367
1998 (29.6) (-13.4) (-2.14) (43.6) (-10.8) (-8.31) (2.78)
(2) 1999 0.027 -0.014 -0.004 0.238 0.110 -0.027 -0.038 0.021 0.397
1999 (30.7) (-16.4) (-1.56) (43.6) (-10.0) (-12.5) (2.36)
(3) 2000 0.016 -0.017 0.002 0.273 0.060 -0.015 -0.022 0.029 0.353
2000 (33.2) (-17.5) (0.96) (30.8) (-5.85) (-9.84) (4.95)
(4) 2001 0.016 -0.012 -0.011 0.268 0.063 -0.019 -0.020 0.017 0.383
2001 (31.4) (-16.3) (-5.67) (35.9) (-9.28) (-9.24) (2.59)
(5) 2002 0.026 -0.012 -0.011 0.278 0.086 -0.019 -0.024 0.034 0.417
2002 (30.4) (-16.3) (-5.31) (37.5) (-7.50) (-7.58) (4.41)
(6) 2003 0.026 -0.012 -0.017 0.213 0.127 -0.034 -0.036 0.021 0.404
2003 (20.9) (-15.1) (-7.32) (37.4) (-11.1) (-8.61) (2.50)
(7) 2004 0.022 -0.010 -0.011 0.239 0.100 -0.023 -0.028 0.003 0.412
2004 (23.4) (-12.8) (-4.56) (36.6) (-9.12) (-8.27) (0.43)
(8) 2005 0.021 -0.010 -0.011 0.216 0.110 -0.025 -0.035 -0.007 0.438
2005 (21.1) (-11.7) (-4.64) (40.7) (-11.0) (-10.1) (-0.97)
(9) 2006 0.023 -0.012 -0.010 0.202 0.121 -0.029 -0.036 -0.015 0.436
2006 (19.6) (-13.0) (-3.65) (40.5) (-11.3) (-9.03) (-1.72)
(10) 2007 0.027 -0.012 -0.007 0.204 0.145 -0.033 -0.054 -0.005 0.468
2007 (20.2) (-12.5) (-2.49) (44.6) (-11.7) (-12.5) (-0.56)
(11) 2008 0.023 -0.012 -0.004 0.204 0.127 -0.030 -0.044 -0.006 0.456
2008 (20.1) (-13.7) (-1.62) (42.5) (-11.9) (-10.9) (-0.68)
89
(12) 2009 0.030 -0.015 -0.007 0.193 0.180 -0.044 -0.067 -0.003 0.432
2009 (17.2) (-16.6) (-2.76) (39.2) (-10.9) (-10.6) (-0.27)
(13) 2010 0.035 -0.014 0.002 0.252 0.152 -0.029 -0.052 0.028 0.470
2010 (23.4) (-15.3) (0.70) (40.2) (-8.61) (-9.50) (2.51)
(14) 2011 0.030 -0.013 -0.001 0.252 0.132 -0.027 -0.044 0.002 0.472
2011 (22.8) (-14.4) (-0.53) (39.6) (-8.95) (-9.16) (0.15)
(15) 2012 0.027 -0.014 -0.006 0.216 0.138 -0.034 -0.041 0.004 0.412
2012 (18.6) (-15.1) (-2.29) (32.8) (-9.33) (-7.44) (0.27)
(16) 2013 0.030 -0.016 -0.002 0.245 0.139 -0.037 -0.039 0.001 0.446
2013 (20.8) (-16.1) (-0.78) (34.4) (-9.94) (-6.88) (0.06)

(17) 2014 0.026 -0.014 -0.004 0.276 0.112 -0.033 -0.031 -0.004 0.491
2014 (23.7) (-15.0) (-1.49) (39.1) (-10.8) (-6.72) (-0.42)
(18) 2015 0.031 -0.016 -0.008 0.307 0.115 -0.032 -0.029 0.003 0.492
2015 (25.4) (-16.0) (-3.30) (38.5) (-8.70) (-5.85) (0.29)
(19) 2016 0.032 -0.015 -0.014 0.322 0.124 -0.031 -0.032 0.004 0.512
2016 (26.0) (-16.2) (-5.49) (38.3) (-8.75) (-5.87) (0.37)
(20) 2017 0.036 -0.017 -0.025 0.302 0.152 -0.036 -0.040 -0.030 0.508
2017 (22.2) (-15.3) (-8.10) (37.1) (-8.16) (-5.78) (-1.93)
The table reports results from annual OLS investment-Q regressions from 1998 to 2017. Regressions are run separately in each year and each regression is purely cross sectional, as
one observation is one firm. The dependent variable in all models is ex post R&D/assets in year t. All RHS variables are ex ante observable in year t − 1. In all, the results below are
based on two models. The first block of four columns is the basic investment-Q regression where R&D/assets is regressed on ex-ante Tobin’s Q and basic controls. The second block
of 9 columns is the conditional model, where R&D/assets is regressed on the life variable interactions with Tobin’s Q (here controls for log age and log assets are included but are not
reported to conserve space). t-statistics are in parentheses.
Table IA.17
SDC Acquisitions/assets Investment-Q Regressions

(1) 1998 0.012 0.009 -0.021 0.019 -0.008 0.017 0.038 0.001 0.025
1998 (7.17) (5.66) (-5.53) (-1.71) (3.37) (6.51) (0.09)
(2) 1999 0.017 0.008 -0.013 0.042 0.016 -0.003 0.031 0.048 0.047
1999 (13.0) (6.28) (-3.92) (3.80) (-0.61) (6.16) (3.19)
(3) 2000 0.017 0.007 -0.022 0.102 0.009 0.008 0.034 0.000 0.107
2000 (20.3) (3.94) (-4.85) (2.66) (1.73) (8.32) (0.01)
(4) 2001 0.003 0.004 -0.007 0.017 0.001 0.004 0.005 0.002 0.017
2001 (5.72) (5.97) (-4.01) (0.44) (1.88) (2.12) (0.25)
(5) 2002 0.002 0.002 -0.003 0.010 -0.001 0.001 0.006 0.009 0.011
2002 (4.49) (4.65) (-2.72) (-0.62) (0.90) (3.18) (2.09)
(6) 2003 0.004 0.003 -0.009 0.013 0.002 -0.002 0.013 0.000 0.015
2003 (3.84) (4.76) (-5.01) (0.52) (-0.64) (3.77) (0.02)
(7) 2004 0.004 0.002 -0.016 0.010 -0.002 0.004 0.016 -0.018 0.012
2004 (3.11) (1.53) (-4.78) (-0.52) (1.12) (3.06) (-1.71)
(8) 2005 0.004 0.002 -0.012 0.011 -0.005 0.001 0.021 0.017 0.016
2005 (3.70) (2.53) (-4.33) (-1.51) (0.17) (4.54) (1.82)
(9) 2006 0.005 0.002 -0.017 0.014 -0.006 0.007 0.017 0.015 0.016
2006 (3.59) (1.93) (-5.58) (-1.50) (1.86) (3.08) (1.25)
(10) 2007 -0.000 0.002 -0.019 0.012 -0.018 -0.008 0.033 0.034 0.024
2007 (-0.03) (1.48) (-6.34) (-4.10) (-2.10) (5.62) (2.73)
(11) 2008 0.001 0.001 -0.004 0.002 -0.005 0.001 0.011 -0.000 0.004
90
2008 (1.43) (2.18) (-2.14) (-2.19) (0.56) (3.23) (-0.06)
(12) 2009 0.004 0.002 -0.005 0.015 0.003 -0.000 0.008 0.011 0.015
2009 (4.41) (4.61) (-4.47) (1.35) (-0.05) (2.28) (1.73)
(13) 2010 0.000 0.001 -0.006 0.002 -0.012 -0.002 0.022 0.003 0.007
2010 (0.33) (1.73) (-2.73) (-3.09) (-0.55) (3.98) (0.24)
(14) 2011 0.002 0.003 -0.004 0.004 -0.005 -0.003 0.021 0.003 0.009
2011 (1.81) (3.29) (-1.83) (-1.25) (-0.88) (3.74) (0.29)
(15) 2012 0.003 0.003 -0.006 0.007 -0.012 0.000 0.027 0.003 0.017
2012 (2.60) (3.44) (-3.05) (-3.19) (0.15) (5.33) (0.24)
(16) 2013 0.001 0.002 -0.008 0.005 -0.007 0.001 0.013 -0.004 0.008

2013 (0.70) (2.10) (-3.73) (-1.90) (0.33) (2.40) (-0.31)
(17) 2014 0.001 0.002 -0.017 0.010 -0.011 0.002 0.017 0.013 0.014
2014 (0.60) (1.64) (-5.31) (-2.64) (0.42) (2.61) (0.88)
(18) 2015 0.003 0.004 -0.008 0.004 -0.012 -0.003 0.034 0.011 0.011
2015 (1.89) (2.95) (-2.59) (-2.74) (-0.63) (4.76) (0.66)
(19) 2016 0.001 0.002 -0.002 0.007 -0.003 0.000 0.006 0.003 0.009
2016 (1.21) (4.60) (-2.05) (-1.53) (0.14) (2.21) (0.44)
The table reports results from annual OLS investment-Q regressions from 1998 to 2017. Regressions are run separately in each year and each regression is purely cross sectional, as one
observation is one firm. The dependent variable in all models is ex post acquisition dummy, which is one if the given firm acquired any assets in the SDC Platinum database in the given
year t. All RHS variables are ex ante observable in year t − 1. In all, the results below are based on two distinct Q-models. The first block of four columns is the basic investment-Q
regression where the SDC acquisition dummy is regressed on ex-ante Tobin’s Q and basic controls. The second block of 9 columns is the conditional model, where the SDC acquisition
dummy is regressed on the life variable interactions with Tobin’s Q (here controls for log age and log assets are included but are not reported to conserve space). t-statistics are in
parentheses.
Table IA.18
SDC Asset Sales/assets DisInvestment-Q Regressions

(1) 1998 -0.002 0.001 0.002 0.003 -0.002 -0.005 0.000 0.012 0.008
1998 (-2.48) (1.32) (1.65) (-1.02) (-2.63) (0.20) (2.07)
(2) 1999 -0.001 0.001 0.003 0.001 0.003 -0.005 -0.003 0.016 0.003
1999 (-0.85) (1.34) (1.34) (1.05) (-1.94) (-0.97) (1.71)
(3) 2000 -0.001 0.001 0.001 0.003 0.000 -0.002 -0.001 0.003 0.006
2000 (-2.14) (2.33) (0.64) (0.06) (-1.74) (-0.42) (1.01)
(4) 2001 -0.000 0.000 0.000 0.001 -0.000 -0.001 0.000 0.005 0.007
2001 (-1.68) (1.43) (0.68) (-0.69) (-1.72) (0.17) (2.19)
(5) 2002 -0.000 0.000 0.000 0.010 -0.001 -0.000 -0.000 0.001 0.016
2002 (-3.52) (4.24) (0.99) (-2.19) (-1.08) (-0.31) (0.95)
(6) 2003 -0.001 0.000 0.001 0.005 -0.000 -0.002 -0.001 0.002 0.006
2003 (-3.29) (1.52) (1.59) (-0.01) (-2.34) (-1.42) (0.99)
(7) 2004 -0.001 0.001 0.001 0.013 -0.000 -0.001 -0.001 0.004 0.017
2004 (-2.60) (4.91) (0.98) (-0.26) (-2.11) (-1.41) (2.62)
(8) 2005 -0.001 0.000 0.000 0.007 -0.002 -0.002 -0.000 0.003 0.015
2005 (-3.92) (1.71) (0.12) (-2.05) (-2.12) (-0.07) (1.14)
(9) 2006 -0.003 0.001 0.003 0.011 -0.004 -0.004 -0.002 0.009 0.016
2006 (-4.55) (1.47) (1.83) (-1.88) (-2.39) (-0.77) (1.55)
(10) 2007 -0.001 0.001 0.002 0.005 -0.002 -0.000 -0.004 0.021 0.013
2007 (-2.14) (2.04) (1.42) (-1.29) (-0.23) (-1.64) (3.97)
(11) 2008 -0.002 -0.001 0.000 0.005 -0.001 -0.001 -0.004 0.009 0.015
91
2008 (-4.17) (-1.86) (0.38) (-1.22) (-0.85) (-2.60) (2.49)
(12) 2009 -0.001 -0.000 0.000 0.006 0.000 -0.001 -0.004 0.001 0.011
2009 (-4.58) (-0.15) (0.67) (0.32) (-1.51) (-3.15) (0.37)
(13) 2010 -0.002 -0.000 0.001 0.005 -0.003 0.001 -0.004 -0.000 0.009
2010 (-3.61) (-0.01) (1.04) (-2.35) (1.13) (-1.95) (-0.09)
(14) 2011 -0.001 0.000 0.000 0.006 -0.001 0.000 -0.002 0.002 0.012
2011 (-3.26) (2.13) (0.48) (-1.56) (0.20) (-1.89) (0.67)
(15) 2012 -0.001 0.000 0.001 0.009 -0.002 -0.000 -0.001 -0.000 0.009
2012 (-2.97) (2.19) (2.05) (-1.89) (-0.24) (-0.36) (-0.05)
(16) 2013 -0.001 0.001 0.000 0.007 -0.001 -0.000 -0.005 0.011 0.013

2013 (-3.52) (1.86) (0.60) (-0.87) (-0.10) (-2.63) (2.51)
(17) 2014 -0.001 0.001 0.001 0.008 -0.001 -0.000 -0.002 0.010 0.014
2014 (-2.21) (2.96) (1.66) (-1.03) (-0.27) (-1.25) (2.72)
(18) 2015 -0.001 0.001 0.001 0.010 -0.002 -0.001 -0.001 0.012 0.017
2015 (-2.63) (3.10) (1.89) (-1.58) (-0.70) (-0.77) (3.16)
(19) 2016 -0.001 0.001 0.001 0.005 -0.001 0.000 -0.001 0.008 0.008
2016 (-2.01) (2.51) (1.20) (-1.66) (0.34) (-0.90) (2.40)
The table reports results from annual OLS divestiture-Q regressions from 1998 to 2017. Regressions are run separately in each year and each regression is purely cross sectional, as one
observation is one firm. The dependent variable in all models is ex post divestiture dummy, which is one if the given firm sold any assets in the SDC Platinum database in the given
year t. All RHS variables are ex ante observable in year t − 1. In all, the results below are based on two distinct Q-models. The first block of four columns is a basic divestiture-Q
regression where the SDC divestiture dummy is regressed on ex-ante Tobin’s Q and basic controls. The second block of 9 columns is the conditional model, where the divestiture dummy
is regressed on the life variable interactions with Tobin’s Q (here controls for log age and log assets are included but are not reported to conserve space). t-statistics are in parentheses.
Table IA.19
Life Cycles and Competition (Testing Abernathy and Utterback 1972)
Instrumented Instrumented Instrumented log log #
Row Instruments Life1 Life2 Life4 assets age obs
Panel A: 10-K Competition Complaints

(1) Tech Instruments 0.112 0.072 0.010 0.000 0.004 66,853
(4.88) (4.12) (0.30) (0.42) (3.08)
(2) Distant Life Instruments 0.112 0.059 0.034 0.000 0.003 66,845
(2.93) (1.06) (1.43) (1.51) (0.78)
Panel B: TNIC Total Similarity
(3) Tech Instruments 3.785 2.793 29.485 0.803 -1.015 66,858
(0.62) (0.64) (2.97) (5.48) (-2.59)
(4) Distant Life Instruments 119.628 90.847 21.621 0.506 5.992 66,850
(3.04) (1.53) (0.90) (1.64) (1.43)
Panel C: One Minus TNIC HHI
92
(5) Tech Instruments 4.360 1.485 0.131 0.016 0.089 66,856
(5.88) (2.56) (0.12) (0.96) (1.97)
(6) Distant Life Instruments 1.632 2.219 -1.645 0.010 0.150 66,848
(1.21) (1.05) (-2.03) (0.92) (1.00)
The table reports results from firm-year instrumental variables investment-Q regressions from 1998 to 2017. The dependent variable is ex post 10-K Competition Complaints (Panels
A), TNIC Total Similarity (Panel B), and one minus the TNIC HHI (Panel C). The key RHS variables being instrumented for are the life cycle stages. The odd numbered rows use the
technology instruments from Table 9 interacted with Tobins Q as the instruments. The even numbered rows use the life cycle stages of distant peers interacted with Tobins Q as the
instruments. Distant peers are firms who are in the focal firm’s TNIC-2 industry but are not in the firm’s TNIC-3 industry. All RHS variables and instruments are ex ante measurable

and are observable in year t − 1. All ratio variables are winsorized at the 1/99% level. The last two columns indicate the adjusted R2 and the number of observations. All regressions

SSRN Id3182158

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

SSRN Id3182158

Uploaded by

Copyright:

Available Formats

Product Life Cycles in Corporate Finance

Marshall School of Business, University of Southern California

Smith School of Business, University of Maryland

Electronic copy available at: https://ssrn.com/abstract=3182158

In this paper, we examine hypotheses motivated by a simple q-model of investment that

Overall, we document economically large variation in q-sensitivities across investment

1. Related Literature and Theory

1.1 Related literature

1.2 An illustrative model

1.2.1 Model solution.

At the beginning of Life2 (time t1 ), the firm selects k2 to maximize value

At time t0 , the firm selects k1 before product development uncertainty is resolved.

V = n1 v1∗ + n2 v2∗ + n3 v3∗ + n4 v4∗ (9)

The firm’s total investment is also the sum over projects.

1.3 Model implications

T otal Investment = a + β1 θ1 Q + β2 θ2 Q + β3 θ3 Q + β4 θ4 Q, (10)

This equation can be estimated in aggregate or separately for each component.

1.3.1 Competition and q -sensitivities.

2. Data and Methods

2.2 The product life cycle

General exclusions: capital expenditure* OR research and development

Life1 list A: product OR products OR service OR services

Life2 list A: cost OR costs OR expense OR expenses

Life3 list A: product OR products OR service OR services

Life4 list A: product OR products OR service OR services OR inventory OR inventories

We examine four investment policies: R&D/assets, CAPX/assets, SDC $ acquisitions/assets,

2.5 Summary statistics and correlations

4. NBER Recessions and Life Cycle Dynamics

In our multivariate regressions displayed on the left-hand side of Table 6 , we regress

+α4 Lif e4i,t Qi,t + α8 Log[Assetsi,t ] + α9 Log[Agei,t ] + λt + µi + i,t (12)

Table 7 also documents an important link between competition and q-sensitivities.

5.1 Age or competition alone

5.2 Instrumental variable models

Distant peer “technological change”: The number of paragraphs in a firm’s 10-K

5.3 Economic magnitudes

Ii,t = a + b1 Qi,t−1 + i,t . (14)

Ii,t+1 = a + b1 Comp1i,t Qi,t + ... + b4 Comp4i,t Qi,t+1 + i,t (15)

To test these predictions, we consider return on assets computed as operating income

6.1 Negative Life3 R&D sensitivity

6.2 Financing policies

Philippon, T. 2009. The bond market’s Q. Quarterly Journal of Economics 124:1011–56.

Stigler, G. 1964. A theory of oligopoly. Journal of Political Economy 72:44–61.

A. Life cycle variables

AR(1) coefficient 0.861 0.874 0.812 0.764

Electronic copy available at: https://ssrn.com/abstract=3182158

A. Firm and year fixed effects

(1) life1 –0.007 .80 68,965

B. Firm and year fixed effects plus controls

(6) life1 –0.008 0.004 0.003 0.000 .80 68,899

Electronic copy available at: https://ssrn.com/abstract=3182158

1 All firms Life1 0.782 0.030 0.076 0.070

5 Low competition Life1 0.661 0.024 0.061 0.060

9 High competition Life1 0.848 0.040 0.095 0.103

13 Low q Life1 0.693 0.017 0.051 0.033

17 High q Life1 0.825 0.058 0.105 0.168

21 Young Life1 0.805 0.041 0.085 0.104

25 Old Life1 0.714 0.020 0.062 0.056

29 Small Life1 0.814 0.055 0.100 0.108

33 Large Life1 0.713 0.015 0.062 0.039

J. FF5 consumer industry

37 Consumer Life1 0.660 0.019 0.056 0.036

K. FF5 health industry

41 Health Life1 0.905 0.116 0.093 0.145

+α4 Lif e4i,t Qi,t + α8 Log[Assetsi,t ] + α9 Log[Agei,t ] + λt + µi + i,t (12)

Ii,t = a + b1 Qi,t−1 + i,t . (14)

Ii,t+1 = a + b1 Comp1i,t Qi,t + ... + b4 Comp4i,t Qi,t+1 + i,t (15)

Investment Variablei,t+1 = a + b1 Qi,t + i,t .