Product Life Cycles in Corporate Finance

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 products mature, and divestitures and product extension investments arise as products 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.

Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online.

Numerous studies in economics, strategy, and finance suggest that a company and its products 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 remain 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 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 example, 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 conditional 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 document 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 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 predictions. 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 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 investment 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 consistent 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

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 Utterback’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 maturity, 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 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 measured as the ratio of the firm’s market value to the cost of reproducing the firm’s assets.⁹q-theory model has been widely studied in finance, both in structural models, such as Hennessy, 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 assumptions 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 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 shuttering 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 |$t_0$|⁠, a product opportunity arrives. The first life cycle stage Life1 begins at |$t_0$| and ends at |$t_1$|⁠, by which time the firm knows whether or not the project will be successful. At time |$t_{0}$|⁠, the firm simultaneously (A) commits a fixed sum |$R>0$| to product development, in which case the probability of success is |$\pi>0$| and (B) decides how much capacity |$k_{1}$| to build at cost |$\frac{\gamma}{2} k_{1}^{2}$|⁠.¹⁰ If the project fails, the capacity |$k_{1}$| has zero scrap value. The success of the project is then revealed just after |$t_1$|⁠. The subscript “1” on |$k_1$| indicates that the investment was made in the first-stage Life1.

The firm reports its investment |$k_1$| at |$t_1$|⁠, and the project’s first-stage q (⁠|$q_1$|⁠) is recorded. Just after the end of the first stage at |$t_{1}$|⁠, if the project is successful, the firm enters Life2 and decides how much additional productive capacity |$k_{2}$| to build at cost |$\frac{\gamma}{2} k_{2}^{2}$|⁠.¹¹ The additional capacity becomes available at the end of Life2 at |$t_2$|⁠. The firm then releases its financial statements and |$q_2$| 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 |$t_{2}$|⁠, the firm has exhausted its organic growth opportunities (as indicated by AU) and the following sequence occurs. The firm enters Life3 and can add |$k_3$| more units of additional capacity through inorganic acquisitions at cost |$\frac{\mu\gamma}{2} k_{3}^{2}$|⁠, where |$\mu>0$| 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 |$\phi (k_{1}+k_{2}+k_3)$|⁠, where |$\phi$| is the net price of output. The firm then releases its financial statements and |$q_3$| is reported.¹³

After time |$t_3$|⁠, 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 revitalize |$k_4$| units of existing capacity at a cost |$\frac{\omega \gamma}{2} k_{4}^{2}$|⁠, where |$\omega$| reflects the differential cost of revitalization. At the same time the remaining units are disposed at zero cost.¹⁴ At the end of time |$t_4$|⁠, 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 |$\delta$| across the dates.

1.2.1 Model solution

At time |$t_{0}$|⁠, the firm selects |$k_{1}$| before product development uncertainty is resolved.

Solving, we obtain |$k_{1}^{\ast} =\frac{\delta ^{3}\phi^{\ast} \pi} {\gamma}$|⁠. As expected, |$k_{1}^{\ast}$| is higher when the probability of success is higher (i.e., when |$\pi$| is larger).

Next, we obtain the |$q$| values for each stage. |$q_4$| can be calculated directly from Equation (2) and equals |$\frac{v_4^*}{k_4^*} = \phi$|⁠. Given the constant returns to scale assumption, the project’s q observed by the market at time |$t_3$| (just before the firm cash flows are distributed from production in Life3) is |$q_3=\frac{\phi^{\ast} (k_{1}^*+k_{2}^*+k_3^*)}{ (k_{1}^*+k_{2}^*+k_3^*)}=\phi^{\ast}$|⁠. This valuation incorporates the continuation value at time |$t_3$|⁠, |$\phi^{\ast}$| rather than |$\phi$|⁠. To relate |$q_4$| to |$q_3$|⁠, the Internet Appendix derives the relation |$\zeta$| (as a function of primitives) between |$\phi$| and |$\phi^{\ast}$|⁠, where |$\phi^{\ast}=\zeta\phi$|⁠, so that |$q_4=\frac{1}{\zeta} q_3$|⁠.

In calculating |$q_1$| the market takes into account the expected value of the project at time |$t_2$| divided by the installed capital base |$k_1$|⁠. To calculate |$q_1$|⁠, the market divides the value of the project |$v_1^{\ast}$| given by Equation (7) by the number of installed units at time |$t_1$|⁠, which is given by |$k_{1}^{\ast}$|⁠. Thus, |$q_1=(\delta^2+\frac{\delta}{2\pi}+\frac{1}{2 \mu \pi})\phi^{\ast}$|⁠. Note that we use “|$A_1$|” henceforth to refer to the quantity |$(\delta^2+\frac{\delta}{2\pi}+\frac{1}{2 \mu \pi})$| for convenience. When the project is very risky (i.e., |$\pi$| is low), upon successful completion, |$q_1$| will be high.¹⁵ Similarly, to obtain |$q_2$|⁠, we divide |$v_2^\ast$| by |$k_1^*+k_2^*$| to obtain |$q_2=\big(\delta+ \frac{1}{2\mu(\delta\pi+1)}\big) \phi^{\ast}=A_2 \phi^{\ast}$|⁠.

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 |$n_i$| projects at each stage. Then the value of the firm, |$V$|⁠, is the sum of the values of each of the firm’s projects.

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 |$\phi$| 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 |$q_i$|⁠, |$i=1,..,4$| (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.

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 competitive 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 |$\gamma$| has two parts (one due to operational adjustment costs and another due to quiet life agency costs): |$\gamma=\gamma_{oper} + \gamma_{quiet}$|⁠. Following Hart (1983) and the broader literature cited above, we assume that agency conflicts are more severe in less competitive industries.

These expressions indicate that q-sensitivities are strictly decreasing as quiet life agency conflicts |$\gamma_{quiet}$| 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 |$\phi$| in competitive relative to concentrated markets. As the equation above illustrates, however, our model predicts that q-sensitivities are not sensitive to |$\phi$|⁠. This is intuitive as a lower |$\phi$| would result in both a lower q and a lower level of investment, and, hence, sensitivity is invariant to |$\phi$|⁠. 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 context 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 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 |$\{{\it Life}1, {\it Life}2, {\it Life}3, {\it Life}4\}$|⁠, 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 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 expenditures 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 transportation OR shipping OR freight OR materials OR overhead OR administrative OR manufacturing 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 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 introduction OR introductions OR new OR introducing OR innovation OR innovations OR expansion 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 |$\{{\it Life}1, {\it Life}2, {\it Life}3, {\it Life}4\}$| 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 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 provide 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 expenses 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 transition 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.

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 observations. 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 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 innovation 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 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 |$({\it Life}1 + {\it Life}2)$|⁠. 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.

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 prediction 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 consistent 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 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 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 portfolios 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 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.

5. Investment and the Product Life Cycle

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 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 significant 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 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: competition 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 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 investment. 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 effects, 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 measured 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 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 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 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 instruments, 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 particularly 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 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 |$R^2$| 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 |$R^2$| for both policies but also its improvement relative to the basic model is also increasing during our sample.

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.

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 |$operating \ profits= a +b_{IO-q} \times q$|⁠. 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 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.

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 fluidity 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 spending, consistent with a reduced fluidity, a strengthening of barriers to entry, and a reduced need to invest in R&D to escape competition. In our context, their findings are also consistent 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 |$\times$|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 substantive, 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 parentheses):³² equity issuance/assets (SSTK/AT), SEO issuance/assets (SDC Platinum SEO dollars 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 competitive 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 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 interactions 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 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.

Acknowledgement

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 analytics 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 Taylor, 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 Commission, Swiss Finance Institute (Lausanne), Swiss Finance Institute (Lugano), Temple University, Toulouse Business School, University of Maryland, University of Southern California, University of Virginia, Vanderbilt 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 excellent comments. All authors contributed equally to this paper. Supplementary data can be found on The Review of Financial Studies web site.

Footnotes

References