r&d spending: dynamic or persistent?public.kenan-flagler.unc.edu/2017msom/sigs/iform...

This is a preliminary version of the manuscript for the discussant of the MSOM SIG.Please do not distribute.

R&D Spending: Dynamic or Persistent?Christophe Pennetier

INSEAD Singapore, [email protected]

Jürgen Mihm,Karan GirotraINSEAD Fontainebleau, [email protected], [email protected]

It is well established that the benefits of R&D spending accrue over several years and that more cumulative R&D spending

benefits the firm more. Here we study how a given amount of R&D spending is best allocated over time to optimize

R&D performance. Under a persistent policy, allocations remain nearly constant irrespective of circumstances; under a

dynamic policy, R&D spending increases (resp., declines) when opportunities arise (resp., fail to materialize). We use

a sample of 3,746 publicly listed companies, observed for an average of seven years between 1982 and 2003, to assess

which of these two R&D allocation policies is preferable. We find that a dynamic allocation strategy is associated with

worse R&D performance—in terms of patent quantity and quality. We identify the originality of an invention and the

firm’s familiarity with the technological base as important factors emphasizing or mitigating the harm that variability

causes. Finally, we find that it is particularly the unpredictable part of any dynamic spending pattern that hurts R&D

performance; the predictable part has no or even positive effects.

Key words: R&D strategy; innovation; variability; originality and familiarity; Arellano-Bond; business analytics

1. Introduction

Should R&D management be persistent in its approach to R&D decisions or rather allow for quick reac-

tion and dynamism? This question is actually disputed by practitioners. To exemplify the perplexity that

abounds on this score, consider two—seemingly contradictory—principles established by the practitioners

Roussel et al. (1991) in their influential book on R&D management. After writing that, “because [...] R&D

management plans are integral to corporate business plans, flexibility and adaptability must characterize

the R&D component of corporate plan” (p. 175), they contradict this principle in claiming that “[j]erking

R&D around in shortsighted response to short-term conditions is particularly destructive. There must be

steadiness of course” (p. 175)—on the same page.

Intuition tells us that both persistence and dynamism can result in good R&D performance. In essence,

they define two general management styles: a relatively stable style that is primarily concerned with long-

term capability building; and a more involved style that is mainly concerned with short-term positioning.

Persistence provides continuity and stability whereas dynamism facilitates rapid adaptation to a changing

landscape of opportunities.

These management styles translate into two competing approaches to allocating R&D resources. On the

1

2 R&D Spending: Dynamic or Persistent?

one hand, a faith in persistence presumes that R&D investments yield their benefits only over a long time

horizon. Any investment induces lagged results, since building expertise requires time as, for example,

researchers must be hired and need to become productive both in their research field and in their organiza-

tional contexts. Yet withdrawing R&D investment may lead to an instantaneous loss of knowledge as, for

example, researchers and their knowledge are immediately lost. It follows that alternating between invest-

ment and withdrawal invariably reduces productivity. The management of R&D must therefore create and

preserve a research environment whose continuity is sustained by persistent investment; variability in R&D

spending should be avoided.

In contrast, the dynamic style assumes that—in an inherently uncertain environment—opportunities for

research are continuously opening and closing. Entire research fields become exciting prospects or lose

their appeal as new discoveries are made. Projects must be slowed down or sped up when technical insights

trigger reorientation. Managers who are of this mind will adjust their R&D allocation in response to the

shifting opportunities encountered by the firm. Dynamism, and hence variability in R&D spend, is a form

of rebalancing the company’s strategy so as to mitigate or exploit the effects of uncertainty.

Achieving a high level of R&D productivity is decisive for many companies. Hence, the academic lit-

erature has established a wealth of factors that impact R&D productivity, i.e. it has discovered a wealth of

factors that influence how R&D spend impacts R&D output. Amongst those are strategic factors such as the

presence of economies of scope (Cockburn and Henderson 2001), the level of a company’s diversification

(Quintana-Garcia and Benavides-Velasco 2008), organizational factors such as the level of R&D central-

ization (Arora et al. 2014) or the use of incentives (Lerner and Wulf 2007) and financial factors such as the

company’s ownership structure (Lerner et al. 2013). However, the question of how the temporal pattern of

R&D spend influences R&D outcomes has received very little attention.

The first aim of our study is to identify which of these two temporal R&D allocation policies (hold-

ing the overall level of R&D spend constant) is preferable with respect to R&D productivity. We structure

our reasoning around a single construct which unifies both dynamism and persistence: variability in R&D

spending. That is, dynamism implies the presence of variability and persistence the lack thereof. Hence

our question asks, whether variability in R&D spend or the lack thereof is conducive for R&D produc-

tivity. Second, we identify specific contexts in which variability is notably more conducive or harmful to

R&D productivity. In particular, we focus on the effects of variability on particularly original (i.e., novel)

inventions inside each firm’s portfolio. We also explore whether the effects of R&D spend variability are

influenced by the technological distance that an invention has with respect to a company’s preexisting port-

folio of activities. Third, we identify managerial actions that can mitigate potentially negative effects of

variability in R&D spend. We seek to understand whether splitting variability into its predictable and un-

predictable components may therefore aid in finding an approach to taking advantage of both dynamism

R&D Spending: Dynamic or Persistent? 3

and persistence without incurring potential cost. The setting for our empirical analysis is publicly traded

US firms for the period 1976–2006. We estimate a firm-level dynamic panel data model via a system GMM

approach (Arellano and Bond 1991, Blundell and Bond 1998) that combines financial information from the

Compustat database with patent outcome details provided by the National Bureau of Economic Research

(NBER).

We find that the R&D outcomes of firms implementing a dynamic allocation strategy are worse in terms

of both quantity and quality of their patents. By interacting variability with originality (i.e., novelty), we

find that the more a company raises its originality level, the more a dynamic allocation hurts performance.

Decomposing the set of original inventions in technologically distant and close subsets, we find that only

the technologically close innovation amplifies the harmful effect of variability on original patents; the tech-

nologically distant innovation component actually enhances the positive effect of a dynamic allocation

strategy. Finally, decomposing variability in its predicted and unpredicted parts, we find that only the un-

predicted part of the variability harms performance; the predicted component actually may have a positive

effect. Our findings are exceptionally robust to alternative variable constructions, model specifications, and

sample choices.

Our paper should not be read as a condemnation of dynamism and adaptation. Strategic reasons may

very well justify a dynamic R&D spend policy, but managers should be aware of its consequences for R&D

productivity; they should know when it is particularly harmful; and they should be aware of mitigating

actions to alleviate any adaptation cost. In sum, beyond establishing a new factor in the theory of R&D

productivity we intend to give managers a framework to conceptualize principles for investment in R&D.

2. Literature Review

Schumpeter’s conjectures on the “relation between innovation, market structure and firm size” (Cohen 2010,

p.131) have spawned many avenues of research; in particular, they have initiated a prolonged and productive

discussion of the determinants of R&D productivity. True to Schumpeter’s original hypothesis the resulting

literature initially has focused on the role of firm size and has established two main insights. First, an in-

crease in absolute R&D spend leads to an increase in absolute R&D output—whether R&D spend is defined

narrowly (Lööf and Heshmati 2006) or takes into account the cost associated with other innovation activi-

ties such as continuous improvement efforts (Hall et al. 2009); whether R&D output is measured in patents

(Crépon et al. 1998, Lööf and Heshmati 2002) or innovation sales (Jefferson et al. 2006); and whether the

setting is manufacturing (Lööf and Heshmati 2002) or a service (Lööf and Heshmati 2006). Second and

more interestingly, this literature has—despite some skeptical voices (Cockburn and Henderson 2001, Hen-

derson and Cockburn 1996)—established that firm R&D productivity falls in firm size (Cohen and Klepper

1996b, Bound et al. 1984, Geroski et al. 1995, Lerner 2006) because larger firms loose managerial control


(Scherer and Ross 1990), focus on more incremental innovation (Prusa and Schmitz 1991, Henderson 1993)

or focus on process rather than product R&D (Cohen and Klepper 1996a).

Eventually, interest has moved beyond the question of firm size and has concentrated on how a multi-

tude of other factors impact the relation between R&D spend and R&D output (i.e., the question of R&D

productivity). Strategic questions feature prominently. Researchers have identified economies of scope as a

factor: The number of research programs in the pharmaceuticals industry (Henderson and Cockburn 1996,

Cockburn and Henderson 2001) as well as the level of diversification of the technology base of US biotech

firms (Quintana-Garcia and Benavides-Velasco 2008) positively impact innovative competence. The strate-

gic intent of the research projects a firm engages in could influence R&D productivity. Ahuja and Lampert

(2001), for example, find a curvilinear relation between the amount of exploration a firm engages in and

the number of breakthrough inventions and Rosenkopf and Nerkar (2001) demonstrate that firms engaged

in broader search efforts also have a broader subsequent impact. The question of how a company’s network

position influences R&D outcomes has been studied (Ahuja 2000), as has the value of different kinds of

links with competitors, customers and suppliers (Lööf and Heshmati 2002, Revilla and Fernández 2012) or

with the science establishment (Lööf and Heshmati 2002, Revilla and Fernández 2012, Cockburn and Hen-

derson 1998). Finally, researchers have begun to address the “make or buy” question in R&D (Cassiman

and Veugelers 2006).

Different organizational factors have been shown to impact R&D productivity. The structure of the R&D

organization affects the link between R&D spend and R&D output in that firms with centralized R&D or-

ganizations patent more per R&D dollar than decentralized firms (Arora et al. 2014) and generate more

frequently and broadly cited patents (Argyres and Silverman 2004)—potentially because centralized R&D

efforts tend be more scientific and broader in scope (Arora et al. 2011) and focus less on imitative innova-

tion (Leiponen and Helfat 2011). If decentralized, firms with a wider reach of their international company

networks have higher innovative sales (Frenz and Ietto-Gillies 2009). The firm’s ability to integrate knowl-

edge across disciplines (Henderson and Cockburn 1994, 1996), different knowledge management practices

including conference attendance (Lööf and Heshmati 2006, Kremp and Mairesse 2004) and the source of

knowledge (Katila 2002) have been associated with R&D productivity. Incentives are beneficial to inno-

vativeness particularly when they are long-term (Lerner and Wulf 2007, Chang et al. 2015, Francis et al.

2011). Even individual characteristics of decision makers such as overconfidence have been (positively)

linked with R&D productivity (Galasso and Simcoe 2011).

Other researchers have been interested in the effects of financial consideration on R&D productivity.

Institutional ownership (Aghion et al. 2013) and Private Equity backing (Lerner et al. 2013) support innova-

tiveness as does geographical dispersion of ownership (Jefferson et al. 2006). Antitakeover provisions make

firms more innovative (Chemmanur and Tian 2016). And, interestingly, Almeida et al. (2013)—in a sample


of US publicly traded firms—find that financial constraints in general may benefit innovation by improving

the efficiency of innovative activities.

Beyond establishing the link between R&D spend and R&D output, there is an extensive body of lit-

erature (for a summary see Hall et al. 2009) that has discussed how the level of R&D spending relates to

overall firm outcomes. Using total factor productivity (TFP) models, researchers have confirmed a positive

relation between R&D investment and productivity (gross value added) (Griffith et al. 2006, Rogers 2010,

Lichtenberg and Siegel 1991). They have also substantiated the link between R&D spend and market value

(Ciftci et al. 2011, Hall and Oriani 2006) and to a lesser extent accounting profit (Ding et al. 2007, Lev

and Sougiannis 1996). And as was true for the relation between R&D spend and R&D productivity, the

literature has identified many factors that moderate the relationship between R&D spend and overall firm

outcomes, such as IT spend (Arkali et al. 2008), commercialization orientation (Lin et al. 2006), or man-

agers’ firm-specific experience (Kor and Mahoney 2005) as well as tactical R&D decisions such as product

line freshness (Terwiesch et al. 1998), early supplier involvement and lead user involvement (Langerak and

Hultink 2005) or detailed project level activities (Kleinschmidt et al. 2007).

However, despite the wealth of factors it has considered, the entire literature has largely neglected one

important class: the temporal pattern of R&D spend (i.e., how a given amount of R&D spend is distributed

over time). The limited literature there is has focused on very specific sub-patterns. In specific Mudambi

and Swift (2014) has looked at how extraordinary compact significant increases in R&D spend the number

and quality of patents published and they find a small positive relation. Kor and Mahoney (2005), in con-

trast, have hypothesized that spending increases in general should lead to increased productivity. Using the

average percentage increase in R&D as their predictor variable, they test their hypothesis in a sample of 218

firms. They fail to establish a significant relation. (Interestingly, their coefficient is negative). In this paper

we look at how the overarching concept of spending variability impacts R&D productivity, controlling for

the overall level of R&D spend. We investigate whether variability in R&D spend affects different important

categories of innovation in differential ways. We identify how potentially negative effects of R&D spend

variability can be mitigated.

3. Theory and Hypotheses

Before building a theory that accounts for the benefits of persistence and dynamism, we must clarify the

most important terms. So: What is variability in R&D spending, and what is R&D performance?

3.1. R&D Spending Variability and R&D Performance

As mentioned previously, we structure our reasoning around a theme that unifies both dynamism and per-

sistence: temporal variability in R&D spending. We conceptualize this variability as the period-to-period

changes in the level of resources allocated to R&D (see Section 4.1 for the mathematical definition). A


firm that consistently spends the same amount on R&D exhibits no variability in such spending; in con-

trast, a firm that either increases or decreases allocations to R&D between periods exhibits R&D spending

variability. Thus dynamism implies the presence of variability and persistence implies the lack of it.

Conceptualizing R&D performance is more complicated because the output of R&D is multifaceted. Re-

search and development explores and develops new opportunities; it provides not only technical knowledge,

production-ready prototypes, and novel production processes but also technical support to manufacturing

functions. It would be impossible to devise a straightforward notion that captures all these aspects of R&D

performance. The existing literature has focused on the two most salient aspects, of which the first is the

sheer quantity of inventions (e.g., Hall et al. 2001, Ahuja 2000, Arora et al. 2011, Cockburn and Henderson

1998, Lerner and Wulf 2007). According to this view of R&D performance, innovation results from the

incremental refinement and improvement of existing ideas (Freeman 1982, Moch and Morse 1977, Norman

and Verganti 2014). Each improvement contributes a small amount to overall progress, and their sum ulti-

mately leads to a significant economic benefit (Hacklin et al. 2004, Hollander 1965). The second frequently

studied aspect of R&D performance is the quality of inventions, which is key because focusing only on

quantity fails to acknowledge the variance in inventions’ importance (Hall et al. 2005, Aghion et al. 2013,

Arora et al. 2011, Lerner et al. 2013, Ahuja and Lampert 2001). In this view, some inventions are dispro-

portionately more beneficial to the firm than are others and so quality considerations must be incorporated

into any discussion of R&D performance. Together, the quantity and quality of inventions form the basis

for discussing our research question.

3.2. Effects of R&D Spending Variability on Quantity

3.2.1. The Negative Argument Variability in the amount of resources allocated to R&D suggests that

the R&D organization must regularly adapt to new budgets. These adaptations entail a cost to the company

in the form of productivity losses. There are many reasons why changes in R&D spending result in poor

performance. The two most prominent ones are discussed next.

Gestation Period of R&D Investment (Time): Frequently starting and then stopping work, especially R&D

work, is inefficient. An R&D organization that starts new activities or raises its level of activity does not im-

mediately attain its optimal steady-state level of knowledge creation. The organization faces an adaptation

period as, for example, new equipment needs to be sourced and brought up, new staff be hired, new outside

partners be found or new work routines and work culture be established. With, on average, more than two

thirds of R&D spend being related to personnel (Cameron 1996), an increase in R&D expenditures often

means that new researchers need to be hired, existing researchers to be reoriented and new organizational

structures be built. This has consequences on an individual and on an organizational level. On an individual

level, when researchers are new to a technical domain, they must immerse themselves in it before becoming

fully productive (Baard et al. 2014, Jones 1986, Smith et al. 2013). Yet they must master not only a highly


specialized technical field but also a wide spectrum of social aspects. Thus newcomers cannot be imme-

diately productive because they must first undergo a socialization phase (Bauer et al. 2007, Van Maanen

and Schein 1979) in which they adapt to new working routines, new corporate cultures, and new teams.

Newcomers must also build informal networks that help them navigate the organization and gain access to

its knowledge; they need to understand how the firm’s chain of command works in practice and how their

role fits in the company (Kim et al. 2005, Klein and Weaver 2000). On an organizational level, entering new

R&D fields requires new organizational routines and often new formal and informal structures. But particu-

larly routines and informal structures require trial and error until they evolve into their most beneficial state.

The R&D personnel hence reaches its steady state level of knowledge creation only with some delay. Anal-

ogous arguments for tangible R&D equipment allows for the conclusion that building R&D competencies

and realizing returns from them requires a gestation period.

In contrast, a reduction in R&D spending interrupts the flow of new knowledge immediately. Researchers

and research managers leave the company or are reoriented and hence are lost for advancing the knowl-

edge agenda in their original field. So is equipment which is reappropriated, sold or decommissioned. This

asymmetry implies that alternating increases and decreases in R&D spending will reduce overall R&D pro-

ductivity: A firm that continuously alternates between “up” and “down” spending will always be needing

to bridge an unproductive phase owing to the asymmetric lags of the resulting effects; hence it will suf-

fer as compared with a firm that spends the same cumulative amount on R&D but employs a persistently

even-keeled investment policy.

Dis-economies of Scale: The second primary reason for the negative effect of variable resource allocation

is that the relation between R&D inputs and R&D outputs is concave. There is ample evidence that, beyond

a threshold of relevance, adding additional resources to a knowledge creation effort scales the output not

proportionally but rather subproportionally. At the project level, it is difficult to partition a given activity into

multiple discrete tasks with no operational and communication interactions. As a result, adding resources

to projects creates additional coordination overhead, from which it follows that increases in output taper off

or may even become negative; thus, productivity declines. This dynamic has been demonstrated repeatedly

in the context of R&D (Edmondson and Nembhard 2009, Ostrom et al. 1999).

Note also that, at the portfolio level, rational R&D managers naturally allocate their limited available

resources to the most promising and profitable R&D projects (Chao and Kavadias 2013, Loch and Kavadias

2002). However, for a relatively larger R&D budget such priorities entail a lower average return per unit

of invested resources. Hence increases in R&D spending can lead to lower R&D performance. Recall that

a concave translation function implies that the functional value of the average output is higher than the

average of any two individual function values. So even if we abstract from the aforementioned issue of

gestation times, it follows from concavity that consistently providing the average level of R&D financial

support is preferable to investing the same overall amount unevenly.


3.2.2. The Positive Argument The argument in favor of variability in R&D spending focuses on un-

certainty as the essence of innovation (Nelson and Winter 1977). Technical uncertainty is constantly being

resolved, either within the company or by outside entities; this progress renders past inventions obsolete or,

alternatively, increases their importance. Such developments may require management to adapt at both the

portfolio level and the project level and hence might end up requiring adjustments in allocations to R&D.

At the portfolio level, innovation is a quest for new territories. In many industries, and especially in high-

technology industries characterized by rapid technological change, survival depends on early entrance into

a new field (Kessler et al. 2007) because technological obsolescence is high (Bernstein and Singh 2008).

Early entrants enjoy more freedom in their R&D pursuits, as the field has yet to be extensively explored by

competitors. The learning curve of such entrants is steep, and they are quick to extend the frontier of current

knowledge. Their inventions follow suit. A firm that seeks rapid entrance into new technology fields—or

that wants to abandon fields that have become unattractive—may need to adjust its R&D budgets, often

quickly. Thus achieving the goal of speedy flexibility may require shifts in R&D allocations.

At the level of individual projects, discoveries made in the course of a project may cast doubt on a

project’s technical feasibility, resource requirements, and (ultimately) the firm’s current course of action.

In particular, truly innovative projects often require a quick readjustment to technical learning (Pich et al.

2002, Sommer et al. 2009). It is only by remaining flexible—adding resources when needed and releasing

them when not—that the discovery process is effective and efficient. Yet because flexibility in content

usually requires flexibility in spending patterns, the company is led to employ a dynamic approach to R&D

allocations.

3.2.3. Trading Off the Negative and Positive Arguments Since R&D performance is the combination

of these two opposite effects, the direction of the total effect is ambiguous, and we consequently provide

two competing hypotheses.

HYPOTHESIS 1a. Variability in R&D spending leads to reduced R&D performance in terms ofquantity.

HYPOTHESIS 1b. Variability in R&D spending leads to increased R&D performance in terms ofquantity.

3.3. Effects of R&D Spending Variability on Quality

Most of the arguments, both negative and positive, about the effect of R&D spending variability on inno-

vation quantity apply also as regards innovation quality—though by a slightly different line of reasoning.

A negative argument we discussed previously is that innovation, whether radical or incremental, requires

time to mature. At the level of the individual, a threshold extent of immersion in the topic is required before

a researcher can develop ideas and innovation of high quality. Chan et al. (2014) find that the quality of


ideas exhibited by those who have related experience in a field is higher than that exhibited by those who

lack such experience (Mecca and Mumford 2014, Rietzschel et al. 2007, Sawyer 2011, Weisberg 2011).

Research on talented inventors has revealed that their most creative and productive inventions are typically

made only after an extensive period of immersion in their field (Jain 2013, Simonton 1997). These tropes

are consistent with findings at the organizational level. For instance, Pruett and Thomas (2008) identify fac-

tors that shape the slope of a learning curve; they find that a company’s past internal innovation experience

fosters both learning and long-term success.

In short, innovation quality requires a gestation period—and perhaps even more so than does innovation

quantity. Hence we again argue for an asymmetric lag in the effects of R&D spending variability. It takes

time for R&D investments to become quality inventions, but the destruction of quality is immediate when

R&D resource allocations are reduced.1

Yet the opposite can also be argued. At the portfolio level, three arguments favor quick adaptations to fos-

ter high-quality inventions. First, a company that enters a nascent field can freely choose its own technology

position (Fleming and Sorenson 2003). A first mover is often able to make “foundational” inventions that

rule (or at least strongly affect) the field for many years to come, often by way of setting basic definitions

and standards. For example, Texas Instruments was the first company to work on commercial applications

of the silicon planar transistor; it thus secured the foundational patent that all other semiconductor manu-

facturers had to license if they wanted to manufacture integrated circuits (Weber 1990). Second, entering

a field early has implications for second- and third-generation products because first-generation experience

has a positive effect on the quality of subsequent inventions (Bilgram et al. 2008, Von Hippel 2009). Third,

occupying a knowledge frontier increases the firm’s attractiveness to new talent. By thus being the pre-

ferred destination of highly qualified staff, the company’s average quality of innovation increases. All three

arguments favor quick entries into new fields as they emerge and hence call for dynamic R&D spending.

At the level of individual projects, agility and fast reactions directly affect the quality of inventions. This

relationship has been demonstrated by Kessler and Bierly (2002), who link development cycle time and

product quality, and by Griffin (1997), who links development cycle time and “newness” (which can be

viewed as a dimension of quality). Again, flexibility in actions usually requires flexibility in means.

Trading off the arguments against and in favor of variability in R&D funding levels, we cannot arrive at

an obvious conclusion regarding innovation quality. We are in the same situation as the one for innovation

quantity. The positive aspects of variability seem to be strong in the case of quality; yet the negative aspects

remain valid. Again, we provide two competing hypotheses.

1 The negative argument based on dis-economies of scale, which we advanced in the context of quantity, applies only indirectly inthe context of quality.


HYPOTHESIS 1c. Variability in R&D spending leads to reduced R&D performance in terms ofquality.

HYPOTHESIS 1d. Variability in R&D spending leads to increased R&D performance in terms ofquality.

3.4. Technology Originality and R&D Spending Variability

Our first hypothesis studies how a given amount of R&D spending is best allocated over time so as to

optimize R&D performance. The argument in favor of persistence maintains first, that adjustments in R&D

spending entail a cost since increases in R&D spending require a gestation period while reductions in

spending take immediate effect and, second, that dis-economies of scale lead to suboptimal additions of

new resources. The argument in favor of dynamism stresses the importance of budget flexibility in the face

of the inherent uncertainty of R&D projects both on the project and the portfolio levels.

But the optimal allocation of R&D resources over time and hence the balance between the two arguments

most certainly will depend on the type of innovation that a focal company engages in. Two dimensions

along which we can classify innovation are of particular interest: the market-wide technological originality

of the invention and each firm’s technology familiarity with the field of the invention (or its search distance).

We take technological originality to characterize the degree to which an invention is new to the world. We

qualify an invention as pioneering if it is based on a new technology that does not build on existing work

inside or outside the company (Ahuja and Lampert 2001, Quintana-Garcia and Benavides-Velasco 2008).

As an example, the modules that enable cars to drive autonomously constitute a pioneering invention while

and improved gearbox is more of an existing invention. Pioneering technology often has the potential to

redefine markets and hence may be of exceptional economic value.

We take technology familiarity to designate the degree to which the company is familiar with the general

technical field of the invention (Freeman and Soete 1997). Some of the literature has characterized this

as search distance (March 1991, Stuart and Podolny 1996, Quintana-Garcia and Benavides-Velasco 2008,

Ahuja and Lampert 2001, Rosenkopf and Nerkar 2001). If a company has mastered the general field, a

technology is close whereas a technology is distant if the field is new to the company. As an example, for

Toyota or Google the autonomous vehicle is a close invention whereas for Pfizer it would be far.

Focusing first on technology originality, we ask how the nature of an invention (pioneering or existing)

influences the balance between the positive and negative arguments both as regards the quantity and the

quality of pioneering inventions.

3.4.1. The Negative Argument When a company engages in pioneering innovation, it needs extensive

time and effort to build its R&D operations. To design new inventions in a pioneering field, a company

will need to acquire new physical and intellectual assets (including its supplier network), which may not


be readily available on the market. It will have to adapt or even create entirely new processes (including

the development process itself), since the old processes may not be suited to the new technology or since

simply there were no old processes. It will have to adapt or even create entirely new mental models and new

routines as such diverse organizational elements as leadership behaviors and invention mindsets need to be

tailored to the new invention. And most importantly, the firm does not understand how exactly it will need

to adapt. It will have to experiment, fail and learn in order to find its path.

Yet, as is the case for all inventions, knowledge decays immediately once funding is cut; however the loss

of knowledge is more definite. When the system of renewed assets, processes, mental models and routines

has not reached a stable new state, interrupting its evolution by cutting funding more thoroughly negates

any achievements. The knowledge gained is less sticky.

As a consequence we expect the gestation argument to become even more pronounced for pioneering

inventions. That is, we expect both the quantity of ideas as well as the quality of ideas to suffer dispropor-

tionately for inventions with a high degree of technological originality.

3.4.2. The Positive Argument However, more technological originality also means more technological

uncertainty. Companies need to deal with this heightened level of uncertainty. At the individual project level,

unexpected events need to be detected, the group needs to build an understanding of the root causes for them

and then different reactions need to be tried and validated. During any phase of the development process

the team may need to react to its environment quickly acquiring or potentially shedding resources. At the

portfolio level more uncertainty makes the outcome of each project less predictable. Projects that proved to

be futile need to be quickly ended; new project alternatives that emerge during the development effort need

to be initiated. Companies need to re-balance their portfolio. In sum, increasing levels of uncertainty require

increasing levels of managerial flexibility which in turn entails an increasing need for budgetary flexibility.

Consequently, for pioneering inventions, the flexibility argument becomes even more relevant. That is we

expect the quality as well as the quantity of ideas to gain from heightened levels of flexibility.

3.4.3. Trading Off the Negative and Positive Arguments So as pioneering technologies are con-

cerned, the main arguments both for and against variability in R&D spending gain in prominence. We,

therefore, cannot determine how the level of technological originality influences the impact of R&D vari-

ability, and we consequently provide two competing hypotheses.

HYPOTHESIS 2a. The interaction between pioneering innovation and variability in R&D spendingleads to reduced R&D performance in terms of both quantity and quality.

HYPOTHESIS 2b. The interaction between pioneering innovation and variability in R&D spendingleads to increased R&D performance in terms of both quantity and quality.


3.5. Variability in R&D Spending, Technology Originality and Technological Familiarity

There is an important distinction to be made when analyzing pioneering innovation. New-to-the-world

(pioneering) technology can be close to a company’s existing technological base or it can be distant from

anything the company has developed before? How does technological familiarity impact the effects of R&D

variability on pioneering performance?

When the pioneering innovation is based on a distant technology base, the company may need to develop

new assets, both physical and intellectual, since existing ones can simply not be reused. Neither can firms

make resort to existing processes, mindsets and routines. They need to develop new ones. In contrast, when

pioneering innovation uses close technologies, the firm may be able to reuse existing assets, processes,

mindsets and routines—certainly not in their current from, but after they have been appropriately adapted.

The consequences of having or not having access to pre-existing assets, processes, mindsets or routines

are ambiguous. An argument can be made that when increasing R&D spend on pioneering inventions the

gestation period is shortened whenever a company can reuse existing assets, processes, mindsets and rou-

tines, since it does not have to develop them from scratch. The productivity loss caused by the gestation

period has been incurred during previous projects. However, the alternative argument that reusing those

aspects of an existing R&D organization will lengthen the gestation period cannot be refuted either: It takes

companies disproportionate amounts of time to realize which well-honed assets processes and mindsets are

not well adapted to the new situation and it will take even more time to redefine them.

But which of the alternative arguments is true. Or more fundamentally: Does variability affect pioneering

inventions differently whether these pioneering inventions are close to the company’s technological base or

when they are far from it? Since we cannot determine the direction of the effect, we consequently provide

two competing hypotheses.

HYPOTHESIS 3a. The interaction between pioneering distant (resp. close) innovation and variabil-ity in R&D spending leads to reduced (resp. increased) R&D performance in terms of both quantityand quality.

HYPOTHESIS 3b. The interaction between pioneering distant (resp. close) innovation and variabil-ity in R&D spending leads to increased (resp. reduced) R&D performance in terms of both quantityand quality.

3.6. Predicted and Unpredicted Variability

If the negative arguments hold—although intellectually and academically appealing—our hypotheses send

a negative message to practitioners: Changes in R&D spending may reduce performance in quantity and

quality. Thus it would seem that adaptation comes at the cost of a decline in R&D performance. Does

this mean that, if our negative arguments hold, managers—or at least those who are focused on R&D


effectiveness and efficiency—should strive to maintain R&D spending at about the same level?2 Taking

another angle, we focus on a method to mitigate the negative effects and foreground the positive effects of

changes in R&D spending. Hence we build a more subtle argument that turns not on the effects of variability

per se but rather on uncertainty.

At the root of our argument is devising the categories of predicted variability versus unpredicted vari-

ability—that is, changes that are (respectively) foreseeable or entirely unexpected. For the case of predicted

variability, management can prepare the organization for impending measures. Thus strategic and forward-

looking managers adapt their behavior to accommodate expectations about the future (Case 2012, March

1994). In particular, R&D managers anticipate expected future spending levels, including any foreseeable

decrease or increase in the resources allocated to R&D. Managers can then, for example, plan hiring in

advance and in that way shorten the time needed to find qualified candidates; they can also limit the damage

of any staff cuts by finding alternative scenarios for employing the most talented among them. Forward-

looking managers can mentally prepare the organization for possible outcomes. They moderate the negative

effect of dis-economies of scale by promoting choices that are the least damaging to individual projects

and to the R&D portfolio overall. In other words, managers can generally protect past R&D investment by

addressing and assessing each potential variation of R&D inputs. Such planning activities mitigate the neg-

ative effects of variability. Thus predictability changes the relative weight of the negative and the positive

arguments that address variation in R&D spending, and it does so in favor of the positive ones with respect

to both the quantity and quality of innovation.

In contrast, unpredicted variability amplifies the asymmetric lag of effects and their resulting coordination

issues because managers cannot anticipate the effect of any change, upward or downward. An upswing in

spending would require the hiring of staff and the purchase of equipment, so if that upswing is unpredicted

then managers may need to rush activities and thereby compromise performance. Unpredicted changes

affect the R&D organization in a more subtle way as well. Past experience with unpredicted changes con-

tributes to the uncertainty of managers as regards any assumptions about future allocations (cf. the notion

of “affective forecasting” discussed by Dane and George (2014)), and that uncertainty inclines managers to

build buffers into their operating plans so they can later deal more easily with shortages. Yet in the absence

of such shortages, those buffers constitute excess spending and thus entail negative effects for research pro-

ductivity. As a result, unpredicted variability exacerbates problems and, in so doing, pushes positive effects

to the background.

In sum, splitting variability into its predicted and unpredicted parts changes the relative weight of positive

and negative arguments and therefore affects the trade-off between them. Our final hypothesis is as follows.

2 We note that reduced R&D effectiveness and efficiency is sometimes an acceptable price to pay for lucrative commercial outcomes.


HYPOTHESIS 4a. Unpredicted variability in R&D spending leads to reduced R&D performance interms of quantity,yet predicted variability in R&D spending has a positive effect.

HYPOTHESIS 4b. Unpredicted variability in R&D spending leads to reduced R&D performance interms of quality,yet predicted variability in R&D spending has a positive effect.

4. Data and Variables

Our sample consists of publicly traded firms featured in the Compustat US industrial annual database for

the 30-year period from 1976 to 2006, which we use to retrieve accounting information including the R&D

spend. According to the SFAS 2 of 1975 all R&D expenses must be expensed in the period incurred and, in

particular, cannot be capitalized. (Of course, in line with general accounting principles tangible and some

limited intangible assets with alternative uses employed by the R&D department can (have to be) capitalized

and depreciated over their economic life.) While there may be some discretion in accounting as to what

the boundaries of R&D spend and other cost categories are, companies tend to be consistent over time in

how they categorize those fringe cases. As a result, the US R&D spend data is particularly amenable to

study temporal aspects of R&D spend such as variability. In fact, the R&D spend data has been at the core

of what the accounting literature calls “real earnings management”, or the practice of manipulating real

world decisions with the primary goal of affecting accounting outcomes (Roychowdhury 2006, Bruns Jr and

Merchant 1990, Graham et al. 2005, Dechow and Sloan 1991). We combine the Compustat data with data

on patent applications and approvals from the NBER Patent Data Project (Hall et al. 2001). This project

provides details on all 2,812,428 US utility patents granted during the period 1976–2006. For each patent

filing, these details include the year of filling/approval, the assignees, and a “corrected” count of citations

(Hall et al. 2001, 2005). The correction is necessary because a raw citation count not only includes self-

citations but also underestimates the citation count of more recent patents, which have had less time to

accumulate citations.

To combine the financial information from the Compustat database with the patent information, we use

the “concordance files” provided by the Patent Data Project.The concordance files resolve two important

issues. First, they correct for spelling discrepancies between the two databases. Second, companies often

hold patents in subsidiaries and special-purpose legal entities with names and addresses that differ from the

parent public corporation covered by the Compustat database. The concordance files consolidate patents

received by all such assignees to their legal parent entity.

After merging the databases, we identify 11,698 firms that have been granted at least one patent during

the period extending from 1976 to 2006. Throughout this 30-year period, we follow the patenting activity

of these firms as well as their financial details; the result is a raw yearly panel data set consisting of 139,867

firm-year events. We clean our data set in three ways. First, in line with existing research we consider only

the data from 1982 onward, since a structural change in the application of patent law renders data from


before and after incommensurate (Hall et al. 2001, Somaya 2003). Second, we follow Hall et al. (2001)

and omit the years 2003–2006 because the correction for future citations is insufficiently accurate in the last

three years of the Project data set. Finally, in accordance with previous assessments of the effects of R&D

investment (Kothari et al. 2002, Pandit et al. 2011), we restrict ourselves to firms with significant R&D

outlays. Thus we consider only firms that spent more than 5% of their sales revenue on R&D at least once

during the period under study, although we consider several other cut-off values for this R&D intensity in

Section 6.1. Our final sample contains 3,711 firms for which we have data from 1982 to 2003—a total of

28,034 firm-year events.

4.1. Definition of Variables

4.1.1. R&D Performance We consider two measures of R&D performance: the number of patents

filed by a firm and its related legal entities, and the number of citations generated by these patents; the

former (resp. latter) measure captures raw R&D performance via a quantity (resp. quality) dimension. Not

surprisingly, both patents and citations are concentrated in the most productive firms; hence the distribution

of these variables across firms is heavily right-skewed. Following the usual practice when analyzing such

variables (Chang et al. 2012), we apply a log-plus-one transformation.

4.1.2. R&D Spending Patterns Our main explanatory variables capture some aspects of the level of

variability in R&D spend. However, first we must define our most important control variable: R&D stock.

Previous research indicates that the level of R&D spending is a key driver of R&D performance (Hall et al.

2009). Our hypothesized effects of variability in R&D spending are in addition to the baseline effect of

R&D allocation levels, so examining the former requires that we first control for the latter.

R&D Stock: Spending on R&D seldom affects firm performance in the same year. More typically, an-

nual R&D spending augments a cumulative body of knowledge that can be exploited for a number of years

before its value decays. Hall et al. (2009) suggest using the notion of R&D stock to capture this cumula-

tive knowledge. Most of the empirical results using R&D stock have assumed a depreciation rate of 15%

(Corrado et al. 2009, Czarnitzki et al. 2014, Griliches and Mairesse 1984, Hall et al. 2010). Formally, with

XRDit the reported R&D expenditures of firm i in year t and k= 15%, we have

R&DStockit =RDSit =XRDit +(1− k)RDSi(t−1) (1)

Aggregate R&D Variability: We define intertemporal R&D variability as an aggregate of the absolute

year-to-year differences in R&D spending. As in the case of R&D stock, we discount past years. Formally,

V RDit =∣∣XRDit −XRDi(t−1)

∣∣+(1− k)V RDi(t−1). (2)

Predicted and Unpredicted Variability: We split aggregate variability into changes that are predicted (or

that follow past trends) and changes that are unpredicted (or at variance with extrapolated past trends).


Specifically, for each given year we calculate the R&D expenses that would result from a linear extrapolation

of the past two years’ trend for such expenses. The predicted change in R&D spending is then the absolute

difference between real past expenses and extrapolated future ones, while the unpredicted change is the

absolute difference between the predicted and actual expenses. (Section 6.1 reports the estimates obtained

when using other predictive models.) As before, we use a discount of 15%:

V RDPit =∣∣XRDP

it −XRDi(t−1)

∣∣+(1− k)V RDPi(t−1),

V RDUit =∣∣XRDit −XRDP

it

∣∣+(1− k)V RDUi(t−1);

4.1.3. Types of Innovation Hypotheses 2 and 3 are based on the interaction between R&D spending

variability and the originality of firm innovations. Following Ahuja and Lampert (2001), we define an inno-

vation as pioneering when a granted patent cites no other patent inside or outside the firm; the underlying

technology is both new-to-the-firm and new-to-the-world. Following this definition, for firm i in year t, we

define a firm’s pioneering ratio PNRit by dividing by the number of pioneering patents by the total number

of patents the firm is granted in that year.

Second, we split the ratio of original patents into the ones that are technologically distant and the ones

that are technologically close. We follow Ahuja and Lampert (2001) by defining a patent as technologically

close if it falls into a category that the firm holds other patents in. If not, it is technologically distant.

Formally, with ncloseit (resp. ndistantit) the number of original patents of firm i in year t granted in a

technology class already (resp. never) granted to firm before, we obtain

PNRit = PNRCit +PNRDit =ncloseitnpatentit

+ndistantitnpatentit

; (3)

In Section 6.1, we test different constructions of pioneering innovation and technological distance.

4.2. Control Variables

Opportunities Available: In addition to variability, the productivity of any R&D investment depends on the

extent of opportunities available to the company. Capital expenditures (CAPEX, as a percentage of sales) is

a well-known proxy for these opportunities (Kothari et al. 2002).

Capital Intensity (CAPIN): Capital intensity—the amount of fixed capital in relation to labor—is a known

predictor of R&D performance (Hall and Ziedonis 2001). We compute capital intensity as the logarithm of

the ratio of property, plant, and equipment (PPE) to the number of employees (EE).

Liquidity (LIQ): A metric that is much used by bankruptcy analysts, liquidity captures the firm’s ability

to cover short-term debts (Hirshleifer et al. 2012). We define liquidity as the ratio of a firm’s working

capital (see Hall and Kruiniker (1995) for more details on why we use working capital instead of cash)


MeanVariables Overall Between Within # Firms # Obs.

Patents (PAT) log--count 0.938 1.474 1.127 0.559 3,711 28,034Citations per patent (CP) log--ratio 1.172 1.553 1.163 0.963 3,711 28,034

R&D Variability (VRD) $ Mn 67.112 315.044 215.603 174.215 3,707 27,974Predicted Variability (VRDP) $ Mn 60.254 291.863 204.117 157.556 3,707 27,974Unpredicted Variability (VRDU) $ Mn 76.055 431.057 313.028 257.033 3,707 27,974

Pioneering Ratio (PNR) ratio 0.017 0.087 0.063 0.075 3,711 28,034Distant Pioneering Ratio (PNRD) ratio 0.005 0.053 0.031 0.049 3,711 28,034Close Pioneering Ratio (PNRC) ratio 0.012 0.068 0.048 0.057 3,711 28,034

R&D Stock (RDS) $ Mn 387.143 1,975.103 1,252.689 937.348 3,711 28,034Total Assets (LAT) log--$ Mn 3.934 2.511 2.211 0.718 3,710 27,762Squared Total Assets (LAT2) sq.log--$ Mn 21.777 23.437 18.688 6.325 3,710 27,762Capital Expenditure (CAPX) % of sales 0.464 18.145 16.312 14.719 3,623 26,516Capital Intensity (CAPIN) log--ratio 3.105 1.063 0.967 0.539 3,603 26,131Liquidity (LIQ) ratio 2.286 53.196 39.465 44.508 3,540 22,890Age (LAGE) log--years 2.368 0.730 0.680 0.330 3,711 28,034Industry Concentration (HERF) ratio 0.151 0.162 0.143 0.070 3,710 28,004Squared Ind. Conc. (HERF2) ratio--sq 0.049 0.200 0.161 0.107 3,710 28,004Industry Sales Growth (SALG) % 0.100 0.343 0.147 0.319 3,711 28,031

Standard Deviation N

Table 1 Summary statistics

to the value of its tangible assets,—excluding intangible, financial, and current assets (Fazzari et al. 1988,

Harhoff 1998).

Firm Age (LAGE): Firm age controls for the maturity of the firm in its innovation process (Chang et al.

2012). For each year we compute the logarithm of the firm’s age, where by “age” we mean the time between

its first appearance in Compustat and the current year.

Firm Size (LAT/LAT2): In line with a wealth of literature on operating performance in the fields of ac-

counting, finance, and economics (Chang et al. 2012), we include the logarithm of the firm’s total assets

(LAT) as well as its squared version as controls for firm size.

Industry Concentration (HERF): Changes in industry concentration, which can result from acquisitions

by competitors, may change the industry’s level of competitiveness and hence its financial strength (Hen-

dricks and Singhal 2008). Our estimates therefore include the normalized Herfindahl index of the applicable

3-digit SIC industry codes and its square term (HERF2) (Atanassov 2013, Chemmanur and Tian 2013) .

Industry Sales Growth (SALG): Different industries experience different cycles of expansion and stag-

nation over time. Expansion is generally associated with industry profitability while stagnation increases

competition and thus erodes profits. Our regressions incorporate industry sales growth at the 3-digit SIC

level (Hendricks and Singhal 2008).

Table 1 provides summary statistics for the variables used in our analysis. There is substantial within-

firm variation in R&D variability, which argues for using a fixed-effects model to isolate the effects of


[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18][1] PAT 1.00[2] CP 0.73 1.00[3] VRD 0.48 0.23 1.00[4] VRDP 0.47 0.22 0.96 1.00[5] VRDU 0.37 0.16 0.94 0.94 1.00[6] PNR 0.17 0.15 0.05 0.04 0.03 1.00[7] PNRD 0.03 0.05 -0.01 -0.01 -0.01 0.63 1.00[8] PNRC 0.20 0.15 0.06 0.06 0.05 0.80 0.03 1.00[9] RDS 0.52 0.25 0.86 0.86 0.74 0.04 -0.01 0.06 1.00

[10] LAT 0.67 0.48 0.45 0.44 0.37 0.11 0.01 0.13 0.45 1.00[11] LAT2 0.75 0.47 0.61 0.60 0.51 0.11 0.00 0.14 0.63 0.90 1.00[12] CAPX 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 -0.01 1.00[13] CAPIN 0.33 0.22 0.19 0.19 0.16 0.07 0.00 0.09 0.18 0.46 0.40 0.03 1.00[14] LIQ 0.00 0.00 -0.01 -0.01 -0.01 0.00 0.00 0.00 -0.01 0.06 -0.01 0.00 -0.01 1.00[15] LAGE 0.30 0.16 0.18 0.18 0.14 0.05 0.00 0.06 0.22 0.30 0.36 -0.02 0.09 -0.03 1.00[16] HERF -0.07 -0.08 -0.03 -0.03 -0.03 -0.02 0.00 -0.02 -0.02 -0.07 -0.04 -0.01 -0.05 0.00 0.05 1.00[17] HERF2 -0.05 -0.05 -0.02 -0.02 -0.02 -0.01 0.00 -0.01 -0.02 -0.04 -0.03 0.00 0.00 0.00 0.04 0.87 1.00[18] SALG -0.02 0.01 -0.01 -0.02 -0.02 -0.01 0.00 -0.01 -0.01 -0.02 -0.02 0.00 0.01 0.00 -0.01 0.10 0.08 1.00

Table 2 Correlation table

variability from the differences across firms. The 0 median values of our main variables reveal that their

distributions are right-skewed. We therefore use the absolute distance when constructing our measures of

R&D variability (so that they will be robust to outliers). Table 2 gives the correlation factors between each

variables of our model. As described in Section 4.1.2, VRDP and VRDU are a decomposition of VRD. This

explains their high correlation factor. However, they are never used in the same estimation.

5. Model Specification and Estimation

Conceptually, for a given firm i at time t , R&D performance is impacted by its own past performance. There

is stickiness of performance because accumulated R&D efforts characterized by the R&D stock of each firm

and companies’ innovation dynamic capabilities (Teece et al. 1997, Teece 2000) induce multi-period results

both in quantity (successive years of patent granting) and in quality (increasing and predominant role in an

ecosystem up to become a reference). In addition, companies which try to optimize their innovation process

and protect their intellectual property may “milk” their R&D investments through a strong and intensive

patenting policy. Finally, some companies set standards their domains. By doing so, they provide references

for followers, mechanically increasing the number of received citations for their granted patents. That is

why we estimate the following dynamic panel data model (DPD):

PATit = α11PATi(t−1) +α12V RDit +α13PNRi(t−1) +α14V RDitPNRi(t−1) +α15Xi(t−1) + εit (4)

CPit = α21CPi(t−1) +α22V RDit +α23PNRi(t−1) +α24V RDitPNRi(t−1) +α25Xi(t−1) + εit (5)

The first regression equation captures the effect of variability in R&D spending on the number of granted

patents, or the quantity aspect of R&D performance; the second focuses on the number of citations per


patent, or the quality aspect. The panel has dimensions N × T and may be unbalanced. V RDit denotes

our predictor variable(s), PNRit the originality ratio variable(s), and Xit our controls. Note that depending

on the tested hypothesis, we may use different constructs for each variable or, in the case of PNRit not

incorporate them at all. In addition, the pioneering ratio variables PNRi(t−1) and all the controls in Xi(t−1)

but age and R&D stock are taken at the beginning of the period to avoid any simultaneity bias due to

feedback effects (see Czarnitzki et al. (2014) for more details).

In both models, firm fixed effects eliminate estimation biases due to unobserved time-invariant R&D-

related differences across firms (e.g., differences in R&D abilities, culture, management); year fixed effects

eliminate any unobserved effects that are due to aggregate trends affecting our entire sample.

A DPD model raises a methodological challenge for the estimation strategy. Applying OLS estimators

to DPD models disregards the fact that the lagged dependent variable are by construction correlated with

the fixed effects in the error term. An OLS estimator would give rise to “dynamic panel bias” (Nickell

1981, Roodman 2009), particularly in the “small T, large N” configuration we face. As a second challenge

for our estimation strategy, our predictor variables may be prone to endogeneity issues. Thus, we use a

general method of moments (GMM) introduced by Arellano and Bond (1991) and augmented by Blundell

and Bond (1998). This approach allows us to use lags of endogenous variables as instruments for the lagged

performance measures as well as all of our predictor variables (and our most important control). It has

the benefit over a 2SLS estimation strategy of being more efficient in the face of heteroskedasticity and in

“small T, large N” configurations.

Both, the difference GMM (D-GMM) introduced by Arellano and Bond (1991) and the system GMM

(S-GMM) proposed by Blundell and Bond (1998) and refined by Windmeijer (2005) address the above

outlined problems and match the requirements of our model. However, our models exhibit lags of dependent

variables whose distributions are close to persistence (reported as the coefficient of the 1-year Lagged

Dependent variable in Table 3). We also find that the instruments for our predictor variables in a D-GMM

framework are weak (not reported). Hence, we use the S-GMM approach as suggested by Blundell and

Bond (1998), Bond (2002).

We run our estimation using the xtabond2 function in Stata that includes the two-step S-GMM estimator

with the Windmeijer correction. We ensure that all our tests (Hansen’s J test to test overall exogeneity,

Difference-in-Hansen tests to test exogeneity for each endogenous variable, Arellano-Bond AR(p) tests to

correctly specify the starting point of the lag structure of the instruments (see Roodman 2007)) are valid,

that we have the lowest number of instruments3 and that we have enough information incorporated through

moment conditions to reach good precision in our estimates.

3 First, we limit the number of lags we use for each endogenous variable by choosing a range of lags from p to p+ q with p ≥1, q ≥ 0. By doing so, we cap the number of instruments per period and make the instrument count, originally quadratic in T ,linear in T . Second, we use the collapse option to reduce the number of instruments to a minimum. Each gmmstyle instrument is


6. Results

Table 3 displays the results of our study. Each column header denotes the hypothesis tested, the first four

columns of results relating to performance in quantity (number of patents) and the last four columns to

performance in quality (number of citations per patent).

Aggregate R&D Variability In Table 3, Panel H1 the coefficients for V RD expose a significant and

negative effect on both patents (−0.205× 10−3, p= 0.02) and citations (−0.277× 10−3, p= 0.015). These

findings support hypotheses H1a and H1c: A certain amount of steadfastness is necessary when investing

in R&D or else the firm’s R&D performance will decline. In contrast, variability in R&D spending results

in a loss of the quantity of patents and the average quality of patents.

For our estimation, we use 3,351 companies, 151 instruments for patents, and 104 instruments for ci-

tations per patents. We keep the ratio of instruments over companies below 5%, assuring reliability and

relevance of our tests. The high p-values our our exogeneity tests strengthen this reliability. Our instru-

ments are valid. For all our Arellano-Bond AR(p) tests, we get p= 3. We therefore instrument our variables

starting at lag 34 We apply that thinking for other hypothesis as well.

To develop a sense of the magnitude of this effect, consider an hypothetical firm that reflects the average

(annual) characteristics of our sample: 23 patents granted and 11.9 citations earned per patent. If this firm

has a low variability strategy equal to the first quartile of V RD, and decides to invest in R&D by way of

highly dynamic spending—that is, with the variability of the third quartile rather than the first—then our

estimates indicate it would be granted 5.92% less patents and garner 26.5% less citations.5

In a follow-on analysis (not shown) we exploit the fact that variability in R&D spending reflects two kinds

of changes: year-to-year increases and decreases in R&D spend. We can use this to (further) ensure that our

results are not due to plain momentum effects: A good momentum produces a self-reinforcing upward trend,

a bad momentum a self-reinforcing downward trend; and the effect of a downward momentum is larger than

that of an upward momentum. Obviously, the specification of our DPD model and the S-GMM estimation

strategy rule out the existence of momentum effects (i.e., serial correlation driving results). But there is also

conceptual evidence: If momentum mechanisms were driving our result then we would expect downward

built by creating one instrument for each variable and lag distance (rather than one for each time period, variable, and lag distance).Practically, instead of having one column per time period in the instrument matrix, we create one column for the whole instrumentset. This gives the same expectation but only one moment condition, hence less conveyed information (see Roodman (2007, 2009),Töngür and Elveren (2012) for more details and applications). The advantage is to retain more information because no lags areactually dropped. This also makes the instrument count linear in T . Finally, we take advantage of the system configuration toinstrument only in levels when the instruments in difference are weak.4 We can start at lag 2 if we instrument in level only.5 To compute these counterfactuals, we use the fact that d

dxln (1 + y) = 1

1+y· dydx

and we introduce the mean values of the number

of patents(npatent

), the number of citations per patent

(ncp

), and the pioneering ratio

(PNR

). This gives us ∆npatent =

1+npatent1−α11

∆V RD ·(α12 +α14 · PNR

)and ∆ncp=

1+ncp1−α21

∆V RD ·(α22 +α24 · PNR

).


1-yr Lagged Dependent Variable 932.3 *** 932.0 *** 944.8 *** 914.5 *** 978.8 *** 979.7 *** 831.8 *** 817.9 ***

R&D Stock (RDS) 0.034 *** -0.002 0.009 -0.008 0.046 ** 0.007 0.001 -0.006

Total Assets (LAT) -1.479 -11.37 * 12.54 -4.253 5.634 -4.218 37.36 *** 34.13 **

Squared Total Assets (LAT2) 4.209 *** 4.462 *** 1.917 5.072 *** -0.135 0.726 1.010 2.776 **

Capital Expenditure (CAPX) 0.141 0.090 0.098 0.134 -0.220 -0.226 -0.193 -0.169Capital Intensity (CAPIN) -16.77 *** -19.25 *** -18.82 *** -16.83 *** -29.53 *** -32.00 *** -28.27 *** -30.45 ***

Liquidity (LIQ) 0.049 0.049 0.017 0.045 0.027 0.031 -0.045 -0.042Age (LAGE) -0.164 -6.604 4.246 0.135 2.218 -5.309 1.044 -8.233Industry Concentration (HERF) -106.5 ** -37.54 -119.1 . -100.3 ** -66.92 6.818 -200.3 * -248.1 **

Squared Ind. Conc. (HERF2) 69.72 ** 40.49 72.38 70.07 ** 50.93 -3.201 116.6 . 144.9 *

Industry Sales Growth (SALG) 4.171 ** 3.980 * 3.736 * 4.372 ** 1.733 2.286 4.231 5.224Pioneering Ratio (PNR) 2,071.1 ** 296.6 1,036.6 * 642.4Distant Pioneering Ratio (PNRD) -5,443.8 . -3,565.3Close Pioneering Ratio (PNRC) 935.8 2,688.4 **

R&D Variability (VRD) -0.205 ** 0.014 -0.034 -0.277 * -0.045 -0.011Interaction VRD x PNR (INTER1) -2.214 ** -1.353 *

Interaction VRD x PNRC (INTER2) -2.89 * -1.923 *

Interaction VRD x PNRD (INTER3) 19.635 ** 5.947Predicted Variability (VRDP) 0.410 *** 0.294 *

Unpredicted Variability (VRDU) -0.294 *** -0.222 ***

GMM Estimation Method system system system system system system system systemCompany Fixed Effects YES YES YES YES YES YES YES YESYear Fixed Effects YES YES YES YES YES YES YES YESNumber of Observations 20,649 20,649 20,649 20,649 20,649 20,649 20,649 20,649Number of Firms 3,351 3,351 3,351 3,351 3,351 3,351 3,351 3,351Number of instruments 151 202 175 115 104 178 184 219Percentage of Instruments 5% 6% 5% 3% 3% 5% 5% 7%

Arellano-Bond Test for AR(1)1 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000Arellano-Bond Test for AR(2)2 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000Arellano-Bond Test for AR(3)3 0.998 0.605 0.554 0.919 0.438 0.379 0.458 0.397

Hansen J-test for Overidentifying Restrictions4

0.271 0.222 0.196 0.362 0.565 0.830 0.370 0.876

Diff.-in-Hansen Test of Exogeneity of:5

Dep. Var. 0.571 0.118 0.433 0.606 0.919 0.795 0.422 0.872VRD 0.867 0.336 0.512 0.426 0.167 0.870RDS 0.623 0.267 0.303 0.225 0.636 0.722 0.394 0.901PNR 0.879 0.416 0.614 0.547INTER1 0.263 0.973PNRD 0.299 0.997PNRC 0.933 0.201INTER2 0.215 0.852INTER3 0.626 0.815VRDP 0.503 0.521VRDU 0.673 0.313

*** -- 0.1% level, ** -- 1% level, * -- 5% level, . -- 10% levelAll controls are lagged by 1 year and coefficients are scaled by a factor of 1000 for better readability1 - H0 = no 1st-order serial correlation in residuals2 - H0 = no 2nd-order serial correlation in residuals3 - H0 = no 3rd-order serial correlation in residuals4 - H0 = model specification is correct5 - H0 = instruments are exogenous

Patents (PAT) Citations per patents (CP)H1.a/b H2.a/b.p H3.a/b.p H4.a H1.c/d H2.a/b.cp H3.a/b.cp H4.b

Interaction w/ Originality

Search Distance

Role of ForecastingVariables Variability in

R&D SpendingInteraction w/

OriginalitySearch

DistanceRole of

ForecastingVariability in

R&D Spending

Table 3 Model estimates


variability to have a negative effect and upward variability to have a positive effect. In contrast, if the

results were driven by the inherent disadvantages of variability then both upward and downward variability

in spending should be detrimental. We estimate the “upward”- and “downward”-effects by decomposing

the R&D variability into its upward and downward components. We find that the upward and downward

components each have significant negative effects on both the number of patents (−0.196× 10−3, p= 0.007

and −0.148× 10−3, p= 0.044, respectively) and the number of citations (−0.185× 10−3, p= 0.072 and

−0.211× 10−3, p= 0.089, respectively). Although the negative effects of downward variability might be

attributed to momentum, the negative effect of upward variability suggests otherwise. That is to say, our

analysis reveals that variability is damaging for R&D productivity (i.e., the number of patents granted or the

number of citations per patent controlling for R&D spend) even when it manifests as additional funding.

Technological Originality Firms fund R&D projects whose level of technological originality ranges

from existing to pioneering. Depending on which end of the spectrum a firm focuses its R&D invest-

ment on, the effects of its temporal allocation pattern may be different. Panels H2.a/b.p and H2.a/b.cp test

the corresponding hypotheses. We find that the interaction between variability and the ratio of pioneering

patents in firms’ portfolio is significant and negative for both the number of patents and the number of

citations per patent (−2.214× 10−3, p = 0.002 and −1.353× 10−3, p = 0.020, respectively). Hypothesis

H2a is strongly supported: The more a firm engages in original invention, the more R&D variability hurts

performance in quantity and in quality. As a side note, when focusing on the main effect of originality,

we find evidence for a sizable effect of originality. The more a company is engaged in pioneering innova-

tion, the better the invention performance both in quantity and in quality (PNR is positive and significant

with 2071.1× 10−3, p = 0.009 and 1036.6× 10−3, p = 0.049, for quantity and quality respectively). We

find here the effects of first-mover advantage where the pioneer can cement technological leadership and

preempt scarce assets (see Lieberman and Montgomery 1988, p.44).

As for the size of the effect, when we take the example from Section 6, assuming 15.85%6 of the portfolio

is dedicated to pioneering patents, going from a first quartile to a third quartile type of dynamism creates

the same 9.67% decrease in the number of patents and a 25.9% decrease in number of citations per patent.

Technological Firm Search Distance Panels H3.a/b.p and H3.a/b.cp of Table 3 test whether the effects

of variability on originality are the same for technologically distant and technologically close inventions.

We thus decompose the fraction of original inventions into its technological distant and close parts. The

interaction between R&D variability and original technologies that are close to the company’s technological

base is negative and significant for both the number of patents (−2.888× 10−3, p = 0.019) and the qual-

ity of patents (−1.923× 10−3, p = 0.027). The same argumentation as for hypothesis H2 applies. When

a company decides to develop novel technologies that are close to its existing technology base, variability

6 Among companies having worked on pioneering technologies, the average PNR is 15.85%.


hurts R&D productivity (The main effect of novelty in close technologies is also positive for both the quan-

tity (935.8× 10−3, p = 0.278) and quality (2688.4× 10−3, p = 0.007) of patents but only significant for

the quality of patents). But interestingly, the interaction between R&D variability and original technologies

that distant from the company’s technological base is positive and significant for the number of patents

(19.64× 10−3, p = 0.002) and positive but not significant for the quality of patents (5.947× 10−3, p =

0.360). Hence, when a company decides to develop novel technologies that are removed from its existing

technology base, flexibility becomes a major driver in the generation of patents and does not harm the qual-

ity of the patents. This insight is consistent with the interpretation that there is so much uncertainty that

innovation teams will make useful mistakes to progress(The main effects in this case are negative but not

significant). Hence, hypothesis H3b is supported for quantity and partially supported for quality. Going on

with the same example as the one used for H2, with a ratio of distant technology of 4.28%7, we obtain an

increase of 16.72% of patents and 0.26% of citations per patent. This is typically a case where the ratio of

distant technologies is high enough to give an overall positive effect. In real, most of the firms have no such

type of research, hence an overall negative variability.

Predicted versus Unpredicted Variability Finally, Hypothesis 4 was borne out of the recognition that

adaptive R&D spending cannot be unrelievedly negative; in other words, there should be a way to at least

mitigate the negative effect of funding variability so that it may be possible for management to limit the

effects of R&D variability if the associated changes are necessary. To test for the existence of a mitigation

method, we decompose variability into anticipated and unanticipated changes in R&D spend. The estimated

effects are reported in Panels H4.a and H4.b of Table 3.

We find that, for both patents and citations, the coefficient for predicted variability is significant and

positive (respectively 0.410× 10−3, p= 0.001 and 0.294× 10−3, p= 0.031) whereas the coefficient for un-

predicted variability is significant and negative (−0.294× 10−3, p= 0.000 and −0.222× 10−3, p= 0.001).

Whereas unanticipated changes in R&D spend reduce performance, anticipated changes actually improve

performance. Managers can anticipate the consequences of a change in R&D spending and thus prepare the

organization for that eventuality. Therefore, Hypotheses 4a and 4b are supported. Under the same assump-

tions as the ones from H1, we obtain a gain from predicted variability of 9.38% for patents and 3.28% for

citations per patent. Unpredicted variability hurts performance by 6.72% for patents and 2.48% for citations

per patent.

In sum, our empirical results suggest that variability in R&D spend has an inherent negative effect that

interacts with the type of innovation strategy. Even so, we find that there are ways to manage any required

changes to R&D resource allocations. By adhering to a predictable method of effecting any necessary

adaptation to funding, managers can at least limit the negative effects of those changes.

7 Among companies having worked on pioneering technologies, the average PNRD is 4.28% and hence the average PNRC is11.57%.


6.1. Robustness Tests

We test the robustness of our results at four levels. First, the models are estimated using alternate DPD

specification with two and three lags of the dependent variables. Second, we launch different types of GMM

estimation and we try different number of instruments in our configuration. Third, we reconstruct the data

set while sub-setting for different industries and clusters of firm size. We also employ different inclusion

criterion, with respect to R&D intensity. Last, we test alternate constructs of our variables. We see that our

results are robust to all these alternate buildings and configurations. Coefficients are unaffected in all these

estimations.

7. Discussion and Implications

This paper provides empirical evidence concerning how temporal aspects can best be accounted for in the

allocation of R&D funding. We show how investing in a persistent rather than in a dynamic fashion helps

improve R&D performance: variation in R&D spend is negatively related to innovation performance in

terms of both quantity and quality: flexibility comes with R&D efficiency loss. The precise nature of this

relation depends on the type of innovation a company engages in. The more original its portfolio, the more

negatively the impact of variation on R&D performance. When splitting the type of innovation further a

more differentiated picture emerges. Research on an original portfolio of inventions close to the company’s

technology base is prone to suffer from variability, but research on an original portfolio of inventions tech-

nologically distant from the company’s technology base does profit from variability. These findings seem

to imply that variability in most of its forms is a detriment to R&D performance. Our subsequent analysis

qualifies such an impression. In particular, if variability in R&D spending is decomposed into its unpre-

dictable and predictable components then only unpredictable variability proves to be harmful; predictable

variability has either a positive effect or no effect. So if the performance of R&D is a priority then managers,

when carrying out a strategic repositioning, should strive to make the associated shifts in R&D funding

as gradual as possible. Our results are not just statistically significant, they are economically relevant: For

realistic scenarios, companies can improve the quantity of their R&D output in between 5% and 10% and

the quality of their R&D output up to 30%.

Our paper clearly contributes to our theoretical understanding of what determines R&D productivity and

hence good R&D management. But the paper has deep implications from the managerial perspective as

well. There are many reasons for which managers may wish to adapt their level of R&D spend, some of

which the result of a positive intent such as chasing technological opportunities some of which the result of

a more negative intent such as the practice of earnings management. Whatever the rationale for the change,

our paper highlights the negative consequences that managers need to consider; it establishes contingencies

for when adaptation is particularly harmful; and it points to policies that can mitigate adaptation pains.


As such it gives give managers a framework to conceptualize principles for how to best invest in R&D.

Particularly as a comment on the practice of “hitting quarterly financial targets” it issues a warning. If R&D

spending is viewed as discretionary when target requirements must be satisfied (as is customary for some

publicly traded companies as documented by Roychowdhury (2006), then one should expect long-term

negative consequences that cannot be reversed simply by later restoring or even increasing R&D investment

as the firm’s financial situation permits.

As with any empirical study based on archival data, there are limits to generalizing the results. Although

our sample is expansive, it considers only public firms. It may be that private firms, which can operate

without the pressure of periodic financial reporting requirements, employ dynamic R&D spending strategies

that lead to better outcomes than the ones described in this study. A similar study on private firms is a

potential avenue for further understanding the effects we observed and describing mechanisms in more

detail.

We use accounting and patent data to draw our conclusions. Both may not be full representations of reality

in that, for example, there may be innovation activities not captured in neither of them and particularly the

patent information may not be an accurate depiction of a company’s innovation qualities. The measures

we use are also based on annual R&D spending data. The implication is that we observe only year-to-year

changes, which might fail to detect equally significant changes that transpire over shorter time periods.

Extant accounting and reporting standards make it virtually impossible to observe changes at a finer level

for a large set of companies. That being said, a collaborative study with a large firm featuring many divisions

and frequent capital budgeting adjustments would likely shed still more light on the nature of variability.

Owing to the nature of lags between R&D spending decisions and observable outcomes, our sample is

focused on a period that goes back almost 40 years. The effects addressed in this study may well be timeless;

yet it could be that, as a consequence of the changing nature of knowledge work and improved human

resource systems, some organizations in fast-moving industries (e.g., information technology) have recently

become better at managing flexibility but that there are too few companies to make the effect salient in data.

In conclusion, we believe that, in addition to furthering our theoretic understanding of R&D productivity,

our study can help foster a managerial discussion about the optimal level of dynamism in R&D spending.

Acknowledgments

ReferencesAghion P, Van Reenen J, Zingales L (2013) Innovation and institutional ownership. The American Economic Review

103(1):277–304.

Ahuja G (2000) Collaboration networks, structural holes, and innovation: A longitudinal study. Administrative science

quarterly 45(3):425–455.


Ahuja G, Lampert CM (2001) Entrepreneurship in the large corporation: A longitudinal study of how established firms

create breakthrough inventions. STRATEGIC MANAGE J 22(6-7):521–543.

Almeida H, Hsu PH, Li D (2013) Less is more: Financial constraints and innovative efficiency .

Arellano M, Bond S (1991) Some tests of specification for panel data: Monte carlo evidence and an application to

employment equations. The Review of Economic Studies 58(2):277–297.

Argyres NS, Silverman BS (2004) R&d, organization structure, and the development of corporate technological knowl-

edge. Strategic Management Journal 25(8-9):929–958.

Arkali G, Bardhan I, Krishnan V (2008) R&d, information technology, and firm performance: A complementarity

model. UCSD Working Paper 1–50.

Arora A, Belenzon S, Rios LA (2011) The organization of r&d in american corporations: the determinants and conse-

quences of decentralization. Technical report, National Bureau of Economic Research.

Arora A, Belenzon S, Rios LA (2014) Make, buy, organize: The interplay between research, external knowledge, and

firm structure. Strategic Management Journal 35(3):317–337.

Atanassov J (2013) Do hostile takeovers stifle innovation? Evidence from antitakeover legislation and corporate patent-

ing. J FINANC 68(3):1097–1131.

Baard SK, Rench TA, Kozlowski SWJ (2014) Performance adaptation a theoretical integration and review. J MANAGE

40(1):48–99.

Bauer T, Bodner T, Erdogan B, Truxillo D, Tucker J (2007) Newcomer adjustment during organizational socialization:

A meta-analytic review of antecedents, outcomes, and methods. J APPL PSYCHOL 92(3):707.

Bernstein B, Singh PJ (2008) Innovation generation process: Applying the adopter categorization model and concept

of "chasm" to better understand social and behavioral issues. EUR J INNOVAT MANAG 11(3):366–388.

Bilgram V, Brem A, Voigt KI (2008) User-centric innovations in new product development - Systematic identification

of lead users harnessing interactive and collaborative online-tools. IJIM 12(03):419–458.

Blundell R, Bond S (1998) Initial conditions and moment restrictions in dynamic panel data models. J ECONOMET-

RICS 87(1):115–143.

Bond SR (2002) Dynamic panel data models: a guide to micro data methods and practice. Portuguese economic journal

1(2):141–162.

Bound J, Cummins C, Griliches Z, Hall BH, Jaffe AB (1984) Who does r&d and who patents? Griliches ZE, ed., R&D

Patents and Productivity (Chicago, IL.: University of Chicago Press for the NBER).

Bruns Jr WJ, Merchant KA (1990) The dangerous morality of managing earnings. Strategic Finance 72(2):22.

Cameron G (1996) On the measurement of real r&d: Divisia price indices for uk business enterprise r&d. Research

Evaluation 6(3):215–219.

Case DO (2012) Looking for information: A survey of research on information seeking, needs and behavior (Bingley,

UK: Emerald Group Publishing).


Cassiman B, Veugelers R (2006) In search of complementarity in innovation strategy: Internal r&d and external knowl-

edge acquisition. Management science 52(1):68–82.

Chan J, Dow SP, Schunn CD (2014) Do the best design ideas (really) come from conceptually distant sources of

inspiration? DES STUD 1(1):1–28.

Chang X, Fu K, Low A, Zhang W (2015) Non-executive employee stock options and corporate innovation. Journal of

financial economics 115(1):168–188.

Chang X, Hilary G, K KJ, W Z (2012) Does accounting conservatism impede corporate innovation? Working Paper,

INSEAD .

Chao RO, Kavadias S (2013) R&d intensity and the new product development portfolio. IEEE Transactions on Engi-

neering Management 60(4):664–675.

Chemmanur TJ, Tian X (2013) Do anti-takeover provisions spur corporate innovation? Available at SSRN 1572219 .

Chemmanur TJ, Tian X (2016) Do anti-takeover provisions spur corporate innovation? .

Ciftci M, Lev B, Radhakrishnan S (2011) Is research and development mispriced or properly risk adjusted? J ACC

AUD FINANCE 26(1):81–116.

Cockburn IM, Henderson RM (1998) Absorptive capacity, coauthoring behavior, and the organization of research in

drug discovery. The Journal of Industrial Economics 46(2):157–182.

Cockburn IM, Henderson RM (2001) Scale and scope in drug development: unpacking the advantages of size in

pharmaceutical research. Journal of health economics 20(6):1033–1057.

Cohen WM (2010) Fifty years of empirical studies of innovative activity and performance. Handbook of the Economics

of Innovation 1:129–213.

Cohen WM, Klepper S (1996a) Firm size and the nature of innovation within industries: the case of process and

product r&d. The review of Economics and Statistics 232–243.

Cohen WM, Klepper S (1996b) A reprise of size and r & d. The Economic Journal 925–951.

Corrado C, Hulten C, Sichel D (2009) Intangible capital and us economic growth. Review of income and wealth

55(3):661–685.

Crépon B, Duguet E, Mairessec J (1998) Research, innovation and productivity: An econometric analysis at the firm

level. Economics of Innovation and new Technology 7(2):115–158.

Czarnitzki D, Hall BH, Hottenrott H (2014) Patents as quality signals? the implications for financing constraints on

r&d. Technical report, National Bureau of Economic Research.

Dane E, George J (2014) Unpacking affective forecasting and its ties to project work in organizations. ACAD MANAGE

REV 39(2):181–201.

Dechow PM, Sloan RG (1991) Executive incentives and the horizon problem: An empirical investigation. Journal of

accounting and Economics 14(1):51–89.


Ding Y, Stolowy H, Tenenhaus M (2007) R&D productivity: An exploratory international study. Review of Accounting

and Finance 6(1):86–101.

Edmondson AC, Nembhard IM (2009) Product development and learning in project teams: The challenges are the

benefits. JPIM 26(2):123–138.

Fazzari SM, Hubbard RG, Petersen BC (1988) Financing constraints and corporate investment. Brooking Papers on

Economic Activity 1:146–206.

Fleming L, Sorenson O (2003) Navigating the technology landscape of innovation. SLOAN MANAGE REV winter:15–

23.

Francis BB, Hasan I, Sharma Z (2011) Incentives and innovation: Evidence from ceo compensation contracts .

Freeman C (1982) The Economics of Industrial Innovation (Cambridge, MA: MIT Press).

Freeman C, Soete L (1997) The economics of industrial innovation (Psychology Press).

Frenz M, Ietto-Gillies G (2009) The impact on innovation performance of different sources of knowledge: Evidence

from the uk community innovation survey. Research Policy 38(7):1125–1135.

Galasso A, Simcoe TS (2011) Ceo overconfidence and innovation. Management Science 57(8):1469–1484.

Geroski PA, et al. (1995) Market structure, corporate performance, and innovative activity. OUP Catalogue .

Graham JR, Harvey CR, Rajgopal S (2005) The economic implications of corporate financial reporting. Journal of

accounting and economics 40(1):3–73.

Griffin A (1997) The effect of project and process characteristics on product development cycle time. J MARKETING

34(1):24–35.

Griffith R, Harrison R, Van Reenen J (2006) How special is the special relationship? Using the impact of U.S. R&D

spillovers on U.K. firms as a test of technology sourcing. AM ECON REV 96(5):1859–1875.

Griliches Z, Mairesse J (1984) Productivity and r&d at the firm level. Griliches Z, ed., Patents, and Productivity

(Chicago: University of Chicago Press).

Hacklin F, Raurich V, Marxt C (2004) How incremental innovation becomes disruptive: The case of technology con-

vergence. Engineering Management Conference, 2004. Proceedings. 2004 IEEE International, volume 1, 32–36

(IEEE).

Hall B, Jaffe A, Trajtenberg M (2005) Market value and patent citations. RAND J ECON 36(1):16–38.

Hall BH, Jaffe AB, Trajtenberg M (2001) The NBER patent citation data file: Lessons, insights and methodological

tools. NBER Working Paper Series No. 8498.

Hall BH, Kruiniker H (1995) The role of working capital in the investment process.

Hall BH, Lotti F, Mairesse J (2009) Innovation and productivity in smes: empirical evidence for italy. Small Business

Economics 33(1):13–33.

Hall BH, Mairesse J, Mohnen P (2010) Measuring the returns to R&D. Handbook of the Economics of Innovation

2:1033–1082.


Hall BH, Oriani R (2006) Does the market value R&D investment by european firms? Evidence from a panel of

manufacturing firms in France, Germany, and Italy. INT J IND ORGAN 24(5):971–993.

Hall BH, Ziedonis RH (2001) The patent paradox revisited: An empirical study of patenting in the u.s. semiconductor

industry, 1979-1995. RAND J ECON 32:101–128.

Harhoff D (1998) R&d and productivity in german manufacturing firms. Economics of Innovation and New Technology

6(1):29–50.

Henderson R (1993) Underinvestment and incompetence as responses to radical innovation: Evidence from the pho-

tolithographic alignment equipment industry. The RAND Journal of Economics 248–270.

Henderson R, Cockburn I (1994) Measuring competence? exploring firm effects in pharmaceutical research. Strategic

management journal 15(S1):63–84.

Henderson R, Cockburn I (1996) Scale, scope, and spillovers: the determinants of research productivity in drug dis-

covery. The Rand journal of economics 32–59.

Hendricks KB, Singhal VR (2008) The effect of product introduction delays on operating performance. MANAGE SCI

54(5):878–892.

Hirshleifer DA, Low A, Teoh SH (2012) Are overconfident ceos better innovators? J FINANC 67:1457–1498.

Hollander S (1965) The Sources of Increased Efficiency: A Study of Du Pont Rayon Plants (Cambridge, MA: MIT

Press).

Jain A (2013) Learning by doing and the locus of innovative capability in biotechnology research. ORGAN SCI

24(6):1683–1700.

Jefferson GH, Bai H, Guan X, Yu X (2006) R&d performance in chinese industry. Economics of Innovation and New

Technology 15(4-5):345–366.

Jones G (1986) Socialization tactics, self-efficacy, and newcomers adjustments to organizations. ACAD MANAGE J

29(2):262–279.

Katila R (2002) New product search over time: past ideas in their prime? Academy of Management Journal 45(5):995–

1010.

Kessler EH, Allocca MA, Rahman N (2007) External knowledge accession and innovation speed in the small and

medium sized enterprise (SME). Small Enterprise Research 15(1):1–21.

Kessler EH, Bierly PE (2002) Is faster really better? An empirical test of the implications of innovation speed. IEEE

T ENG MANAGE 49(1):2–12.

Kim T, Cable D, Kim S (2005) Socialization tactics, employee productivity, and person-organization fit. J APPL

PSYCHOL 90(2):232.

Klein HJ, Weaver NA (2000) The effectiveness of an organizational-level orientation training program in the social-

ization of new hires. PERS PSYCHOL 53(1):47–66.


Kleinschmidt EJ, De Brentani U, Salomo S (2007) Performance of global new product development programs: A

resource-based view. JPIM 24(5):419–441.

Kor YY, Mahoney JT (2005) How dynamics, management, and governance of resource deployments influence firm-

level performance. STRATEGIC MANAGE J 26(5):489–496.

Kothari SP, Laguerre TE, Leone AJ (2002) Capitalization versus expensing: Evidence on the uncertainty of future

earnings from capital expenditures versus R&D outlays. REV ACC STUD 7(4):355–382.

Kremp E, Mairesse J (2004) Knowledge management, innovation, and productivity: A firm level exploration based on

french manufacturing cis3 data. Technical report, National bureau of economic research.

Langerak F, Hultink EJ (2005) The impact of new product development acceleration approaches on speed and prof-

itability: Lessons for pioneers and fast followers. IEEE T ENG MANAGE 52(1):30–42.

Leiponen A, Helfat CE (2011) Location, decentralization, and knowledge sources for innovation. Organization Science

22(3):641–658.

Lerner J (2006) The new new financial thing: The origins of financial innovations. Journal of Financial Economics

79(2):223–255.

Lerner J, Sorensen M, Stromberg P (2013) Private equity and investment in innovation: Evidence from patents. Journal

of Applied Corporate Finance 25(2):95–102.

Lerner J, Wulf J (2007) Innovation and incentives: Evidence from corporate R&D. REV ECON STAT 89(4):634–644.

Lev B, Sougiannis T (1996) The capitalization, amortization, and value-relevance of R&D. J ACCOUNT ECON

21(1):107–138.

Lichtenberg FR, Siegel D (1991) The impact of R&D investment on productivity - new evidence using linked R&D-lrd

data. Econ. Inq. 29(2):203.

Lieberman MB, Montgomery DB (1988) First-mover advantages. STRATEGIC MANAGE J 9(S1):41–58.

Lin BW, Lee Y, Hung SC (2006) R&d intensity and commercialization orientation effects on financial performance.

Journal of Business Research 59(6):679–685.

Loch CH, Kavadias S (2002) Dynamic portfolio selection of NPD programs using marginal returns. MANAGE SCI

48(10):1227–1241.

Lööf H, Heshmati A (2002) Knowledge capital and performance heterogeneity:: A firm-level innovation study. Inter-

national Journal of Production Economics 76(1):61–85.

Lööf H, Heshmati A (2006) On the relationship between innovation and performance: A sensitivity analysis. Eco-

nomics of Innovation and New Technology 15(4-5):317–344.

March JG (1991) Exploration and exploitation in organizational learning. ORGAN SCI 2(1):71–87.

March JG (1994) A Primer on Decision Making: How Decision Happens (New York, NY: Free Press).

Mecca JT, Mumford MD (2014) Imitation and creativity: Beneficial effects of propulsion strategies and specificity.

The Journal of Creative Behavior 48(3):209–236.


Moch M, Morse EV (1977) Size, centralization and organizational adoption of innovations. AM SOCIOL REV

42(5):716–725.

Mudambi R, Swift T (2014) Knowing when to leap: Transitioning between exploitative and explorative r&d. STRATE-

GIC MANAGE J 35(1):126–145.

Nelson RR, Winter SG (1977) In search of useful theory of innovation. RES POLICY 6(1):36–76.

Nickell S (1981) Biases in dynamic models with fixed effects. ECONOMETRICA 1417–1426.

Norman DA, Verganti R (2014) Incremental and radical innovation: Design research vs. technology and meaning

change. DES ISSUES 30(1):78–96.

Ostrom E, Burger J, Field CB, Norgaard RB, Policansky D (1999) Revisiting the commons: local lessons, global

challenges. SCIENCE 284(5412):278–282.

Pandit S, Wasley CE, Zach T (2011) The effect of R&D inputs and outputs on the relation between the uncertainty of

future operating performance and R&D expenditures. J ACC AUD FINANCE 26(1):121–144.

Pich MT, Loch CH, DeMeyer A (2002) On uncertainty, ambiguity and complexity in project management. MANAGE

SCI 48(8):1008–1023.

Pruett M, Thomas H (2008) Experience-based learning in innovation and production. R&D Management 38(2):141–

153.

Prusa TJ, Schmitz JA (1991) Are new firms an important source of innovation?: Evidence from the pc software indus-

try. Economics Letters 35(3):339–342.

Quintana-Garcia C, Benavides-Velasco CA (2008) Innovative competence, exploration and exploitation: The influence

of technological diversification. RES POLICY 37(3):492–507.

Revilla AJ, Fernández Z (2012) The relation between firm size and r&d productivity in different technological regimes.

Technovation 32(11):609–623.

Rietzschel EF, Nijstad BA, Stroebe W (2007) Relative accessibility of domain knowledge and creativity: The effects of

knowledge activation on the quantity and originality of generated ideas. J EXP SOC PSYCHOL 43(6):933–946.

Rogers M (2010) R&D and productivity: Using UK firm-level data to inform policy. EMPIRICA 37(3):329–359.

Roodman D (2007) A short note on the theme of too many instruments. Working paper 125, Center for Global Devel-

opment.

Roodman D (2009) How to do xtabond2: An introduction to difference and system gmm in stata. The Stata Journal

9(1):86–136.

Rosenkopf L, Nerkar A (2001) Beyond local search: boundary-spanning, exploration, and impact in the optical disk

industry. STRATEGIC MANAGE J 22(4):287–306.

Roussel PA, Saad KN, Erickson TJ (1991) Third Generation R&D (Watertown, MA: Havard Business School Press).

Roychowdhury S (2006) Earnings management through real activities manipulation. Journal of accounting and eco-

nomics 42(3):335–370.


Sawyer RK (2011) Explaining creativity: The science of human innovation (Oxford, UK: Oxford University Press).

Scherer FM, Ross D (1990) Industrial market structure and economic performance .

Simonton DK (1997) Creative productivity: A predictive and explanatory model of career trajectories and landmarks.

PSYCHOL REV 104(1):66–89.

Smith LGE, Amiot CE, Smith JR, Callan VJ, Terry DJ (2013) The social validation and coping model of organizational

identity development a longitudinal test. J MANAGE 39(7):1952–1978.

Somaya D (2003) Strategic determinants of decisions not to settle patent litigation. STRATEGIC MANAGE J 24(1):17–

38.

Sommer SC, Loch CH, Dong J (2009) Managing complexity and unforeseeable uncertainty in startup companies: An

empirical study. ORGAN SCI 20(1):118–133.

Stuart TE, Podolny JM (1996) Local search and the evolution of technological capabilities. Strategic Management

Journal 17(S1):21–38.

Teece DJ (2000) Strategies for managing knowledge assets: the role of firm structure and industrial context. Long

Range Plann. 33(1):35–54.

Teece DJ, Pisano G, Shuen A (1997) Dynamic capabilities and strategic management. STRATEGIC MANAGE J

18(7):509–533.

Terwiesch C, Loch CH, Niederkofler M (1998) When product development performance makes a difference. JPIM

15(1):3–15.

Töngür Ü, Elveren AY (2012) Military expenditures, inequality, and welfare and political regimes: A dynamic panel

data analysis.

Van Maanen J, Schein E (1979) Towards a theory of organizational socialization. RES ORGAN BEHAV 1:209–264.

Von Hippel E (2009) Democratizing innovation: The evolving phenomenon of user innovation. IJIS 1(1):29–40.

Weber J (1990) Toshiba will pay TI royalty on chip patents. Los Angeles Times December 12.

Weisberg RW (2011) Frank Lloyd Wright’s Fallingwater: A case study in inside-the-box creativity. CREATIVITY RES

J 23(4):296–312.

Windmeijer F (2005) A finite sample correction for the variance of linear efficient two-step gmm estimators. J

ECONOMETRICS 126(1):25–51.

r&d spending: dynamic or persistent?public.kenan-flagler.unc.edu/2017msom/sigs/iform...

Documents